stationary statistical properties of dissipative … approach continuity and perturbation results...

81
Statistical Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang, Xiaoming [email protected] Florida State University Stochastic and Probabilistic Methods in Ocean-Atmosphere Dynamics, Victoria, July 2008 Wang, Xiaoming [email protected] Stationary Statistical Properties of Dissipative Systems

Upload: vandien

Post on 07-May-2018

218 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistical Properties of DissipativeSystems

Wang Xiaomingwxmmathfsuedu

Florida State University

Stochastic and Probabilistic Methods in Ocean-AtmosphereDynamics Victoria July 2008

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Outline

1 Statistical Approach

2 Continuity and perturbation results

3 Temporal approximation

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Logistic map

T (x) = 4x(1minus x) x isin [0 1]

Figure Sensitive dependence

0 20 40 60 80 100 120 140 160 180 2000

01

02

03

04

05

06

07

08

09

1

Orbits of the Logistic map with close ICs

Figure Statistical coherence

-02 0 02 04 06 08 1 120

100

200

300

400

500

600

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Lorenz 96 modelEdward Norton Lorenz 1917-2008

Lorenz96 duj

dt= (uj+1 minus ujminus2)ujminus1 minus uj + F

j = 0 1 middot middot middot J J = 5 F = minus12

Figure Sensitive dependence Figure Statistical coherence

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Lorenz 96 modelEdward Norton Lorenz 1917-2008

Lorenz96 duj

dt= (uj+1 minus ujminus2)ujminus1 minus uj + F

j = 0 1 middot middot middot J J = 5 F = minus12

Figure Sensitive dependence

(Sensitive dependence)

Figure Statistical coherence

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

L96_demoavi
Media File (videoavi)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistical approach

dudt

= F(u) u isin H

Long time average

lt Φ gt= limTrarrinfin

1T

int T

0Φ(v(t)) dt

Spatial averages

lt Φ gtt=

intH

Φ(v) dmicrot(v)

microt t ge 0 statistical solutions

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistical approach

dudt

= F(u) u isin H

Long time average

lt Φ gt= limTrarrinfin

1T

int T

0Φ(v(t)) dt

Spatial averages

lt Φ gtt=

intH

Φ(v) dmicrot(v)

microt t ge 0 statistical solutions

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistical approach

dudt

= F(u) u isin H

Long time average

lt Φ gt= limTrarrinfin

1T

int T

0Φ(v(t)) dt

Spatial averages

lt Φ gtt=

intH

Φ(v) dmicrot(v)

microt t ge 0 statistical solutions

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Figure Joseph Liouville1809-1882

Figure Eberhard Hopf1902-1983

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Figure George DavidBirkhoff 1884-1944 Definition

micro is ergodic if micro(E) = 0 or 1 for allinvariant sets E

Theorem (Birkhoffrsquos ErgodicTheorem)

If micro is invariant and ergodic thetemporal and spatial averages areequivalent ie

limTrarrinfin

1T

int T

0ϕ(S(t)u) dt =

intH

ϕ(u) dmicro(u) as

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

More on statistical theories for complex dynamical systems can befound

Majda AJ and Wang X Nonlinear Dynamics and StatisticalTheories for Basic Geophysical Flows Cambridge UniversityPress 2006Majda AJ Abramov R and Grote M Information theory andstochastics for multiscale nonlinear systems CRM monographseries American Mathematical Society 2005Foias C Manley O Rosa R Temam R Navier-StokesEquations and Turbulence Cambridge University PressCambridge UK 2001

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Dissipative system

Definition

A dynamical system S(t) t ge 0 on a phase space H is calleddissipative if there exists a global attractor A such that

A is invariant under S(t)A is compactA attracts any bounded set B in H ie

limtrarrinfin

dist(S(t)BA) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and time averages

Theorem (IM and time averages W 08)

Assume H reflexive S dissipativeforallu0 isin HforallLIM rArr existmicro isin IM such that

LIMTrarrinfin1T

int T

0ϕ(S(t)u0) dt =

intH

ϕ(u) dmicro(u)forallϕ isin C(H)

Earlier work under existence of a compact absorbing set or smallerclass of weakly continuous functionals (Foias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Ergodicity and extremal points

Theorem (Ergodicity and extremal points W 08)

Let IM be the set of all invariant probability measures of adissipative dynamical system S(t) t ge 0 Then an invariantmeasure micro is ergodic if micro is an extreme point of IM Moreover if thedynamical system is injective on the global attractor A then everyergodic invariant measure must be an extremal point of IM

Other versions with group (ODE) assumption is well-known

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 2: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Outline

1 Statistical Approach

2 Continuity and perturbation results

3 Temporal approximation

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Logistic map

T (x) = 4x(1minus x) x isin [0 1]

Figure Sensitive dependence

0 20 40 60 80 100 120 140 160 180 2000

01

02

03

04

05

06

07

08

09

1

Orbits of the Logistic map with close ICs

Figure Statistical coherence

-02 0 02 04 06 08 1 120

100

200

300

400

500

600

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Lorenz 96 modelEdward Norton Lorenz 1917-2008

Lorenz96 duj

dt= (uj+1 minus ujminus2)ujminus1 minus uj + F

j = 0 1 middot middot middot J J = 5 F = minus12

Figure Sensitive dependence Figure Statistical coherence

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Lorenz 96 modelEdward Norton Lorenz 1917-2008

Lorenz96 duj

dt= (uj+1 minus ujminus2)ujminus1 minus uj + F

j = 0 1 middot middot middot J J = 5 F = minus12

Figure Sensitive dependence

(Sensitive dependence)

Figure Statistical coherence

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

L96_demoavi
Media File (videoavi)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistical approach

dudt

= F(u) u isin H

Long time average

lt Φ gt= limTrarrinfin

1T

int T

0Φ(v(t)) dt

Spatial averages

lt Φ gtt=

intH

Φ(v) dmicrot(v)

microt t ge 0 statistical solutions

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistical approach

dudt

= F(u) u isin H

Long time average

lt Φ gt= limTrarrinfin

1T

int T

0Φ(v(t)) dt

Spatial averages

lt Φ gtt=

intH

Φ(v) dmicrot(v)

microt t ge 0 statistical solutions

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistical approach

dudt

= F(u) u isin H

Long time average

lt Φ gt= limTrarrinfin

1T

int T

0Φ(v(t)) dt

Spatial averages

lt Φ gtt=

intH

Φ(v) dmicrot(v)

microt t ge 0 statistical solutions

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Figure Joseph Liouville1809-1882

Figure Eberhard Hopf1902-1983

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Figure George DavidBirkhoff 1884-1944 Definition

micro is ergodic if micro(E) = 0 or 1 for allinvariant sets E

Theorem (Birkhoffrsquos ErgodicTheorem)

If micro is invariant and ergodic thetemporal and spatial averages areequivalent ie

limTrarrinfin

1T

int T

0ϕ(S(t)u) dt =

intH

ϕ(u) dmicro(u) as

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

More on statistical theories for complex dynamical systems can befound

Majda AJ and Wang X Nonlinear Dynamics and StatisticalTheories for Basic Geophysical Flows Cambridge UniversityPress 2006Majda AJ Abramov R and Grote M Information theory andstochastics for multiscale nonlinear systems CRM monographseries American Mathematical Society 2005Foias C Manley O Rosa R Temam R Navier-StokesEquations and Turbulence Cambridge University PressCambridge UK 2001

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Dissipative system

Definition

A dynamical system S(t) t ge 0 on a phase space H is calleddissipative if there exists a global attractor A such that

A is invariant under S(t)A is compactA attracts any bounded set B in H ie

limtrarrinfin

dist(S(t)BA) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and time averages

Theorem (IM and time averages W 08)

Assume H reflexive S dissipativeforallu0 isin HforallLIM rArr existmicro isin IM such that

LIMTrarrinfin1T

int T

0ϕ(S(t)u0) dt =

intH

ϕ(u) dmicro(u)forallϕ isin C(H)

Earlier work under existence of a compact absorbing set or smallerclass of weakly continuous functionals (Foias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Ergodicity and extremal points

Theorem (Ergodicity and extremal points W 08)

Let IM be the set of all invariant probability measures of adissipative dynamical system S(t) t ge 0 Then an invariantmeasure micro is ergodic if micro is an extreme point of IM Moreover if thedynamical system is injective on the global attractor A then everyergodic invariant measure must be an extremal point of IM

Other versions with group (ODE) assumption is well-known

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 3: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Logistic map

T (x) = 4x(1minus x) x isin [0 1]

Figure Sensitive dependence

0 20 40 60 80 100 120 140 160 180 2000

01

02

03

04

05

06

07

08

09

1

Orbits of the Logistic map with close ICs

Figure Statistical coherence

-02 0 02 04 06 08 1 120

100

200

300

400

500

600

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Lorenz 96 modelEdward Norton Lorenz 1917-2008

Lorenz96 duj

dt= (uj+1 minus ujminus2)ujminus1 minus uj + F

j = 0 1 middot middot middot J J = 5 F = minus12

Figure Sensitive dependence Figure Statistical coherence

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Lorenz 96 modelEdward Norton Lorenz 1917-2008

Lorenz96 duj

dt= (uj+1 minus ujminus2)ujminus1 minus uj + F

j = 0 1 middot middot middot J J = 5 F = minus12

Figure Sensitive dependence

(Sensitive dependence)

Figure Statistical coherence

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

L96_demoavi
Media File (videoavi)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistical approach

dudt

= F(u) u isin H

Long time average

lt Φ gt= limTrarrinfin

1T

int T

0Φ(v(t)) dt

Spatial averages

lt Φ gtt=

intH

Φ(v) dmicrot(v)

microt t ge 0 statistical solutions

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistical approach

dudt

= F(u) u isin H

Long time average

lt Φ gt= limTrarrinfin

1T

int T

0Φ(v(t)) dt

Spatial averages

lt Φ gtt=

intH

Φ(v) dmicrot(v)

microt t ge 0 statistical solutions

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistical approach

dudt

= F(u) u isin H

Long time average

lt Φ gt= limTrarrinfin

1T

int T

0Φ(v(t)) dt

Spatial averages

lt Φ gtt=

intH

Φ(v) dmicrot(v)

microt t ge 0 statistical solutions

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Figure Joseph Liouville1809-1882

Figure Eberhard Hopf1902-1983

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Figure George DavidBirkhoff 1884-1944 Definition

micro is ergodic if micro(E) = 0 or 1 for allinvariant sets E

Theorem (Birkhoffrsquos ErgodicTheorem)

If micro is invariant and ergodic thetemporal and spatial averages areequivalent ie

limTrarrinfin

1T

int T

0ϕ(S(t)u) dt =

intH

ϕ(u) dmicro(u) as

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

More on statistical theories for complex dynamical systems can befound

Majda AJ and Wang X Nonlinear Dynamics and StatisticalTheories for Basic Geophysical Flows Cambridge UniversityPress 2006Majda AJ Abramov R and Grote M Information theory andstochastics for multiscale nonlinear systems CRM monographseries American Mathematical Society 2005Foias C Manley O Rosa R Temam R Navier-StokesEquations and Turbulence Cambridge University PressCambridge UK 2001

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Dissipative system

Definition

A dynamical system S(t) t ge 0 on a phase space H is calleddissipative if there exists a global attractor A such that

A is invariant under S(t)A is compactA attracts any bounded set B in H ie

limtrarrinfin

dist(S(t)BA) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and time averages

Theorem (IM and time averages W 08)

Assume H reflexive S dissipativeforallu0 isin HforallLIM rArr existmicro isin IM such that

LIMTrarrinfin1T

int T

0ϕ(S(t)u0) dt =

intH

ϕ(u) dmicro(u)forallϕ isin C(H)

Earlier work under existence of a compact absorbing set or smallerclass of weakly continuous functionals (Foias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Ergodicity and extremal points

Theorem (Ergodicity and extremal points W 08)

Let IM be the set of all invariant probability measures of adissipative dynamical system S(t) t ge 0 Then an invariantmeasure micro is ergodic if micro is an extreme point of IM Moreover if thedynamical system is injective on the global attractor A then everyergodic invariant measure must be an extremal point of IM

Other versions with group (ODE) assumption is well-known

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 4: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Lorenz 96 modelEdward Norton Lorenz 1917-2008

Lorenz96 duj

dt= (uj+1 minus ujminus2)ujminus1 minus uj + F

j = 0 1 middot middot middot J J = 5 F = minus12

Figure Sensitive dependence Figure Statistical coherence

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Lorenz 96 modelEdward Norton Lorenz 1917-2008

Lorenz96 duj

dt= (uj+1 minus ujminus2)ujminus1 minus uj + F

j = 0 1 middot middot middot J J = 5 F = minus12

Figure Sensitive dependence

(Sensitive dependence)

Figure Statistical coherence

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

L96_demoavi
Media File (videoavi)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistical approach

dudt

= F(u) u isin H

Long time average

lt Φ gt= limTrarrinfin

1T

int T

0Φ(v(t)) dt

Spatial averages

lt Φ gtt=

intH

Φ(v) dmicrot(v)

microt t ge 0 statistical solutions

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistical approach

dudt

= F(u) u isin H

Long time average

lt Φ gt= limTrarrinfin

1T

int T

0Φ(v(t)) dt

Spatial averages

lt Φ gtt=

intH

Φ(v) dmicrot(v)

microt t ge 0 statistical solutions

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistical approach

dudt

= F(u) u isin H

Long time average

lt Φ gt= limTrarrinfin

1T

int T

0Φ(v(t)) dt

Spatial averages

lt Φ gtt=

intH

Φ(v) dmicrot(v)

microt t ge 0 statistical solutions

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Figure Joseph Liouville1809-1882

Figure Eberhard Hopf1902-1983

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Figure George DavidBirkhoff 1884-1944 Definition

micro is ergodic if micro(E) = 0 or 1 for allinvariant sets E

Theorem (Birkhoffrsquos ErgodicTheorem)

If micro is invariant and ergodic thetemporal and spatial averages areequivalent ie

limTrarrinfin

1T

int T

0ϕ(S(t)u) dt =

intH

ϕ(u) dmicro(u) as

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

More on statistical theories for complex dynamical systems can befound

Majda AJ and Wang X Nonlinear Dynamics and StatisticalTheories for Basic Geophysical Flows Cambridge UniversityPress 2006Majda AJ Abramov R and Grote M Information theory andstochastics for multiscale nonlinear systems CRM monographseries American Mathematical Society 2005Foias C Manley O Rosa R Temam R Navier-StokesEquations and Turbulence Cambridge University PressCambridge UK 2001

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Dissipative system

Definition

A dynamical system S(t) t ge 0 on a phase space H is calleddissipative if there exists a global attractor A such that

A is invariant under S(t)A is compactA attracts any bounded set B in H ie

limtrarrinfin

dist(S(t)BA) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and time averages

Theorem (IM and time averages W 08)

Assume H reflexive S dissipativeforallu0 isin HforallLIM rArr existmicro isin IM such that

LIMTrarrinfin1T

int T

0ϕ(S(t)u0) dt =

intH

ϕ(u) dmicro(u)forallϕ isin C(H)

Earlier work under existence of a compact absorbing set or smallerclass of weakly continuous functionals (Foias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Ergodicity and extremal points

Theorem (Ergodicity and extremal points W 08)

Let IM be the set of all invariant probability measures of adissipative dynamical system S(t) t ge 0 Then an invariantmeasure micro is ergodic if micro is an extreme point of IM Moreover if thedynamical system is injective on the global attractor A then everyergodic invariant measure must be an extremal point of IM

Other versions with group (ODE) assumption is well-known

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 5: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Lorenz 96 modelEdward Norton Lorenz 1917-2008

Lorenz96 duj

dt= (uj+1 minus ujminus2)ujminus1 minus uj + F

j = 0 1 middot middot middot J J = 5 F = minus12

Figure Sensitive dependence

(Sensitive dependence)

Figure Statistical coherence

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

minus20 minus10 0 10 200

2000

4000

6000

8000

10000

12000

14000

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

L96_demoavi
Media File (videoavi)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistical approach

dudt

= F(u) u isin H

Long time average

lt Φ gt= limTrarrinfin

1T

int T

0Φ(v(t)) dt

Spatial averages

lt Φ gtt=

intH

Φ(v) dmicrot(v)

microt t ge 0 statistical solutions

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistical approach

dudt

= F(u) u isin H

Long time average

lt Φ gt= limTrarrinfin

1T

int T

0Φ(v(t)) dt

Spatial averages

lt Φ gtt=

intH

Φ(v) dmicrot(v)

microt t ge 0 statistical solutions

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistical approach

dudt

= F(u) u isin H

Long time average

lt Φ gt= limTrarrinfin

1T

int T

0Φ(v(t)) dt

Spatial averages

lt Φ gtt=

intH

Φ(v) dmicrot(v)

microt t ge 0 statistical solutions

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Figure Joseph Liouville1809-1882

Figure Eberhard Hopf1902-1983

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Figure George DavidBirkhoff 1884-1944 Definition

micro is ergodic if micro(E) = 0 or 1 for allinvariant sets E

Theorem (Birkhoffrsquos ErgodicTheorem)

If micro is invariant and ergodic thetemporal and spatial averages areequivalent ie

limTrarrinfin

1T

int T

0ϕ(S(t)u) dt =

intH

ϕ(u) dmicro(u) as

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

More on statistical theories for complex dynamical systems can befound

Majda AJ and Wang X Nonlinear Dynamics and StatisticalTheories for Basic Geophysical Flows Cambridge UniversityPress 2006Majda AJ Abramov R and Grote M Information theory andstochastics for multiscale nonlinear systems CRM monographseries American Mathematical Society 2005Foias C Manley O Rosa R Temam R Navier-StokesEquations and Turbulence Cambridge University PressCambridge UK 2001

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Dissipative system

Definition

A dynamical system S(t) t ge 0 on a phase space H is calleddissipative if there exists a global attractor A such that

A is invariant under S(t)A is compactA attracts any bounded set B in H ie

limtrarrinfin

dist(S(t)BA) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and time averages

Theorem (IM and time averages W 08)

Assume H reflexive S dissipativeforallu0 isin HforallLIM rArr existmicro isin IM such that

LIMTrarrinfin1T

int T

0ϕ(S(t)u0) dt =

intH

ϕ(u) dmicro(u)forallϕ isin C(H)

Earlier work under existence of a compact absorbing set or smallerclass of weakly continuous functionals (Foias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Ergodicity and extremal points

Theorem (Ergodicity and extremal points W 08)

Let IM be the set of all invariant probability measures of adissipative dynamical system S(t) t ge 0 Then an invariantmeasure micro is ergodic if micro is an extreme point of IM Moreover if thedynamical system is injective on the global attractor A then everyergodic invariant measure must be an extremal point of IM

Other versions with group (ODE) assumption is well-known

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 6: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistical approach

dudt

= F(u) u isin H

Long time average

lt Φ gt= limTrarrinfin

1T

int T

0Φ(v(t)) dt

Spatial averages

lt Φ gtt=

intH

Φ(v) dmicrot(v)

microt t ge 0 statistical solutions

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistical approach

dudt

= F(u) u isin H

Long time average

lt Φ gt= limTrarrinfin

1T

int T

0Φ(v(t)) dt

Spatial averages

lt Φ gtt=

intH

Φ(v) dmicrot(v)

microt t ge 0 statistical solutions

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistical approach

dudt

= F(u) u isin H

Long time average

lt Φ gt= limTrarrinfin

1T

int T

0Φ(v(t)) dt

Spatial averages

lt Φ gtt=

intH

Φ(v) dmicrot(v)

microt t ge 0 statistical solutions

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Figure Joseph Liouville1809-1882

Figure Eberhard Hopf1902-1983

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Figure George DavidBirkhoff 1884-1944 Definition

micro is ergodic if micro(E) = 0 or 1 for allinvariant sets E

Theorem (Birkhoffrsquos ErgodicTheorem)

If micro is invariant and ergodic thetemporal and spatial averages areequivalent ie

limTrarrinfin

1T

int T

0ϕ(S(t)u) dt =

intH

ϕ(u) dmicro(u) as

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

More on statistical theories for complex dynamical systems can befound

Majda AJ and Wang X Nonlinear Dynamics and StatisticalTheories for Basic Geophysical Flows Cambridge UniversityPress 2006Majda AJ Abramov R and Grote M Information theory andstochastics for multiscale nonlinear systems CRM monographseries American Mathematical Society 2005Foias C Manley O Rosa R Temam R Navier-StokesEquations and Turbulence Cambridge University PressCambridge UK 2001

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Dissipative system

Definition

A dynamical system S(t) t ge 0 on a phase space H is calleddissipative if there exists a global attractor A such that

A is invariant under S(t)A is compactA attracts any bounded set B in H ie

limtrarrinfin

dist(S(t)BA) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and time averages

Theorem (IM and time averages W 08)

Assume H reflexive S dissipativeforallu0 isin HforallLIM rArr existmicro isin IM such that

LIMTrarrinfin1T

int T

0ϕ(S(t)u0) dt =

intH

ϕ(u) dmicro(u)forallϕ isin C(H)

Earlier work under existence of a compact absorbing set or smallerclass of weakly continuous functionals (Foias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Ergodicity and extremal points

Theorem (Ergodicity and extremal points W 08)

Let IM be the set of all invariant probability measures of adissipative dynamical system S(t) t ge 0 Then an invariantmeasure micro is ergodic if micro is an extreme point of IM Moreover if thedynamical system is injective on the global attractor A then everyergodic invariant measure must be an extremal point of IM

Other versions with group (ODE) assumption is well-known

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 7: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistical approach

dudt

= F(u) u isin H

Long time average

lt Φ gt= limTrarrinfin

1T

int T

0Φ(v(t)) dt

Spatial averages

lt Φ gtt=

intH

Φ(v) dmicrot(v)

microt t ge 0 statistical solutions

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistical approach

dudt

= F(u) u isin H

Long time average

lt Φ gt= limTrarrinfin

1T

int T

0Φ(v(t)) dt

Spatial averages

lt Φ gtt=

intH

Φ(v) dmicrot(v)

microt t ge 0 statistical solutions

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Figure Joseph Liouville1809-1882

Figure Eberhard Hopf1902-1983

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Figure George DavidBirkhoff 1884-1944 Definition

micro is ergodic if micro(E) = 0 or 1 for allinvariant sets E

Theorem (Birkhoffrsquos ErgodicTheorem)

If micro is invariant and ergodic thetemporal and spatial averages areequivalent ie

limTrarrinfin

1T

int T

0ϕ(S(t)u) dt =

intH

ϕ(u) dmicro(u) as

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

More on statistical theories for complex dynamical systems can befound

Majda AJ and Wang X Nonlinear Dynamics and StatisticalTheories for Basic Geophysical Flows Cambridge UniversityPress 2006Majda AJ Abramov R and Grote M Information theory andstochastics for multiscale nonlinear systems CRM monographseries American Mathematical Society 2005Foias C Manley O Rosa R Temam R Navier-StokesEquations and Turbulence Cambridge University PressCambridge UK 2001

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Dissipative system

Definition

A dynamical system S(t) t ge 0 on a phase space H is calleddissipative if there exists a global attractor A such that

A is invariant under S(t)A is compactA attracts any bounded set B in H ie

limtrarrinfin

dist(S(t)BA) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and time averages

Theorem (IM and time averages W 08)

Assume H reflexive S dissipativeforallu0 isin HforallLIM rArr existmicro isin IM such that

LIMTrarrinfin1T

int T

0ϕ(S(t)u0) dt =

intH

ϕ(u) dmicro(u)forallϕ isin C(H)

Earlier work under existence of a compact absorbing set or smallerclass of weakly continuous functionals (Foias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Ergodicity and extremal points

Theorem (Ergodicity and extremal points W 08)

Let IM be the set of all invariant probability measures of adissipative dynamical system S(t) t ge 0 Then an invariantmeasure micro is ergodic if micro is an extreme point of IM Moreover if thedynamical system is injective on the global attractor A then everyergodic invariant measure must be an extremal point of IM

Other versions with group (ODE) assumption is well-known

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 8: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistical approach

dudt

= F(u) u isin H

Long time average

lt Φ gt= limTrarrinfin

1T

int T

0Φ(v(t)) dt

Spatial averages

lt Φ gtt=

intH

Φ(v) dmicrot(v)

microt t ge 0 statistical solutions

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Figure Joseph Liouville1809-1882

Figure Eberhard Hopf1902-1983

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Figure George DavidBirkhoff 1884-1944 Definition

micro is ergodic if micro(E) = 0 or 1 for allinvariant sets E

Theorem (Birkhoffrsquos ErgodicTheorem)

If micro is invariant and ergodic thetemporal and spatial averages areequivalent ie

limTrarrinfin

1T

int T

0ϕ(S(t)u) dt =

intH

ϕ(u) dmicro(u) as

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

More on statistical theories for complex dynamical systems can befound

Majda AJ and Wang X Nonlinear Dynamics and StatisticalTheories for Basic Geophysical Flows Cambridge UniversityPress 2006Majda AJ Abramov R and Grote M Information theory andstochastics for multiscale nonlinear systems CRM monographseries American Mathematical Society 2005Foias C Manley O Rosa R Temam R Navier-StokesEquations and Turbulence Cambridge University PressCambridge UK 2001

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Dissipative system

Definition

A dynamical system S(t) t ge 0 on a phase space H is calleddissipative if there exists a global attractor A such that

A is invariant under S(t)A is compactA attracts any bounded set B in H ie

limtrarrinfin

dist(S(t)BA) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and time averages

Theorem (IM and time averages W 08)

Assume H reflexive S dissipativeforallu0 isin HforallLIM rArr existmicro isin IM such that

LIMTrarrinfin1T

int T

0ϕ(S(t)u0) dt =

intH

ϕ(u) dmicro(u)forallϕ isin C(H)

Earlier work under existence of a compact absorbing set or smallerclass of weakly continuous functionals (Foias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Ergodicity and extremal points

Theorem (Ergodicity and extremal points W 08)

Let IM be the set of all invariant probability measures of adissipative dynamical system S(t) t ge 0 Then an invariantmeasure micro is ergodic if micro is an extreme point of IM Moreover if thedynamical system is injective on the global attractor A then everyergodic invariant measure must be an extremal point of IM

Other versions with group (ODE) assumption is well-known

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 9: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Figure Joseph Liouville1809-1882

Figure Eberhard Hopf1902-1983

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Figure George DavidBirkhoff 1884-1944 Definition

micro is ergodic if micro(E) = 0 or 1 for allinvariant sets E

Theorem (Birkhoffrsquos ErgodicTheorem)

If micro is invariant and ergodic thetemporal and spatial averages areequivalent ie

limTrarrinfin

1T

int T

0ϕ(S(t)u) dt =

intH

ϕ(u) dmicro(u) as

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

More on statistical theories for complex dynamical systems can befound

Majda AJ and Wang X Nonlinear Dynamics and StatisticalTheories for Basic Geophysical Flows Cambridge UniversityPress 2006Majda AJ Abramov R and Grote M Information theory andstochastics for multiscale nonlinear systems CRM monographseries American Mathematical Society 2005Foias C Manley O Rosa R Temam R Navier-StokesEquations and Turbulence Cambridge University PressCambridge UK 2001

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Dissipative system

Definition

A dynamical system S(t) t ge 0 on a phase space H is calleddissipative if there exists a global attractor A such that

A is invariant under S(t)A is compactA attracts any bounded set B in H ie

limtrarrinfin

dist(S(t)BA) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and time averages

Theorem (IM and time averages W 08)

Assume H reflexive S dissipativeforallu0 isin HforallLIM rArr existmicro isin IM such that

LIMTrarrinfin1T

int T

0ϕ(S(t)u0) dt =

intH

ϕ(u) dmicro(u)forallϕ isin C(H)

Earlier work under existence of a compact absorbing set or smallerclass of weakly continuous functionals (Foias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Ergodicity and extremal points

Theorem (Ergodicity and extremal points W 08)

Let IM be the set of all invariant probability measures of adissipative dynamical system S(t) t ge 0 Then an invariantmeasure micro is ergodic if micro is an extreme point of IM Moreover if thedynamical system is injective on the global attractor A then everyergodic invariant measure must be an extremal point of IM

Other versions with group (ODE) assumption is well-known

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 10: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Figure Joseph Liouville1809-1882

Figure Eberhard Hopf1902-1983

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Figure George DavidBirkhoff 1884-1944 Definition

micro is ergodic if micro(E) = 0 or 1 for allinvariant sets E

Theorem (Birkhoffrsquos ErgodicTheorem)

If micro is invariant and ergodic thetemporal and spatial averages areequivalent ie

limTrarrinfin

1T

int T

0ϕ(S(t)u) dt =

intH

ϕ(u) dmicro(u) as

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

More on statistical theories for complex dynamical systems can befound

Majda AJ and Wang X Nonlinear Dynamics and StatisticalTheories for Basic Geophysical Flows Cambridge UniversityPress 2006Majda AJ Abramov R and Grote M Information theory andstochastics for multiscale nonlinear systems CRM monographseries American Mathematical Society 2005Foias C Manley O Rosa R Temam R Navier-StokesEquations and Turbulence Cambridge University PressCambridge UK 2001

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Dissipative system

Definition

A dynamical system S(t) t ge 0 on a phase space H is calleddissipative if there exists a global attractor A such that

A is invariant under S(t)A is compactA attracts any bounded set B in H ie

limtrarrinfin

dist(S(t)BA) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and time averages

Theorem (IM and time averages W 08)

Assume H reflexive S dissipativeforallu0 isin HforallLIM rArr existmicro isin IM such that

LIMTrarrinfin1T

int T

0ϕ(S(t)u0) dt =

intH

ϕ(u) dmicro(u)forallϕ isin C(H)

Earlier work under existence of a compact absorbing set or smallerclass of weakly continuous functionals (Foias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Ergodicity and extremal points

Theorem (Ergodicity and extremal points W 08)

Let IM be the set of all invariant probability measures of adissipative dynamical system S(t) t ge 0 Then an invariantmeasure micro is ergodic if micro is an extreme point of IM Moreover if thedynamical system is injective on the global attractor A then everyergodic invariant measure must be an extremal point of IM

Other versions with group (ODE) assumption is well-known

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 11: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Figure Joseph Liouville1809-1882

Figure Eberhard Hopf1902-1983

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Figure George DavidBirkhoff 1884-1944 Definition

micro is ergodic if micro(E) = 0 or 1 for allinvariant sets E

Theorem (Birkhoffrsquos ErgodicTheorem)

If micro is invariant and ergodic thetemporal and spatial averages areequivalent ie

limTrarrinfin

1T

int T

0ϕ(S(t)u) dt =

intH

ϕ(u) dmicro(u) as

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

More on statistical theories for complex dynamical systems can befound

Majda AJ and Wang X Nonlinear Dynamics and StatisticalTheories for Basic Geophysical Flows Cambridge UniversityPress 2006Majda AJ Abramov R and Grote M Information theory andstochastics for multiscale nonlinear systems CRM monographseries American Mathematical Society 2005Foias C Manley O Rosa R Temam R Navier-StokesEquations and Turbulence Cambridge University PressCambridge UK 2001

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Dissipative system

Definition

A dynamical system S(t) t ge 0 on a phase space H is calleddissipative if there exists a global attractor A such that

A is invariant under S(t)A is compactA attracts any bounded set B in H ie

limtrarrinfin

dist(S(t)BA) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and time averages

Theorem (IM and time averages W 08)

Assume H reflexive S dissipativeforallu0 isin HforallLIM rArr existmicro isin IM such that

LIMTrarrinfin1T

int T

0ϕ(S(t)u0) dt =

intH

ϕ(u) dmicro(u)forallϕ isin C(H)

Earlier work under existence of a compact absorbing set or smallerclass of weakly continuous functionals (Foias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Ergodicity and extremal points

Theorem (Ergodicity and extremal points W 08)

Let IM be the set of all invariant probability measures of adissipative dynamical system S(t) t ge 0 Then an invariantmeasure micro is ergodic if micro is an extreme point of IM Moreover if thedynamical system is injective on the global attractor A then everyergodic invariant measure must be an extremal point of IM

Other versions with group (ODE) assumption is well-known

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 12: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Statistics Solutions

microt t ge 0 dvdt

= F(v) S(t) t ge 0

Pull-backmicro0(Sminus1(t)(E)) = microt(E)

Push-forward (Φ suitable test functional)intH

Φ(v) dmicrot(v) =

intH

Φ(S(t)v) dmicro0(v)

Finite ensemble example

micro0 =Nsum

j=1

pjδv0j (v) microt =Nsum

j=1

pjδvj (t)(v) vj(t) = S(t)v0j

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Figure Joseph Liouville1809-1882

Figure Eberhard Hopf1902-1983

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Figure George DavidBirkhoff 1884-1944 Definition

micro is ergodic if micro(E) = 0 or 1 for allinvariant sets E

Theorem (Birkhoffrsquos ErgodicTheorem)

If micro is invariant and ergodic thetemporal and spatial averages areequivalent ie

limTrarrinfin

1T

int T

0ϕ(S(t)u) dt =

intH

ϕ(u) dmicro(u) as

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

More on statistical theories for complex dynamical systems can befound

Majda AJ and Wang X Nonlinear Dynamics and StatisticalTheories for Basic Geophysical Flows Cambridge UniversityPress 2006Majda AJ Abramov R and Grote M Information theory andstochastics for multiscale nonlinear systems CRM monographseries American Mathematical Society 2005Foias C Manley O Rosa R Temam R Navier-StokesEquations and Turbulence Cambridge University PressCambridge UK 2001

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Dissipative system

Definition

A dynamical system S(t) t ge 0 on a phase space H is calleddissipative if there exists a global attractor A such that

A is invariant under S(t)A is compactA attracts any bounded set B in H ie

limtrarrinfin

dist(S(t)BA) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and time averages

Theorem (IM and time averages W 08)

Assume H reflexive S dissipativeforallu0 isin HforallLIM rArr existmicro isin IM such that

LIMTrarrinfin1T

int T

0ϕ(S(t)u0) dt =

intH

ϕ(u) dmicro(u)forallϕ isin C(H)

Earlier work under existence of a compact absorbing set or smallerclass of weakly continuous functionals (Foias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Ergodicity and extremal points

Theorem (Ergodicity and extremal points W 08)

Let IM be the set of all invariant probability measures of adissipative dynamical system S(t) t ge 0 Then an invariantmeasure micro is ergodic if micro is an extreme point of IM Moreover if thedynamical system is injective on the global attractor A then everyergodic invariant measure must be an extremal point of IM

Other versions with group (ODE) assumption is well-known

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 13: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Figure Joseph Liouville1809-1882

Figure Eberhard Hopf1902-1983

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Figure George DavidBirkhoff 1884-1944 Definition

micro is ergodic if micro(E) = 0 or 1 for allinvariant sets E

Theorem (Birkhoffrsquos ErgodicTheorem)

If micro is invariant and ergodic thetemporal and spatial averages areequivalent ie

limTrarrinfin

1T

int T

0ϕ(S(t)u) dt =

intH

ϕ(u) dmicro(u) as

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

More on statistical theories for complex dynamical systems can befound

Majda AJ and Wang X Nonlinear Dynamics and StatisticalTheories for Basic Geophysical Flows Cambridge UniversityPress 2006Majda AJ Abramov R and Grote M Information theory andstochastics for multiscale nonlinear systems CRM monographseries American Mathematical Society 2005Foias C Manley O Rosa R Temam R Navier-StokesEquations and Turbulence Cambridge University PressCambridge UK 2001

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Dissipative system

Definition

A dynamical system S(t) t ge 0 on a phase space H is calleddissipative if there exists a global attractor A such that

A is invariant under S(t)A is compactA attracts any bounded set B in H ie

limtrarrinfin

dist(S(t)BA) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and time averages

Theorem (IM and time averages W 08)

Assume H reflexive S dissipativeforallu0 isin HforallLIM rArr existmicro isin IM such that

LIMTrarrinfin1T

int T

0ϕ(S(t)u0) dt =

intH

ϕ(u) dmicro(u)forallϕ isin C(H)

Earlier work under existence of a compact absorbing set or smallerclass of weakly continuous functionals (Foias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Ergodicity and extremal points

Theorem (Ergodicity and extremal points W 08)

Let IM be the set of all invariant probability measures of adissipative dynamical system S(t) t ge 0 Then an invariantmeasure micro is ergodic if micro is an extreme point of IM Moreover if thedynamical system is injective on the global attractor A then everyergodic invariant measure must be an extremal point of IM

Other versions with group (ODE) assumption is well-known

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 14: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Figure George DavidBirkhoff 1884-1944 Definition

micro is ergodic if micro(E) = 0 or 1 for allinvariant sets E

Theorem (Birkhoffrsquos ErgodicTheorem)

If micro is invariant and ergodic thetemporal and spatial averages areequivalent ie

limTrarrinfin

1T

int T

0ϕ(S(t)u) dt =

intH

ϕ(u) dmicro(u) as

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

More on statistical theories for complex dynamical systems can befound

Majda AJ and Wang X Nonlinear Dynamics and StatisticalTheories for Basic Geophysical Flows Cambridge UniversityPress 2006Majda AJ Abramov R and Grote M Information theory andstochastics for multiscale nonlinear systems CRM monographseries American Mathematical Society 2005Foias C Manley O Rosa R Temam R Navier-StokesEquations and Turbulence Cambridge University PressCambridge UK 2001

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Dissipative system

Definition

A dynamical system S(t) t ge 0 on a phase space H is calleddissipative if there exists a global attractor A such that

A is invariant under S(t)A is compactA attracts any bounded set B in H ie

limtrarrinfin

dist(S(t)BA) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and time averages

Theorem (IM and time averages W 08)

Assume H reflexive S dissipativeforallu0 isin HforallLIM rArr existmicro isin IM such that

LIMTrarrinfin1T

int T

0ϕ(S(t)u0) dt =

intH

ϕ(u) dmicro(u)forallϕ isin C(H)

Earlier work under existence of a compact absorbing set or smallerclass of weakly continuous functionals (Foias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Ergodicity and extremal points

Theorem (Ergodicity and extremal points W 08)

Let IM be the set of all invariant probability measures of adissipative dynamical system S(t) t ge 0 Then an invariantmeasure micro is ergodic if micro is an extreme point of IM Moreover if thedynamical system is injective on the global attractor A then everyergodic invariant measure must be an extremal point of IM

Other versions with group (ODE) assumption is well-known

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 15: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Figure George DavidBirkhoff 1884-1944 Definition

micro is ergodic if micro(E) = 0 or 1 for allinvariant sets E

Theorem (Birkhoffrsquos ErgodicTheorem)

If micro is invariant and ergodic thetemporal and spatial averages areequivalent ie

limTrarrinfin

1T

int T

0ϕ(S(t)u) dt =

intH

ϕ(u) dmicro(u) as

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

More on statistical theories for complex dynamical systems can befound

Majda AJ and Wang X Nonlinear Dynamics and StatisticalTheories for Basic Geophysical Flows Cambridge UniversityPress 2006Majda AJ Abramov R and Grote M Information theory andstochastics for multiscale nonlinear systems CRM monographseries American Mathematical Society 2005Foias C Manley O Rosa R Temam R Navier-StokesEquations and Turbulence Cambridge University PressCambridge UK 2001

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Dissipative system

Definition

A dynamical system S(t) t ge 0 on a phase space H is calleddissipative if there exists a global attractor A such that

A is invariant under S(t)A is compactA attracts any bounded set B in H ie

limtrarrinfin

dist(S(t)BA) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and time averages

Theorem (IM and time averages W 08)

Assume H reflexive S dissipativeforallu0 isin HforallLIM rArr existmicro isin IM such that

LIMTrarrinfin1T

int T

0ϕ(S(t)u0) dt =

intH

ϕ(u) dmicro(u)forallϕ isin C(H)

Earlier work under existence of a compact absorbing set or smallerclass of weakly continuous functionals (Foias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Ergodicity and extremal points

Theorem (Ergodicity and extremal points W 08)

Let IM be the set of all invariant probability measures of adissipative dynamical system S(t) t ge 0 Then an invariantmeasure micro is ergodic if micro is an extreme point of IM Moreover if thedynamical system is injective on the global attractor A then everyergodic invariant measure must be an extremal point of IM

Other versions with group (ODE) assumption is well-known

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 16: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Figure George DavidBirkhoff 1884-1944 Definition

micro is ergodic if micro(E) = 0 or 1 for allinvariant sets E

Theorem (Birkhoffrsquos ErgodicTheorem)

If micro is invariant and ergodic thetemporal and spatial averages areequivalent ie

limTrarrinfin

1T

int T

0ϕ(S(t)u) dt =

intH

ϕ(u) dmicro(u) as

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

More on statistical theories for complex dynamical systems can befound

Majda AJ and Wang X Nonlinear Dynamics and StatisticalTheories for Basic Geophysical Flows Cambridge UniversityPress 2006Majda AJ Abramov R and Grote M Information theory andstochastics for multiscale nonlinear systems CRM monographseries American Mathematical Society 2005Foias C Manley O Rosa R Temam R Navier-StokesEquations and Turbulence Cambridge University PressCambridge UK 2001

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Dissipative system

Definition

A dynamical system S(t) t ge 0 on a phase space H is calleddissipative if there exists a global attractor A such that

A is invariant under S(t)A is compactA attracts any bounded set B in H ie

limtrarrinfin

dist(S(t)BA) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and time averages

Theorem (IM and time averages W 08)

Assume H reflexive S dissipativeforallu0 isin HforallLIM rArr existmicro isin IM such that

LIMTrarrinfin1T

int T

0ϕ(S(t)u0) dt =

intH

ϕ(u) dmicro(u)forallϕ isin C(H)

Earlier work under existence of a compact absorbing set or smallerclass of weakly continuous functionals (Foias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Ergodicity and extremal points

Theorem (Ergodicity and extremal points W 08)

Let IM be the set of all invariant probability measures of adissipative dynamical system S(t) t ge 0 Then an invariantmeasure micro is ergodic if micro is an extreme point of IM Moreover if thedynamical system is injective on the global attractor A then everyergodic invariant measure must be an extremal point of IM

Other versions with group (ODE) assumption is well-known

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 17: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Liouville and Hopf equations

Liouville type equation

ddt

intH

Φ(v) dmicrot(v) =

intH

lt Φprime(v) F(v) gt dmicrot(v)

Φ good test functionals eg

Φ(v) = φ((v v1) middot middot middot (v vN))

Liouville equation (finite d)

part

parttp(v t) +nabla middot (p(v t)F(v)) = 0

Hopfrsquos equation (special case of Liouville type)

ddt

intH

ei(vg) dmicrot(v) =

intH

i lt F(v) g gt ei(vg) dmicrot(v)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Figure George DavidBirkhoff 1884-1944 Definition

micro is ergodic if micro(E) = 0 or 1 for allinvariant sets E

Theorem (Birkhoffrsquos ErgodicTheorem)

If micro is invariant and ergodic thetemporal and spatial averages areequivalent ie

limTrarrinfin

1T

int T

0ϕ(S(t)u) dt =

intH

ϕ(u) dmicro(u) as

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

More on statistical theories for complex dynamical systems can befound

Majda AJ and Wang X Nonlinear Dynamics and StatisticalTheories for Basic Geophysical Flows Cambridge UniversityPress 2006Majda AJ Abramov R and Grote M Information theory andstochastics for multiscale nonlinear systems CRM monographseries American Mathematical Society 2005Foias C Manley O Rosa R Temam R Navier-StokesEquations and Turbulence Cambridge University PressCambridge UK 2001

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Dissipative system

Definition

A dynamical system S(t) t ge 0 on a phase space H is calleddissipative if there exists a global attractor A such that

A is invariant under S(t)A is compactA attracts any bounded set B in H ie

limtrarrinfin

dist(S(t)BA) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and time averages

Theorem (IM and time averages W 08)

Assume H reflexive S dissipativeforallu0 isin HforallLIM rArr existmicro isin IM such that

LIMTrarrinfin1T

int T

0ϕ(S(t)u0) dt =

intH

ϕ(u) dmicro(u)forallϕ isin C(H)

Earlier work under existence of a compact absorbing set or smallerclass of weakly continuous functionals (Foias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Ergodicity and extremal points

Theorem (Ergodicity and extremal points W 08)

Let IM be the set of all invariant probability measures of adissipative dynamical system S(t) t ge 0 Then an invariantmeasure micro is ergodic if micro is an extreme point of IM Moreover if thedynamical system is injective on the global attractor A then everyergodic invariant measure must be an extremal point of IM

Other versions with group (ODE) assumption is well-known

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 18: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Figure George DavidBirkhoff 1884-1944 Definition

micro is ergodic if micro(E) = 0 or 1 for allinvariant sets E

Theorem (Birkhoffrsquos ErgodicTheorem)

If micro is invariant and ergodic thetemporal and spatial averages areequivalent ie

limTrarrinfin

1T

int T

0ϕ(S(t)u) dt =

intH

ϕ(u) dmicro(u) as

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

More on statistical theories for complex dynamical systems can befound

Majda AJ and Wang X Nonlinear Dynamics and StatisticalTheories for Basic Geophysical Flows Cambridge UniversityPress 2006Majda AJ Abramov R and Grote M Information theory andstochastics for multiscale nonlinear systems CRM monographseries American Mathematical Society 2005Foias C Manley O Rosa R Temam R Navier-StokesEquations and Turbulence Cambridge University PressCambridge UK 2001

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Dissipative system

Definition

A dynamical system S(t) t ge 0 on a phase space H is calleddissipative if there exists a global attractor A such that

A is invariant under S(t)A is compactA attracts any bounded set B in H ie

limtrarrinfin

dist(S(t)BA) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and time averages

Theorem (IM and time averages W 08)

Assume H reflexive S dissipativeforallu0 isin HforallLIM rArr existmicro isin IM such that

LIMTrarrinfin1T

int T

0ϕ(S(t)u0) dt =

intH

ϕ(u) dmicro(u)forallϕ isin C(H)

Earlier work under existence of a compact absorbing set or smallerclass of weakly continuous functionals (Foias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Ergodicity and extremal points

Theorem (Ergodicity and extremal points W 08)

Let IM be the set of all invariant probability measures of adissipative dynamical system S(t) t ge 0 Then an invariantmeasure micro is ergodic if micro is an extreme point of IM Moreover if thedynamical system is injective on the global attractor A then everyergodic invariant measure must be an extremal point of IM

Other versions with group (ODE) assumption is well-known

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 19: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Figure George DavidBirkhoff 1884-1944 Definition

micro is ergodic if micro(E) = 0 or 1 for allinvariant sets E

Theorem (Birkhoffrsquos ErgodicTheorem)

If micro is invariant and ergodic thetemporal and spatial averages areequivalent ie

limTrarrinfin

1T

int T

0ϕ(S(t)u) dt =

intH

ϕ(u) dmicro(u) as

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

More on statistical theories for complex dynamical systems can befound

Majda AJ and Wang X Nonlinear Dynamics and StatisticalTheories for Basic Geophysical Flows Cambridge UniversityPress 2006Majda AJ Abramov R and Grote M Information theory andstochastics for multiscale nonlinear systems CRM monographseries American Mathematical Society 2005Foias C Manley O Rosa R Temam R Navier-StokesEquations and Turbulence Cambridge University PressCambridge UK 2001

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Dissipative system

Definition

A dynamical system S(t) t ge 0 on a phase space H is calleddissipative if there exists a global attractor A such that

A is invariant under S(t)A is compactA attracts any bounded set B in H ie

limtrarrinfin

dist(S(t)BA) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and time averages

Theorem (IM and time averages W 08)

Assume H reflexive S dissipativeforallu0 isin HforallLIM rArr existmicro isin IM such that

LIMTrarrinfin1T

int T

0ϕ(S(t)u0) dt =

intH

ϕ(u) dmicro(u)forallϕ isin C(H)

Earlier work under existence of a compact absorbing set or smallerclass of weakly continuous functionals (Foias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Ergodicity and extremal points

Theorem (Ergodicity and extremal points W 08)

Let IM be the set of all invariant probability measures of adissipative dynamical system S(t) t ge 0 Then an invariantmeasure micro is ergodic if micro is an extreme point of IM Moreover if thedynamical system is injective on the global attractor A then everyergodic invariant measure must be an extremal point of IM

Other versions with group (ODE) assumption is well-known

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 20: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Figure George DavidBirkhoff 1884-1944 Definition

micro is ergodic if micro(E) = 0 or 1 for allinvariant sets E

Theorem (Birkhoffrsquos ErgodicTheorem)

If micro is invariant and ergodic thetemporal and spatial averages areequivalent ie

limTrarrinfin

1T

int T

0ϕ(S(t)u) dt =

intH

ϕ(u) dmicro(u) as

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

More on statistical theories for complex dynamical systems can befound

Majda AJ and Wang X Nonlinear Dynamics and StatisticalTheories for Basic Geophysical Flows Cambridge UniversityPress 2006Majda AJ Abramov R and Grote M Information theory andstochastics for multiscale nonlinear systems CRM monographseries American Mathematical Society 2005Foias C Manley O Rosa R Temam R Navier-StokesEquations and Turbulence Cambridge University PressCambridge UK 2001

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Dissipative system

Definition

A dynamical system S(t) t ge 0 on a phase space H is calleddissipative if there exists a global attractor A such that

A is invariant under S(t)A is compactA attracts any bounded set B in H ie

limtrarrinfin

dist(S(t)BA) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and time averages

Theorem (IM and time averages W 08)

Assume H reflexive S dissipativeforallu0 isin HforallLIM rArr existmicro isin IM such that

LIMTrarrinfin1T

int T

0ϕ(S(t)u0) dt =

intH

ϕ(u) dmicro(u)forallϕ isin C(H)

Earlier work under existence of a compact absorbing set or smallerclass of weakly continuous functionals (Foias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Ergodicity and extremal points

Theorem (Ergodicity and extremal points W 08)

Let IM be the set of all invariant probability measures of adissipative dynamical system S(t) t ge 0 Then an invariantmeasure micro is ergodic if micro is an extreme point of IM Moreover if thedynamical system is injective on the global attractor A then everyergodic invariant measure must be an extremal point of IM

Other versions with group (ODE) assumption is well-known

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 21: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Figure George DavidBirkhoff 1884-1944 Definition

micro is ergodic if micro(E) = 0 or 1 for allinvariant sets E

Theorem (Birkhoffrsquos ErgodicTheorem)

If micro is invariant and ergodic thetemporal and spatial averages areequivalent ie

limTrarrinfin

1T

int T

0ϕ(S(t)u) dt =

intH

ϕ(u) dmicro(u) as

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

More on statistical theories for complex dynamical systems can befound

Majda AJ and Wang X Nonlinear Dynamics and StatisticalTheories for Basic Geophysical Flows Cambridge UniversityPress 2006Majda AJ Abramov R and Grote M Information theory andstochastics for multiscale nonlinear systems CRM monographseries American Mathematical Society 2005Foias C Manley O Rosa R Temam R Navier-StokesEquations and Turbulence Cambridge University PressCambridge UK 2001

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Dissipative system

Definition

A dynamical system S(t) t ge 0 on a phase space H is calleddissipative if there exists a global attractor A such that

A is invariant under S(t)A is compactA attracts any bounded set B in H ie

limtrarrinfin

dist(S(t)BA) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and time averages

Theorem (IM and time averages W 08)

Assume H reflexive S dissipativeforallu0 isin HforallLIM rArr existmicro isin IM such that

LIMTrarrinfin1T

int T

0ϕ(S(t)u0) dt =

intH

ϕ(u) dmicro(u)forallϕ isin C(H)

Earlier work under existence of a compact absorbing set or smallerclass of weakly continuous functionals (Foias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Ergodicity and extremal points

Theorem (Ergodicity and extremal points W 08)

Let IM be the set of all invariant probability measures of adissipative dynamical system S(t) t ge 0 Then an invariantmeasure micro is ergodic if micro is an extreme point of IM Moreover if thedynamical system is injective on the global attractor A then everyergodic invariant measure must be an extremal point of IM

Other versions with group (ODE) assumption is well-known

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 22: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Stationary Statistics Solutions (IM)

Invariant measure (IM) micro isin PM(H)

micro(Sminus1(t)(E) = micro(E)

Stationary statistical solutions essentiallyintH

lt F (v)Φprime(v) gt dmicro(v) = 0forallΦ

Stationary statistical solutions IM are not necessarily supportedon steady state solutions or periodic orbitsUncertainty in both initial condition and parameter(s) (modelerror)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Figure George DavidBirkhoff 1884-1944 Definition

micro is ergodic if micro(E) = 0 or 1 for allinvariant sets E

Theorem (Birkhoffrsquos ErgodicTheorem)

If micro is invariant and ergodic thetemporal and spatial averages areequivalent ie

limTrarrinfin

1T

int T

0ϕ(S(t)u) dt =

intH

ϕ(u) dmicro(u) as

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

More on statistical theories for complex dynamical systems can befound

Majda AJ and Wang X Nonlinear Dynamics and StatisticalTheories for Basic Geophysical Flows Cambridge UniversityPress 2006Majda AJ Abramov R and Grote M Information theory andstochastics for multiscale nonlinear systems CRM monographseries American Mathematical Society 2005Foias C Manley O Rosa R Temam R Navier-StokesEquations and Turbulence Cambridge University PressCambridge UK 2001

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Dissipative system

Definition

A dynamical system S(t) t ge 0 on a phase space H is calleddissipative if there exists a global attractor A such that

A is invariant under S(t)A is compactA attracts any bounded set B in H ie

limtrarrinfin

dist(S(t)BA) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and time averages

Theorem (IM and time averages W 08)

Assume H reflexive S dissipativeforallu0 isin HforallLIM rArr existmicro isin IM such that

LIMTrarrinfin1T

int T

0ϕ(S(t)u0) dt =

intH

ϕ(u) dmicro(u)forallϕ isin C(H)

Earlier work under existence of a compact absorbing set or smallerclass of weakly continuous functionals (Foias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Ergodicity and extremal points

Theorem (Ergodicity and extremal points W 08)

Let IM be the set of all invariant probability measures of adissipative dynamical system S(t) t ge 0 Then an invariantmeasure micro is ergodic if micro is an extreme point of IM Moreover if thedynamical system is injective on the global attractor A then everyergodic invariant measure must be an extremal point of IM

Other versions with group (ODE) assumption is well-known

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 23: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Figure George DavidBirkhoff 1884-1944 Definition

micro is ergodic if micro(E) = 0 or 1 for allinvariant sets E

Theorem (Birkhoffrsquos ErgodicTheorem)

If micro is invariant and ergodic thetemporal and spatial averages areequivalent ie

limTrarrinfin

1T

int T

0ϕ(S(t)u) dt =

intH

ϕ(u) dmicro(u) as

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

More on statistical theories for complex dynamical systems can befound

Majda AJ and Wang X Nonlinear Dynamics and StatisticalTheories for Basic Geophysical Flows Cambridge UniversityPress 2006Majda AJ Abramov R and Grote M Information theory andstochastics for multiscale nonlinear systems CRM monographseries American Mathematical Society 2005Foias C Manley O Rosa R Temam R Navier-StokesEquations and Turbulence Cambridge University PressCambridge UK 2001

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Dissipative system

Definition

A dynamical system S(t) t ge 0 on a phase space H is calleddissipative if there exists a global attractor A such that

A is invariant under S(t)A is compactA attracts any bounded set B in H ie

limtrarrinfin

dist(S(t)BA) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and time averages

Theorem (IM and time averages W 08)

Assume H reflexive S dissipativeforallu0 isin HforallLIM rArr existmicro isin IM such that

LIMTrarrinfin1T

int T

0ϕ(S(t)u0) dt =

intH

ϕ(u) dmicro(u)forallϕ isin C(H)

Earlier work under existence of a compact absorbing set or smallerclass of weakly continuous functionals (Foias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Ergodicity and extremal points

Theorem (Ergodicity and extremal points W 08)

Let IM be the set of all invariant probability measures of adissipative dynamical system S(t) t ge 0 Then an invariantmeasure micro is ergodic if micro is an extreme point of IM Moreover if thedynamical system is injective on the global attractor A then everyergodic invariant measure must be an extremal point of IM

Other versions with group (ODE) assumption is well-known

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 24: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

More on statistical theories for complex dynamical systems can befound

Majda AJ and Wang X Nonlinear Dynamics and StatisticalTheories for Basic Geophysical Flows Cambridge UniversityPress 2006Majda AJ Abramov R and Grote M Information theory andstochastics for multiscale nonlinear systems CRM monographseries American Mathematical Society 2005Foias C Manley O Rosa R Temam R Navier-StokesEquations and Turbulence Cambridge University PressCambridge UK 2001

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Dissipative system

Definition

A dynamical system S(t) t ge 0 on a phase space H is calleddissipative if there exists a global attractor A such that

A is invariant under S(t)A is compactA attracts any bounded set B in H ie

limtrarrinfin

dist(S(t)BA) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and time averages

Theorem (IM and time averages W 08)

Assume H reflexive S dissipativeforallu0 isin HforallLIM rArr existmicro isin IM such that

LIMTrarrinfin1T

int T

0ϕ(S(t)u0) dt =

intH

ϕ(u) dmicro(u)forallϕ isin C(H)

Earlier work under existence of a compact absorbing set or smallerclass of weakly continuous functionals (Foias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Ergodicity and extremal points

Theorem (Ergodicity and extremal points W 08)

Let IM be the set of all invariant probability measures of adissipative dynamical system S(t) t ge 0 Then an invariantmeasure micro is ergodic if micro is an extreme point of IM Moreover if thedynamical system is injective on the global attractor A then everyergodic invariant measure must be an extremal point of IM

Other versions with group (ODE) assumption is well-known

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 25: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

More on statistical theories for complex dynamical systems can befound

Majda AJ and Wang X Nonlinear Dynamics and StatisticalTheories for Basic Geophysical Flows Cambridge UniversityPress 2006Majda AJ Abramov R and Grote M Information theory andstochastics for multiscale nonlinear systems CRM monographseries American Mathematical Society 2005Foias C Manley O Rosa R Temam R Navier-StokesEquations and Turbulence Cambridge University PressCambridge UK 2001

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Dissipative system

Definition

A dynamical system S(t) t ge 0 on a phase space H is calleddissipative if there exists a global attractor A such that

A is invariant under S(t)A is compactA attracts any bounded set B in H ie

limtrarrinfin

dist(S(t)BA) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and time averages

Theorem (IM and time averages W 08)

Assume H reflexive S dissipativeforallu0 isin HforallLIM rArr existmicro isin IM such that

LIMTrarrinfin1T

int T

0ϕ(S(t)u0) dt =

intH

ϕ(u) dmicro(u)forallϕ isin C(H)

Earlier work under existence of a compact absorbing set or smallerclass of weakly continuous functionals (Foias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Ergodicity and extremal points

Theorem (Ergodicity and extremal points W 08)

Let IM be the set of all invariant probability measures of adissipative dynamical system S(t) t ge 0 Then an invariantmeasure micro is ergodic if micro is an extreme point of IM Moreover if thedynamical system is injective on the global attractor A then everyergodic invariant measure must be an extremal point of IM

Other versions with group (ODE) assumption is well-known

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 26: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

More on statistical theories for complex dynamical systems can befound

Majda AJ and Wang X Nonlinear Dynamics and StatisticalTheories for Basic Geophysical Flows Cambridge UniversityPress 2006Majda AJ Abramov R and Grote M Information theory andstochastics for multiscale nonlinear systems CRM monographseries American Mathematical Society 2005Foias C Manley O Rosa R Temam R Navier-StokesEquations and Turbulence Cambridge University PressCambridge UK 2001

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Dissipative system

Definition

A dynamical system S(t) t ge 0 on a phase space H is calleddissipative if there exists a global attractor A such that

A is invariant under S(t)A is compactA attracts any bounded set B in H ie

limtrarrinfin

dist(S(t)BA) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and time averages

Theorem (IM and time averages W 08)

Assume H reflexive S dissipativeforallu0 isin HforallLIM rArr existmicro isin IM such that

LIMTrarrinfin1T

int T

0ϕ(S(t)u0) dt =

intH

ϕ(u) dmicro(u)forallϕ isin C(H)

Earlier work under existence of a compact absorbing set or smallerclass of weakly continuous functionals (Foias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Ergodicity and extremal points

Theorem (Ergodicity and extremal points W 08)

Let IM be the set of all invariant probability measures of adissipative dynamical system S(t) t ge 0 Then an invariantmeasure micro is ergodic if micro is an extreme point of IM Moreover if thedynamical system is injective on the global attractor A then everyergodic invariant measure must be an extremal point of IM

Other versions with group (ODE) assumption is well-known

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 27: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

More on statistical theories for complex dynamical systems can befound

Majda AJ and Wang X Nonlinear Dynamics and StatisticalTheories for Basic Geophysical Flows Cambridge UniversityPress 2006Majda AJ Abramov R and Grote M Information theory andstochastics for multiscale nonlinear systems CRM monographseries American Mathematical Society 2005Foias C Manley O Rosa R Temam R Navier-StokesEquations and Turbulence Cambridge University PressCambridge UK 2001

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Dissipative system

Definition

A dynamical system S(t) t ge 0 on a phase space H is calleddissipative if there exists a global attractor A such that

A is invariant under S(t)A is compactA attracts any bounded set B in H ie

limtrarrinfin

dist(S(t)BA) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and time averages

Theorem (IM and time averages W 08)

Assume H reflexive S dissipativeforallu0 isin HforallLIM rArr existmicro isin IM such that

LIMTrarrinfin1T

int T

0ϕ(S(t)u0) dt =

intH

ϕ(u) dmicro(u)forallϕ isin C(H)

Earlier work under existence of a compact absorbing set or smallerclass of weakly continuous functionals (Foias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Ergodicity and extremal points

Theorem (Ergodicity and extremal points W 08)

Let IM be the set of all invariant probability measures of adissipative dynamical system S(t) t ge 0 Then an invariantmeasure micro is ergodic if micro is an extreme point of IM Moreover if thedynamical system is injective on the global attractor A then everyergodic invariant measure must be an extremal point of IM

Other versions with group (ODE) assumption is well-known

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 28: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

More on statistical theories for complex dynamical systems can befound

Majda AJ and Wang X Nonlinear Dynamics and StatisticalTheories for Basic Geophysical Flows Cambridge UniversityPress 2006Majda AJ Abramov R and Grote M Information theory andstochastics for multiscale nonlinear systems CRM monographseries American Mathematical Society 2005Foias C Manley O Rosa R Temam R Navier-StokesEquations and Turbulence Cambridge University PressCambridge UK 2001

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Dissipative system

Definition

A dynamical system S(t) t ge 0 on a phase space H is calleddissipative if there exists a global attractor A such that

A is invariant under S(t)A is compactA attracts any bounded set B in H ie

limtrarrinfin

dist(S(t)BA) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and time averages

Theorem (IM and time averages W 08)

Assume H reflexive S dissipativeforallu0 isin HforallLIM rArr existmicro isin IM such that

LIMTrarrinfin1T

int T

0ϕ(S(t)u0) dt =

intH

ϕ(u) dmicro(u)forallϕ isin C(H)

Earlier work under existence of a compact absorbing set or smallerclass of weakly continuous functionals (Foias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Ergodicity and extremal points

Theorem (Ergodicity and extremal points W 08)

Let IM be the set of all invariant probability measures of adissipative dynamical system S(t) t ge 0 Then an invariantmeasure micro is ergodic if micro is an extreme point of IM Moreover if thedynamical system is injective on the global attractor A then everyergodic invariant measure must be an extremal point of IM

Other versions with group (ODE) assumption is well-known

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 29: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Maximum entropy for conservative case

Maximum entropy principle most probable pdf maximize theShannon entropy S = minus

intp ln p

Entropy is conserved in the deterministic case with Liouvilleproperty (nabla middot F = 0)

d ~Xdt

= ~F (~X ) + εd ~Wdt

Fokker-Planck equation (Kolmogorov Smoluchowski)

partppartt

+ ~F middot nabla~X p minus ε2

2∆~X p = 0

Equation for the density of Shannon entropy Q = minusp ln p

partQpartt

+nabla~X middot (~FQ)minus ε2

2∆~X Q =

ε2

2p|nablap|2

Monotonicity of Shannon entropy (noise increases uncertainty)ddtS(p(t)) ge 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

More on statistical theories for complex dynamical systems can befound

Majda AJ and Wang X Nonlinear Dynamics and StatisticalTheories for Basic Geophysical Flows Cambridge UniversityPress 2006Majda AJ Abramov R and Grote M Information theory andstochastics for multiscale nonlinear systems CRM monographseries American Mathematical Society 2005Foias C Manley O Rosa R Temam R Navier-StokesEquations and Turbulence Cambridge University PressCambridge UK 2001

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Dissipative system

Definition

A dynamical system S(t) t ge 0 on a phase space H is calleddissipative if there exists a global attractor A such that

A is invariant under S(t)A is compactA attracts any bounded set B in H ie

limtrarrinfin

dist(S(t)BA) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and time averages

Theorem (IM and time averages W 08)

Assume H reflexive S dissipativeforallu0 isin HforallLIM rArr existmicro isin IM such that

LIMTrarrinfin1T

int T

0ϕ(S(t)u0) dt =

intH

ϕ(u) dmicro(u)forallϕ isin C(H)

Earlier work under existence of a compact absorbing set or smallerclass of weakly continuous functionals (Foias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Ergodicity and extremal points

Theorem (Ergodicity and extremal points W 08)

Let IM be the set of all invariant probability measures of adissipative dynamical system S(t) t ge 0 Then an invariantmeasure micro is ergodic if micro is an extreme point of IM Moreover if thedynamical system is injective on the global attractor A then everyergodic invariant measure must be an extremal point of IM

Other versions with group (ODE) assumption is well-known

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 30: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

More on statistical theories for complex dynamical systems can befound

Majda AJ and Wang X Nonlinear Dynamics and StatisticalTheories for Basic Geophysical Flows Cambridge UniversityPress 2006Majda AJ Abramov R and Grote M Information theory andstochastics for multiscale nonlinear systems CRM monographseries American Mathematical Society 2005Foias C Manley O Rosa R Temam R Navier-StokesEquations and Turbulence Cambridge University PressCambridge UK 2001

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Dissipative system

Definition

A dynamical system S(t) t ge 0 on a phase space H is calleddissipative if there exists a global attractor A such that

A is invariant under S(t)A is compactA attracts any bounded set B in H ie

limtrarrinfin

dist(S(t)BA) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and time averages

Theorem (IM and time averages W 08)

Assume H reflexive S dissipativeforallu0 isin HforallLIM rArr existmicro isin IM such that

LIMTrarrinfin1T

int T

0ϕ(S(t)u0) dt =

intH

ϕ(u) dmicro(u)forallϕ isin C(H)

Earlier work under existence of a compact absorbing set or smallerclass of weakly continuous functionals (Foias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Ergodicity and extremal points

Theorem (Ergodicity and extremal points W 08)

Let IM be the set of all invariant probability measures of adissipative dynamical system S(t) t ge 0 Then an invariantmeasure micro is ergodic if micro is an extreme point of IM Moreover if thedynamical system is injective on the global attractor A then everyergodic invariant measure must be an extremal point of IM

Other versions with group (ODE) assumption is well-known

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 31: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Dissipative system

Definition

A dynamical system S(t) t ge 0 on a phase space H is calleddissipative if there exists a global attractor A such that

A is invariant under S(t)A is compactA attracts any bounded set B in H ie

limtrarrinfin

dist(S(t)BA) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and time averages

Theorem (IM and time averages W 08)

Assume H reflexive S dissipativeforallu0 isin HforallLIM rArr existmicro isin IM such that

LIMTrarrinfin1T

int T

0ϕ(S(t)u0) dt =

intH

ϕ(u) dmicro(u)forallϕ isin C(H)

Earlier work under existence of a compact absorbing set or smallerclass of weakly continuous functionals (Foias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Ergodicity and extremal points

Theorem (Ergodicity and extremal points W 08)

Let IM be the set of all invariant probability measures of adissipative dynamical system S(t) t ge 0 Then an invariantmeasure micro is ergodic if micro is an extreme point of IM Moreover if thedynamical system is injective on the global attractor A then everyergodic invariant measure must be an extremal point of IM

Other versions with group (ODE) assumption is well-known

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 32: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and time averages

Theorem (IM and time averages W 08)

Assume H reflexive S dissipativeforallu0 isin HforallLIM rArr existmicro isin IM such that

LIMTrarrinfin1T

int T

0ϕ(S(t)u0) dt =

intH

ϕ(u) dmicro(u)forallϕ isin C(H)

Earlier work under existence of a compact absorbing set or smallerclass of weakly continuous functionals (Foias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Ergodicity and extremal points

Theorem (Ergodicity and extremal points W 08)

Let IM be the set of all invariant probability measures of adissipative dynamical system S(t) t ge 0 Then an invariantmeasure micro is ergodic if micro is an extreme point of IM Moreover if thedynamical system is injective on the global attractor A then everyergodic invariant measure must be an extremal point of IM

Other versions with group (ODE) assumption is well-known

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 33: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and attractors

Theorem (IM and the global attractors W 08)1 IM is a convex compact set (with respect to the weak topology)2 suppmicro sub Aforallmicro isin IM

Singular nature of invariant measureEarlier work under existence of a compact absorbing set (CiprianFoias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and time averages

Theorem (IM and time averages W 08)

Assume H reflexive S dissipativeforallu0 isin HforallLIM rArr existmicro isin IM such that

LIMTrarrinfin1T

int T

0ϕ(S(t)u0) dt =

intH

ϕ(u) dmicro(u)forallϕ isin C(H)

Earlier work under existence of a compact absorbing set or smallerclass of weakly continuous functionals (Foias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Ergodicity and extremal points

Theorem (Ergodicity and extremal points W 08)

Let IM be the set of all invariant probability measures of adissipative dynamical system S(t) t ge 0 Then an invariantmeasure micro is ergodic if micro is an extreme point of IM Moreover if thedynamical system is injective on the global attractor A then everyergodic invariant measure must be an extremal point of IM

Other versions with group (ODE) assumption is well-known

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 34: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

IM and time averages

Theorem (IM and time averages W 08)

Assume H reflexive S dissipativeforallu0 isin HforallLIM rArr existmicro isin IM such that

LIMTrarrinfin1T

int T

0ϕ(S(t)u0) dt =

intH

ϕ(u) dmicro(u)forallϕ isin C(H)

Earlier work under existence of a compact absorbing set or smallerclass of weakly continuous functionals (Foias et al)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Ergodicity and extremal points

Theorem (Ergodicity and extremal points W 08)

Let IM be the set of all invariant probability measures of adissipative dynamical system S(t) t ge 0 Then an invariantmeasure micro is ergodic if micro is an extreme point of IM Moreover if thedynamical system is injective on the global attractor A then everyergodic invariant measure must be an extremal point of IM

Other versions with group (ODE) assumption is well-known

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 35: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Ergodicity and extremal points

Theorem (Ergodicity and extremal points W 08)

Let IM be the set of all invariant probability measures of adissipative dynamical system S(t) t ge 0 Then an invariantmeasure micro is ergodic if micro is an extreme point of IM Moreover if thedynamical system is injective on the global attractor A then everyergodic invariant measure must be an extremal point of IM

Other versions with group (ODE) assumption is well-known

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 36: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 37: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 38: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 39: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Regular perturbation

Theorem (Conv of IM regular version W 08)

Assume for S(t ε)1 (uniformly dissipativity) pre-compactness of K =

⋃0lt|ε|leε0

2 (finite time u-conv)limεrarr0 supuisinAε

S(t ε)uminus S(t 0)uH = 0forallt ge 0

Thenlimεrarr0 microε = micro0 microε isin IMε micro0 isin IM0

Beyond continuity linear response

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 40: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Two time scale set-up

Two-time-scale problem X1 X2 Hilbert spaces

ε(du1

dt+ g(u1 u2)) = f1(u1 u2) u1(0) = u10

du2

dt= f2(u1 u2) u2(0) = u20

Limit problem ( ε = 0)

0 = f1(u01 u0

2)

du02

dt= f2(u0

1 u02) u0

2(0) = u20

y = f1(u1 u2) hArr u1 = F1(u2 y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 41: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 42: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 43: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 44: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 45: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 46: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Conv of stat prop 2 time scale case

Theorem (Conv of IM singular version W 08)

Assume1 (uniform dissipativity) pre-compactness of K =

⋃0ltεltε0

2 (dissipativity of the limit system) A0 in X23 (conv of the slow variable)

limεrarr0 supu2isinP2AεP2S(t ε)(F1(u2 0) u2)minus S(t 0)u2X2 =

0 forall t ge 0

4 (smallness of the perturbation)limεrarr0 sup(u1u2)isinAε

ε( du1dt + g(u1 u2))X1 = 0

5 (continuity of the slave relation) u1 = F1(u2 y)

Thenmicroε Lmicro0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 47: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Rayleigh and Beacutenard

Figure Lord Rayleigh (JohnWilliam Strutt) 1842-1919

Figure Henri Beacutenard (left)1874-1939

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 48: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 49: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 50: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to RBC W CPAM 07

Boussinesq system for Rayleigh-Beacutenard convection

1Pr

(partupartt

+ (u middot nabla)u) +nablap = ∆u + Ra kθ nabla middot u = 0 u|z=01 = 0

partθ

partt+ u middot nablaθ minus u3 = ∆θ θ|z=01 = 0

Infinite Prandtl number model for convection

nablap0 = ∆u0 + Ra kθ0 nabla middot u0 = 0 u0|z=01 = 0

partθ0

partt+ u0 middot nablaθ0 minus u0

3 = ∆θ0 θ0|z=01 = 0

X1 = H X2 = L2 ε = 1Pr and F1(θ y) = Ra Aminus1(kθ)minus Aminus1(y)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 51: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

RBC set-up and numerics

Figure RBC set-up Figure Numerical simulation

(infin Pr simulation)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

m1mpg
Media File (videompeg)

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 52: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 53: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 54: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 55: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Possible deficiency of classical schemes

dudt

= F(u) u isin H

classical scheme of order m

u(n∆t)minus un le C(n∆t)∆tm

Dependence on TC(T ) = exp(αT )

Error in approximation of long time averages

| lim supNrarrinfin

1N

Nsumn=1

(Φ(u(n∆t))minus Φ(un))|

le c lim supNrarrinfin

∆tm exp((N + 1)α∆t)minus exp(α∆t)exp(α∆t)minus 1

= infin

Classical schemes may not be able to capture the climatealthough they may work very well for weather

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 56: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 57: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Difficulty with large Rayleigh number

Infinite Pr number model

partθ0

partt+ Ra Aminus1(kθ0) middot nablaθ0 minus Ra Aminus1(kθ0)3 = ∆θ0

A Stokes operatorAlternative form with s = Ra t

partθ0

parts+ Aminus1(kθ0) middot nablaθ0 minus Aminus1(kθ0)3 =

1Ra

∆θ0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 58: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 59: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 60: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 61: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 62: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Theorem (General result W 08)

Assume1 (Uniform dissipativity) K =

⋃0ltklek0

Ak

2 (Finite time uniform convergence)supuisinAk nkisin[t0T ] Sn

k uminus S(nk)u rarr 0 as k rarr 0

3 (Uniform continuity of the continuous system)limtrarrT supuisinK S(t)uminus S(T )u = 0

Then1 (Conv of stationary stat prop)

microk micro microk isin IMk micro isin IM

2 (Conv of attractors)

limkrarr0

dist(Ak A) = 0

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 63: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 64: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Application to infin Pr model

infin Prandtl number model (alternative form)

partθ

partt+ Ra Aminus1(kθ) middot nablaθ minus Ra Aminus1(kθ)3 = ∆θ θ|z=01 = 0

Semi-implicit scheme

θn+1 minus θn

k+ Ra Aminus1(kθn) middot nablaθn+1 + Ra Aminus1(kθn)3 = ∆θn+1

Equivalent form (T = θ + 1minus z)

T n+1 minus T n

k+ Ra Aminus1(kT n) middot nablaT n+1 = ∆T n+1

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 65: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 66: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number (recast)

For the infinite Prandtl number model

Nu = supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

|nablaT (x s)|2 dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kT (x s))3T (x s) dxds

= 1 + Ra supθ0isinL2

lim suptrarrinfin

1tLxLy

int t

0

intΩ

Aminus1(kθ(x s))3θ(x s) dxds

For the scheme

Nuk = 1 + Ra supθ0isinL2

lim supNrarrinfin

1NLxLy

Nsumn=1

intΩ

Aminus1(kθn(x))3θn(x) dx

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 67: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 68: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Nusselt number limit

Convergence of Nusselt number

lim supkrarr0

Nuk le Nu

Complements variational approach (Constantinamp Doering)

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 69: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 70: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 71: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 72: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 73: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Summary

Invariant measures of appropriate numerical approximationsconverge to invariant measures of the continuous in timedynamical system under approximationSpecific stationary statistical properties (such as Nusseltnumber) also convergeSame idea can be applied to many dissipative dynamicalsystemsUniform dissipativity is crucialLong way to go to reach our goal

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 74: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 75: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 76: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 77: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 78: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 79: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 80: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation
Page 81: Stationary Statistical Properties of Dissipative … Approach Continuity and perturbation results Temporal approximation Stationary Statistical Properties of Dissipative Systems Wang,

Statistical ApproachContinuity and perturbation results

Temporal approximation

Questions

Continuity instead of upper semi-continuity Linear responseStatistical hysteresisUniqueness of physical invariant measure Noise effectBalancing mixing rate and errorConvergence rate At least for certain good statisticalquantitiesHigh order schemesExplicit time stepping (Perhaps with posterior approach)Fully discrete schemesOther schemesPartiallyweakly dissipative systems

Wang Xiaoming wxmmathfsuedu Stationary Statistical Properties of Dissipative Systems

  • Statistical Approach
  • Continuity and perturbation results
  • Temporal approximation