numerical bifurcation analysis of dynamical systems: recent
TRANSCRIPT
![Page 1: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/1.jpg)
Numerical bifurcation analysis ofdynamical systems:
Recent progress and perspectivesYuri A. Kuznetsov
Department of Mathematics
Utrecht University, The Netherlands
UTwente – p. 1/58
![Page 2: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/2.jpg)
Content:
1. Mathematical analysis of deterministic systems
2. Equilibria of ODEs and their bifurcations
3. Limit cycles of ODEs and their local bifurcations
4. Bifurcations of homoclinic orbits
5. Open problems
6. References
UTwente – p. 2/58
![Page 3: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/3.jpg)
1. Mathematical analysis of deterministic systems
and
modelling theoryDynamical systems
Numerical analysis
software tools
Mathematical
⇔ ⇔
⇔
UTwente – p. 3/58
![Page 4: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/4.jpg)
Example: Hodgkin-Huxley [1952] axon equations
CV = I − gNam3h(V − VNa) − gKn
4(V − VK) − gL(V − VL),
m = φ((1 −m)αm −mβm),
h = φ((1 − h)αh − hβh),
n = φ((1 − n)αn − nβn),
where
φ = 3(T−6.3)/10, ψαm= (25 − V )/10, ψαn
= (10 − V )/10,
αm =ψαm
exp(ψαm) − 1
, αh = 0.07 exp(−V/20), αn = 0.1ψαn
exp(ψαn) − 1
,
βm = 4 exp(−V/18), βh =1
1 + exp((30 − V )/10), βn = 0.125 exp(−V/80).
Other models: Connor et al. [1977], Moris-Lecar [1981]; Traub-Miles
[1991]UTwente – p. 4/58
![Page 5: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/5.jpg)
Numerical analysis of dynamical systems
• Simulation at fixed parameter values
• initial-value problems;
• spectral analysis;
• Lyapunov exponents.
• Bifurcation analysis of parameter-dependent systems
• stability boundaries;
• sensitive dependence on control parameters;
• bifurcation diagrams.
UTwente – p. 5/58
![Page 6: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/6.jpg)
Bifurcation analysis of smooth dynamical systems
• Continuation of orbits:
• Equilibria (fixed points) and cycles
• Orbits in invariant manifolds of equilibria and cycles
UTwente – p. 6/58
![Page 7: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/7.jpg)
Bifurcation analysis of smooth dynamical systems
• Continuation of orbits:
• Equilibria (fixed points) and cycles
• Orbits in invariant manifolds of equilibria and cycles
• Continuation of bifurcations:
• Bifurcations of equilibria and cycles
• Bifurcations of homoclinic and heteroclinic orbits
UTwente – p. 6/58
![Page 8: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/8.jpg)
Bifurcation analysis of smooth dynamical systems
• Continuation of orbits:
• Equilibria (fixed points) and cycles
• Orbits in invariant manifolds of equilibria and cycles
• Continuation of bifurcations:
• Bifurcations of equilibria and cycles
• Bifurcations of homoclinic and heteroclinic orbits
• Combined center manifold reduction and normalization:
• Normal forms for bifurcations of equilibria
• Periodic normal forms for bifurcations of cycles
UTwente – p. 6/58
![Page 9: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/9.jpg)
Bifurcation analysis of smooth dynamical systems
• Continuation of orbits:
• Equilibria (fixed points) and cycles
• Orbits in invariant manifolds of equilibria and cycles
• Continuation of bifurcations:
• Bifurcations of equilibria and cycles
• Bifurcations of homoclinic and heteroclinic orbits
• Combined center manifold reduction and normalization:
• Normal forms for bifurcations of equilibria
• Periodic normal forms for bifurcations of cycles
• Branch switching at bifurcations
UTwente – p. 6/58
![Page 10: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/10.jpg)
Software tools for bifurcation analysis
Standard bifurcation and continuation software:
• LOCBIF [1986-1992]
• AUTO97 (HOMCONT[1994-1997], SLIDECONT[2001-2005])
• CONTENT [1993-1998]
• MATCONT [2000-]
UTwente – p. 7/58
![Page 11: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/11.jpg)
Strategy of local bifurcation analysis of ODEs
dx
dt= f(x, α), x ∈ R
n, α ∈ Rm
GPDCPCZH HHBP CP BT GH BPC R3R1 R4 CH LPNS PDNS
PD
EP
O
LP NS
LC
H LPC1
0
codim
2 R2 NSNS LPPD
UTwente – p. 8/58
![Page 12: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/12.jpg)
2. Equilibria of ODEs and their bifurcations
dx
dt= f(x, α), x ∈ R
n, α ∈ Rm.
An equilibriumx0 satisfies
f(x0, α0) = 0
and has eigenvalues{λ1, λ2, . . . , λn} = σ(fx(x0, α0))
0Re λ
Im λ
UTwente – p. 9/58
![Page 13: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/13.jpg)
2. Equilibria of ODEs and their bifurcations
dx
dt= f(x, α), x ∈ R
n, α ∈ Rm.
An equilibriumx0 satisfies
f(x0, α0) = 0
and has eigenvalues{λ1, λ2, . . . , λn} = σ(fx(x0, α0))
0Re λ
Im λ
• Fold (LP): λ1 = 0;
UTwente – p. 9/58
![Page 14: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/14.jpg)
2. Equilibria of ODEs and their bifurcations
dx
dt= f(x, α), x ∈ R
n, α ∈ Rm.
An equilibriumx0 satisfies
f(x0, α0) = 0
and has eigenvalues{λ1, λ2, . . . , λn} = σ(fx(x0, α0))
+iω0
−iω0
0Re λ
Im λ
• Fold (LP): λ1 = 0;
• Andronov-Hopf (H): λ1,2 = ±iω0.
UTwente – p. 9/58
![Page 15: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/15.jpg)
Continuation of LP bifurcation in two parameters
• Defining system:(x, α) ∈ Rn × R2
f(x, α) = 0,
g(x, α) = 0,
whereg is computed by solving thebordered system[Griewank
& Reddien, 1984; Govaerts, 2000] A(x, α) w1
vT1 0
v
g
=
0
1
whereA(x, α) = fx(x, α).
UTwente – p. 10/58
![Page 16: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/16.jpg)
Continuation of LP bifurcation in two parameters
• Defining system:(x, α) ∈ Rn × R2
f(x, α) = 0,
g(x, α) = 0,
whereg is computed by solving thebordered system[Griewank
& Reddien, 1984; Govaerts, 2000] A(x, α) w1
vT1 0
v
g
=
0
1
whereA(x, α) = fx(x, α).
• Vectorsv1, w1 ∈ Rn are adapted along the LP-curve to make the
linear system nonsingular.
UTwente – p. 10/58
![Page 17: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/17.jpg)
Continuation of LP bifurcation in two parameters
• Defining system:(x, α) ∈ Rn × R2
f(x, α) = 0,
g(x, α) = 0,
whereg is computed by solving thebordered system[Griewank
& Reddien, 1984; Govaerts, 2000] A(x, α) w1
vT1 0
v
g
=
0
1
whereA(x, α) = fx(x, α).
• Vectorsv1, w1 ∈ Rn are adapted along the LP-curve to make the
linear system nonsingular.
• (gy, gα) can be computed efficiently using the adjoint linear
system.UTwente – p. 10/58
![Page 18: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/18.jpg)
Generic fold (LP) bifurcation: λ1 = 0
• Smooth normal form on CM:
ξ = β + bξ2 +O(ξ3), b 6= 0.
• Topological normal form on CM:
ξ = β + bξ2, b 6= 0.
W c0
β(α) < 0 β(α) = 0 β(α) > 0
W cβ
W cβ
O− O+
Collision and disappearance of two equilibria:O− +O+ → ∅.
UTwente – p. 11/58
![Page 19: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/19.jpg)
Critical LP-coefficient b
Write following [Coullet & Spiegel, 1983]
F (H) := f(x0 +H,α0) = AH +1
2B(H,H) +O(‖H‖3),
and locally represent the center manifoldW c0 as the graph of a function
H : R → Rn,
x = H(ξ) = ξq +1
2h2ξ
2 +O(ξ3), ξ ∈ R, h2 ∈ Rn.
The restriction ofx = F (x) toW c0 is
ξ = G(ξ) = bξ2 +O(ξ3).
The invariance of the center manifoldW c0 impliesHξ ξ = F (H(ξ)).
UTwente – p. 12/58
![Page 20: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/20.jpg)
A(ξq +1
2h2ξ
2) +1
2B(ξq, ξq) +O(|ξ|3) = bξ2q +
1
2h2ξ
2 +O(|ξ|3)
• Theξ-terms give the identity:Aq = 0.
• Theξ2-terms give the equation forh2:
Ah2 = −B(q, q) + 2bq.
It is singular and itsFredholm solvabilityimplies
b =1
2〈p,B(q, q)〉,
whereAq = ATp = 0, 〈q, q〉 = 〈p, q〉 = 1.
UTwente – p. 13/58
![Page 21: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/21.jpg)
3. Limit cycles of ODEs and their local bifurcations
dx
dt= f(x, α), x ∈ R
n, α ∈ Rm.
A limit cycle C0 corresponds to a periodic solutionx0(t+ T0) = x0(t)
and has Floquet multipliers{µ1, µ2, . . . , µn−1, µn = 1} = σ(M(T0)),
where
M(t) − fx(x0(t), α0)M(t) = 0, M(0) = In.
Re µ
Im µ
UTwente – p. 14/58
![Page 22: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/22.jpg)
3. Limit cycles of ODEs and their local bifurcations
dx
dt= f(x, α), x ∈ R
n, α ∈ Rm.
A limit cycle C0 corresponds to a periodic solutionx0(t+ T0) = x0(t)
and has Floquet multipliers{µ1, µ2, . . . , µn−1, µn = 1} = σ(M(T0)),
where
M(t) − fx(x0(t), α0)M(t) = 0, M(0) = In.
Re µ
Im µ
• Fold (LPC): µ1 = 1;
UTwente – p. 14/58
![Page 23: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/23.jpg)
3. Limit cycles of ODEs and their local bifurcations
dx
dt= f(x, α), x ∈ R
n, α ∈ Rm.
A limit cycle C0 corresponds to a periodic solutionx0(t+ T0) = x0(t)
and has Floquet multipliers{µ1, µ2, . . . , µn−1, µn = 1} = σ(M(T0)),
where
M(t) − fx(x0(t), α0)M(t) = 0, M(0) = In.
Re µ
Im µ
• Fold (LPC): µ1 = 1;
• Flip (PD): µ1 = −1;
UTwente – p. 14/58
![Page 24: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/24.jpg)
3. Limit cycles of ODEs and their local bifurcations
dx
dt= f(x, α), x ∈ R
n, α ∈ Rm.
A limit cycle C0 corresponds to a periodic solutionx0(t+ T0) = x0(t)
and has Floquet multipliers{µ1, µ2, . . . , µn−1, µn = 1} = σ(M(T0)),
where
M(t) − fx(x0(t), α0)M(t) = 0, M(0) = In.
Re µ
Im µ
θ0
• Fold (LPC): µ1 = 1;
• Flip (PD): µ1 = −1;
• Torus (NS): µ1,2 = e±iθ0 .
UTwente – p. 14/58
![Page 25: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/25.jpg)
Simple bifurcation points
Φ(τ) − Tfx(u(τ), α0)Φ(τ) = 0, Φ(0) = In,
Ψ(τ) + TfTx (u(τ), α0)Ψ(τ) = 0, Ψ(0) = In.
• LPC:
(Φ(1)−In)q0 = 0, (Φ(1)−In)q1 = q0, (Ψ(1)−In)p0 = 0, (Ψ(1)−In)p1 =
• PD:
(Φ(1) + In)q2 = 0, (Ψ(1) + In)p2 = 0.
• NS:κ = cos θ0
(Φ(1)− eiθ0In)(q3 + iq4) = 0, (Ψ(1)− e−iθ0In)(p3 + ip4) = 0.
We have(In − 2κΦ(1) + Φ2(1))q3,4 = 0.
UTwente – p. 15/58
![Page 26: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/26.jpg)
Continuation of bifurcations in two parameters
• PD and LPC:(u, T, α) ∈ C1([0, 1],Rn) × R × R2
u(τ) − Tf(u(τ), α) = 0, τ ∈ [0, 1],
u(0) − u(1) = 0,∫ 10 〈
˙u(τ), u(τ)〉 dτ = 0,
G[u, T, α] = 0.
• NS: (u, T, α, κ) ∈ C1([0, 1],Rn) × R × R2 × R
u(τ) − Tf(u(τ), α) = 0, τ ∈ [0, 1],
u(0) − u(1) = 0,∫ 10 〈
˙u(τ), u(τ)〉 dτ = 0,
G11[u, T, α, κ] = 0,
G22[u, T, α, κ] = 0.
UTwente – p. 16/58
![Page 27: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/27.jpg)
PD-continuation
• There existv01, w01 ∈ C0([0, 1],Rn), andw02 ∈ Rn, such that
N1 : C1([0, 1],Rn) × R → C0([0, 1],Rn) × Rn × R,
N1 =
D − Tfx(u, α) w01
δ0 − δ1 w02
Intv010
,
is one-to-one and onto near a simple PD bifurcation point.
• DefineG by solving N1
v
G
=
0
0
1
.
UTwente – p. 17/58
![Page 28: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/28.jpg)
• The BVP for(v,G) can be written in the “classical form”
v(τ) − Tfx(u(τ), α)v(τ) +Gw01(τ) = 0, τ ∈ [0, 1],
v(0) + v(1) +Gw02 = 0,∫ 10 〈v01(τ), v(τ)〉dτ − 1 = 0.
• If G = 0 thenΦ(1) has eigenvalueµ1 = −1.
• One can take
w02 = 0
and
w01(τ) = Ψ(τ)p2, v01(τ) = Φ(τ)q2.
UTwente – p. 18/58
![Page 29: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/29.jpg)
LPC-continuation
• There existv01, w01 ∈ C0([0, 1],Rn), w02 ∈ Rn, and
v02, w03 ∈ R such that
N2 : C1([0, 1],Rn) × R2 → C0([0, 1],Rn) × Rn × R2,
N2 =
D − Tfx(u, α) −f(u, α) w01
δ0 − δ1 0 w02
Intf(u,α) 0 w03
Intv01v02 0
,
is one-to-one and onto near a simple LPC bifurcation point.
• DefineG by solving N2
v
S
G
=
0
0
0
1
.
UTwente – p. 19/58
![Page 30: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/30.jpg)
NS-continuation
• There existv01, v02, w11, w12 ∈ C0([0, 2],Rn), and
w21, w22 ∈ Rn, such that
N3 : C1([0, 2],Rn) × R2 → C0([0, 2],Rn) × Rn × R2,
N3 =
D − Tfx(u, α) w11 w12
δ0 − 2κδ1 + δ2 w21 w22
Intv010 0
Intv020 0
,
is one-to-one and onto near a simple NS bifurcation point.
• DefineGjk by solving N3
r s
G11 G12
G21 G22
=
0 0
0 0
1 0
0 1
.
UTwente – p. 20/58
![Page 31: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/31.jpg)
Remarks on continuation of bifurcations• After discretization via orthogonal collocation, all linear BVPs
for G’s have sparsity structure that is identical to that of the
linearization of the BVP for limit cycles.
• For each defining system holds:Simplicity of the bifurcation +
Transversality⇒ Regularity of the defining BVP.
• Jacobian matrix of each (discretized) defining BVP can be
efficiently computed using adjoint linear BVP.
• Border adaptation using solutions of the adjoint linear BVPs.
• Actually implemented in MatCont, also with compiledC-codes
for the Jacobian matrices.
UTwente – p. 21/58
![Page 32: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/32.jpg)
Periodic normalization on center manifolds• Parameter-dependent periodic normal forms for LPC, PD, and
NS [Iooss, 1988]
• Computation of critical normal form coefficients
UTwente – p. 22/58
![Page 33: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/33.jpg)
Generic LPC-bifurcation
T0-periodic normal form onW cα:
dτ
dt= 1 + ν(α) − ξ + a(α)ξ2 + O(ξ3),
dξ
dt= β(α) + b(α)ξ2 + O(ξ3),
wherea, b ∈ R.
β(α) = 0
W c0
β(α) < 0 β(α) > 0
W cβ
W cβ
C−1
C+2 C0
UTwente – p. 23/58
![Page 34: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/34.jpg)
Generic PD-bifurcation
2T0-periodic normal form onW cα:
dτ
dt= 1 + ν(α) + a(α)ξ2 + O(ξ4),
dξ
dt= β(α)ξ + c(α)ξ3 + O(ξ4),
wherea, c ∈ R.
β(α) < 0
W c0
β(α) = 0 β(α) > 0
W cβ W c
β
C−1 C0 C+
1C−
2
UTwente – p. 24/58
![Page 35: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/35.jpg)
Generic NS-bifurcation
T0-periodic normal form onW cα:
dτ
dt= 1 + ν(α) + a(α)|ξ|2 + O(|ξ|4),
dξ
dt=
(β(α) +
iθ(α)
T (α)
)ξ + d(α)ξ|ξ|2 + O(|ξ|4),
wherea ∈ R, d ∈ C.
β(α) < 0 β(α) = 0 β(α) > 0
C+1C−
1 C0
T2
UTwente – p. 25/58
![Page 36: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/36.jpg)
Critical normal form coefficients
At a codimension-one point write
f(x0(t)+v, α0) = f(x0(t), α0)+A(t)v+1
2B(t; v, v)+
1
6C(t; v, v, v)+O(‖v‖4),
whereA(t) = fx(x0(t), α0) and the components of the multilinear
functionsB andC are given by
Bi(t;u, v) =
n∑
j,k=1
∂2fi(x, α0)
∂xj∂xk
∣∣∣∣x=x0(t)
ujvk
and
Ci(t;u, v, w) =
n∑
j,k,l=1
∂3fi(x, α0)
∂xj∂xk∂xl
∣∣∣∣x=x0(t)
ujvkwl,
for i = 1, 2, . . . , n. These areT0-periodic int.
UTwente – p. 26/58
![Page 37: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/37.jpg)
Fold (LPC): µ1 = 1
• Critical center manifoldW c0 : τ ∈ [0, T0], ξ ∈ R
W c0
C0
ξ
τ
x = x0(τ) + ξv(τ) +H(τ, ξ),
whereH(T0, ξ) = H(0, ξ),
H(τ, ξ) =1
2h2(τ)ξ
2 +O(ξ3)
• Critical periodic normal form onW c0 :
dτ
dt= 1 − ξ + aξ2 + O(ξ3),
dξ
dt= bξ2 + O(ξ3),
wherea, b ∈ R, while theO(ξ3)-terms areT0-periodic inτ .
UTwente – p. 27/58
![Page 38: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/38.jpg)
LPC: Eigenfunctions
v(τ) −A(τ)v(τ) − f(x0(τ), α0) = 0, τ ∈ [0, T0],
v(0) − v(T0) = 0,∫ T0
0 〈v(τ), f(x0(τ), α0)〉dτ = 0,
implying ∫ T0
0〈ϕ∗(τ), f(x0(τ), α0)〉 dτ = 0,
whereϕ∗ satisfies
ϕ∗(τ) +AT(τ)ϕ∗(τ) = 0, τ ∈ [0, T0],
ϕ∗(0) − ϕ∗(T0) = 0,∫ T0
0 〈ϕ∗(τ), v(τ)〉dτ − 1 = 0.
UTwente – p. 28/58
![Page 39: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/39.jpg)
LPC: Computation of b
• Substitute intodx
dt=∂x
∂ξ
dξ
dt+∂x
∂τ
dτ
dt
• Collect
ξ0 : x0 = f(x0, α0),
ξ1 : v −A(τ)v = x0,
ξ2 : h2 −A(τ)h2 = B(τ ; v, v) − 2af(x0, α0) + 2v − 2bv.
• Fredholm solvability condition
b =1
2
∫ T0
0〈ϕ∗(τ), B(τ ; v(τ), v(τ)) + 2A(τ)v(τ)〉 dτ.
UTwente – p. 29/58
![Page 40: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/40.jpg)
Remarks on numerical periodic normalization
• Only the derivatives off(x, α0) are used, not those of the
Poincaré mapP(y, α0).
• Detection of codim 2 points is easy.
• After discretization via orthogonal collocation, all linear BVPs
involved have the standard sparsity structure.
• One can re-use solutions to linear BVPs appearing in the
continuation to compute the normal form coefficients.
• Actually implemented in MatCont for LPC, PD, and NS.
UTwente – p. 30/58
![Page 41: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/41.jpg)
Example: Oscillations in peroxidase-oxidase reaction
• 2Y H2 +O2 + 2H+ → 2Y H+ + 2H2O
A+B +Xk1→ 2X, Y
k5→ Q,
2Xk2→ 2Y, X0
k6→ X,
A+B + Yk3→ 2X, A0
k7↔ A,
Xk4→ P, B0
k8→ B
UTwente – p. 31/58
![Page 42: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/42.jpg)
Example: Oscillations in peroxidase-oxidase reaction
• 2Y H2 +O2 + 2H+ → 2Y H+ + 2H2O
A+B +Xk1→ 2X, Y
k5→ Q,
2Xk2→ 2Y, X0
k6→ X,
A+B + Yk3→ 2X, A0
k7↔ A,
Xk4→ P, B0
k8→ B
• Steinmetz & Larter (1991):
A = −k1ABX − k3ABY + k7 − k−7A,
B = −k1ABX − k3ABY + k8,
X = k1ABX − 2k2X2 + 2k3ABY − k4X + k6,
Y = −k3ABY + 2k2X2 − k5Y.
UTwente – p. 31/58
![Page 43: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/43.jpg)
MatCont
ACM Trans. Math. Software29 (2003), 141 - 164UTwente – p. 32/58
![Page 44: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/44.jpg)
Bifurcation curves
0 1 2 3 4 5 6 70.3
0.4
0.5
0.6
0.7
0.8
0.9
1
k7
k8
GH
GH
LPC
NS
H
ACM Trans. Math. Software24 (1998), 418-436
SIAM J. Numer. Anal.38 (2000), 329-346
SIAM J. Sci. Comp.27 (2005), 231-252
UTwente – p. 33/58
![Page 45: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/45.jpg)
Bifircation curves (zoom)
1 1.5 2 2.50.55
0.6
0.65
0.7
0.75
0.8
0.85
0.9
0.95
k8
LPC
NS
1 : 1
k7
1 : 1
H
SIAM J. Numer. Anal.41 (2003), 401-435
SIAM J. Numer. Anal.43 (2005), 1407-1435
Physica D237(2008), 3061-3068
UTwente – p. 34/58
![Page 46: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/46.jpg)
4. Bifurcations of homoclinic orbits• Consider a family of smooth ODEs
x = f(x, α), x ∈ Rn, α ∈ R
m,
having a hyperbolic equilibriumx0 with eigenvalues
ℜ(µnS) ≤ ... ≤ ℜ(µ1) < 0 < ℜ(λ1) ≤ ... ≤ ℜ(λnU
)
of A(x0, α) = fx(x0, α).
• Homoclinic problem:
f(x0, α) = 0,
x(t) − f(x(t), α) = 0,
limt→±∞
x(t) − x0 = 0, t ∈ R,∫
∞
−∞
˙x(t)T(x(t) − x(t))dt = 0,
wherex is a reference solution.UTwente – p. 35/58
![Page 47: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/47.jpg)
Homoclinic orbits
x(t)
x(0)
x(t)
Γ
WU (x0)
WS(x0) x0
Tx(t)WS(x0)
Tx(t)WU (x0)
Homoclinic orbits to hyperbolic equilibria have codim 1.
UTwente – p. 36/58
![Page 48: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/48.jpg)
Defining BVP [Beyn, 1993; Doedel & Friedman 1994]
• Truncate withprojection boundary conditions:
f(x0, α) = 0,
x(t) − f(x(t), α) = 0, t ∈ [−T, T ]
〈x(−T ) − x0, q0,nU+i〉 = 0, i = 1, 2, . . . , nS
〈x(+T ) − x0, q1,nS+i〉 = 0, i = 1, 2, . . . , nU∫ T
−T
˙x(t)T(x(t) − x(t))dt = 0,
where the columns ofQU⊥ = [q0,nU+1, . . . , q0,nU+nS] and
QU⊥ = [q1,nS+1, . . . , q1,nS+nU] span the orthogonal
complements toTx0WU (x0) andTx0
WS(x0), resp.
UTwente – p. 37/58
![Page 49: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/49.jpg)
Smooth Schur Block Factorization
A(s) = Q(s)
R11(s) R12(s)
0 R22(s)
QT(s) ∈ R
n×n,
whereQ(s) = [Q1(s) Q2(s)] such that
• Q(s) is orthogonal, i.e.QT(s)Q(s) = In;
• the columns ofQ1(s) ∈ Rn×m span an eigenspaceE(s) of A(s);
• the columns ofQ2(s) ∈ Rn×(n−m) spanE⊥(s);
• eigenvalues ofR11 are the eigenvalues ofA(s) corresponding to
E(s);
• Q(s) andRij(s) have the same smoothness asA(s).
Then holds theinvariant subspace relation:
QT2 (s)A(s)Q1(s) = 0.
UTwente – p. 38/58
![Page 50: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/50.jpg)
CIS-algorithm [Dieci & Friedman, 2001]
• Define T11(s) T12(s)
T21(s) T22(s)
= QT(0)A(s)Q(0)
for small |s|, whereT11(s) ∈ Rm×m.
• ComputeY ∈ R(n−m)×m from theRiccati matrix equation
Y T11(s) − T22(s)Y + Y T12(s)Y = T21(s).
• ThenQ(s) = Q(0)U(s) whereU(s) = [U1(s) U2(s)] with
U1(s) =
Im
Y
(In−m+Y TY )−
1
2 , U2(s) =
−Y T
In−m
(In−m+Y Y T)−
1
2 ,
UTwente – p. 39/58
![Page 51: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/51.jpg)
• The columns of
Q1(s) = Q(0)U1(s) and Q2(s) = Q(0)U2(s)
form orthogonalbases inE(s) andE⊥(s).
• The columns of
Q(0)
Im
Y (s)
and Q(0)
−Y (s)T
In−m
form bases inE(s) andE⊥(s), which are in general
non-orthogonal.
UTwente – p. 40/58
![Page 52: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/52.jpg)
Continuation of homoclinic orbits in MatCont
x(t) − 2Tf(x(t), α) = 0,
f(x0, α) = 0,∫ 1
0˙x(t)T(x(t) − x(t))dt = 0,
〈x(0) − x0, q0,nU+i〉 = 0, i = 1, 2, . . . , nS
〈x(1) − x0, q1,nS+i〉 = 0, i = 1, 2, . . . , nU
T22UYU − YUT11U + T21U − YUT12UYU = 0,
T22SYS − YST11S + T21S − YST12SYS = 0,
‖x(0) − x0‖ − ǫ0 = 0,
‖x(1) − x0‖ − ǫ1 = 0,
[q0,nU+1 q0,nU+2 · · · q0,nU+nS] = QU (0)
−Y T
U
InS
[q1,nS+1 q1,nS+2 · · · q1,nS+nU] = QS(0)
−Y T
S
InU
.
UTwente – p. 41/58
![Page 53: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/53.jpg)
Example: Complex nerve pulses
• The slow subsystem of the Hodgkin-Huxley PDEs is approximated by
the FitHugh-Nagumo [1962] system
ut = uxx − u(u− a)(u− 1) − v,
vt = bu,
where0 < a < 1, b > 0.
UTwente – p. 42/58
![Page 54: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/54.jpg)
Example: Complex nerve pulses
• The slow subsystem of the Hodgkin-Huxley PDEs is approximated by
the FitHugh-Nagumo [1962] system
ut = uxx − u(u− a)(u− 1) − v,
vt = bu,
where0 < a < 1, b > 0.
• Traveling wavesu(x, t) = U(ξ), v(x, t) = V (ξ) with ξ = x+ ct satisfy
U = W,
W = cW + U(U − a)(U − 1) + V,
V =b
cU,
wherec is the propagation speed.
UTwente – p. 42/58
![Page 55: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/55.jpg)
Homoclinic orbits define traveling impulses
(a)
(b)
c
c
UTwente – p. 43/58
![Page 56: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/56.jpg)
Homoclinic bifurcation curves (HomCont/MatCont)
Db3Db2
Db1
c
a
P(1)b1
P(1)b2
P(1)b3
Int. J. Bifurcation & Chaos4 (1994), 795-822
UTwente – p. 44/58
![Page 57: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/57.jpg)
Double pulses
c
P(1)b
P(2)b,j
A1
A2Db
a
c
Selecta Math. Sovetica13 (1994), 128-142
SIAM J. Appl. Math.62 (2001), 462-487
UTwente – p. 45/58
![Page 58: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/58.jpg)
5. Open problems
• Computation of normal forms for bifurcations of equilibriain
ODEs with delays.
• Bifurcation analysis of spatially-distributed systems with delays,
e.g. neural fields.
• Computing codim 2 periodic normal forms for limit cycles.
• Location and continuation of homoclinic and heteroclinic orbits
to limit cycles.
• Starting homoclinic codim 1 bifurcations from codim 2 points.
• ...
UTwente – p. 46/58
![Page 59: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/59.jpg)
Example: Delay-differential equations
• Model of two interacting layers of neurons [Visser et al., 2010]:
x1(t) = −x1(t) − aG(bx1(t− τ1)) + cG(dx2(t− τ2)),
x2(t) = −x2(t) − aG(bx2(t− τ1)) + cG(dx1(t− τ2)),
whereG(x) = (tanh(x− 1) + tanh(1))cosh2(1) andxj is the
population averaged neural activity in layerj = 1, 2.
• For b = 2, d = 1.2, τ1 = 12.99, τ2 = 20.15 there is adouble
Hopf (HH) bifurcation at
(abG′(0), cdG′(0)) = (0.559667, 0.688876)
that gives rise to a stable quasi-periodic behaviour with two base
frequences (2-torus).
UTwente – p. 47/58
![Page 60: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/60.jpg)
Example: Neural field model
The two-population neural network model:
1
αE
∂uE(x, t)
∂t= −uE(x, t)
+∫∞
−∞wEE(y)fE(uE(x− y, t−
|y|
vE)) dy
−∫∞
−∞wEI(y)fI(uI(x− y, t−
|y|
vI)) dy,
1
αI
∂uI(x, t)
∂t= −uI(x, t)
+∫∞
−∞wIE(y)fE(uE(x− y, t−
|y|
vE)) dy
−∫∞
−∞wII(y)fI(uI(x− y, t−
|y|
vI)) dy.
Turing instabilities and pattern formation, cf. [Blomquist et al., 2005;
Venkov & Coombes, 2007; Wyller et al., 2007]. Partial results but no
systematic bifurcation analysis.
UTwente – p. 48/58
![Page 61: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/61.jpg)
Cusp bifurcation of limit cycles (codim 2)
• Critical center manifoldW c0 : τ ∈ [0, T0], ξ ∈ R
x = x0(τ) + ξv(τ) +H(τ, ξ),
whereH(T0, ξ) = H(0, ξ),
H(τ, ξ) =1
2h2(τ)ξ
2 +O(ξ3)
• Critical periodic normal form onW c0 :
dτ
dt= 1 − ξ + a1ξ
2 + a2ξ3 + O(ξ4),
dξ
dt= eξ3 + O(ξ4),
wherea1,2, e ∈ R, while theO(ξ3)-terms areT0-periodic inτ .
UTwente – p. 49/58
![Page 62: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/62.jpg)
Example: Swallow-tail bifurcation in Lorenz-84
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 210
11
12
13
14
15
16
17
18
19
20
GG
FF
CPC
CPC
CPC
CPC
CPC
CPC
CPC
CPC
x = −y2 − z2 − ax+ aF,
y = xy − bxz − y +G,
z = bxy + xz − z.
a = 0.25, b ∈ [2.95, 4.0].
UTwente – p. 50/58
![Page 63: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/63.jpg)
Computation of cycle-to-cycle connecting orbits in 3D
x+
x+0
u(1)x−
x−0
u(0)
O−
w−(0)
w−
w+(0)
w+
f−0
O+
u
f+0
• ODEs for both cycles, their (adjoint) eigenfunctions, and the
connection;
• Projection boundary conditions in orthogonal planes at base
points.
UTwente – p. 51/58
![Page 64: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/64.jpg)
Example: Bifurcations and chaos in ecology
• The tri-trophic food chain model [Hogeweg & Hesper, 1978]:
x1 = rx1
(1 −
x1
K
)−
a1x1x2
1 + b1x1,
x2 = e1a1x1x2
1 + b1x1−
a2x2x3
1 + b2x2− d1x2,
x3 = e2a2x2x3
1 + b2x2− d2x3,
where
x1 prey biomass
x2 predator biomass
x3 super−predator biomass
UTwente – p. 52/58
![Page 65: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/65.jpg)
Global bifurcation diagram
SIAM J. Appl. Math.62 (2001), 462-487
UTwente – p. 53/58
![Page 66: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/66.jpg)
Local bifurcations
B T
MTC T
TCeHp
�TCeTe� H+
H�d1
d 2
0.90.70.50.30.10.020.0160.0120.008
Math. Biosciences134(1996), 1-33
UTwente – p. 54/58
![Page 67: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/67.jpg)
Local and key global bifurcations
T G=eF 2G= G6= F 1T TC T
� H�d1
d 2
0.50.40.30.20.10.0160.0140.0120.010.008
Int. J. Bifurcation & Chaos18 (2008), 1889-1903
Int. J. Bifurcation & Chaos19 (2009), 159-169
UTwente – p. 55/58
![Page 68: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/68.jpg)
Trends• Larger dimensions and codimensions;
UTwente – p. 56/58
![Page 69: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/69.jpg)
Trends• Larger dimensions and codimensions;
• Semi-local and global phenomena in phase and parameter spaces;
UTwente – p. 56/58
![Page 70: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/70.jpg)
Trends• Larger dimensions and codimensions;
• Semi-local and global phenomena in phase and parameter spaces;
• Non-standard dynamical models (non-smooth, hybrid, constrained,
spatially-distributted delays, etc.);
UTwente – p. 56/58
![Page 71: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/71.jpg)
Trends• Larger dimensions and codimensions;
• Semi-local and global phenomena in phase and parameter spaces;
• Non-standard dynamical models (non-smooth, hybrid, constrained,
spatially-distributted delays, etc.);
• Implication of connection topology on network dynamics;
UTwente – p. 56/58
![Page 72: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/72.jpg)
6. References
UTwente – p. 57/58
![Page 73: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/73.jpg)
Book projects
• Yu.A. Kuznetsov, O. Diekmann, W.-J. Beyn.Dynamical Systems
Essentials: An application oriented introduction to ideas, concepts,
examples, methods, and results. Springer (polishing stage).
UTwente – p. 58/58
![Page 74: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/74.jpg)
Book projects
• Yu.A. Kuznetsov, O. Diekmann, W.-J. Beyn.Dynamical Systems
Essentials: An application oriented introduction to ideas, concepts,
examples, methods, and results. Springer (polishing stage).
• W. Govaerts and Yu.A. Kuznetsov.Bifurcation Analysis of Dynamical
Systems in MATLAB. Springer (advanced stage).
UTwente – p. 58/58
![Page 75: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/75.jpg)
Book projects
• Yu.A. Kuznetsov, O. Diekmann, W.-J. Beyn.Dynamical Systems
Essentials: An application oriented introduction to ideas, concepts,
examples, methods, and results. Springer (polishing stage).
• W. Govaerts and Yu.A. Kuznetsov.Bifurcation Analysis of Dynamical
Systems in MATLAB. Springer (advanced stage).
• Yu.A. Kuznetsov and H.G.E. MeijerCodimension Two Bifurcations of
Iterated Maps. Cambridge University Press.
UTwente – p. 58/58
![Page 76: Numerical bifurcation analysis of dynamical systems: Recent](https://reader030.vdocuments.mx/reader030/viewer/2022021307/62074f8649d709492c302747/html5/thumbnails/76.jpg)
Book projects
• Yu.A. Kuznetsov, O. Diekmann, W.-J. Beyn.Dynamical Systems
Essentials: An application oriented introduction to ideas, concepts,
examples, methods, and results. Springer (polishing stage).
• W. Govaerts and Yu.A. Kuznetsov.Bifurcation Analysis of Dynamical
Systems in MATLAB. Springer (advanced stage).
• Yu.A. Kuznetsov and H.G.E. MeijerCodimension Two Bifurcations of
Iterated Maps. Cambridge University Press.
• A.R. Champneys and Yu.A. Kuznetsov.Homoclinic Connections:
Localized phenomena in science and engineering. Springer.
UTwente – p. 58/58