analytical methods for engineering - slides of the course (lessons 1-8)
DESCRIPTION
An introduction to Elliptical EquationsTRANSCRIPT
Mathematical Methods in the Engineering Sciences
Elena Beretta
Dipartimento di MatematicaPolitecnico di Milano (ITALY)
1 / 82
Main goals of the course
Many constitutive laws of physics are governed by partial differentialequations wThe qualitative study of solutions to these equations from amathematical viewpoint give an insight of the physical processes orsystems. In general additional information is necessary to select orpredict a unique solution. This information is often supplied in the formof initial and/or boundary data wWe will analyze the following fundamental equations: Laplaceequation,Poisson equation, Heat equation, Schroedinger equationanalyzing some boundary value problems and Cauchy problems. Inparticular we will investigate well-posedness (existence, uniqueness,stability) of such problems using several tools of functional analysis.Finally we will study two mathematical models, one arising in medicalimaging (Optical Tomography) and the second in quantum mechanics(Equilibrium of atoms)
2 / 82
Part I
Prerequisits
3 / 82
Banach Spaces
X vector space on R
A norm on X is a function ‖ · ‖ : X → [0,+∞) s.t.‖x‖ ≥ 0 and ‖x‖ = 0 if and only if x = 0 (positivity)‖λx‖ = |λ|‖x‖, ∀x ∈ X , ∀λ ∈ R (homogeneity)‖x + y‖ ≤ ‖x‖+ ‖y‖, ∀x , y ∈ X (triangular inequality)
(X , ‖ · ‖) is a normed space
(X , ‖ · ‖) is a metric space (X ,d) w.r.t. d induced byd(x , y) = ‖x − y‖, ∀x , y ∈ X
xn → x∗ in X if ‖xn − x∗‖ → 0 as n→ +∞(strong convergence)
A Banach space is a complete normed space(any Cauchy sequence is convergent in X )
4 / 82
Examples
(Rn, ‖ · ‖n) and (Cn, ‖ · ‖n) are Banach spaces with respect to
‖x‖ =√∑n
i=1 x2i where x = (x1, . . . , xn).
C0(K ) = u : Ω ⊂ Rn → R continuous on K, K is compact‖u‖C0(Ω) = ‖u‖∞ = sup
x∈Ω
|u(x)|,
(C0(K ), ‖ · ‖∞) is a Banach space
Lp(Ω) = u :(∫
Ω |u(x)|pdx)1/p
< +∞, 1 ≤ p < +∞
‖u‖p =(∫
Ω |u(x)|pdx)1/p
(Lp(Ω), ‖ · ‖p) is a Banach space
5 / 82
Examples
xk = (x1, x2, . . . ), xk ∈ R1 ≤ p < +∞,
`p =
x = xk :
∞∑k=1
|xk |p < +∞
(`p, ‖ · ‖) is a Banach space with ‖x‖p = (∑+∞
k=1 |xk |p)1/p
6 / 82
Hilbert Spaces
X vector space on R
An inner product on X is a function (·, ·) : X × X → R s.t.(x , x) ≥ 0, ∀x ∈ X , (x , x) = 0 iff x = 0 (positivity)(y , x) = (x , y), ∀x , y ∈ X (symmetry)(λx + µy , z) = λ(x , z) + µ(y , z), ∀x , y , z ∈ X , ∀λ, µ ∈ R (bilinearity)
(X , (·, ·)) is an inner product space
If H is a vector space on C then symmetry is substituted by(x , y) = (y , x)
7 / 82
Hilbert Spaces
(X , (·, ·)) is a normed space (X , ‖ · ‖) w.r.t. ‖ · ‖ induced by‖x‖ = (x , x)1/2, ∀x ∈ X
|(x , y)| ≤ ‖x‖‖y‖, ∀x , y ∈ X Cauchy-Schwarz inequality
‖x + y‖2 + ‖x − y‖2 = 2‖x‖2 + 2‖y‖2 Parallelogram law
A Hilbert space is a complete inner product space
8 / 82
Examples of Hilbert Spaces
Rn (x , y) =∑n
i=1 xiyi ,Cn (x , y) =
∑ni=1 xiy i
L2(Ω) is a Hilbert space(u, v) =
∫Ω u(x)v(x)dx , (u, v) =
∫Ω u(x)v(x)dx
`2 is a Hilbert space
(x,y) =∞∑
k=1
xkyk
∞∑k=1
xkyk
9 / 82
Orthogonal projections
V ⊂ H is closed in H if it contains all limit points of sequences inV .
V⊥ = u ∈ H : (u, v) = 0, ∀ v ∈ Vorthogonal complement of V ⊂ H, H Hilbert space
(Projection Theorem) (with proof) If V is a closed subspace of Hthen ∃! decompositionx = u + v , u ∈ V , v ∈ V⊥, ∀ x ∈ H
PV x = u orthogonal projection of x onto V
QV x = v orthogonal projection of x onto V⊥
‖x‖2 = ‖PV x‖2 + ‖x − PV x‖2, ‖PV x‖ ≤ ‖x‖
10 / 82
Hilbert Spaces: separability
H Hilbert space, Y ⊂ H is dense if ∀ x ∈ H ∃ yn ⊂ Y :yn → x
A Hilbert space is separable if there exists a countable and denseY ⊂ X .
Examples of separable Hilbert spaces:
(Rn, ‖ · ‖n) and (Cn, ‖ · ‖n)
(L2(Ω), ‖ · ‖2)
(`2, ‖ · ‖2)
11 / 82
Bases in Hilbert Spaces
ej : (ei ,ej) = δij orthonormal (countable) set in H
ej, orthonormal set, is a (countable) basis for H if
x =∞∑
j=1
(x ,ej)ej , ∀ x ∈ H
Generalized Fourier Series
cj = (x ,ej)
Fourier coefficients
‖x‖2 =∞∑
j=1
c2j
(follows from Pythagora’s Theorem)
H is separable iff H has a countable basis12 / 82
Examples of orthonormal bases
Rn, Cn
e1 = (1,0, . . . ,0), . . . ,e2 = (0,1,0 . . . ,0) . . .
L2(0,2π)e0 = 1√
2π,e1 = 1√
πcos x ,e2 = 1√
πsin x ,e3 = 1√
πcos 2x . . .
`2
e1 = (1,0, . . . ), e2 = (0,1, . . . ) . . .
13 / 82
Linear operators on Hilbert spaces
L : (H1, ‖ · ‖H1)→ (H2, ‖ · ‖H2)
L is linear if L(λx + µz) = λLx + µLz, ∀ x , z ∈ H1
Kernel of L (N (L))
N (L) = x ∈ H1 : Lx = 0
Range of L (R(L))
R(L) = y ∈ H2 : ∃x ∈ H1 : Lx = y
L is bounded if ∃M > 0 : ‖Lx‖Y ≤ M‖x‖H1 , ∀ x ∈ H1
L is continuous if xn → x in H1 ⇒ Lxn → Lx in H2
14 / 82
Linear operators on Hilbert spaces
L(H1,H2) set of all bounded linear operator from X to Y
L(H1,H2) vector space
‖L‖L(H1,H2) = supx 6=0
‖Lx‖H2
‖x‖H1
= sup‖x‖H1
=1‖Lx‖H2
‖Lx‖H2 ≤ ‖L‖L(H1,H2)‖x‖H1 , ∀ x ∈ H1
15 / 82
Examples of linear bounded operators
L : Rn → Rm
Lx = Ax
where A is an m × n real valued matrix.
V ⊂ H, closed subset of a Hilbert space H.L1 : H → H with L1x = PV x and L2 : H → H with L1x = QV x
16 / 82
Results on linear operators
TheoremL(H1,H2) is a Banach space
TheoremL : (H1, ‖ · ‖H1)→ (H2, ‖ · ‖H2), L linear.L is continuous ⇔ L is bounded.
(with proof)
17 / 82
Linear functionals and dual spaces
L : (H, ‖ · ‖H)→ R bounded and linear: functional on H
L(H,R) is the dual space of H: H∗ or H ′
H∗ = L(H,R), ‖L‖H∗ = ‖f‖L(H,R) = supx 6=0
|Lx |‖x‖H
(H∗, ‖ · ‖H∗) is a Banach space (R is a Banach space)
H∗〈·, ·〉H : H∗ × H → R H∗〈L, x〉H = Lx is a bilinear form
H∗〈Lx〉H or 〈L, x〉 (duality)
|〈L, x〉| ≤ ‖L‖H∗‖x‖H
18 / 82
Example of linear, bounded functionals on Hilbertspaces
H = Rn, L : Rn → R, let a ∈ Rn,
Lx = (a, x)
H = L2(Ω), Ω ⊂ Rn, L : L2(Ω)→ R, fix g ∈ L2(Ω)
Lf =
∫Ω
fg dx
Fix y ∈ H and let L : H → R
Lx = (x , y)
19 / 82
Dual spaces of Hilbert spaces
TheoremRiesz representation TheoremLet H be a real Hilbert space. Then, ∀L ∈ H∗ ⇒ ∃! yL ∈ H :Lx = (x , yL),∀ x ∈ H and ‖L‖H∗ = ‖yL‖H
with proof
RemarkRiesz Theorem ⇒ we identify H and H∗ endowing H∗ with theinner product (L1,L2) = (yL1 , yL2).
Exercise: Prove that the kernel of a linear bounded operator is aclosed vector space.Exercise: Show that the Representation Theorem holds also incomplex Hilbert spaces.
20 / 82
Bilinear Forms
DefinitionLet V be a real vector space. a : V × V → R is a bilinear form on V if∀y ∈ V the function x → a(x , y) is linear∀x ∈ V the function y → a(x , y) is linear
ExampleThe inner product on a real vector space is a bilinear form
In complex inner product spaces we define sesquilinear formssubstituting the linearity with respect to the second component withantilinearity
21 / 82
Lax-Milgram Theorem
Problem (P):Let F ∈ H∗. Find u ∈ H such that
a(u, v) = 〈F , v〉, ∀v ∈ H
22 / 82
TheoremLax-Milgram Theorem Let H be a real Hilbert space, a be a bilinearform such that
a is continuous i.e.∃M > 0:
|a(u, v)| ≤ M‖u‖‖v‖, ∀u, v ∈ H
a is coercive i.e. ∃α > 0:
a(v , v) ≥ α‖v‖2, ∀v ∈ H.
Then, there exists a unique solution u to P and
‖u‖ ≤ 1α‖F‖
23 / 82
RemarkIf ∃M > 0:
|a(u, v)| ≤ M‖u‖‖v‖, ∀u, v ∈ H
then a(u, v) is continuous in H × H
RemarkLax-Milgram Theorem still holds in complex Hilbert spaces forsequilinear forms which are continuous and coercive.
24 / 82
Minimization Problem
If the bilinear form a is symmetric then Problem (P) is equivalent to
minv∈H
E(v) =12
a(v , v)− 〈F , v〉
If the bilinear form a is symmetric ( antisymmetric in the complexcase)
(u, v)H = a(u, v)
is an inner product and Lax Milgram Theorem derives from RieszTheorem.Exercise: check that this is true.
24 / 82
Galerkin Approximation Method
H Hilbert space, H = ∪Vk , Vk finite dimensional space for all k .Key ideas:
Project Problem (P)
a(u, v) = 〈F , v〉, v ∈ H
on Vk i.e. find solution uk of Problem (Pk )
a(uk , v) = 〈F , v〉, v ∈ Vk
Show‖uk − u‖H → 0
as k → +∞.
25 / 82
Galerkin Approximation Method
PropositionAssume the assumptions of the Lax-Milgram Theorem are satisfiedand let u be the unique solution to Problem (P). If uk is the uniquesolution of Problem (Pk ) then
‖u − uk‖ ≤Mα
infv∈Vk‖u − v‖
Theorem
uk → u in V as k → +∞
with proof
26 / 82
Weak and strong convergence
DefinitionH Hilbert space, a sequence xn of elements in H converges stronglyto an element x ∈ H
xn → x
if‖xn − x‖H → 0
as n→ +∞
DefinitionLet H be a Hilbert space, xn, x ∈ H.xn x (weak convergence) if (xn, y)→ (x , y) for any y ∈ H.
27 / 82
Strong and weak convergence
TheoremH Hilbert space, xn, x ∈ H.xn → x (strong convergence) ⇒ xn x (weak convergence)
Proof.If xn → x strongly then
|(xn, y)−(x , y)| = |(xn−x , y)| ≤ ‖xn−x‖‖y‖ → 0, ∀y ∈ H ⇒ xn x
Exercise: Prove that if H finite dimensional Hilbert space, xn, x ∈ H.xn → x ⇔ xn x
28 / 82
Counterexample
Example
H = `2 y ∈ `2
y ∈ `2 ⇒∞∑
k=1
y2k < +∞ ⇒ yk → 0
ek orthonormal basis
(ek ,y) = yk → 0, ∀y ∈ `2 ⇒ ek 0
ek 9 0 strongly in `2
(ek is not a Cauchy sequence: ‖ek − ej‖`2 =√
2, k 6= j )
29 / 82
Uniqueness of weak limit
TheoremH Hilbert space, xn ∈ H.If xn x ∈ H then x is unique
Proof.Assume xn x . Then
(x − x , y) = (x − xn, y)− (x − xn, y)→ 0, ∀y ∈ H
Hence(x − x , y) = 0, ∀y ∈ H
which impliesx = x .
30 / 82
Boundedness of weak convergent sequences
TheoremH Hilbert space, xn ∈ H.If xn x ∈ H then xn is bounded
31 / 82
An estimate of the norm of the weak limit
TheoremH Hilbert space, xn ∈ H.If xn x ∈ H then ‖x‖ ≤ lim infn→∞‖xn‖
Proof.xn x ⇒
‖x‖2 = (x , x) = limn→∞(xn, x) = lim infn→∞(xn, x)
≤ lim infn→∞‖xn‖‖x‖ = ‖x‖lim infn→∞‖xn‖
32 / 82
Compactness
E ⊂ X is compact if every open cover of E contains a finite subcover
E ⊂ X is compact if and only if every bounded sequence in E containsa convergent subsequence in E (sequentially compact ).
33 / 82
Weak convergence and compactness Theorems
TheoremA sequentially compact subset E of a normed space is closed andbounded.
Exercise: Prove the theorem.
RemarkIf E is a subset of a finite dimensional normed space (Rn) then
E sequentially compact ⇐⇒ Eclosed and bounded
In particularB1 = x ∈ Rn : ‖x‖ ≤ 1
is compact.
TheoremA normed space is finite dimensional ⇐⇒ B1 is compact.
34 / 82
Weak convergence and compactness Theorems
ExampleH infinite dimensional separable Hilbert space. and ej∞1 anothonormal basis. Then
‖ej‖H = 1 ∀j
so that ej ⊂ B1 but
‖en − em‖H =√
2 ∀n,m, n 6= m
so ej∞1 is not a Cauchy sequence.
TheoremAny bounded sequence in a Hilbert space H has a weak convergentsubsequence in H.
35 / 82
Distributions
DefinitionGiven a continuous function ψ : Ω ⊂ Rn → R the support of ψ is the set
supp ψ = x ∈ Ω : ψ(x) 6= 0
DefinitionWe indicate by C∞0 (Ω) the set of functions in C∞(Ω) compactlysupported in Ω
36 / 82
Distributions
We now endow C∞0 (Ω) with a suitable notion of convergence (notinducing any metric)
Definitionψk∞k=1 ⊂ C∞0 (Ω) , ψ ∈ C∞0 (Ω).
ψk → ψ in C∞0 (Ω) as k →∞
ifDαψk → Dαψ, ∀α uniformly in Ω
and there exists a compact set K ⊂ Ω which contains the support of allthe ψk ’s.
37 / 82
Distributions
ExampleLet ak ⊂ R, ak → 0 as k →∞, let ψ ∈ C∞0 (Ω). Then
ψk = akψ → 0, k →∞
in the above sense.
38 / 82
Distributions
C∞0 (Ω) endowed with this convergence will be denoted as D(Ω).
DefinitionA distribution f is a linear continuous functional on D(Ω) (f ∈ D′(Ω))i.e. such that
< f , αψ1 + βψ2 >= α < f , ψ1 > +β < f , ψ2 >, ∀ψ1, ψ2 ∈D(Ω), α, β ∈ R< f , ψk >→< f , ψ > as ψk → ψ in D(Ω) as k → +∞.
39 / 82
Distributions
Examplesf ∈ C0(Ω) defines a distribution
< f , ψ >=
∫Ω
fψ
since|∫
Ωfψ| ≤ ‖f‖K sup
Ω|ψ|
f ∈ L2(Ω) defines a distribution
< f , ψ >=
∫Ω
fψ
since|∫
Ωfψ| ≤ ‖f‖L2(Ω)‖ψ‖L2(Ω)
40 / 82
Distributions
Examplesf = δ0 is a distribution
< f , ψ >= ψ(0)
since|ψ(0)| ≤ sup
Ω|ψ|
f ∈ L1loc(Ω) is a distribution
< f , ψ >=
∫Ω
fψ
since|∫
Ωfψ| ≤ ‖f‖L1(K ) sup
Ω|ψ|
Exercise: Verify that for p < n the function f (x) = 1|x |p ∈ D
′(Rn)
41 / 82
Distributions
TheoremThe map I : L2(Ω)→ D′(Ω) where
< I(f ), ψ >=
∫Ω
fψ dx
is injective
with proof
42 / 82
Distributional derivatives
Let f ∈ C1(R) with and ψ ∈ D(R). Then∫R
f ′ψ dx = fψ|+∞−∞ −∫R
fψ′ dx = −∫R
fψ′ dx
Hence< f ′, ψ >= − < f , ψ′ >
DefinitionLet f ∈ D′(Ω) where Ω ⊂ Rn. The derivative ∂xi f is the distributiondefined by
< ∂xi f , ψ >= − < f , ∂xiψ >
Every distribution possesses (distributional) derivatives of any order
< Dαf , ψ >= (−1)|α| < f ,Dαψ > ∀α
43 / 82
Examples of distributional derivatives
Examplesf = δ0
< ∂xi δ0, ψ >= −∂xiψ(0)
f = H (Heaviside function)
< H, ψ >= − < H, ψ >=
∫RHψ′ dx = −
∫ +∞
0ψ′ dx = ψ(0)
Hence< H′, ψ >= ψ(0) =< δ, ψ >
Exercise: Compute the derivative of f (x) = |x | ∈ D(R)
44 / 82
Some important Sobolev spaces: H1(Ω)
DefinitionWe denote by H1(Ω)
H1(Ω) = v ∈ L2(Ω) : ∇v ∈ L2(Ω)
i.e.< ∂xi v , ψ >= −
∫Ω
v∂xiψ dx , ∀ψ ∈ D′(Ω)
TheoremH1(Ω) is a Hilbert space with inner product
(u, v)1 =
∫Ω
uv dx +
∫Ω∇u · ∇v dx
and
‖u‖1 =
√∫Ω|v |2 dx +
∫Ω|∇v |2 dx
45 / 82
Some important Sobolev spaces: H10 (Ω)
DefinitionWe denote by H1
0 (Ω) the closure of D(Ω) in H1(Ω).
H10 (Ω) ⊂ H1(Ω) and elements of H1
0 (Ω) have zero trace on ∂Ω
TheoremPoincaré inequality. Let Ω ⊂ Rn. There exists a positive constant CPsuch that
‖u‖0 ≤ CP‖∇u‖0
We can choose(u, v)1 = (∇u,∇v)0
as inner product
46 / 82
Some important Sobolev spaces: H−1(Ω)
DefinitionWe denote by H−1(Ω) the dual of H1
0 (Ω) with
‖F‖−1 = sup|Fv | : v ∈ H10 (Ω), ‖v‖1 ≤ 1
TheoremH−1(Ω) is the set of distributions of the form
F = f0 + div f
where f0 ∈ L2(Ω) and f ∈ L2(Ω)
47 / 82
Part II
Elliptic and parabolic PDE’s
48 / 82
Elliptic equations
Laplace equation
∆u = 0
Stationary diffusion equation, conductivity equation, equilibrium of anelastic membrane
Poisson equation
∆u = f
f external force or source e.g. density of electric charges, load
49 / 82
Harmonic functions
DefinitionA solution of
∆u = 0
is called harmonic.
Examplesz = ρeiθ, <zm and =zm, m ∈ N are harmonic functionsx , y , xy , x2 − y2, x3 − 3xy2, . . . are harmonic functionsuζ(x) = eζ·x , ζ ∈ Cn : ζ · ζ = 0, x ∈ Rn
50 / 82
Mean Value Theorem
TheoremIf u is harmonic in Ω ⊂ Rn, n ≥ 2. Then, for any ball BR(x) ⊂ Ω
u(x) =1
|BR(x)|
∫BR(x)
u(y)dy
u(x) =1
|∂BR(x)|
∫∂BR(x)
u(y)dσy
TheoremIf u ∈ C0(Ω) has the mean value property then u is C∞(Ω) and isharmonic in Ω.
51 / 82
Maximum principle
TheoremIf u ∈ C0(Ω) has the mean value property then, if u attains a maximumor a minimum at x0 ∈ Ω then u is constant. In particular if Ω isbounded, u ∈ C0(Ω) and u is not constant then, ∀x ∈ Ω
min∂Ω
u < u(x) < max∂Ω
u
52 / 82
Application of maximum principle
Uniqueness and continuous dependence of boundary value problems
53 / 82
Boundary value problems
Dirichlet Problem∆u = f in Ω ⊂ Rn,n ≥ 2
u = g on ∂Ω.
Neumann Problem∆u = f in Ω ⊂ Rn,n ≥ 2∂u∂n = h on ∂Ω.
Robin Problem∆u = f in Ω ⊂ Rn,n ≥ 2
∂u∂ν + αu = h on ∂Ω.
54 / 82
Well-posedness
DefinitionA problem is well posed if
there exists a solution (existence)the solution is unique (uniqueness)the solution depends continuously upon the data (stability)
55 / 82
Uniqueness
TheoremΩ ⊂ Rn bounded and smooth. There exists a unique solutionu ∈ C2(Ω) ∩ C1(Ω) of the Dirichlet and of the Robin problem. In thecase of the Neumann problem the solution is unique up to an additiveconstant.
with proof
56 / 82
Existence
Existence is much more complicated to prove. For particulargeometries of the domain Ω existence can be proved via method ofseparation of variables
57 / 82
Existence via method of separation of variables: anexample
Let Ω = BR(P) ball centered at P of radius R. Then
TheoremPoisson formula The unique solution to the Dirichlet problem
∆u = 0 in BR(P) ⊂ R2
u = g on ∂BR(P).
is given by
u(x) =R2 − |x − P|2
2πR
∫∂BR(P)
g(y)
|x − y |2dσy
with proof
58 / 82
Existence via singular solutions
u harmonic function. Then, for fixed y ∈ Rn
v(x) = u(x − y)
is harmonic (translation invariance of ∆)
v(x) = u(Mx)
for M orthogonal (MT = M−1) is harmonic (rotation invariance of ∆).
59 / 82
The fundamental solution
Radially symmetric harmonic solutions: u = u(r), r = |x |.n = 2
∂2u∂r2 +
1r∂u∂u
= 0
Thenu(r) = C1 ln r + C2
n = 3∂2u∂r2 +
2r∂u∂u
= 0
Thenu(r) = C1
1r
+ C2
60 / 82
The fundamental solution
The fundamental solution for the Laplacian
Φ(x) =
1
2π ln |x |, n = 2− 1
4π|x | , n = 3
is solution to∆Φ = δ0
61 / 82
The representation formula
TheoremRepresentation FormulaΩ ⊂ Rn smooth and bounded, u ∈ C2(Ω). Then, ∀x ∈ Ω
u(x) =
∫Ω
Φ(x − y)∆u dy −∫∂Ω
Φ(x − y)∂u∂ν
dσy +
∫∂Ω
u∂Φ(x − y)
∂νdσy
62 / 82
The Green’s function
DefinitionFix y ∈ Ω
∆G(·, y) = δy in ΩG(·, y) = 0 on ∂Ω.
ANSATZ:
G(x , y) = Φ(x − y)− ψ(x , y)
where ∆ψ(·, y) = 0 in Ωψ(·, y) = Φ(x − y) on ∂Ω.
63 / 82
Properties of Green’s function
G < 0 (negativity)G(x , y) = G(y , x) (symmetry)|G(x , y)| → +∞ as x → y (singularity)
For particular geometries of Ω the Green’s function can be computedexplicitly.
64 / 82
Representation formula
TheoremConsider
∆u = f in Ωu = g on ∂Ω.
Then,
u(x) =
∫Ω
G(x , y)f (y) dy +
∫∂Ω
u∂G(x , y)
∂νdσy
If u is harmonicu(x) =
∫∂Ω
u∂G(x , y)
∂νdσy
65 / 82
Representation formula
TheoremConsider
∆u = f in Ω∂u∂ν = h on ∂Ω.
Then,
u(x) =
∫Ω
N(x , y)f (y) dy −∫∂Ω
hN(x , y) dσy
where N(x , y) is the Neumann function solution to∆N(·, y) = δy in Ω
∂N(·,y)∂ν = 1
|∂Ω| on ∂Ω.
66 / 82
More general equations
Electrical Impedence Tomographydiv(σ∇u) = f in Ω ⊂ R3
u = g on ∂Ω,
σ conductivity, u electrostatic potential, g boundary potential, f externalsource.
Optical Tomography−div(k∇ψ) + µψ = f in Ω ⊂ R3
∂ψ∂n + αψ = h on ∂Ω,
k diffusion coefficient, µ absorption coefficient, ψ optical density, hillumination function.
67 / 82
More general elliptic equations
DefinitionConsider Ω ⊂ Rn, A(x) = (aij(x))n
i,j=1, c(x) = (ci(x))ni=1, a0(x) and f (x)
scalar functions. An equation of the form
−div (A∇u) + c · ∇u + a0u = f in Ω
is elliptic in Ω if
A(x)ξ · ξ > 0, ∀x ∈ Ω,∀ξ ∈ Rn, ξ 6= 0
68 / 82
Variational formulation: Dirichlet Problem
Consider the Dirichlet problem−div(a∇u) + a0u = f in Ω
u = 0 on ∂Ω.
Suppose a, a0 and f smooth and u ∈ C2(Ω) ∩ C0(Ω) is a solution. Letv ∈ C1
0(Ω), multiplying the equation by v and integrating by parts wehave ∫
Ωa∇u · ∇v + a0uv =
∫Ω
fv
Viceversa, integrating back last equation implies that u is solution ofthe Dirichlet problem.
69 / 82
Variational formulation
Problem (DP)Find u ∈ H1
0 (Ω) such that∫Ω
a∇u · ∇v + a0uv =
∫Ω
fv , ∀v ∈ H10 (Ω)
LetB(u, v) :=
∫Ω
a∇u · ∇v + a0uv
and the linear functional
〈F , v〉 =
∫Ω
fv
Problem (DP) is equivalent to the abstract variational problem
B(u, v) = 〈F , v〉, ∀v ∈ H10 (Ω)
⇒ well-posedness follows from Lax-Milgram Theorem if
a ≥ λ, and a0 ≥ 0
70 / 82
Well-posedness
TheoremLet f ∈ L2(Ω) and λ−1
0 ≤ a ≤ λ0, 0 ≤ a0(x) ≤ γ0 in Ω. Then, Problem(DP) has a unique solution u ∈ H1
0 (Ω). Moreover
‖∇u‖0 ≤ C‖f‖0
71 / 82
Remarks
Problem (DP) is equivalent to the minimization in H10 (Ω) of the
functional (total energy)
E(u) =
∫Ω
a|∇u|2 dx +
∫Ω
a0u2 dx −∫
Ωfu dx
In the case of non homogeneous boundary conditions ifg ∈ H1(Ω) and u − g ∈ H1
0 (Ω) then we reduce to the previouscase considering w = u − g.
72 / 82
Variational formulation: Neumann Problem
Consider the Neumann problem−div(a∇u) + a0u = f in Ω
a∂u∂ν = g on ∂Ω.
Suppose a0 and f smooth and u ∈ C2(Ω) ∩ C0(Ω) is a solution. Letv ∈ C1(Ω), multiplying the equation by v and integrating by parts wehave
73 / 82
Variational formulation: Neumann Problem
∫Ω
a∇u · ∇v + a0uv =
∫Ω
fv +
∫∂Ω
gv , ∀v ∈ C1(Ω)
Assume last relation holds and integrate by parts we get∫Ω
(−div(a∇u) + a0u − f )v +
∫∂Ω
a∂u∂ν
v =
∫∂Ω
gv , ∀v ∈ C1(Ω)
Choose v ∈ C10(Ω) and we get
−div(a∇u) + a0u − f = 0
and we have ∫∂Ω
a∂u∂ν
v =
∫∂Ω
gv ∀v ∈ C1(Ω)
which gives
a∂u∂ν
= g
74 / 82
Variational formulation: Neumann Problem
Problem (NP)Find u ∈ H1
0 (Ω) such that∫Ω
a∇u · ∇v + a0uv =
∫Ω
fv +
∫∂Ω
gv , ∀v ∈ H1(Ω)
LetB(u, v) :=
∫Ω
a∇u · ∇v + a0uv
and the linear functional
〈F , v〉 =
∫Ω
fv +
∫∂Ω
gv
Problem (DP) is equivalent to the abstract variational problem
B(u, v) = 〈F , v〉, ∀v ∈ H1(Ω)
⇒ well-posedness follows from Lax-Milgram Theorem if
a ≥ λ−10 > 0, a0 ≥ γ−1
0 > 0
75 / 82
Well-posedness
TheoremLet f ∈ L2(Ω), g ∈ L2(∂Ω), 0 < λ−1
0 ≤ a(x) ≤ λ0 and0 < γ−1
0 ≤ a0(x) ≤ γ0 in Ω. Then, Problem (NP) has a unique solutionu ∈ H1
0 (Ω). Moreover
‖u‖1 ≤1
minλ−10 , γ−1
0 (‖f‖0 + C‖g‖L2(∂Ω))
76 / 82
Well-posedness
RemarkThe condition a0 ≥ γ−1
0 > 0 is necessary for existence anduniqueness!
If a0 = 0 then uniqueness fails and a normalization condition is neededto restore uniqueness.Existence of solutions requires that the compatibility condition∫
Ωf +
∫∂Ω
g = 0
is satisfied.
77 / 82
Eigenvalues
DefinitionLet Ω ⊂ Rn bounded. A nontrivial weak solution u ∈ H1
0 (Ω) to theproblem
−∆u = λu in Ωu = 0 on ∂Ω.
is called Dirichlet eigenfunction and λ the correspondingDirichlet eigenvalue
TheoremThere exists in L2(Ω) an orthonormal basis uk consisting of Dirichleteigenfunctions. The corresponding eigenvalues are an increasingsequence
0 < λ1 < λ2 ≤ · · · ≤ λk ≤ . . .
with λk → +∞. The sequence uk/λk is an orthonormal basis inH1
0 (Ω).78 / 82
Eigenvalues
RemarkIf u ∈ L2(Ω) then u =
∑ckuk and ‖u‖2 =
∑c2
k . So
‖∇uk‖2 = (∇uk ,∇uk )0 = λk (uk ,uk )0 = λk
Thus, u ∈ H10 (Ω) iff
‖∇u‖20 =∑
c2kλk <∞
79 / 82
Eigenvalues
By‖∇u‖20 =
∑c2
kλk <∞
and monotonicity of eigenvalues we get
‖∇u‖20 ≥ λ1‖u‖20
and we derive the following variational problem
λ1 = min
∫Ω |∇u|2∫
Ω u2 : u ∈ H10 (Ω),u 6= 0
or equivalently
λ1 = min∫
Ω|∇u|2 : u ∈ H1
0 (Ω), ‖u‖0 = 1
80 / 82
Eigenvalues
Let F (u) =∫
Ω |∇u|2 and G(u) =∫
Ω u2 − 1
TheoremLet u ∈ H1
0 (Ω) be a local extremum of the functional F subject to thecondition G(u) = 0, then u is an eigenfunction and λ = F (u) is thecorresponding eigenvalue.
Proof. Follows by using Lagrange multiplier Theorem
81 / 82
Eigenvalues
TheoremThere exists a global minimum u ∈ H1
0 (Ω) of the functional F subject tothe condition G(u) = 0.
Main steps of the proof.
82 / 82