Você está na página 1de 23

The Lax-Milgram Theorem

http://www.math.unibas.ch/∼beilina

1
Definition 1.
A bilinear form a(·, ·) on a normed linear space H is said to be bounded
or continuous, if ∃C < ∞ such that
|a(v, w)| ≤ CkvkH kwkH ∀v, w ∈ H, and coercive on V ⊂ H if
∃α > 0 such that a(v, v) ≥ αkvk2H ∀v ∈ V .
Propositon 1 Let H be a Hilbert space, and suppose that a(·, ·) is a
symmetric bilinear form that is continuous on H and coercive on V ⊂ H.
Then, V, a(·, ·) is a Hilbert space.

2
Proof
Because a(·, ·) is coercive → a(·, ·) is an inner product on V .
p
Next, let kvkE = a(v, v) and suppose that {vn } is a Cauchy sequence
in (V, k · kE ). By coercivity, vn is also Cauchy sequence in (H, k · kH ).
Since H is complete, ∃v ∈ H : vn → v in the k · kH norm. Since V is
closed in H, v ∈ V .

Since a(·, ·) is bounded → kv − vn kH ≤ c1 kv − vn kH . Hense, vn → v
in the k · kE norm and (V, k · kE ) is complete.
2

3
A symmetric variational problem
If the following conditions are valid
1. H, (·, ·) is a Hilbert space,
2. V is closed subspace of H,
3. a(·, ·) is bounded, symmetric form that is coercive on V ,
then the symmetric variational problem is the following:
Given F ∈ V 0 find u ∈ V such that a(u, v) = F (v) ∀v ∈ V.
Theorem
Suppose that conditions (1)-(3) above hold. Then ∃!u ∈ V solving
a(u, v) = F (v) ∀v ∈ V.
Proof
Proposition 1 implies, that a(·, ·) is an inner product on V and V, a(·, ·) is
a Hilbert space. Then we apply the Riesz Theorem. 2

4
The Ritz-Galerkin approximation
Given a finite-dimensional subspace Vh ⊂ V and F ∈ V 0 , find uh ∈ Vh
such that a(uh , v) = F (v) ∀v ∈ Vh .
Theorem
Under the conditions (1)-(3) there exists a unique uh that solves
a(uh , v) = F (v) ∀v ∈ Vh .
Proof
Vh , a(·, ·) is a Hilbert space, and F ∈ Vh0 . Then proof follows from Riesz
theorem.
Fundamental Orthogonality
Let u, uh be the solutions to a(u, v) = F (v) and a(uh , v) = F (v),
respectively, then a(u − uh , v) = 0 ∀v ∈ Vh .

5
The Ritz method
In the symmetric case, uh minimizes the quadratic functional
Q(v) = a(v, v) − 2F (v) ∀v ∈ Vh .

6
Formulation of nonsymmetric variational problems

1. H, (·, ·) is a Hilbert space.


2. V is closed subspace of H.

3. a(·, ·) is a bilinear form on V , not necessarily symmetric.


4. a(·, ·) is continuous (bounded) on V .

5. a(·, ·) is coercive on V .

then the nonsymmetric variational problem is the following:


Given F ∈ V 0 find u ∈ V such that a(u, v) = F (v) ∀v ∈ V.
Galerkin approximation problem
Given a finite-dimensional subspace Vh ⊂ V and F ∈ V 0 , find uh ∈ Vh
such that a(uh , v) = F (v) ∀v ∈ Vh .

7
The Lax-Milgram Theorem

We would like to prove the existence and uniqueness of the solution of the
(nonsymmetric) variational problem.
Lemma. Contraction mapping Principle
Given a Banach space V and a mapping T : V → V , satisfying
||T v1 − T v2 || ≤ M ||v1 − v2 || ∀v1 , v2 ∈ V and fixed M, 0 ≤ M < 1,
there ∃!u ∈ V such that
u = T u,
or the contraction mapping T has a unique fixed point u.
Proof
1) We show uniqueness. Let T v1 = v1 and T v2 = v2 . Since T is a
contraction mapping, ||T v1 − T v2 || ≤ M ||v1 − v2 || for some
M, 0 ≤ M < 1. But ||T v1 − T v2 || = ||v1 − v2 ||, therefore,

8
||v1 − v2 || ≤ M ||v1 − v2 ||, and this implies ||v1 − v2 || = 0 (otherwise
1 ≤ M ). Then v1 = v2 .
2) Existence. We take v0 ∈ V and define
v1 = T v0 , v2 = T v − 1, ..., vk+1 = T vk . Note that
||vk+1 − vk || = ||T vk − T vk−1 || ≤ M ||vk − vk−1 ||. By induction,
||vk − vk−1 || ≤ M k−1 ||v1 − v0 ||. Therefore, ∀N > n
N
X
||vN − vn || = || vk − vk−1 ||
k=n+1
N
X
≤ ||v1 − v0 || M k−1
k=n+1
(1)
Mn
≤ ||v1 − v0 ||
1−M
Mn
= ||T v0 − v0 ||,
1−M

9
from what follows that {vn } is a Cauchy sequence. Since V is complete
and limn→∞ vn =: v, we have
v = lim vn+1
n→∞
= lim T vn
n→∞
(2)
= T ( lim vn ) (T is continuous)
n→∞
= T v,

or there exists a fixed point.


2

10
The Lax-Milgram Theorem
Given a Hilbert space V, (·, ·), a continuous, coercive bilinear form a(·, ·)
and a continuous linear functional F ∈ V 0 , there exists a unique u ∈ V
such that
a(u, v) = F (v) ∀v ∈ V. (3)

Proof
For any u ∈ V we define a functional Au(v) = a(u, v) ∀v ∈ V . Au is
linear since
Au(αv1 + βv2 ) = a(u, αv1 + βv2 ))
= αa(u, v1 ) + βa(u, v2 )
= αAu(v1 ) + βAu(v2 ) ∀v1 , v2 ∈ V, α, β ∈ R.

Au is continuous, because ∀v ∈ V

|Au(v)| = |a(u, v)| ≤ C||u||||v|| ∀v ∈ V.

11
Therefore,
Au(v)
||Au(v)||V 0 = sup ≤ C||u|| < ∞.
v6=0 ||v||
Thus, Au ∈ V 0 . Similarly one can show that the mapping u → Au is a
linear map V → V 0 .
By Riesz representation theorem, for ∀ϕ ∈ V 0 there ∃!τ ϕ ∈ V such that
ϕ(v) = (τ ϕ, v) ∀v ∈ V . We must find a unique u such that
Au(v) = F (v) ∀v ∈ V , in other words, we want to find a unique u
such that Au = f in V 0 or τ Au = τ F in V , since τ : V 0 → V is
one-to-one mapping.

12
Now we use Contraction Mapping Principle. We want to find ρ 6= 0 such
that the mapping T : V → V is a contraction mapping, where T is
defined by:
T v := v − ρ(τ Av − τ F ) ∀v ∈ V (4)
If T is a contraction mapping, then by Contraction Mapping Principle
there ∃!u ∈ V such that T u = u − ρ(τ Au − τ F ) = u, that is
ρ(τ Au − τ F ) = 0 or τ Au = τ F . It remains to show that there ∃ρ 6= 0.

13
For ∀v1 , v2 ∈ V let v = v1 − v2 , then
||T v1 − T v2 ||2 = ||v1 − v2 − ρ(τ Av1 − τ Av2 )||2
= ||v − ρ(τ Av)||2
= ||v||2 − 2ρ(τ Av, v) + ρ2 ||τ Av||2
= ||v||2 − 2ρAv(v) + ρ2 Av(τ Av)
(definition of τ , (τ Av, v) = Av(v))
= ||v||2 − 2ρa(v, v) + ρ2 a(v, τ Av)
(definition of A)
≤ ||v||2 − 2ρα||v||2 + ρ2 C||v|| · ||τ Av||
(coercivity and continuity of A)
≤ (1 − 2ρα + ρ2 C 2 )||v||2
(A bounded, τ isometric and ||τ Av|| = ||Av|| ≤ C||v||)
≤ (1 − 2ρα + ρ2 C 2 )||v1 − v2 ||2
= M 2 ||v1 − v2 ||2 . 14
We need 1 − 2ρα + ρ2 C 2 < 1 for some ρ. If we choose ρ ∈ (0, 2α/C 2 )
then M < 1 and the proof is complete.
2
Corollary
The nonsymmetric variational problem has a unique solution.
Theorem (Céa)
Suppose that conditions for nonsymmetric variational problem hold and
that u solves a(u, v) = F (v) ∀v ∈ V. For the finite element variational
problem a(uh , v) = F (v) ∀v ∈ Vh we have
C
||u − uh ||V ≤ min ||u − v||V ,
α v∈V h

where C is the continuity constant and α is the coercivity constant of


a(·, ·).

15
Proof
Since a(u, v) = F (v) ∀v ∈ V and a(uh , v) = F (v) ∀v ∈ Vh we
have a(u − uh , v) = 0 ∀v ∈ Vh . For ∀v ∈ Vh

α||u − uh ||2V ≤ a(u − uh , u − uh )


= a(u − uh , u − v) + a(u − uh , v − uh )
= a(u − uh , u − v)(since v − uh ∈ Vh and use orthogonality)
≤ C||u − uh ||V · ||u − v||V .
C
Hence, ||u − uh ||V ≤ α ||u − v||V . Therefore,
C
||u − uh ||V ≤ inf ||u − v||V
α v∈Vh
C
= min ||u − v||V (since Vh is closed).
α v∈Vh

16
Applications of the Lax-Milgram theorem

A problem with Neumann boundary conditions We consider Poisson’s


equation in 1D with an absorption term together with Neumann boundary
conditions:
−u00 + u = f in Ω
(5)
∂x u|x=0,1 = 0,
where Ω ⊂ R. The variational form is:
Z
a(v, w) = (v 0 · w 0 + vw)dx,
ZΩ
L(v) = f vdx, (6)

Z
V = {v : (|v 0 |2 + v 2 )dx < ∞}.

We will verify that assumptions of the Lax-Milgram theorem are satisfied

17
with these choices.
1. V has natural scalar product and norm defined as
Z
(v, w)V := (v 0 w0 + vw)dx,
ZΩ (7)
||v||V := (|v 0 |2 + v 2 )1/2 dx.

Then, we see that V is complete, and therefore, V is Hilbert space.


2. a(v, v)V is V -elliptic-coercive and continuous - trivially holds.
3. L(·) is continuous. We note, that

|L(v)| ≤ ||f ||L2 (Ω) ||u||L2 (Ω) ≤ ||f ||L2 (Ω) ||u||V , (8)

which means that we can take constant in definition of continuity as


k = ||f ||L2(Ω) .
Then we can apply Lax-Milgram theorem to (5).

18
The spaces H 1 (Ω) and H01 (Ω)
The space defined in (13) has a special notation
Z
H 1 (Ω) = {v : (|v 0 |2 + v 2 )dx < ∞}, (9)

while the scalar product and norm are defined by


Z
(v, w)H 1 (Ω) = (v 0 w0 + vw)dx,
ZΩ (10)
||v||H 1 (Ω) := (|v 0 |2 + v 2 )1/2 dx

We also use subspase H01 (Ω) of H 1 (Ω):

H01 (Ω) = {v ∈ H 1 (Ω) : v = 0 on Γ}. (11)

19
Poisson’s equation in 1D with Dirichlet boundary conditions
We consider Poisson’s equation in 1D with with Dirichlet boundary
conditions:
−u00 = f in Ω
(12)
u(0) = u(1) = 0,

where Ω ⊂ R. The variational form is:


Z
a(v, w) = v 0 · w 0 dx,
ZΩ
L(v) = f vdx, (13)

V = H01 (Ω).

In this case the V -ellipticity of a does not follow automatically from the
definition of norm V = H01 (Ω) as above (a(v, w) does not contain the
R 2
term Ω v dx). To verify the ellipticity, we use the

20
Poincare-Friedrichs inequality

||v||L2 (0,1) ≤ 2(v(0)2 + ||v 0 ||2L2 (0,1) ) (14)

Proof
Z x Z 1
2 0 2 2
v (x) = (v(0) + v (y)dy) ≤ 2(v (0) + (v 0 (y))2 dy) (15)
0 0

for 0 ≤ x ≤ 1 ( we use integrating, Caushy’s inequality and the fact that


(a + b)2 ≤ 2(a2 + b2 )). 2
For functions v ∈ H 1 (Ω) with v(0) = v(1) = 0, or ||v||L2 (Γ) = 0,
Poincare-Friedrichs inequality means

||v||H 1 (Ω) = ||v 0 ||2L2 (0,1) +||v||2L2 (0,1) ≤ (1+C)||v 0 ||2L2 (0,1) = (1+C)a(v, v),
(16)
1
which proves V -ellipticity with constant in definition k1 = 1+C . In 1D
constant C = 2.

21
Continuity of a(·, ·) and L(·) follow exactly as for the case of Neumann
boundary conditions.
Remains to show that H01 (Ω) is a well defined Hilbert space, or that a
function in H01 (Ω) has well defined values on the boundary. In general it
is impossible to uniquely define the boundary values of v ∈ L2 (Ω),
because v changes only close to the boundary. However, if we change
L2 (Ω) to H 1 (Ω), then the trace of v ∈ H 1 (Ω) on the boundary is well
defined. This expressed in the trace inequality:
Theorem
If Ω is a bounded domain with boundray Γ, then there is a constant C
such that ∀v ∈ H 1 (Ω)

||v||L2 (Γ) ≤ C||v||H 1 (Ω) (17)

Proof

22
|x|2
We take ϕ = 2 such that ϕ00 = 1. The we use the fact that
Z 1 Z 1
v 2 ϕ00 dx = v 2 ∂x ϕ|10 − 2vv 0 ϕ0 dx (18)
0 0

and choose ∂x ϕ = 1 to get from (18):


Z 1 Z 1
v 2 dx = v 2 |10 − 2vv 0 ϕ0 dx, (19)
0 0

therefore,
Z 1 Z 1
v 2 |10 ≤ 2vv 0 ϕ0 dx + v 2 dx, (20)
0 0
and applying Poincare-Friedrichs inequality we get
Z 1 Z 1
||v 2 ||L2 (0,1) ≤ C( v 02 dx + v 2 dx) = C||v||H 1 (Ω) . (21)
0 0
2

23

Você também pode gostar