Optimal Control of Partial Differential Equations

Optimal Control of Partial Differential Equations Optimal Control of Partial Differential Equations

math.univ.toulouse.fr
from math.univ.toulouse.fr More from this publisher
29.03.2013 Views

52 CHAPTER 5. CONTROL OF THE HEAT EQUATION 5.3 Neumann boundary control In this section, we study problems in which the control variable acts through a Neumann boundary condition ∂z ∂t − ∆z = f in Q, ∂z ∂n = u on Σ, z(x, 0) = z0 in Ω. (5.3.10) Theorem 5.3.1 Set Z = L2 (Ω), D(A) = {z ∈ H2 (Ω) | ∂z = 0}, Az = ∆z. The operator ∂n (A, D(A)) is the infinitesimal generator of a strongly continuous semigroup of contractions in L2 (Ω). Proof. The proof still relies on the Hille-Yosida theorem. It is very similar to the proof of Theorem 5.2.1 and is left to the reader. The operator (A, D(A)) is selfadjoint in Z. Equation ∂z ∂t may be written in the form − ∆z = f in Q, ∂z ∂n = 0 on Σ, z(x, 0) = z0 in Ω, (5.3.11) z ′ = Az + f, z(0) = z0. (5.3.12) A function z ∈ L2 (0, T ; L2 (Ω)) is a weak solution to equation (5.3.12) if for all ζ ∈ D(A) the mapping t ↦→ 〈z(t), ζ〉 belongs to H1 (0, T ), 〈z(0), ζ〉 = 〈z0, ζ〉, and d z(t)ζ = 〈z, Aζ〉 + 〈f, ζ〉 = z(t)∆ζ + f(t)ζ. dt Ω Theorem 5.3.2 For every φ ∈ L 2 (Q) and every z0 ∈ L 2 (Ω), equation (5.3.11) admits a unique weak solution z(φ, z0) in L 2 (0, T ; L 2 (Ω)), moreover the operator Ω (φ, z0) ↦→ z(φ, z0) is linear and continuous from L 2 (Q) × L 2 (Ω) into W (0, T ; H 1 (Ω), (H 1 (Ω)) ′ ). Recall that W (0, T ; H 1 (Ω), (H 1 (Ω)) ′ ) = z ∈ L 2 (0, T ; H 1 (Ω)) | dz dt ∈ L2 (0, T ; (H 1 (Ω)) ′ ) . Proof. The existence in C([0, T ]; L 2 (Ω)) follows from Theorem 5.3.1. The regularity in W (0, T ; H 1 (Ω), (H 1 (Ω)) ′ ) can be proved as for Theorem 5.2.2. Similarly we would like to say that a function z ∈ L2 (0, T ; L2 (Ω)) is a weak solution to equation (5.3.10) if for all ζ ∈ D(A) the mapping t ↦→ 〈z(t), ζ〉 belongs to H1 (0, T ), 〈z(0), ζ〉 = 〈z0, ζ〉, and d z(t)ζ = z(t)∆ζ + fζ + uζ. dt Ω Ω Ω Γ Unfortunately the mapping ζ ↦→ Γ uζ is not an element of L2 (0, T ; L2 (Ω)), it only belongs to L2 (0, T ; (H1 (Ω)) ′ ). One way to study equation (5.3.10) consists in using (A∗ 1) ∗ (see Chapter 4), the extension of A to (D(A∗ )) ′ = (D(A)) ′ (A is selfadjoint). We can directly improve this Ω

5.3. NEUMANN BOUNDARY CONTROL 53 result in the following way. We set Z = (H 1 (Ω)) ′ . We endow (H 1 (Ω)) ′ with the dual norm of the H 1 -norm. We can check that the corresponding inner product in (H 1 (Ω)) ′ is defined by z, ζ (H1 (Ω)) ′ = Ω z(−∆ + I) −1 ζ = where (−∆ + I) −1 ζ is the function w ∈ H 1 (Ω) obeying −∆w + w = ζ in Ω, ∂w ∂n Ω (−∆ + I) −1 z ζ, = 0 on Γ. We define the unbounded operator A in (H1 (Ω)) ′ by D( A) = H1 (Ω), and 〈 Az, ζ〉 (H1 (Ω)) ′ ,H1 (Ω) = − ∇z∇ζ = Ω Az, ζ (H1 (Ω)) ′ . Theorem 5.3.3 The operator ( A, D( A)) is the infinitesimal generator of a strongly continuous semigroup of contractions in (H 1 (Ω)) ′ . Proof. The proof still relies on the Hille-Yosida theorem. It is more complicated than the previous ones. It is left to the reader. We write equation (5.3.10) in the form z ′ = Az + f + û, z(0) = z0, (5.3.13) where û ∈ L2 (0, T ; (H1 (Ω)) ′ ) is defined by û ↦→ Γ uζ for all ζ ∈ H1 (Ω). Due to Theo- rem 5.3.3 equation (5.3.13), or equivalently equation (5.3.10), admits a unique solution in L 2 (0, T ; (H 1 (Ω)) ′ ) and this solution belongs to C([0, T ]; (H 1 (Ω)) ′ ). To establish regularity properties of solutions to equation (5.3.10) we need to construct solutions by an approximation process. Approximation by regular solutions. Recall that the solution to equation satisfies the estimate ∆w − w = 0 in Ω, ∂w ∂n = v on Γ, (5.3.14) w H 2 (Ω) ≤ Cv H 1/2 (Γ). (5.3.15) Let u be in L 2 (Σ) and let (un)n be a sequence in C 1 ([0, T ]; H 1/2 (Γ)), converging to u in L 2 (Σ). Denote by wn(t) the solution to equation (5.3.14) corresponding to v = un(t). With estimate (5.3.15) we can prove that wn belongs to C 1 ([0, T ]; H 2 (Ω)) and that wn C 1 ([0,T ];H 2 (Ω)) ≤ Cun C 1 ([0,T ];H 1/2 (Γ)). Let zn be the solution to equation (5.3.10) corresponding to (f, un, z0). Then yn = zn − wn is the solution to ∂y ∂t − ∆y = f − ∂wn ∂t + ∆wn in Q, ∂y ∂n = 0 on Σ, y(x, 0) = (z0 − wn(0))(x) in Ω.

52 CHAPTER 5. CONTROL OF THE HEAT EQUATION<br />

5.3 Neumann boundary control<br />

In this section, we study problems in which the control variable acts through a Neumann<br />

boundary condition<br />

∂z<br />

∂t<br />

− ∆z = f in Q,<br />

∂z<br />

∂n = u on Σ, z(x, 0) = z0 in Ω. (5.3.10)<br />

Theorem 5.3.1 Set Z = L2 (Ω), D(A) = {z ∈ H2 (Ω) | ∂z = 0}, Az = ∆z. The operator<br />

∂n<br />

(A, D(A)) is the infinitesimal generator <strong>of</strong> a strongly continuous semigroup <strong>of</strong> contractions in<br />

L2 (Ω).<br />

Pro<strong>of</strong>. The pro<strong>of</strong> still relies on the Hille-Yosida theorem. It is very similar to the pro<strong>of</strong> <strong>of</strong><br />

Theorem 5.2.1 and is left to the reader.<br />

The operator (A, D(A)) is selfadjoint in Z. Equation<br />

∂z<br />

∂t<br />

may be written in the form<br />

− ∆z = f in Q,<br />

∂z<br />

∂n = 0 on Σ, z(x, 0) = z0 in Ω, (5.3.11)<br />

z ′ = Az + f, z(0) = z0. (5.3.12)<br />

A function z ∈ L2 (0, T ; L2 (Ω)) is a weak solution to equation (5.3.12) if for all ζ ∈ D(A) the<br />

mapping t ↦→ 〈z(t), ζ〉 belongs to H1 (0, T ), 〈z(0), ζ〉 = 〈z0, ζ〉, and<br />

<br />

<br />

d<br />

z(t)ζ = 〈z, Aζ〉 + 〈f, ζ〉 = z(t)∆ζ + f(t)ζ.<br />

dt<br />

Ω<br />

Theorem 5.3.2 For every φ ∈ L 2 (Q) and every z0 ∈ L 2 (Ω), equation (5.3.11) admits a<br />

unique weak solution z(φ, z0) in L 2 (0, T ; L 2 (Ω)), moreover the operator<br />

Ω<br />

(φ, z0) ↦→ z(φ, z0)<br />

is linear and continuous from L 2 (Q) × L 2 (Ω) into W (0, T ; H 1 (Ω), (H 1 (Ω)) ′ ).<br />

Recall that<br />

W (0, T ; H 1 (Ω), (H 1 (Ω)) ′ ) =<br />

<br />

z ∈ L 2 (0, T ; H 1 (Ω)) | dz<br />

dt ∈ L2 (0, T ; (H 1 (Ω)) ′ <br />

) .<br />

Pro<strong>of</strong>. The existence in C([0, T ]; L 2 (Ω)) follows from Theorem 5.3.1. The regularity in<br />

W (0, T ; H 1 (Ω), (H 1 (Ω)) ′ ) can be proved as for Theorem 5.2.2.<br />

Similarly we would like to say that a function z ∈ L2 (0, T ; L2 (Ω)) is a weak solution to equation<br />

(5.3.10) if for all ζ ∈ D(A) the mapping t ↦→ 〈z(t), ζ〉 belongs to H1 (0, T ), 〈z(0), ζ〉 = 〈z0, ζ〉,<br />

and<br />

<br />

d<br />

z(t)ζ = z(t)∆ζ + fζ + uζ.<br />

dt Ω<br />

Ω<br />

Ω Γ<br />

Unfortunately the mapping ζ ↦→ <br />

Γ uζ is not an element <strong>of</strong> L2 (0, T ; L2 (Ω)), it only belongs<br />

to L2 (0, T ; (H1 (Ω)) ′ ). One way to study equation (5.3.10) consists in using (A∗ 1) ∗ (see Chapter<br />

4), the extension <strong>of</strong> A to (D(A∗ )) ′ = (D(A)) ′ (A is selfadjoint). We can directly improve this<br />

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!