20.07.2013 Views

Estimation optimale du gradient du semi-groupe de la chaleur sur le ...

Estimation optimale du gradient du semi-groupe de la chaleur sur le ...

Estimation optimale du gradient du semi-groupe de la chaleur sur le ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

L. Mil<strong>le</strong>r / Journal of Functional Analysis 236 (2006) 592–608 605<br />

the control<strong>la</strong>bility cost of a system is not changed by taking its tensor pro<strong>du</strong>ct with a unitary<br />

group). It applies in particu<strong>la</strong>r when M is a rectang<strong>le</strong> or an infinite strip in the p<strong>la</strong>ne control<strong>le</strong>d<br />

from one si<strong>de</strong> (this control<strong>la</strong>bility prob<strong>le</strong>m in a rectang<strong>le</strong> with other boundary conditions was<br />

solved in [11] without the cost estimate, which was ad<strong>de</strong>d <strong>la</strong>ter at the end of [23]). N.b. in this<br />

examp<strong>le</strong>, the condition LΓ < ∞ of [5] required in Theorem 2 is not satisfied.<br />

Corol<strong>la</strong>ry 3. Let ˜M <strong>de</strong>note another smooth comp<strong>le</strong>te Riemannian manifold. For all ρ ∈ (0, 2),<br />

there are C1 > 0 and C2 > 0 such that, for all L>0 and T ∈ (0, 1] for all ζ0 and ζ1, there is<br />

an input function u such that the solution ζ of (20) with M = (−L,L) × ˜M and Γ ={L}×∂ ˜M<br />

satisfies ζ(T)= ˙ζ(T)= 0 and the cost estimate:<br />

T<br />

0<br />

u 2<br />

L2 dt C2 exp C1L 2 /T ζ0 2<br />

H 1 +ζ1 2<br />

H −1<br />

<br />

.<br />

Proof. Let (s, y) <strong>de</strong>note the variab<strong>le</strong> on M = (−L,L) × ˜M. Denoting respectively by s<br />

and y the Dirich<strong>le</strong>t Lap<strong>la</strong>cians on the segment (−L,L) and on ˜M, wehave = s + y.<br />

Since is boun<strong>de</strong>dly invertib<strong>le</strong>, (20) may also be restated as a first-or<strong>de</strong>r system on X =<br />

H −1 (M) × H −1 (M) by setting ξ(t) = (ζ(t), ˙ζ(t)). Then the <strong>semi</strong>group generator A of the<br />

<strong>du</strong>al homogeneous system (7) becomes:<br />

<br />

0 1<br />

A = R with R =<br />

, and e<br />

−1 −ρ<br />

tA = e tsR tyR tyR tsR<br />

e = e e .<br />

The observation operator Cs <strong>de</strong>fined by Csx = χΓ ∂νz = ∂sz⌉s=L commutes with e tyR . We shall<br />

estimate the cost by the <strong>du</strong>ality in Lemma 1. Fix the initial state x0 ∈ X and T>0. Applying to<br />

s ↦→ (e TyR x0)(s, y) for fixed y the observability inequality corresponding to Theorem 3 yields,<br />

with C ′ T := C2 exp(C1L 2 /T):<br />

<br />

0<br />

L<br />

<br />

e TsR<br />

<br />

TyR<br />

e x02<br />

ds C ′ T<br />

T<br />

0<br />

<br />

0<br />

L<br />

<br />

Cse tsR<br />

<br />

TyR<br />

e x02<br />

dsdt.<br />

Integrating this inequality over ˜M yields (the first and <strong>la</strong>st step use Fubini’s theorem and the<br />

commutation of operators acting separately on s and y, the second step uses that e tyR is a<br />

contraction):<br />

<br />

M<br />

<br />

e T A x0<br />

<br />

2 dsdy C ′ T<br />

C ′ T<br />

T<br />

0<br />

T<br />

0<br />

L<br />

0<br />

L<br />

0<br />

<br />

˜M<br />

<br />

˜M<br />

<br />

e TyR<br />

Cse tsR<br />

<br />

x02<br />

dydsdt<br />

<br />

e tyR<br />

Cse tsR<br />

<br />

x02<br />

dydsdt = C ′ T<br />

This is the observability inequality corresponding to Corol<strong>la</strong>ry 3. ✷<br />

T<br />

0<br />

<br />

M<br />

<br />

Cse tA <br />

x02<br />

dsdydt.

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

Saved successfully!

Ooh no, something went wrong!