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

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

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

1.3. Proof of Theorem 1<br />

The <strong>la</strong>st ingredient of this proof is the following cost estimate corresponding to the control<br />

operator B = I proved in [2].<br />

Proposition 2. (Avalos–Lasiecka, 2003) For all ρ>0, for all α ∈ (0, 1), there are positive constants<br />

c1 and c2 such that for all T ∈ (0, 1] the solutions of (5) satisfy:<br />

∀z0 ∈ H2, ∀z1 ∈ H0,<br />

<br />

z(T ) 2 2 + ˙z(T ) 2 c2<br />

0 T c1<br />

T<br />

0<br />

<br />

˙z(t) 2 dt. (12)<br />

(In<strong>de</strong>ed, [2] specifies how the power c1 <strong>de</strong>pends on α.)<br />

In a first step, from the stationary condition in Definition 1, Proposition 2 and the <strong>du</strong>ality in<br />

Lemma 1, we <strong>de</strong><strong>du</strong>ce the “control<strong>la</strong>bility of low mo<strong>de</strong>s at exponential cost” in the corresponding<br />

dynamics. In a second step, combining it with the <strong>de</strong>cay bound in Proposition 1 according to the<br />

iterative control strategy intro<strong>du</strong>ced by Lebeau and Robbiano in [15], we prove the control<strong>la</strong>bility<br />

of all mo<strong>de</strong>s. We estimate the control<strong>la</strong>bility cost as the control time tends to zero, like in [20],<br />

in the <strong>la</strong>st step.<br />

First step. With the notations intro<strong>du</strong>ced in Section 1.1, the observation inequality (12) in Proposition<br />

2 writes:<br />

∀x0 ∈ X,<br />

<br />

e T A x0<br />

<br />

2 c2<br />

T<br />

T c1<br />

0<br />

<br />

tA<br />

Πe x0<br />

2 dt. (13)<br />

Let τ ∈ (0, 1], μ 1 and x0 ∈ 1A γ μX. For all t ∈[0,τ], we may apply (4) to Πe tA x0 since it<br />

is in 1A γ μH0:<br />

<br />

Πe tA <br />

x02<br />

0 D2 0e2D1μ CΠe tA <br />

x02<br />

.<br />

First integrating on [0,τ], then using (13) yields:<br />

<br />

e T A x0<br />

<br />

2 D 2 0<br />

c2<br />

e2D1μ<br />

τ c1<br />

τ<br />

0<br />

<br />

Ce tA <br />

x02<br />

dt.<br />

This “low mo<strong>de</strong>s fast observability for e tA at exponential cost” is equiva<strong>le</strong>nt, by the same <strong>du</strong>ality<br />

as in Lemma 1, to the control<strong>la</strong>bility property: for all τ ∈ (0, 1] and μ>1, there is a boun<strong>de</strong>d<br />

operator S τ μ : X → L2 (0,τ; U) such that, for all ξ0 ∈ 1A γ μX, the solution ξ ∈ C(R+,X) of<br />

(6) with input function u = S τ μ ξ0 satisfies 1A γ μξ(τ) = 0, and ∃d3 > 0, S τ μ (d3/τ c1/2 )e D1μ<br />

(cost estimate).<br />

Second step. The hypothesis on α implies that the γ ′ of Proposition 1 is lower than 1. We<br />

intro<strong>du</strong>ce a dyadic sca<strong>le</strong> of mo<strong>de</strong>s μk = 2 k (k ∈ N) and a sequence of time intervals τk = σδT/μ δ k

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

Saved successfully!

Ooh no, something went wrong!