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) 592–608 599<br />

where δ ∈ (0,γ ′−1 − 1) and σδ = (2 <br />

k∈N 2−kδ ) −1 > 0, so that the sequence of times <strong>de</strong>fined<br />

recursively by T0 = 0 and Tk+1 = Tk + 2τk converges to T . The strategy consists in steering the<br />

initial state ξ0 to 0, through the sequence of states ξk = ξ(Tk) ∈ 1Aγ >μk−1X composed of ever<br />

higher mo<strong>de</strong>s, by applying recursively the input function uk = S τk<br />

μkξk to ξk <strong>du</strong>ring a time τk and<br />

no input <strong>du</strong>ring a time τk. Intro<strong>du</strong>cing the notations<br />

εk =ξk, Ck = D2e D1μk /τ c1/2<br />

k , and ρk =<br />

the cost estimate of the previous step writes S τk<br />

μk Ck and implies:<br />

u 2<br />

L 2 (0,T ;U)<br />

= <br />

k∈N<br />

uk 2<br />

L 2 (0,τk;U)<br />

<br />

k∈N<br />

Ck+1εk+1<br />

Ckεk<br />

Since τk T 1, the estimate (10) between the times Tk and Tk + τk implies<br />

<br />

ξ(Tk + τk) 2 2 1 + K1C 2 2<br />

k εk .<br />

2<br />

, (14)<br />

C 2 k ε2 k . (15)<br />

Since 1Aγ μkξ(Tk ′<br />

−rμ1/γ + τk) = 0 and Proposition 1 imply εk+1 e k<br />

τkξ(Tk + τk), we <strong>de</strong><strong>du</strong>ce<br />

ε 2 k+1<br />

1/γ ′ <br />

2e−2rτkμk 1 + K1C 2 2<br />

k εk .<br />

Since Ck+1/Ck = 2 δc1/2 e D1μk , we <strong>de</strong><strong>du</strong>ce that, for any D3 > 4D1, there is a D4 > 0 such that<br />

ρk 2 1+δc1<br />

Since γ ′−1 − δ>1, this implies:<br />

<br />

e −2D1μk<br />

K1D<br />

+ 2 2<br />

τ c1<br />

<br />

1/γ ′<br />

4D1μk−2rτkμ<br />

e k <br />

k<br />

D4<br />

T c1 eD3μk−2rσδTμ γ ′−1−δ k . (16)<br />

∀ρ ∈ (0, 1), ∃N ∈ N, k N ⇒ ρk ρ.<br />

Therefore limk εk = 0 and the <strong>la</strong>st series in (15) converges. This comp<strong>le</strong>tes the proof of the<br />

control<strong>la</strong>bility in Theorem 1.<br />

Third step. The control<strong>la</strong>bility cost CT , formally <strong>de</strong>fined after Lemma 1, satisfies:<br />

Since<br />

l μl,<br />

<br />

0kl−1<br />

C 2 T C2 0<br />

<br />

1 + <br />

<br />

l1 0kl−1<br />

μk μl and <br />

0kl−1<br />

ρk<br />

<br />

. (17)<br />

μ γ ′−1 −δ<br />

k<br />

μ γ ′−1−δ l−1 /2,

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

Saved successfully!

Ooh no, something went wrong!