• Keine Ergebnisse gefunden

Convergence of the stochastic data

tν}Lµεpf1ăăf2q ´f1ăăLµεf2qptq}Cγ`β´2

p pGε1ptqρ2ptqq À }f1}Lp,tν,γpGε1q}f2ptq}CβpGε2ptqq. where Lµε“ Bt´Lεµ is a discrete diffusion operator induced by some µPµpωq as in Definition 3.3.3. The involved constants are independent of ε.

Proof. Again we can almost follow along the lines of the proof in [GP15b, Lemma 6.5]

with the only difference that in the derivation of the second estimate the application of the “product rule” ofLµε does not yield a term ´2∇făă∇f2 but a more complex object, namely

ż

Rd

dµpyq

ε2 Dyεf1ăăDεyf2, (4.7) where Dεyf1pt, xq “ f1pt, x`εyq ´f1pt, xq and similar for f2. The bound on (4.7) follows from Lemma 4.1.5 once we can show

}Dyεϕ}Cpγ´1pGε1q À }ϕ}CpγpGε1q|y|ε (4.8) for any γ PR. Note that due to Lemma 3.1.11 we can write

jDεyϕ “

´Ψ˜ε,j¨`εy´Ψ˜ε,j

¯

˚Gε ϕ , whereΨ˜ε,j “EεΨGε,j “2jdϕxjyεp2j¨qwith ϕxjyε PSωpRdq. With

Ψ˜ε,jx`εy´Ψ˜ε,jx “2j ż1

0

2jdϕxjyεp2jpx`ζεyqqdζ¨yε we get (4.8) by applying Lemma 3.1.7.

Morally, the reason why the diffusion operator Lµε can be pulled on the second factor f2 in the product f1ăăf2 is that f2 describes the small-scale behavior of this object which is the regime whereLµε acts.

4.2 Convergence of the stochastic data

Let Gε be, as in Definition 3.1.2, a sequence of refining lattices build from some Bravais latticeGin dimensiond. Let further`

ξεpzq˘

zPGε be a discrete approximation to white noise onGε as in Section 3.4.

Fix a symmetricχPCω,c8 pR2q, independent ofε, which is0on 14¨Gpand 1outside

where lµε is as in (3.32) the multiplier of the diffusion operator Lµε associated to µ. Note that LµεYµε “ ´LεµYµε “ χpDGεε :“ FG´1ε pχ ¨ FGεξεq so that Yµε is a time independent solution to the heat (or Poisson) equation on Gε induced by our operator Lµε. Note that χ is not scaled by ε and only serves as a cut-off for the

“Fourier modes” around0, where l1ε

µ diverges.

Our first task will be to measure the regularity of the sequencespξεq, pYµεqin the discrete Besov spaces introduced in Subsection 3.1.2.

As an application of the discrete Wick calculus we introduced in Section 3.4 we can bound the moments of ξε and Yµε in Besov spaces. We also want to control the resonant productYµε˝ξε, for which we introduce the renormalization constant

cεµ:“

We define arenormalized resonance product by Yµε‚ξε :“Yµε˝ξε´cεµ.

Remark 4.2.1. Since lεµ «|¨|2 (Lemma 3.3.5 together with the easy estimate lµε À

|¨|2) we have c嵫 ´logε in dimension 2.

Using Lemma 3.4.1 we can derive the following bounds.

Lemma 4.2.2. Let ξε, Yε and Yµε‚ξε be defined on Gε as above with pξ ě4 (where The involved constant is independent of ε.

Proof. Let us bound the regularity ofYµε. Recall that by Lemma 3.1.9 we have the continuous embedding (with norm uniformly bounded in ε) Bζ`d{ppξ,pξ ξpGε,xxy´κq Ď

96 4.2 Convergence of the stochastic data By assumption we have κpξ ą d and can bound ř

zPGε|Gε|p1` |z|q´κpξ À 1 uni-formly in ε (for example by Lemma 3.5.1). It thus suffices to derive a bound for Er|∆GjεYµεpxq|pεs, uniform in ε and x. Note that by (3.6)

GjεYµεpxq “ ÿ

zPGε

|Gε|Kjεpx´zqξεpzq

withKjε “FG´1εφGjεχ{lεµ so that Lemma 3.4.1, Parseval’s identity (3.5) andlµε Á|¨|2 onxGε (from Lemma 3.3.5) imply

Er|∆GjεYµεpxq|pξs À }Kjε}pLξ2pGεqÀ2jpξpd{2´2q,

which proves the bound forYµε. The bound for ξε follows from the same arguments or with Lemma 3.3.4.

Now let us turn to Yµε‚ξε. A short computation shows that ErpYµε˝ξεqpxqs “ErpYµε¨ξεqpxqs “cεµ, xP Gε, and, by a similar argument as above, it now suffices to boundYµε‚ξε in

Bpβξ{2,pξ{2pRd,xxy´2κqfor β ă2´d. We are therefore left with the task of bounding the pξ{2-th moment of ř

|i´j|ď1GiεYµεGjεξε´Er∆GiεYµεGjεξεs. But

GiεYµεpxq∆Gjεξεpxq ´Er∆GiεYµεpxq∆Gjεξεpxqs

“ ÿ

z1,z2

|Gε|2Kiεpx´z1jpx´z2q pξεpz1εpz2q ´Erξεpz1εpz2qsq

“ ÿ

z1,z2

|Gε|2Kiεpx´z1jpx´z2εpz1q ˛ξεpz2q, so that Lemma 3.4.1 yields

E

”ˇ

ˇ∆GiεYµεGjεξε´Er∆GiεYµεGjεξεsˇ ˇ

pξ{2ı

À }Kiε}pLξ2{2pGεqj}pLξ2{2pGεq

À2ipd{2´2qpξ{22jd{2¨pξ{2 »2jpd´2q¨pξ{2, where we used Parseval’s identity, lεµÁ|¨|2 onxGε, and that |i´j| ď1.

By the compact embedding result in Lemma 2.1.26 we see that the sequences pEεξεq, pEεYµεq, and pEεpYµε ‚ξεqq have convergent subsequences in distribution.

We will see in Lemma 4.2.3 below that Eεξε converges to the white noise ξ on R2. Consequently, the solution Yµε to ´LεµYµε “ χpDGεε should, if all goes well, approach the solution of ´LµYµ“χpDRdqξ :“FR´1d`

χFRdξ˘ , i.e.

Yµ“ 1

p2πq2}DR2}2µχpDR2qξ“Kµ0˚ξ, Kµ0 :“FR´1d χ p2πq2∥¨∥2µ

. (4.11)

where } ¨ }µ is defined as in Definition 3.3.1. The limit of EεpYµε‚ξεq will turn out to be the distribution

Yµ‚ξpφq:“

ż

Rd

ż

R2

Kµ0pz1´z2qφpz1qξpdz1q ˛ξpdz2q ´ pYµăξ`ξăYµqpφq (4.12) for φ P SωpRdq, where the first term on the right hand side denotes as in Section 3.4 the second order Wiener-Itô (or Skorohod) integral with respect to the Gaussian stochastic measure ξpdzq induced by the white noise ξ. Note that Yµ‚ ξ is not a continuous functional of ξ, so the last convergence is not a trivial consequence of the convergence for Eεξε. To identify the limit of EεpYµε ‚ξεq we could use a diagonal sequence argument that first approximates the bilinear functional by a continuous bilinear functional as in [MW17b, HS15, CGP17]. However, having already established the machinery in Section 3.4 we can apply Lemma 3.4.2 instead.

Lemma 4.2.3. In the setup of Lemma 4.2.2 with ξ, Yµ and Yµ‚ξ defined as above and with ζ, κ as in Lemma 4.2.2 we have for dă4

pEεξε,EεYµε,EεpYµε‚ξεqqÝÑ pξ, YεÑ0 µ, Yµ‚ξq

in distribution inCζ´2pRd,xxy´κq ˆCζpRd,xxy´κq ˆC2ζ´2pRd,xxy´2κq.

Proof. Recall that the extension operator Eε is constructed from ψε “ψpε¨qwhere the smear functionψ is symmetric and satisfies in particularψ P Cω,c8 pRdqandψ “1 on some ball around 0.

Since from Lemma 4.2.2 we already know that the sequencepEεξε,EεYµε,EεpYµε‚ ξεqqis tight inCζ´2pRd,xxy´κqˆCζpRd,xxy´κqˆC2ζ´2pRd,xxy´2κq, it suffices to prove the convergence after testing againstφPSωpRdq:

pEεξεpφq,EεYµεpφq,EεpYµε‚ξεqpφqqÑ pd ξpφq, Yµpφq, Yµ‚ξpφqq. (4.13) We can even restrict ourselves to those φ P SωpRdq with FRdφ P Cω,c8 pRdq, which implies suppFRdφ Ď Gxε and FR´1dεFRdφq “ φ for ε small enough, which we will assume from now on. Note that suppFRdφĎxGε implies

FGεφ“FRdφ|Gxε (4.14)

since by definition ofFG´1ε

FG´1εFRdφ“ pFR´1dFRdφq|Gε “φ|Gε.

Let us first show the convergence of (4.13) in every component.

98 4.2 Convergence of the stochastic data To show the convergence of Eεξε to white noise note that we have from (3.21) the following formula

Eεξεpφq “ ÿ

zPGε

|Gε| pFR´1dψε˚φqpzqξεpzq “ ÿ

zPGε

|Gε|FR´1dεFRdφqpzqξεpzq

“ ÿ

zPGε

|Gε|φpzqξεpzq

where we used in the first step that ψε is symmetric and in the last step that FR´1dεFRdφq “φby our choice of φandε. Using Lemma 3.4.2 and relation (4.14) the convergence of Eεξεpφq toξpφq follows.

For the limit ofEεYε we use the following formula, which is derived by the same argument as above:

EεYµεpφq “ ÿ

z1, z2PGε

|Gε|2φpz1qKµεpz2´z1εpz2q with Kµε“FG´1ε

χ

lεµ. In view of Lemma 3.4.2 it then suffices to note that fˆε :“FGεpφ˚GεKµεq “FGεφ¨lχε

µ

(4.14)

“ FRdφ¨lχε

µ is due to Lemma 3.3.5 dominated by χ{|¨|2 onGxε and converges to

FRdφ χ{pp2πq2∥¨∥2µq˘ by the explicit formula forlεµ in (3.32).

We are left with the convergence of the third component. Since Eεξε Ñ ξ and EεYµεÑYµ we obtain via the (E)-Property of the paraproduct

limεÑ0EεpYµεăξεq “ lim

εÑ0EεYµεăEεξε“Yµăξ

and similarly one gets EεεăYµεq Ñ ξăYµ. We can therefore show instead Eε`

Yµεξε´ErYµεξε

pφq Ñ pYµ‚ξ`ξăYµ`Yµăξqpφq. (4.15) Note that we have the representations

Eε`

Yµεξε´ErYµεξε

pφq “ ÿ

z1,z2PGε

|Gε|2φpz1qKµεpz1 ´z2εpz1q ˛ξεpz2q, pYµ‚ξ`ξăYµ`Yµăξqpφq “

ż

R2

ż

R2

φpz1qKµ0pz1´z2qξpdz1q ˛ξpdz2q

withKµεas above andKµ0as in (4.11). ThepGεq2-Fourier transform ofφpz1qKµεpz1´ z2q is φˆextpx1 ´ x2qχpx2q{lµεpx2q for x1, x2 P xGε, where φˆext denotes the periodic

extension from (3.10) for FRdφ|Gxε P Cω,c8 pxGεq (recall again that suppFRdφ Ď Gxε).

We can therefore apply Lemma 3.4.2 since for dă4the function pχpx2q{lεpx2qq2 À 1|x|Á1{|x|4 is integrable on xGε and thus we obtain (4.15).

We have shown the convergence in distribution of all the components in (4.13).

By Lemma 3.4.2 we can take any linear combination of these components and still get the convergence from the same estimates, so that (4.13) follows from the Cramér-Wold Theorem.