Gaussian fluctuations for the stochastic wave equation with drift

We show that both sides have the same Fourier transforms. On the left side,

On the right side,

(i) By (?), it follows that K_t(x) \leq T for any x \in \mathbb{R} and t \in [0,T]. Therefore,

The conclusion follows by applying (?) with \ell=1.

(ii) Using (?) and (?), it follows that for all \varepsilon > 0,

We fix p' \in (q,1). We apply Hölder’s inequality, followed by Young’s inequality:

Note that b=\frac{p'}{2p'-1}>1 since p'\in (\frac12,1).

To estimate \|K_t^{\alpha}\|_{L^{c}(\mathbb{R}^d)}, we consider separately the two cases. In case (a), we apply Hölder’s inequality with \frac{1}{p_1}+\frac{1}{p_2}=1 and p_1=\frac{\ell}{b}>1. (This means that \frac{1}{\ell}<\frac{1}{b}=2 - \frac{1}{p'}, i.e. p'>\frac{\ell}{2\ell -1}=q.) Then

We use (Proof) and the compact support of K_t. Then, \|K_t^{\alpha}\|_{L^{c}(\mathbb{R}^d)} \leq C_{T,\varepsilon}\int_{|x| < 2t} |x|^{-\alpha \varepsilon c} dx < \infty, provided that we choose \varepsilon > 0 so that \alpha \varepsilon c< d. This implies that

In case (b), using (Proof) and the compact support of K_t, we have:

provided that we choose \varepsilon>0 such that (\beta+\alpha\varepsilon ) b< d. This means that

This is possible since p'>\frac{d}{2d-\beta} =q. Then,

Using Heisenberg commutation rule (2.3) followed by the chain rule (2.4), we infer that the sequence (Du_n)_{n\geq 0} satisfies the following recurrence relation:

for any n\geq 0. Note that D_{r,z}u_{0}(t,x)=0 since u_0(t,x)=1.

We use Burkholder-Davis-Gundy (BDG) inequality and Minkowski inequality to estimate the p-th moment of the stochastic integral. Using (1.2), it follows that for any 0<r<t\leq T, x,z \in \mathbb{R}^d and n\geq 0,

where B_{t} is a constant that depends on (t,p,\|b'\|_{\infty},\|\sigma'\|_{\infty}) and is non-decreasing in t.

We will prove below that for any 0<r<t\leq T, x,z \in \mathbb{R}^d and n\geq 1,

The conclusion will follow by Lemma A.1 of [BS26], since \{Du_n(t,x)\}_{n\geq 1} converges to Du(t,x) in the weak topology L^2(\Omega;\mathcal{H}) (by Lemma 1.2.3 of [nualart06]).

Case 1. d=1. For the first integral on the RHS of (Proof), we use Hölder’s inequality with respect to the measure \mu_t(ds,dy)=G_{t-s}(x-y)dsdy on [r,t] \times \mathbb{R} whose total mass is (t-r)^2/2. For the second integral of (Proof), we use (?). We obtain that for any 0<r<t\leq T, x,z \in \mathbb{R}, n\geq 0,

where q is given by (?). Using the fact that G_t^{2q}(x)=c_{q} G_t^{2}(x) and applying Hölder’s inequality with respect to the measure G_{t-s}^2(x-y)dy whose total mass is (t-s)/2, it follows that:

Hence

where A_T=B_{T} \big(1+\frac{T^2}{2}+c_{q}^{q} \frac{T}{2}\big). By induction on n\geq 1, it follows that

for any 0<r<t\leq T and x,z \in \mathbb{R}, where

with \pmb{x}_k=(x_1,\ldots,x_k), \pmb{t}_k=(t_1,\ldots,t_k), t_{k+1}=t and x_{k+1}=x. Using the fact that G_t^2(x)=\frac{1}{2}G_t(x) and inequality

it follows that

Hence, for any 0<r<t\leq T, x,z \in \mathbb{R} and n\geq 1,

Case 2. d=2. We proceed similarly to Lemma A.2 of [BS26]. We fix r \in [0,T] and z \in \mathbb{R}^2. By induction on n \geq 1, D_{r,z}u_n(t,x)=0 if |x-z|\geq t-r. We denote

We suppose first that \widetilde{f}_n(t)<\infty for all t \in [r,T], and we derive a recurrence relation for \widetilde{f}_{n+1}.

We bound the terms on the right-hand side of (Proof). We start with I, the term due to the drift. We have:

where for the second line we use the identity (2.1).

Considering separately the cases \theta \in [r,\frac{t+r}{2}] and \theta \in [\frac{t+r}{2},t-r], and using Fubini’s theorem, it follows that

Using Caucy-Schwarz inequality, we obtain that for any \theta \in [0,t-r],

Hence,

By direct calculation,

Indeed,

It follows that

We now bound the term II, due to the noise. Using the definition of \widetilde{f}_n(t) and inequality (?) with q given by (?), we have:

We apply Hölder’s inequality, with \frac{1}{p}+\frac{1}{p'}=1:

We now apply Lemma 4.3 of [BNZ] with \delta=p'/q, which means that we must choose 1<p'<\frac{q}{2q-1}. Note that this lemma remains valid for any \delta(1-2q)+1>0 and q \in (\frac{1}{2},1). We obtain that

for some constant C_{f,q}>1.

We put together the bounds from (Proof) and (Proof). First, by Hölder’s inequality,

Secondly, we use the fact that G_{t-r}^{\delta(2q-1)}(x-z) \leq (t-r)^{2-\delta(2q-1)} G_{t-r}^{2}(x-z), due to the trivial bound:

Hence,

We divide by G_{t-r}^2(x-z) and we take the supremum over x with |x-z|<t-r. Then, we take power p. We obtain that:

This relation is similar to (A.10) of [BS26]. Proceeding as in Step 2 and 3 of the proof of Lemma A.1 of [BS26], we conclude that \widetilde{f}_{n}(t) \leq C_{f,q} C_t where C_t is a constant that is non-decreasing in t.

We first note that D_{r,z}u(t,x) = 0 whenever t < r by (?). Furthermore, we use the convention G_t(x) = 0 whenever t < 0. Let 0\leq r,\theta \leq t \leq t' \leq T. We denote \rho = r \vee \theta.

Taking the first and then second Malliavin derivative of the mild solution (?) using (2.4) and (2.3),

Using BDG inequality and Minkowski’s inequality,

Using Theorem 3.1, Hölder’s inequality, and the boundedness of b',b'',\sigma' and \sigma'',

where

We compute each term individually. For \mathcal{B}_1 we use (?), noting that 2p'<2 if d=2:

For \mathcal{B}_2 we use Hölder’s inequality with respect to the finite measure \mathbf 1_{\{\rho < s < t\}} \mathbf 1_{\{|x-y| \leq T\}}dyds and exponent 2p' > 1:

Applying (?) with respect to dx,

Applying (?) with respect to dz and dy then integrating in ds,

Similarly for \mathcal{B}_3, we use Hölder’s inequality with respect to the measure \mathbf 1_{\{\rho \leq s \leq t\}} \mathbf 1_{\{|x-y| \leq T\}}dyds:

Applying (?) with respect to dx,

Relabelling,

Replacing the dx integral (which depends on t) by its supremum over s \in [\rho, t],

For 0 \leq t \leq t' \leq T we have 1 \leq T^{1-p'}(t'-t)^{p'-1}. We will applying this trivial bound so that \mathcal{B}_3 matches the estimate for \mathcal{B}_5 below,

For \mathcal{B}_4 we use Lemma A.1 with respect to the measure \mathbf 1_{\{|x-z| < T\}}dx, if p'< 1. (If p' = 1, the relation below is trivial. We obtain:

Identifying the convolution K_{t-s}(y-y'), taking the supremum over t \in [\rho,t'], and using the fact that t \to K_t(x) is non-decreasing, we have:

We apply again Lemma A.1 with respect to the measure \mathbf 1_{\{|z-w| < 2T\}}dz, if p' < 1. (If p'=1, the relation below is trivial.) We obtain:

Identifying the convolution K_{s-r}(y-y'), and bounding it by K_{t'}(y-y'), we obtain:

By Lemma 2.2,

Using (?) then integrating with respect to ds,

For \mathcal{B}_5, we note that Corollary 3.2 introduces the indicator \mathbf 1_{\{|x-z| < T\}}. We apply again Lemma A.1 for the measure \mathbf 1_{\{|x-z| < T\}}dx, if p'< 1. (If p'=1, the relation below is trivial.) We obtain:

Applying Lemma 2.2, we get:

The dy integral depends on s. Taking the supremum of similar integrals for s' \in [\rho,s], we obtain:

We define

Then M_{r,z,\theta,w}(t) is non-decreasing in t. Therefore, by Lemma A.2, we have:

Combining the bounds gives the recursion,

Let

Applying the above recursion twice and integrating yields,

Applying the classical Gronwall lemma gives the desired bound,

By Lemma 4.2 of [BZ24], it suffices to show that for any k\geq 1, and b_1, \ldots , b_k, \zeta_1, \ldots, \zeta_k\in\mathbb{R}^d and g \in \{x \to \sin(x), x \to \cos(x)\}, the following relation holds:

We denote

and

By Poincaré inequality (2.5), (?), and Minkowski’s inequality,

By the definition of X_{b,\zeta}(t,x), the chain rule (2.4), and (4.1),

Therefore, by the above and Hölder’s inequality,

For R \geq t, since |x| \leq \sqrt{d}R, we have |\zeta_j - y| \leq (1+\sqrt{d})R for each j = 1,\cdots,k. Therefore,

Bounding [0,R]^d by \mathbb{R}^d and integrating in the order of dx, dy then dr yields,

Since \gamma < 2d this proves Lemma 4.1 and in turn, Theorem 1.1.

Let p=\frac{1}{p'}>1 and F(x)=\big(\phi(x)\big)^{p'}. By Hölder’s inequality,

The conclusion follows taking power 1/p.

The p'= 1 case is trivial. We suppose that p'\in (\frac12,1). By a change of variable,

We denote H(s)=M(t-s) for s \in [0,t-r] and L=t-r. Then H is non-increasing and left-continuous. It is enough to prove that

By Minkowski’s inequality,

where H^{-1}(y)=\sup\{x \in [0,L];H(x)\geq y\}. On the other hand, by Fubini theorem,

Relation (Proof) follows.