arXiv is now an independent nonprofit! Learn more
License: CC BY 4.0
arXiv:2608.19753v1 [math.PR] 20 Aug 2026

Uniform large deviation principles for the stochastic heat equation over unbounded sets of initial dataThanks: The first author was supported by Simons Foundation Grant 962543. The second author was supported by the VICI subsidy VI.C.212.027 of the Netherlands Organisation for Scientific Research (NWO)

Michael Salins Michael SalinsBoston University
United States of America.
Email address: msalins@bu.edu
and Esmée Theewis Esmée TheewisDelft University of Technology
The Netherlands.
Email address: e.s.theewis@tudelft.nl
Date: August 20, 2026
Abstract.

We study small-noise large deviations for the stochastic heat equation (SHE) on the torus with unbounded, multiplicative space-time white noise. We establish uniform large deviation principles (ULDPs) over unbounded sets of initial data, alongside novel well-posedness and regularity results. The ULDPs are obtained for both continuous-in-space and LpL^{p} initial data. For these two classes, the ULDP holds uniformly over LqL^{q}-bounded subsets, with q<q<\infty and q<pq<p allowed respectively, so these sets are highly unbounded in the state space. The admissible range of qq is dictated by the growth of the noise coefficient in the SHE. Crucially, our methods allow us to reach down to q=1q=1. This yields ULDPs over L1L^{1}-bounded subsets, enabling the study of exit times in physical systems with mass conservation.

Key words and phrases: 
Uniform large deviation principle, stochastic heat equation
2020 Mathematics Subject Classification
Primary: 60H15, Secondary: 60F10, 35K58

Introduction

We consider the following stochastic heat equation (SHE) on the one-dimensional torus D/2πD\coloneqq\mathbb{R}/2\pi\mathbb{Z}, for small ε>0\varepsilon>0:

(0.1) {Xxεt(t,ξ)=ΔXxε(t,ξ)+b(Xxε(t,ξ))+εσ(Xxε(t,ξ))W˙(t,ξ),ξD,t(0,T],Xxε(0,ξ)=x(ξ),ξD,\begin{cases}\frac{\partial X_{x}^{\varepsilon}}{\partial t}(t,\xi)=\Delta X_{x}^{\varepsilon}(t,\xi)+b(X_{x}^{\varepsilon}(t,\xi))+\sqrt{\varepsilon}\sigma(X_{x}^{\varepsilon}(t,\xi))\dot{W}(t,\xi),\quad\xi\in D,t\in(0,T],\\ X_{x}^{\varepsilon}(0,\xi)=x(\xi),\quad\xi\in D,\end{cases}

where WW is a space-time white noise. In this paper, we study uniform large deviation principles (ULDPs) for families of solutions (Xxε)ε,x(X^{\varepsilon}_{x})_{\varepsilon,x} to (0.1), as ε0\varepsilon\to 0. We prove ULDPs in the C([0,T],Lp(D))C([0,T];L^{p}(D)) topology for p2p\geq 2 and the C([0,T]×D)C([0,T]\times D) topology on the path space of solutions to (0.1). Unlike previous results for this problem, our main results prove that the ULDP holds uniformly over initial data that are bounded in Lq(D)L^{q}(D) norm, for some q<pq<p. In this regime, the set {xLp(D):xLq(D)R}\{x\in L^{p}(D):\|x\|_{L^{q}(D)}\leq R\} is unbounded in Lp(D)L^{p}(D). Previous results mostly prove uniformity over bounded sets or compact sets of initial data [9, 5].

A large deviation principle formally quantifies the exponential decay rates for probabilities of rare events. In the context of the SHE, large deviation principles allow us to accurately study the asymptotic behavior of solutions XxεX_{x}^{\varepsilon}, for the small-noise limit ε0\varepsilon\to 0. Beyond guaranteeing almost sure convergence of XxεX_{x}^{\varepsilon} to the deterministic solution Xx0X_{x}^{0}, they provide the precise rates of decay for probabilities of relevant events. A powerful application of this theory is the Freidlin–Wentzell exit time problem, due to Freidlin and Wentzell [23]. For GC(D)G\subset C(D) (or GLp(D)G\subset L^{p}(D)), using large deviation principles, one can characterize the exponential growth rates of the corresponding exit times for the solutions XxεX_{x}^{\varepsilon}, defined as

τxεinf{t>0:Xxε(t,)G}.\tau^{\varepsilon}_{x}\coloneqq\inf\{t>0:X_{x}^{\varepsilon}(t,\cdot)\notin G\}.

For these exit time applications, it is crucial to establish an LDP that holds uniformly with respect to the initial data in GG, i.e. a ULDP (see Definition 1.2). We refer the reader to [23, 18] for details on the general proof strategy used in such applications.

A highly influential development for proving ULDPs for small-noise SPDEs is the weak convergence approach by Budhiraja, Dupuis, and Maroulas [5], which yields uniformity over compact sets of initial data. This approach has been applied in a large body of literature on LDPs [4, 28, 27, 25, 46, 1, 2, 8, 16, 19, 29, 30, 31, 32, 34, 36, 37, 38, 44, 45, 48, 49, 35] and ULDPs [3, 6, 24, 21]. However, for the exit problem above, one typically wants to consider sets GG with non-empty interior. Because such sets are non-compact in infinite-dimensional Banach spaces and one needs uniformity over initial data in GG, the weak convergence approach cannot be applied directly. In [39], the first author showed that the variational principle underlying [5] can nevertheless be used to establish uniform large deviation principles on non-compact, and even unbounded, sets of initial data, provided that the controlled equations satisfy a stronger mode of convergence than weak convergence.

Concerning the SHE and more general reaction-diffusion equations, the earliest LDP results include [22, 43, 26] and treat Lipschitz coefficients bb and σ\sigma. In [9], Cerrai and Röckner relaxed the assumptions of those works by removing Lipschitz continuity from reaction terms and removing ellipticity conditions on the multiplicative noise terms, and they were the first to prove a ULDP over supremum-norm-bounded subsets of initial data. More recently, the work [40] by the first author further relaxed the assumptions of [9] by removing local Lipschitz and polynomial growth conditions for the reaction terms, while establishing a ULDP that still holds uniformly over supremum-norm-bounded subsets of initial data, and even uniformly over all initial data in case σ\sigma is bounded.

The purpose of the current work is to establish, for unbounded noise coefficients σ\sigma, a ULDP that holds uniformly over sets of initial data that are unbounded in C(D)C(D), i.e. sets for which we do not have pointwise-in-space control. More precisely, we will prove large deviation principles that hold uniformly over LqL^{q}-bounded subsets of initial data, where q<q<\infty. We assume that bb and σ\sigma are Lipschitz and grow (sub)linearly:

Assumption 0.1.

T(0,)T\in(0,\infty), b,σ:b,\sigma\colon\mathbb{R}\to\mathbb{R} are Lipschitz continuous, λ(0,1]\lambda\in(0,1], and there is a constant CC such that for all zz\in\mathbb{R}:

(0.2) |b(z)|C(1+|z|) and |σ(z)|C(1+|z|λ).\displaystyle\begin{split}&|b(z)|\leq C(1+|z|)\quad\text{ and }\quad|\sigma(z)|\leq C(1+|z|^{\lambda}).\end{split}

While we anticipate that our results generalize to multi-dimensional systems and non-Lipschitz dissipative drifts, our primary focus in this work is establishing uniformity with respect to unbounded sets.

Our first main result is the following. The precise notions of mild solution and ULDP will be discussed in Section 1.

Theorem 0.2 (ULDP on C([0,T],C(D))C([0,T];C(D))).

Let Assumption 0.1 hold. Suppose that q[1,)q\in[1,\infty) and

2λ<q,2\lambda<q,

where λ\lambda is the growth rate of σ\sigma from (0.1). Then, the family {Xxε:ε>0,xC(D)}\{X_{x}^{\varepsilon}:\varepsilon>0,x\in C(D)\} of mild solutions to (0.1) satisfies the ULDP on C([0,T],C(D))C([0,T];C(D)), uniformly over initial data in Lq(D)L^{q}(D)-bounded subsets of C(D)C(D), with rate functions Ix:C([0,T],C(D))[0,+]I_{x}\colon C([0,T];C(D))\to[0,+\infty] given by

(0.3) Ix(φ)12inf{uL2(0,T,L2(D))2:uL2(0,T;L2(D)),φ=Xx0,u},\displaystyle\begin{split}I_{x}(\varphi)\coloneqq\frac{1}{2}\inf\{\|u&\|_{L^{2}(0,T;L^{2}(D))}^{2}:u\in L^{2}(0,T;L^{2}(D)),\,\varphi=X_{x}^{0,u}\},\end{split}

where inf+\inf\varnothing\coloneqq+\infty and Xx0,uX_{x}^{0,u} is the mild solution to

{Xt(t,ξ)=ΔX(t,ξ)+b(X(t,ξ))+σ(X(t,ξ))u(t,ξ),ξD,t(0,T],X(0,ξ)=x(ξ),ξD.\begin{cases}\frac{\partial X}{\partial t}(t,\xi)=\Delta X(t,\xi)+b(X(t,\xi))+\sigma(X(t,\xi))u(t,\xi),\quad\xi\in D,t\in(0,T],\\ X(0,\xi)=x(\xi),\quad\xi\in D.\end{cases}
Remark 0.3.

If λ(0,1/2)\lambda\in(0,1/2), then Theorem 0.2 establishes the ULDP uniformly over L1(D)L^{1}(D)-bounded subsets of initial data in C(D)C(D), in the notably strong C([0,T],C(D))C([0,T];C(D)) topology.

In many applications, the solutions to the SHE are nonnegative and the solutions Xxε(t,ξ)X^{\varepsilon}_{x}(t,\xi) represent mass densities of diffusing molecules. In this setting, the total mass of the system at time tt is the L1L^{1} norm DXxε(t,ξ)𝑑ξ\int_{D}X^{\varepsilon}_{x}(t,\xi)\,\mathrm{d}\xi. Thus, taking q=1q=1, Theorem 0.2 shows that when λ<1/2\lambda<1/2, the solutions XxεX^{\varepsilon}_{x} satisfy a ULDP in the uniform topology that is uniform over sets of initial data xC(D)x\in C(D) with bounded mass. This is a big improvement over previous results that could only prove a ULDP that was uniform over sets of densities bounded in the LL^{\infty} norm. Returning to the motivation discussed above, this improvement paves the way for deriving probabilistic exit time estimates for mass-conserving systems.

Besides initial data in C(D)C(D), we will also consider initial data in Lp(D)L^{p}(D) for p[2,)p\in[2,\infty) and study uniform large deviations in the C([0,T],Lp(D))C([0,T];L^{p}(D)) topology. In this setting, one expects the ULDP to hold in the C([0,T],Lp(D))C([0,T];L^{p}(D)) topology uniformly over Lp(D)L^{p}(D)-bounded sets of initial data, and consequently uniformly over Lq(D)L^{q}(D)-bounded sets for qpq\geq p (since Lq(D)Lp(D)L^{q}(D)\hookrightarrow L^{p}(D)). We cover this expected regime. The core novelty of the current work, however, lies in establishing the ULDP for the regime where q<pq<p and the Lq(D)L^{q}(D) balls are unbounded in the Lp(D)L^{p}(D) topology. Our result is as follows.

Theorem 0.4 (ULDP on C([0,T],Lp(D))C([0,T];L^{p}(D))).

Let Assumption 0.1 hold. Suppose that p[2,),q[1,)p\in[2,\infty),q\in[1,\infty), and

2λpp+2<q,\frac{2\lambda p}{p+2}<q,

where λ\lambda is the growth rate of σ\sigma from (0.1). Then, the family {Xxε:ε>0,xLp(D)}\{X_{x}^{\varepsilon}:\varepsilon>0,x\in L^{p}(D)\} of mild solutions to (0.1) satisfies the ULDP on C([0,T],Lp(D))C([0,T];L^{p}(D)), uniformly over initial data in Lq(D)L^{q}(D)-bounded subsets of Lp(D)L^{p}(D), with rate functions Ix:C([0,T],Lp(D))[0,+]I_{x}\colon C([0,T];L^{p}(D))\to[0,+\infty] given by the formula (0.3).

The lower bound for qq satisfies 2λpp+2<λpp\frac{2\lambda p}{p+2}<\lambda p\leq p, thus indeed, Theorem 0.4 yields an admissible range of q<pq<p for which we have uniform large deviations over Lq(D)L^{q}(D)-bounded – but Lp(D)L^{p}(D)-unbounded – sets of initial data. In particular, we obtain uniformity over L1(D)L^{1}(D)-bounded initial data for all pp if the growth rate λ\lambda is at most 1/21/2, and for a λ\lambda-dependent range of pp if λ(1/2,1)\lambda\in(1/2,1):

Remark 0.5.

If λ(0,1/2]\lambda\in(0,1/2], then the lower bound for qq in Theorem 0.4 is automatically satisfied for q=1q=1 and any p[2,)p\in[2,\infty). Thus we have a ULDP on C([0,T],Lp(D))C([0,T];L^{p}(D)) over L1(D)L^{1}(D)-bounded subsets of initial data in Lp(D)L^{p}(D) for all p[2,)p\in[2,\infty).

If λ(1/2,1)\lambda\in(1/2,1), then we also have the ULDP on C([0,T],Lp(D))C([0,T];L^{p}(D)) with uniformity over L1(D)L^{1}(D)-bounded subsets of Lp(D)L^{p}(D), for all pp in the non-empty range [2,22λ1)[2,\frac{2}{2\lambda-1}). In particular, p=2p=2 is always admissible.

We prove the ULDP theorems by studying the convergence of controlled mild solutions. We study

Xxε,u(t,ξ)=[S(t)x](ξ)+Vxε,u(t,ξ)+Yxε,u(t,ξ)+εZxε,u(t,ξ),\displaystyle X_{x}^{\varepsilon,u}(t,\xi)=[S(t)x](\xi)+V^{\varepsilon,u}_{x}(t,\xi)+Y_{x}^{\varepsilon,u}(t,\xi)+\sqrt{\varepsilon}Z_{x}^{\varepsilon,u}(t,\xi),

where

Vxε,u(t,ξ)=0tDG(ts,ξη)b(Xxε,u(s,η))dηds,Yxε,u(t,ξ)=0tDG(ts,ξη)σ(Xxε,u(s,η))u(s,η)dηds,Zxε,u(t,ξ)=0tDG(ts,ξη)σ(Xxε,u(s,η))W(dηds).\displaystyle\begin{split}V^{\varepsilon,u}_{x}(t,\xi)&=\int_{0}^{t}\int_{D}G(t-s,\xi-\eta)b(X_{x}^{\varepsilon,u}(s,\eta))\,\mathrm{d}\eta\,\mathrm{d}s,\\ Y_{x}^{\varepsilon,u}(t,\xi)&=\int_{0}^{t}\int_{D}G(t-s,\xi-\eta)\sigma(X_{x}^{\varepsilon,u}(s,\eta))u(s,\eta)\,\mathrm{d}\eta\,\mathrm{d}s,\\ Z_{x}^{\varepsilon,u}(t,\xi)&=\int_{0}^{t}\int_{D}G(t-s,\xi-\eta)\sigma(X_{x}^{\varepsilon,u}(s,\eta))W(\mathrm{d}\eta\,\mathrm{d}s).\end{split}

Here, S(t)S(t) is the heat semigroup and G(t,ξ)G(t,\xi) is the heat kernel defined in Subsection 1.1. A key ingredient in our analysis is the semigroup regularization estimate

S(t)xLp(D)t12(1q1p)xLq(D),1q<p.\|S(t)x\|_{L^{p}(D)}\lesssim t^{-\frac{1}{2}(\frac{1}{q}-\frac{1}{p})}\|x\|_{L^{q}(D)},\qquad 1\leq q<p\leq\infty.

The above time singularity shows that S()xLr([0,T],Lp(D))S(\cdot)x\in L^{r}([0,T];L^{p}(D)) for all rr satisfying r2(1q1p)<1\frac{r}{2}(\frac{1}{q}-\frac{1}{p})<1. We demonstrate that under the assumptions of Theorem 0.4 or 0.2, the integral terms Vxε,u,Yxε,uV^{\varepsilon,u}_{x},Y^{\varepsilon,u}_{x}, and Zxε,uZ^{\varepsilon,u}_{x} all belong to C([0,T],Lp(D))C([0,T];L^{p}(D)) (resp. C([0,T]×D)C([0,T]\times D)), and establish important bounds that depend on the Lq(D)L^{q}(D) norm of the initial data xx. These calculations enable us to prove the ULDP and also yield novel existence and uniqueness results for rough initial data. The latter are presented in Theorems 2.5 and 2.7. Smoothing properties of the SHE with rough initial data have previously been studied in [10, 11, 12, 13, 7, 14, 20, 42, 15]. Our focus differs from those results because we establish Lp(D)L^{p}(D) estimates that are uniform over Lq(D)L^{q}(D)-bounded initial data, and we show how the growth rate λ\lambda of σ\sigma affects the regularization.

Building on our well-posedness results and the proofs of the ULDP theorems above, we establish in Section 5 additional ULDPs that hold uniformly over arbitrary bounded sets of Lq(D)L^{q}(D)-initial data. That is, the sets of initial data are no longer assumed to be subsets of C(D)C(D) or Lp(D)L^{p}(D). Because of this, the solutions no longer take values in C([0,T],C(D))C([0,T];C(D)) or C([0,T],Lp(D))C([0,T];L^{p}(D)), so the ULDP in these topologies is inherently ill defined. However, if we subtract off the irregular semigroup term, we can prove that (XxεS()x)(X^{\varepsilon}_{x}-S(\cdot)x) satisfies a ULDP in the C([0,T],Lp(D))C([0,T];L^{p}(D)) (resp. C([0,T]×D)C([0,T]\times D)) topology, see Corollary 5.1. As a further consequence, we also establish ULDPs for (Xxε)(X^{\varepsilon}_{x}) in weaker topologies, such as Lr(0,T,Lp(D))L^{r}(0,T;L^{p}(D))-type topologies, while allowing for initial data belonging only to Lq(D)L^{q}(D).

The organization of this paper is as follows. In Section 1, we review the relevant background theory, specify the solution notion that we adopt, and state sufficient conditions from [39] that we will use to prove our ULDP results. In Section 2, we establish deterministic and stochastic convolution estimates for the stochastic heat equation and for associated control problems that are relevant for the ULDP. Using these estimates, we establish new well-posedness and regularity results in Theorems 2.5 and 2.7. In Sections 3 and 4, we prove our main ULDP results (Theorems 0.2 and 0.4) using the estimates of Section 2 together with an efficient Grönwall scheme. In Section 5, we derive further ULDPs based on these results. In particular, we establish ULDPs for initial data belonging only to Lq(D)L^{q}(D), using alternative topologies.

Acknowledgement

The authors are grateful to Mark Veraar for encouraging and supporting E.T.’s research visit to M.S. at Boston University, where this work was carried out. They also thank Boston University for hosting this visit.

1. Preliminaries

The following notation will be used throughout the paper.

  • The notation ‘\lesssim’ means ‘less than or equal to a constant multiple of’. Subscripts, as in ‘T,p\lesssim_{T,p}’, indicate the parameters on which the implicit constant may depend. In statements of the form ‘\lesssim … a.s.’, the constant is uniform with respect to a.e. ωΩ\omega\in\Omega.

    Although the measure |D|=2π|D|=2\pi of the torus DD is fixed throughout the paper, we retain |D||D| in such subscripts to emphasize when the finite measure of the domain is used.

  • * denotes the convolution operator over the domain DD for the spatial variable, or over [0,T][0,T] for the time variable, as will be clear from the context.

  • \diamond denotes the stochastic convolution with respect to space-time white noise.

In this section, we discuss the theoretical background that is relevant for our ULDP results. From now on, we fix an arbitrary filtered probability space (Ω,,,(t))(\Omega,{\mathscr{F}},{\mathbb{P}},({\mathscr{F}}_{t})) satisfying the usual conditions.

1.1. Heat semigroup

Let G:[0,)×DG\colon[0,\infty)\times D\to\mathbb{R} be the periodic heat kernel defined by

G(t,ξ)12π+k=11πe|k|2tcos(kξ),G(t,\xi)\coloneqq\frac{1}{2\pi}+\sum_{k=1}^{\infty}{\frac{1}{\pi}}e^{-|k|^{2}t}\cos(k\xi),

and define SS through

[S(t)x](ξ)[G(t,)x](ξ)=DG(t,ξη)x(η)𝑑η,t[0,),ξD.[S(t)x](\xi)\coloneqq[G(t,\cdot)*x](\xi)=\int_{D}G(t,\xi-\eta)x(\eta)\,\mathrm{d}\eta,\qquad t\in[0,\infty),\xi\in D.

The kernel GG satisfies the following (see [41, Lem. 2.1]):

(1.1) G(t,)L1(D)=1,\displaystyle\|G(t,\cdot)\|_{L^{1}(D)}=1,
(1.2) G(t,)L(D)C(1+t1/2)C(1+T1/2)t1/2 for all T(0,) and t(0,T],\displaystyle\|G(t,\cdot)\|_{L^{\infty}(D)}\leq C(1+t^{-1/2})\leq C(1+T^{1/2})t^{-1/2}\text{ for all $T\in(0,\infty)$ and $t\in(0,T]$,}

where CC is a constant independent of TT and tt.

As a consequence of (1.1) and Young’s convolution inequality, S(t)(Lp)S(t)\in\mathcal{L}(L^{p}) for all p[1,]p\in[1,\infty]. Moreover, one can show that SS is a strongly continuous semigroup on Lp(D)L^{p}(D), and on C(D)C(D). For all these domains, we call the resulting semigroup the heat semigroup and denote it simply by SS.

The kernel bounds above and interpolation (or Hölder’s inequality) yield for all r[1,]r\in[1,\infty]:

(1.3) G(t,)Lr(D)G(t,)L(D)11rG(t,)L1(D)1rr(1+T12(11r))t12(11r).\|G(t,\cdot)\|_{L^{r}(D)}\leq\|G(t,\cdot)\|_{L^{\infty}(D)}^{1-\frac{1}{r}}\|G(t,\cdot)\|_{L^{1}(D)}^{\frac{1}{r}}\lesssim_{r}(1+T^{\frac{1}{2}(1-\frac{1}{r})})t^{-\frac{1}{2}(1-\frac{1}{r})}.

Combining (1.3) with Young’s convolution inequality, the following smoothing estimate holds for the heat semigroup, for any T(0,)T\in(0,\infty), t(0,T]t\in(0,T], 1qp11\leq q\leq p_{1}\leq\infty and xLq(D)x\in L^{q}(D):

(1.4) S(t)xLp1(D)=G(t,)xLp1(D)G(t,)Lr(D)xLq(D)CT,q,p1t12(1q1p1)xLq(D),\displaystyle\|S(t)x\|_{L^{p_{1}}(D)}=\|G(t,\cdot)*x\|_{L^{p_{1}}(D)}\leq\|G(t,\cdot)\|_{L^{r}(D)}\|x\|_{L^{q}(D)}\leq C_{T,q,p_{1}}t^{-\frac{1}{2}(\frac{1}{q}-\frac{1}{p_{1}})}\|x\|_{L^{q}(D)},

where r[1,]r\in[1,\infty] is such that 1+1p1=1r+1q1+\frac{1}{p_{1}}=\frac{1}{r}+\frac{1}{q}.

Finally, using the semigroup property of SS and the fundamental lemma of the calculus of variations for instance, one can show that the kernel GG satisfies for all 0st<0\leq s\leq t<\infty and ξD\xi\in D:

(1.5) DG(ts,ξη)G(s,η)𝑑η=G(t,ξ).\int_{D}G(t-s,\xi-\eta)G(s,\eta)\,\mathrm{d}\eta=G(t,\xi).

1.2. Space-time white noise

We will use both Walsh’s definition and the functional analytic definition of space-time white noise.

Let T>0T>0, let (ek)(e_{k}) be an orthonormal basis for L2(D)L^{2}(D) and let (Bk)(B_{k}) be a sequence of independent standard (t)({\mathscr{F}}_{t})-Brownian motions. For any progressively measurable process φ:Ω×[0,T]×D\varphi\colon\Omega\times[0,T]\times D\to\mathbb{R} such that φL2(Ω×[0,T]×D)\varphi\in L^{2}(\Omega\times[0,T]\times D), we define

(1.6) 0tDφ(s,η)W(dη𝑑s)k0t(Dφ(s,η)ek(η)𝑑η)dBk(s),t[0,T],\int_{0}^{t}\int_{D}\varphi(s,\eta)W(\mathrm{d}\eta\,\mathrm{d}s)\coloneqq\sum_{k\in\mathbb{N}}\int_{0}^{t}\Big(\int_{D}\varphi(s,\eta)e_{k}(\eta)\,\mathrm{d}\eta\Big)\,\mathrm{d}B_{k}(s),\qquad t\in[0,T],

where the dBk(s)\mathrm{d}B_{k}(s) integral denotes the Itô integral. We call WW the space-time white noise measure.

We can identify WW with an L2(D)L^{2}(D)-cylindrical Brownian motion W(L2(0,T,L2(D)),L2(Ω))W\in\mathcal{L}(L^{2}(0,T;L^{2}(D)),L^{2}(\Omega)), where the latter satisfies W(𝟏(0,t]ek)=Bk(t)W({{\bf 1}}_{(0,t]}\otimes e_{k})=B_{k}(t) for all t(0,)t\in(0,\infty) and kk\in\mathbb{N}.

Let p,p~[2,)p,\tilde{p}\in[2,\infty) and let φ:Ω×[0,T]×DLp(D)\varphi\colon\Omega\times[0,T]\times D\to L^{p}(D) be a progressively measurable process such that φLp~(Ω,Lp(D,L2(0,T,L2(D))))\varphi\in L^{\tilde{p}}(\Omega;L^{p}(D;L^{2}(0,T;L^{2}(D)))), identifying φ(ω,s,η)(ξ)=φ(ω)(ξ)(s)(η)\varphi(\omega,s,\eta)(\xi)=\varphi(\omega)(\xi)(s)(\eta). We can define the Walsh integral 0tDφ(s,η)(ξ)W(dη𝑑s)\int_{0}^{t}\int_{D}\varphi(s,\eta)(\xi)W(\mathrm{d}\eta\,\mathrm{d}s) pointwise in ξ\xi through (1.6) for t[0,T]t\in[0,T]. Then, the following identity holds a.s. in Lp(D)L^{p}(D) (with respect to ξ\xi):

0tDφ(s,η)(ξ)W(dη𝑑s)=[0tΦ𝑑W](ξ),\int_{0}^{t}\int_{D}\varphi(s,\eta)(\xi)W(\mathrm{d}\eta\,\mathrm{d}s)=\Big[\int_{0}^{t}\Phi\,\mathrm{d}W\Big](\xi),

where the dW\mathrm{d}W integral is the stochastic integral in the framework of UMD spaces (see [47]), and

[Φ(ω)g](ξ)0TDφ(ω,s,η)(ξ)g(s,η)𝑑η𝑑s,gL2(0,T,L2(D)).[\Phi(\omega)g](\xi)\coloneqq\int_{0}^{T}\int_{D}\varphi(\omega,s,\eta)(\xi)\,g(s,\eta)\,\mathrm{d}\eta\,\mathrm{d}s,\qquad g\in L^{2}(0,T;L^{2}(D)).

By the γ\gamma-Fubini isomorphism (see [47, (3.1), (3.2)]), Φ\Phi is a random operator such that

Φ(ω)γ(L2(0,T,L2(D)),Lp(D))\Phi(\omega)\in\gamma\bigl(L^{2}(0,T;L^{2}(D)),L^{p}(D)\bigr)

for a.e. ωΩ\omega\in\Omega. Moreover,

Φ(ω)γ(L2(0,T,L2(D)),Lp(D))p(0TD|φ(ω,s,η)()|2dηds)1/2Lp(D).\|\Phi(\omega)\|_{\gamma(L^{2}(0,T;L^{2}(D)),L^{p}(D))}\simeq_{p}\bigg\|\Big(\int_{0}^{T}\int_{D}|\varphi(\omega,s,\eta)(\cdot)|^{2}\,\mathrm{d}\eta\,\mathrm{d}s\Big)^{1/2}\bigg\|_{L^{p}(D)}.

The BDG inequality in the UMD Banach space Lp(D)L^{p}(D) (see [47, (5.5)]) and the isomorphism above yield

(1.7) 𝔼[supt[0,T]0tΦ(s)dW(s)Lp(D)p~]p,p~𝔼[(D(0TD|φ(s,η)(ξ)|2dηds)p2dξ)p~p].{\mathbb{E}}\Big[\sup_{t\in[0,T]}\Big\|\int_{0}^{t}\Phi(s)\mathrm{d}W(s)\Big\|_{L^{p}(D)}^{\tilde{p}}\Big]\simeq_{p,\tilde{p}}{\mathbb{E}}\Big[\Big(\int_{D}\Big(\int_{0}^{T}\int_{D}|\varphi(s,\eta)(\xi)|^{2}\,\mathrm{d}\eta\,\mathrm{d}s\Big)^{\frac{p}{2}}\,\mathrm{d}\xi\Big)^{\frac{\tilde{p}}{p}}\Big].

1.3. Solution notion

For N[0,)N\in[0,\infty), we define

(1.8) 𝒜N{uL2(Ω×[0,T]×D): u is progressively measurable and (uL2(0,T,L2(D))N)=1}.\displaystyle\begin{split}{\mathcal{A}}_{N}\coloneqq\big\{u\in&L^{2}(\Omega\times[0,T]\times D):\\ &\text{ $u$ is progressively measurable and }{\mathbb{P}}\big(\|u\|_{L^{2}(0,T;L^{2}(D))}\leq N\big)=1\big\}.\end{split}

For u𝒜Nu\in{\mathcal{A}}_{N} and ε[0,)\varepsilon\in[0,\infty), we consider the following implicit equation

(1.9) Xxε,u(t,ξ)=[S(t)x](ξ)+Vxε,u(t,ξ)+Yxε,u(t,ξ)+εZxε,u(t,ξ),\displaystyle X_{x}^{\varepsilon,u}(t,\xi)=[S(t)x](\xi)+V^{\varepsilon,u}_{x}(t,\xi)+Y_{x}^{\varepsilon,u}(t,\xi)+\sqrt{\varepsilon}Z_{x}^{\varepsilon,u}(t,\xi),

where

(1.10) Vxε,u(t,ξ)=0tDG(ts,ξη)b(Xxε,u(s,η))dηds,Yxε,u(t,ξ)=0tDG(ts,ξη)σ(Xxε,u(s,η))u(s,η)dηds,Zxε,u(t,ξ)=0tDG(ts,ξη)σ(Xxε,u(s,η))W(dηds).\displaystyle\begin{split}V^{\varepsilon,u}_{x}(t,\xi)&=\int_{0}^{t}\int_{D}G(t-s,\xi-\eta)b(X_{x}^{\varepsilon,u}(s,\eta))\,\mathrm{d}\eta\,\mathrm{d}s,\\ Y_{x}^{\varepsilon,u}(t,\xi)&=\int_{0}^{t}\int_{D}G(t-s,\xi-\eta)\sigma(X_{x}^{\varepsilon,u}(s,\eta))u(s,\eta)\,\mathrm{d}\eta\,\mathrm{d}s,\\ Z_{x}^{\varepsilon,u}(t,\xi)&=\int_{0}^{t}\int_{D}G(t-s,\xi-\eta)\sigma(X_{x}^{\varepsilon,u}(s,\eta))W(\mathrm{d}\eta\,\mathrm{d}s).\end{split}

Here, the dη\mathrm{d}\eta and ds\mathrm{d}s integrals are Lebesgue integrals, and the W(dηds)W(\mathrm{d}\eta\,\mathrm{d}s) integral is the stochastic integral with respect to the space-time white noise WW as defined in Subsection 1.2.

Observe that if u=0u=0, then the equation for Xxε,0X_{x}^{\varepsilon,0} corresponds to the mild formulation of the original stochastic heat equation (0.1) discussed in the introduction. Furthermore, if ε=0\varepsilon=0 and uL2(0,T,L2(D))u\in L^{2}(0,T;L^{2}(D)) (so uu is ω\omega-independent), then the equation for Xx0,uX_{x}^{0,u} corresponds to the so-called skeleton equation that appears in the rate functions IxI_{x} in Theorems 0.2 and 0.4 (see (0.3)).

We define solutions in our framework as follows.

Definition 1.1 (Mild solution).

Let ε,N[0,)\varepsilon,N\in[0,\infty), u𝒜Nu\in{\mathcal{A}}_{N}, xL1(D)x\in L^{1}(D) and let Xxε,u:Ω×[0,T]×DX_{x}^{\varepsilon,u}\colon\Omega\times[0,T]\times D\to\mathbb{R} be a progressively measurable process. If a.s., (1.9) and (1.10) are satisfied for Lebesgue-a.e. t[0,T]t\in[0,T] and ξD\xi\in D (in particular, the integrals appearing therein are well-defined), then we call Xxε,uX_{x}^{\varepsilon,u} a mild solution to (1.9).

If u=0u=0, and Xxε,0X_{x}^{\varepsilon,0} is a mild solution to (1.9) in the sense above, then we call XxεXxε,0X_{x}^{\varepsilon}\coloneqq X_{x}^{\varepsilon,0} a mild solution to (0.1).

1.4. ULDP and sufficient criteria

The Freidlin–Wentzell uniform large deviation principle (with speed ε\varepsilon) is defined as follows.

Definition 1.2 (ULDP).

Let \mathcal{E} be a Polish space, completely metrized by a metric dd. Let (Ω,,)(\Omega,\mathcal{F},\mathbb{P}) be a probability space. Let 0\mathcal{E}_{0} be any set, and let {Xxε:ε>0,x0}\{X^{\varepsilon}_{x}:\varepsilon>0,x\in\mathcal{E}_{0}\} be a family of \mathcal{E}-valued random variables. Let {Ix:x0}\{I_{x}:x\in\mathcal{E}_{0}\} be a collection of lower-semicontinuous maps Ix:[0,+]I_{x}:\mathcal{E}\to[0,+\infty]. Let 𝒜\mathscr{A} be a collection of subsets of 0\mathcal{E}_{0}. Then, the family {Xxε:ε>0,x0}\{X^{\varepsilon}_{x}:\varepsilon>0,x\in\mathcal{E}_{0}\} is said to satisfy a Freidlin–Wentzell uniform large deviation principle (ULDP) with respect to the rate functions IxI_{x} uniformly over 𝒜\mathscr{A}, if

  • For any A𝒜A\in\mathscr{A}, s0>0s_{0}>0, and δ>0\delta>0,

    lim infε0infxAinfφΦx(s0)(εlog(d(Xxε,φ)<δ)+Ix(φ))0.\liminf_{\varepsilon\to 0}\inf_{x\in A}\inf_{\varphi\in\Phi_{x}(s_{0})}\big(\varepsilon\log\mathbb{P}(d(X^{\varepsilon}_{x},\varphi)<\delta)+I_{x}(\varphi)\big)\geq 0.
  • For any A𝒜A\in\mathscr{A}, s0>0s_{0}>0, and δ>0\delta>0,

    lim supε0supxAsups[0,s0](εlog(distd(Xxε,Φx(s))δ)+s)0,\limsup_{\varepsilon\to 0}\sup_{x\in A}\sup_{s\in[0,s_{0}]}\big(\varepsilon\log\mathbb{P}(\mathrm{dist}_{d}(X^{\varepsilon}_{x},\Phi_{x}(s))\geq\delta)+s\big)\leq 0,

where Φx(s){φ:Ix(φ)s}\Phi_{x}(s)\coloneqq\{\varphi\in\mathcal{E}:I_{x}(\varphi)\leq s\}.

From now on, we specialize to XxεX^{\varepsilon}_{x} being a solution to the stochastic heat equation (0.1), and we will let 0=C(D)\mathcal{E}_{0}=C(D), =C([0,T],C(D))\mathcal{E}=C([0,T];C(D)), or 0=Lp(D)\mathcal{E}_{0}=L^{p}(D), =C([0,T],Lp(D))\mathcal{E}=C([0,T];L^{p}(D)), respectively. In these settings, we will consider large deviation principles that hold uniformly over families of LqL^{q}-bounded subsets of C(D)C(D) and Lp(D)L^{p}(D). For q[1,)q\in[1,\infty) and R[0,)R\in[0,\infty), we define

(1.11) BRq{xLq(D):xLq(D)R}.\displaystyle B_{R}^{q}\coloneqq\{x\in L^{q}(D):\|x\|_{L^{q}(D)}\leq R\}.

The next result follows from [39, Th. 2.13]. It provides sufficient criteria for the ULDPs stated in Theorems 0.2 and 0.4, and Corollary 5.4. We will use these criteria to prove the latter theorems. Here, we denote by Xxε,uX_{x}^{\varepsilon,u} the unique mild solution that will be provided by Corollary 2.8.

Theorem 1.3 (Special case of [39, Th. 2.13]).

Let Assumption 0.1 hold. Then:

  1. (1)

    If p,q,p,q, and λ\lambda are as in Theorem 0.2, and for all N,R,δ(0,)N,R,\delta\in(0,\infty),

    limε0supxBRqC(D)supu𝒜N(Xxε,uXx0,uC([0,T],C(D))>δ)=0,\lim_{\varepsilon\downarrow 0}\sup_{x\in B_{R}^{q}\cap C(D)}\sup_{u\in{\mathcal{A}}_{N}}{\mathbb{P}}(\|X_{x}^{\varepsilon,u}-X_{x}^{0,u}\|_{C([0,T];C(D))}>\delta)=0,

    then the family {Xxε,0:ε>0,xC(D)}\{X_{x}^{\varepsilon,0}:\varepsilon>0,x\in C(D)\} satisfies the ULDP uniformly over {BRqC(D):R[0,)}\{B_{R}^{q}\cap C(D):R\in[0,\infty)\}, with the rate functions IxI_{x} stated in Theorem 0.2.

  2. (2)

    If p,q,p,q, and λ\lambda are as in Theorem 0.4 or Corollary 5.4, and for all N,R,δ(0,)N,R,\delta\in(0,\infty),

    limε0supxBRqLp(D)supu𝒜N(Xxε,uXx0,uC([0,T],Lp(D))>δ)=0,\lim_{\varepsilon\downarrow 0}\sup_{x\in B_{R}^{q}\cap L^{p}(D)}\sup_{u\in{\mathcal{A}}_{N}}{\mathbb{P}}(\|X_{x}^{\varepsilon,u}-X_{x}^{0,u}\|_{C([0,T];L^{p}(D))}>\delta)=0,

    then the family {Xxε,0:ε>0,xLp(D)}\{X_{x}^{\varepsilon,0}:\varepsilon>0,x\in L^{p}(D)\} satisfies the ULDP uniformly over {BRqLp(D):R[0,)}\{B_{R}^{q}\cap L^{p}(D):R\in[0,\infty)\}, with the rate functions IxI_{x} stated in Theorem 0.4.

2. Convolution estimates, existence, uniqueness, and regularity

First, in Subsection 2.1, we prove deterministic and stochastic convolution results for the heat semigroup, which will then be used for our well-posedness results in Subsection 2.2 and are also essential for the ULDP proofs later on.

2.1. Deterministic and stochastic convolution estimates

The next lemma demonstrates a regularizing effect – specifically, an improvement in spatial integrability – for the mapping φuS(φu)\varphi u\mapsto S*(\varphi u). This estimate will serve as a crucial tool for bounding terms Yxε,uY_{x}^{\varepsilon,u} and Yx0,uY_{x}^{0,u} in our subsequent proofs.

Lemma 2.1.

Let T,N(0,)T,N\in(0,\infty) and suppose that p1(2,]p_{1}\in(2,\infty], p2[p1,]p_{2}\in[p_{1},\infty], or p1=2p_{1}=2, p2[2,)p_{2}\in[2,\infty). Let uL2(0,T,L2(D))u\in L^{2}(0,T;L^{2}(D)) be such that uL2(0,T,L2(D))N\|u\|_{L^{2}(0,T;L^{2}(D))}\leq N. Let

(2.1) p~(412p1+2p2,)\tilde{p}\in\Big(\frac{4}{1-\frac{2}{p_{1}}+\frac{2}{p_{2}}},\infty\Big)

and let φLp~(0,T,Lp1(D))\varphi\in L^{\tilde{p}}(0,T;L^{p_{1}}(D)). Define

Y(t,ξ)[S[φ()u()]](t)(ξ)=0tDG(ts,ξη)φ(s,η)u(s,η)𝑑η𝑑s.Y(t,\xi)\coloneqq\big[S*[\varphi(\cdot)u(\cdot)]\big](t)(\xi)=\int_{0}^{t}\int_{D}G(t-s,\xi-\eta)\varphi(s,\eta)u(s,\eta)d\eta\,\mathrm{d}s.

Then, it holds that YC([0,T],Lp2(D))Y\in C([0,T];L^{p_{2}}(D)), and for any t[0,T]t\in[0,T],

Y(t,)Lp2(D)p~p~Np~(1+Tp~21)0tφ(s,)Lp1(D)p~ds.\|Y(t,\cdot)\|_{L^{p_{2}}(D)}^{\tilde{p}}\lesssim_{{\tilde{p}}}N^{{\tilde{p}}}(1+T^{\frac{\tilde{p}}{2}-1})\int_{0}^{t}\|\varphi(s,\cdot)\|_{L^{p_{1}}(D)}^{{\tilde{p}}}\,\mathrm{d}s.

Moreover, if p2=p_{2}=\infty, then YC([0,T],C(D))Y\in C([0,T];C(D)).

Proof.

First, by the Cauchy–Schwarz inequality in L2(0,T,L2(D))L^{2}(0,T;L^{2}(D)), we have for all t[0,T]t\in[0,T] and ξD\xi\in D:

|Y(t,ξ)|0tD|G(ts,ξη)φ(s,η)u(s,η)|𝑑η𝑑sN(0tDG2(ts,ξη)|φ(s,η)|2𝑑η𝑑s)12.|Y(t,\xi)|\leq\int_{0}^{t}\int_{D}|G(t-s,\xi-\eta)\varphi(s,\eta)u(s,\eta)|\,\mathrm{d}\eta\,\mathrm{d}s\leq N\Big(\int_{0}^{t}\int_{D}G^{2}(t-s,\xi-\eta)|\varphi(s,\eta)|^{2}\,\mathrm{d}\eta\,\mathrm{d}s\Big)^{\frac{1}{2}}.

If p1[2,)p_{1}\in[2,\infty), then we apply Minkowski’s inequality, followed by Young’s convolution inequality (with powers p22,p12,\frac{p_{2}}{2},\frac{p_{1}}{2}, and ρ11+2p22p1\rho\coloneqq\frac{1}{1+\frac{2}{p_{2}}-\frac{2}{p_{1}}}) and the bound (1.3) for GG, giving

Y(t,)Lp2(D)2\displaystyle\|Y(t,\cdot)\|_{L^{p_{2}}(D)}^{2} =Y(t,)2Lp22(D)\displaystyle=\|Y(t,\cdot)^{2}\|_{L^{\frac{p_{2}}{2}}(D)}
N2ξ0t[G2(ts,)|φ(s,)|2](ξ)dsLp22(D)\displaystyle\leq N^{2}\Big\|\xi\mapsto\int_{0}^{t}\big[G^{2}(t-s,\cdot)*|\varphi(s,\cdot)|^{2}\big](\xi)\,\mathrm{d}s\Big\|_{L^{\frac{p_{2}}{2}}(D)}
N20tG2(ts,)|φ(s,)|2Lp22(D)𝑑s\displaystyle\leq N^{2}\int_{0}^{t}\|G^{2}(t-s,\cdot)*|\varphi(s,\cdot)|^{2}\|_{L^{\frac{p_{2}}{2}}(D)}\,\mathrm{d}s
N20tG(ts,)L2ρ(D)2|φ(s,)|2Lp12(D)𝑑s\displaystyle\leq N^{2}\int_{0}^{t}\|G(t-s,\cdot)\|_{L^{2\rho}(D)}^{2}\||\varphi(s,\cdot)|^{2}\|_{L^{\frac{p_{1}}{2}}(D)}\,\mathrm{d}s
N2(1+T12+1p11p2)0t(ts)121p1+1p2φ(s,)Lp1(D)2𝑑s.\displaystyle\lesssim N^{2}(1+T^{\frac{1}{2}+\frac{1}{p_{1}}-\frac{1}{p_{2}}})\int_{0}^{t}(t-s)^{-\frac{1}{2}-\frac{1}{p_{1}}+\frac{1}{p_{2}}}\|\varphi(s,\cdot)\|_{L^{p_{1}}(D)}^{2}\,\mathrm{d}s.

Note that 121p1+1p2>1-\frac{1}{2}-\frac{1}{p_{1}}+\frac{1}{p_{2}}>-1 by the assumed ranges of p1p_{1} and p2p_{2}.

Consequently, putting θ12+1p11p2\theta\coloneqq\frac{1}{2}+\frac{1}{p_{1}}-\frac{1}{p_{2}} and applying Hölder’s inequality with rp~p~2,r=p~2r\coloneqq\frac{\tilde{p}}{\tilde{p}-2},r^{\prime}=\frac{\tilde{p}}{2}, we have for all t[0,T]t\in[0,T]:

Y(t,)Lp2(D)p~\displaystyle\|Y(t,\cdot)\|_{L^{p_{2}}(D)}^{{\tilde{p}}} Np~(1+Tθ)p~2(0t(ts)θφ(s,)Lp1(D)2𝑑s)p~2\displaystyle\lesssim N^{{\tilde{p}}}(1+T^{\theta})^{\frac{\tilde{p}}{2}}\Big(\int_{0}^{t}(t-s)^{-\theta}\|\varphi(s,\cdot)\|_{L^{p_{1}}(D)}^{2}\,\mathrm{d}s\Big)^{\frac{\tilde{p}}{2}}
Np~(1+Tθ)p~2(0t(ts)θr𝑑s)1rp~2(0tφ(s,)Lp1(D)p~𝑑s)\displaystyle\leq N^{{\tilde{p}}}(1+T^{\theta})^{\frac{\tilde{p}}{2}}\Big(\int_{0}^{t}(t-s)^{-\theta r}\,\mathrm{d}s\Big)^{\frac{1}{r}\frac{\tilde{p}}{2}}\Big(\int_{0}^{t}\|\varphi(s,\cdot)\|_{L^{p_{1}}(D)}^{\tilde{p}}\,\mathrm{d}s\Big)
p~Np~(1+Tθ)p~2Tp~(1θ)220tφ(s,)Lp1(D)p~ds\displaystyle\lesssim_{{\tilde{p}}}N^{{\tilde{p}}}(1+T^{\theta})^{\frac{\tilde{p}}{2}}T^{\frac{\tilde{p}(1-\theta)-2}{2}}\int_{0}^{t}\|\varphi(s,\cdot)\|_{L^{p_{1}}(D)}^{{\tilde{p}}}\,\mathrm{d}s
(2.2) p~Np~(1+Tp~21)0tφ(s,)Lp1(D)p~ds,\displaystyle\lesssim_{\tilde{p}}N^{{\tilde{p}}}(1+T^{\frac{\tilde{p}}{2}-1})\int_{0}^{t}\|\varphi(s,\cdot)\|_{L^{p_{1}}(D)}^{{\tilde{p}}}\,\mathrm{d}s,

where we used that θr>1-\theta r>-1 if and only if

p~>21θ=412p1+2p2,\tilde{p}>\frac{2}{1-\theta}=\frac{4}{1-\frac{2}{p_{1}}+\frac{2}{p_{2}}},

i.e. if and only if (2.1) holds.

If p1=p_{1}=\infty, then p2=p_{2}=\infty and the above still applies with 2=\frac{\infty}{2}=\infty and ρ=1\rho=1. Thus, in either case, we conclude that YL(0,T,Lp2(D))Y\in L^{\infty}(0,T;L^{p_{2}}(D)) and the claimed bound holds.

Next, we prove that YC([0,T],Lp2(D))Y\in C([0,T];L^{p_{2}}(D)) if p2<p_{2}<\infty, by approximating uu and φ\varphi. Assume that uL([0,T]×D)u\in L^{\infty}([0,T]\times D) and φLp~(0,T,Lp2(D))\varphi\in L^{\tilde{p}}(0,T;L^{p_{2}}(D)). Then, φ()u()Lp~(0,T,Lp2(D))\varphi(\cdot)u(\cdot)\in L^{\tilde{p}}(0,T;L^{p_{2}}(D)) and SS is a strongly continuous semigroup on Lp2(D)L^{p_{2}}(D), so by classical semigroup theory we have Y=S(φu)C([0,T],Lp2(D))Y=S*(\varphi u)\in C([0,T];L^{p_{2}}(D)). Consequently, the following bilinear operator is well-defined:

Ψ:L([0,T]×D)×Lp~(0,T,Lp2(D))C([0,T],Lp2(D)):(u,φ)S(φu).\Psi\colon L^{\infty}([0,T]\times D)\times L^{\tilde{p}}(0,T;L^{p_{2}}(D))\to C([0,T];L^{p_{2}}(D))\colon(u,\varphi)\mapsto S*(\varphi u).

By the bound (2.2) (take N=uL2(0,T,L2(D))N=\|u\|_{L^{2}(0,T;L^{2}(D))}), we have

(2.3) Ψ(u,φ)C([0,T],Lp2(D))p~,TuL2(0,T,L2(D))φLp~(0,T,Lp1(D)).\|\Psi(u,\varphi)\|_{C([0,T];L^{p_{2}}(D))}\lesssim_{\tilde{p},T}\|u\|_{L^{2}(0,T;L^{2}(D))}\|\varphi\|_{L^{\tilde{p}}(0,T;L^{p_{1}}(D))}.

Thus, by density of L([0,T]×D)L^{\infty}([0,T]\times D) in L2(0,T,L2(D))L^{2}(0,T;L^{2}(D)) and Lp~(0,T,Lp2(D))L^{\tilde{p}}(0,T;L^{p_{2}}(D)) in Lp~(0,T,Lp1(D))L^{\tilde{p}}(0,T;L^{p_{1}}(D)), Ψ\Psi extends to a bounded bilinear operator on L2(0,T,L2(D))×Lp~(0,T,Lp1(D))L^{2}(0,T;L^{2}(D))\times L^{\tilde{p}}(0,T;L^{p_{1}}(D)), proving the claimed continuity for the case p2<p_{2}<\infty.

It remains to prove that YC([0,T],C(D))Y\in C([0,T];C(D)) when p2=p_{2}=\infty. Since DD is bounded, we may assume without loss of generality that p1<p_{1}<\infty while (2.1) is still satisfied and φLp~(0,T,Lp1(D))\varphi\in L^{\tilde{p}}(0,T;L^{p_{1}}(D)). For u,φC([0,T],C(D))u,\varphi\in C([0,T];C(D)), we have Ψ(u,φ)=S(φu)C([0,T],C(D))\Psi(u,\varphi)=S*(\varphi u)\in C([0,T];C(D)) by classical strongly continuous semigroup theory. Then, from (2.2) and density of C([0,T],C(D))C([0,T];C(D)) in L2(0,T,L2(D))L^{2}(0,T;L^{2}(D)) and Lp~(0,T,Lp1(D))L^{\tilde{p}}(0,T;L^{p_{1}}(D)), it follows that Ψ\Psi extends uniquely to a bounded bilinear map L2(0,T,L2(D))×Lp~(0,T,Lp1(D))C([0,T],C(D))L^{2}(0,T;L^{2}(D))\times L^{\tilde{p}}(0,T;L^{p_{1}}(D))\to C([0,T];C(D)), completing the proof. ∎

Next, we prove a C([0,T],Lp(D))C([0,T];L^{p}(D))-norm moment estimate for stochastic convolutions with the heat semigroup. Here, we denote by \diamond the stochastic convolution in time with respect to the space-time white noise WW, viewed as an L2(D)L^{2}(D)-cylindrical Brownian motion (see Subsection 1.2).

Proposition 2.2.

Let T(0,)T\in(0,\infty), p[2,)p\in[2,\infty) and p~(4,)\tilde{p}\in(4,\infty). Let Φ:(0,T)×ΩLp(D)\Phi\colon(0,T)\times\Omega\to L^{p}(D) be a progressively measurable process such that 𝔼[0TΦ(s)Lp(D)p~𝑑s]<{\mathbb{E}}[\int_{0}^{T}\|\Phi(s)\|_{L^{p}(D)}^{\tilde{p}}\,\mathrm{d}s]<\infty. Let

Z(t)[SΦ](t)=0tS(ts)Φ(s)𝑑W(s),Z(t)\coloneqq[S\diamond\Phi](t)=\textstyle{\int_{0}^{t}}S(t-s)\Phi(s)\,\mathrm{d}W(s),

where SS is the heat semigroup on Lp(D)L^{p}(D). Then, ZLp~(Ω,C([0,T],Lp(D)))Z\in L^{\tilde{p}}(\Omega;C([0,T];L^{p}(D))), and for all T1(0,T]T_{1}\in(0,T],

𝔼[ZC([0,T1],Lp(D))p~]p,p~(1+Tp~4)Tp~41𝔼[0T1Φ(s)Lp(D)p~ds].{\mathbb{E}}\big[\|Z\|_{C([0,T_{1}];L^{p}(D))}^{{\tilde{p}}}\big]\lesssim_{p,\tilde{p}}(1+T^{\frac{\tilde{p}}{4}})T^{\frac{\tilde{p}}{4}-1}{\mathbb{E}}\Big[\int_{0}^{T_{1}}\|\Phi(s)\|_{L^{p}(D)}^{\tilde{p}}\,\mathrm{d}s\Big].
Proof.

It suffices to prove the bound with T1=TT_{1}=T, since one can afterwards apply the bound to Φ𝟏[0,T1]\Phi{{\bf 1}}_{[0,T_{1}]} for T1<TT_{1}<T. Furthermore, it suffices to prove the bound for pointwise bounded Φ\Phi. Indeed, ΦSΦ:Lp~(Ω,Lp~(0,T1,Lp(D)))Lp~(Ω,C([0,T1],Lp(D)))\Phi\mapsto S*\Phi\colon L^{\tilde{p}}(\Omega;L^{\tilde{p}}(0,T_{1};L^{p}(D)))\to L^{\tilde{p}}(\Omega;C([0,T_{1}];L^{p}(D))) is a linear operator, and pointwise bounded, progressively measurable processes are dense in the class of progressively measurable processes in Lp~(Ω,Lp~(0,T1,Lp(D)))L^{\tilde{p}}(\Omega;L^{\tilde{p}}(0,T_{1};L^{p}(D))). Frow now on, we assume that Φ\Phi is pointwise bounded.

We employ the factorization method from [17]. We fix an arbitrary α(1/p~,1/4)\alpha\in(1/\tilde{p},1/4), recalling that p~>4\tilde{p}>4. By (1.1) and Young’s convolution inequality, we have S(t)xLp(D)xLp(D).\|S(t)x\|_{L^{p}(D)}\leq\|x\|_{L^{p}(D)}. Therefore, since p~>1{\tilde{p}}>1 and α>1/p~\alpha>{1}/{\tilde{p}}, [17, Prop. 5.9] yields that

(2.4) Gα(Lp~(0,T;Lp(D)),C([0,T];Lp(D))),Gαp~,αTα1p~,G_{\alpha}\in\mathcal{L}\big(L^{{\tilde{p}}}(0,T;L^{p}(D)),C([0,T];L^{p}(D))\big),\quad\|G_{\alpha}\|\lesssim_{\tilde{p},\alpha}T^{\alpha-\frac{1}{\tilde{p}}},

where Gα(h)(t)0t(ts)α1S(ts)h(s)𝑑sG_{\alpha}(h)(t)\coloneqq\int_{0}^{t}(t-s)^{\alpha-1}S(t-s)h(s)\,\mathrm{d}s. Furthermore, if α(0,1)\alpha\in(0,1), then a.s.

Z=SΦ=sin(απ)πGα(()αSΦ)=sin(απ)πGα(Zα),Z=S\diamond\Phi=\tfrac{\sin(\alpha\pi)}{\pi}G_{\alpha}\big((\cdot)^{-\alpha}S\diamond\Phi\big)=\tfrac{\sin(\alpha\pi)}{\pi}G_{\alpha}(Z_{\alpha}),

where \diamond denotes the stochastic convolution with respect to the space-time white noise ww, and

Zα(t,ξ)[[(()αS)Φ](t)](ξ).Z_{\alpha}(t,\xi)\coloneqq[[((\cdot)^{-\alpha}S)\diamond\Phi](t)](\xi).

Combining the formula for SΦS\diamond\Phi above with (2.4), we have

(2.5) SΦLp~(Ω,C([0,T],Lp(D)))p~p~,αTp~α1ZαLp~(Ω,Lp~(0,T,Lp(D)))p~.\|S\diamond\Phi\|_{L^{\tilde{p}}(\Omega;C([0,T];L^{p}(D)))}^{\tilde{p}}\lesssim_{\tilde{p},\alpha}T^{\tilde{p}\alpha-1}\|Z_{\alpha}\|_{L^{\tilde{p}}(\Omega;L^{{\tilde{p}}}(0,T;L^{p}(D)))}^{\tilde{p}}.

Now, we first apply the BDG inequality in the UMD space Lp(D)L^{p}(D) (see (1.7)) for fixed t[0,T]t\in[0,T]. After that, we apply the bound (1.2) for GG, followed by Minkowski’s inequality (using that p[2,)p\in[2,\infty)) and Young’s convolution inequality in space, giving

Zα(t,)Lp~(Ω,Lp(D))p~\displaystyle\|Z_{\alpha}(t,\cdot)\|_{L^{\tilde{p}}(\Omega;L^{p}(D))}^{\tilde{p}} p,p~𝔼[(0tD(ts)2αG2(ts,η)|Φ(s,η)|2dηds)12Lp(D)p~]\displaystyle\lesssim_{p,\tilde{p}}{\mathbb{E}}\bigg[\Big\|\Big(\int_{0}^{t}\int_{D}(t-s)^{-2\alpha}G^{2}(t-s,\cdot-\eta)|\Phi(s,\eta)|^{2}\,\mathrm{d}\eta\,\mathrm{d}s\Big)^{\frac{1}{2}}\Big\|_{L^{p}(D)}^{\tilde{p}}\bigg]
=𝔼[0tD(ts)2αG2(ts,η)|Φ(s,η)|2dηdsLp2(D)p~2]\displaystyle={\mathbb{E}}\bigg[\Big\|\int_{0}^{t}\int_{D}(t-s)^{-2\alpha}G^{2}(t-s,\cdot-\eta)|\Phi(s,\eta)|^{2}\,\mathrm{d}\eta\,\mathrm{d}s\Big\|_{L^{\frac{p}{2}}(D)}^{\frac{\tilde{p}}{2}}\bigg]
(1+T12)p~2𝔼[0t(ts)2α12[G(ts,)|Φ(s,)|2]𝑑sLp2(D)p~2]\displaystyle\lesssim(1+T^{\frac{1}{2}})^{\frac{\tilde{p}}{2}}{\mathbb{E}}\bigg[\Big\|\int_{0}^{t}(t-s)^{-2\alpha-\frac{1}{2}}[G(t-s,\cdot)*|\Phi(s,\cdot)|^{2}]\,\mathrm{d}s\Big\|_{L^{\frac{p}{2}}(D)}^{\frac{\tilde{p}}{2}}\bigg]
p~(1+Tp~4)𝔼[(0t(ts)2α12G(ts,)|Φ(s,)|2Lp2(D)ds)p~2]\displaystyle\lesssim_{\tilde{p}}(1+T^{\frac{\tilde{p}}{4}}){\mathbb{E}}\bigg[\Big(\int_{0}^{t}(t-s)^{-2\alpha-\frac{1}{2}}\big\|G(t-s,\cdot)*|\Phi(s,\cdot)|^{2}\big\|_{L^{\frac{p}{2}}(D)}\,\mathrm{d}s\Big)^{\frac{\tilde{p}}{2}}\bigg]
(1+Tp~4)𝔼[(0t(ts)2α12G(ts,)L1(D)|Φ(s,)|2Lp2(D)𝑑s)p~2]\displaystyle\leq(1+T^{\frac{\tilde{p}}{4}}){\mathbb{E}}\bigg[\Big(\int_{0}^{t}(t-s)^{-2\alpha-\frac{1}{2}}\big\|G(t-s,\cdot)\|_{L^{1}(D)}\||\Phi(s,\cdot)|^{2}\|_{L^{\frac{p}{2}}(D)}\,\mathrm{d}s\Big)^{\frac{\tilde{p}}{2}}\bigg]
(1+Tp~4)𝔼[(0t(ts)2α12Φ(s,)Lp(D)2𝑑s)p~2].\displaystyle\lesssim(1+T^{\frac{\tilde{p}}{4}}){\mathbb{E}}\bigg[\Big(\int_{0}^{t}(t-s)^{-2\alpha-\frac{1}{2}}\|\Phi(s,\cdot)\|_{L^{p}(D)}^{2}\,\mathrm{d}s\Big)^{\frac{\tilde{p}}{2}}\bigg].

Applying Tonelli’s theorem twice, and applying the above, we obtain

ZαLp~(Ω,Lp~(0,T,Lp(D)))p~\displaystyle\|Z_{\alpha}\|_{L^{\tilde{p}}(\Omega;L^{{\tilde{p}}}(0,T;L^{p}(D)))}^{{\tilde{p}}} =ZαLp~(Ω,Lp(D))p~L1(0,T)\displaystyle=\|\|Z_{\alpha}\|_{L^{\tilde{p}}(\Omega;L^{p}(D))}^{\tilde{p}}\|_{L^{1}(0,T)}
p,p~(1+Tp~4)0T𝔼[(0t(ts)2α12Φ(s,)Lp(D)2ds)p~2]dt\displaystyle\lesssim_{p,\tilde{p}}(1+T^{\frac{\tilde{p}}{4}})\int_{0}^{T}{\mathbb{E}}\bigg[\Big(\int_{0}^{t}(t-s)^{-2\alpha-\frac{1}{2}}\|\Phi(s,\cdot)\|_{L^{p}(D)}^{2}\,\mathrm{d}s\Big)^{\frac{\tilde{p}}{2}}\bigg]\,\mathrm{d}t
=(1+Tp~4)𝔼[0T(0t(ts)2α12Φ(s,)Lp(D)2𝑑s)p~2𝑑t]\displaystyle=(1+T^{\frac{\tilde{p}}{4}}){\mathbb{E}}\bigg[\int_{0}^{T}\Big(\int_{0}^{t}(t-s)^{-2\alpha-\frac{1}{2}}\|\Phi(s,\cdot)\|_{L^{p}(D)}^{2}\,\mathrm{d}s\Big)^{\frac{\tilde{p}}{2}}\,\mathrm{d}t\bigg]
=(1+Tp~4)𝔼[()2α12ΦLp(D)2Lp~2(0,T)p~2]\displaystyle=(1+T^{\frac{\tilde{p}}{4}}){\mathbb{E}}\Big[\big\|(\cdot)^{-2\alpha-\frac{1}{2}}*\|\Phi\|_{L^{p}(D)}^{2}\big\|_{L^{\frac{\tilde{p}}{2}}(0,T)}^{\frac{\tilde{p}}{2}}\Big]
(1+Tp~4)𝔼[()2α12L1(0,T)p~2ΦLp~(0,T,Lp(D))p~]\displaystyle\leq(1+T^{\frac{\tilde{p}}{4}}){\mathbb{E}}\big[\|(\cdot)^{-2\alpha-\frac{1}{2}}\|_{L^{1}(0,T)}^{\frac{\tilde{p}}{2}}\|\Phi\|_{L^{{\tilde{p}}}(0,T;L^{p}(D))}^{\tilde{p}}\big]
α(1+Tp~4)T(2α+12)p~2𝔼[ΦLp~(0,T,Lp(D))p~].\displaystyle\lesssim_{\alpha}(1+T^{\frac{\tilde{p}}{4}})T^{(-2\alpha+\frac{1}{2})\frac{\tilde{p}}{2}}{\mathbb{E}}\big[\|\Phi\|_{L^{\tilde{p}}(0,T;L^{p}(D))}^{\tilde{p}}\big].

In the second-last line, we applied Young’s convolution inequality in time with 1+1p~/2=1+1p~/21+\frac{1}{\tilde{p}/2}=1+\frac{1}{\tilde{p}/2}. In the last line, we used that α<1/4\alpha<1/4. Combining the last estimate with (2.5), we conclude that

𝔼[ZC([0,T],Lp(D))p~]\displaystyle{\mathbb{E}}[\|Z\|_{C([0,T];L^{p}(D))}^{\tilde{p}}] p~,αTp~α1𝔼[ZαLp~([0,T],Lp(D))p~]p,p~,α(1+Tp~4)Tp~41𝔼[ΦLp~(0,T,Lp(D))p~].\displaystyle\lesssim_{\tilde{p},\alpha}T^{\tilde{p}\alpha-1}{\mathbb{E}}[\|Z_{\alpha}\|_{L^{\tilde{p}}([0,T];L^{p}(D))}^{\tilde{p}}]\lesssim_{p,\tilde{p},\alpha}(1+T^{\frac{\tilde{p}}{4}})T^{\frac{\tilde{p}}{4}-1}{\mathbb{E}}\big[\|\Phi\|_{L^{\tilde{p}}(0,T;L^{p}(D))}^{\tilde{p}}\big].

Choosing, for instance, α=12(1p~+14)\alpha=\frac{1}{2}(\frac{1}{\tilde{p}}+\frac{1}{4}), the implicit constant depends only on pp and p~\tilde{p}. ∎

The following estimate for the stochastic convolution, uniform in both time and space, is due to [41, Th. 1.2]. The continuity of ZZ in time and space follows from the proof of [41, Th. 1.2] and the factorization method [17, Prop. 5.9, Th. 5.10] applied with E1=C(D)E_{1}=C(D), E2=Lp~(D)E_{2}=L^{\tilde{p}}(D), p=p~p=\tilde{p}, and r=1/(2p~)r=1/(2\tilde{p}).

Theorem 2.3 ([41, Th. 1.2]).

Let p~(6,){\tilde{p}}\in(6,\infty). Let Φ:Ω×[0,T]Lp~(D)\Phi\colon\Omega\times[0,T]\to L^{\tilde{p}}(D) be a progressively measurable process such that 𝔼[0TΦ(t,)Lp~(D)p~𝑑t]<{\mathbb{E}}\big[\int_{0}^{T}\|\Phi(t,\cdot)\|_{L^{\tilde{p}}(D)}^{\tilde{p}}\,\mathrm{d}t\big]<\infty. Let

Z(t,)[SΦ](t)=0tS(ts)Φ(s)𝑑W(s),Z(t,\cdot)\coloneqq[S\diamond\Phi](t)=\int_{0}^{t}S(t-s)\Phi(s)\,\mathrm{d}W(s),

where SS is the heat semigroup on Lp~(D)L^{\tilde{p}}(D). Then, ZLp~(Ω,C([0,T],C(D)))Z\in L^{\tilde{p}}(\Omega;C([0,T];C(D))), and for all T1(0,T]T_{1}\in(0,T],

𝔼[ZC([0,T1],C(D))p~]p~Tp~432𝔼[0T1Φ(s,)Lp~(D)p~ds].{\mathbb{E}}[\|Z\|_{C([0,T_{1}];C(D))}^{\tilde{p}}]\lesssim_{\tilde{p}}T^{\frac{\tilde{p}}{4}-\frac{3}{2}}{\mathbb{E}}\Big[\int_{0}^{T_{1}}\|\Phi(s,\cdot)\|_{L^{\tilde{p}}(D)}^{\tilde{p}}\,\mathrm{d}s\Big].

2.2. Existence, uniqueness, and regularity for solutions with initial data in LqL^{q}

With the results of the previous subsection at hand, we are ready to begin our study of the controlled SHE for u𝒜Nu\in{\mathcal{A}}_{N}, i.e.

Xxε,u(t,ξ)=[S(t)x](ξ)+Vxε,u(t,ξ)+Yxε,u(t,ξ)+εZxε,u(t,ξ),\displaystyle X_{x}^{\varepsilon,u}(t,\xi)=[S(t)x](\xi)+V^{\varepsilon,u}_{x}(t,\xi)+Y_{x}^{\varepsilon,u}(t,\xi)+\sqrt{\varepsilon}Z_{x}^{\varepsilon,u}(t,\xi),

where Vxε,u(t,ξ),Yxε,u(t,ξ),V^{\varepsilon,u}_{x}(t,\xi),Y_{x}^{\varepsilon,u}(t,\xi), and Zxε,u(t,ξ)Z_{x}^{\varepsilon,u}(t,\xi) are as in (1.10). In the main results of this subsection, we allow for ε[0,)\varepsilon\in[0,\infty), thereby treating the skeleton equation (ε=0\varepsilon=0), the controlled SHE (ε>0\varepsilon>0), and the original SHE (ε>0,u=0\varepsilon>0,u=0) simultaneously.

Throughout this subsection, we let

(2.6) ETC([0,T],Lp(D)),ETC([0,T],Lp(D)),FTC([0,T],C(D)),FTC([0,T],C(D)).\displaystyle\begin{aligned} E_{T}&\coloneqq C([0,T];L^{p}(D)),&\qquad\|\cdot\|_{E_{T}}&\coloneqq\|\cdot\|_{C([0,T];L^{p}(D))},\\ F_{T}&\coloneqq C([0,T];C(D)),&\qquad\|\cdot\|_{F_{T}}&\coloneqq\|\cdot\|_{C([0,T];C(D))}.\end{aligned}

Our goal is to prove well-posedness and regularity results for initial data belonging only to Lq(D)L^{q}(D), allowing q<pq<p and q<q<\infty. In this case, solutions Xxε,uX_{x}^{\varepsilon,u} cannot be ETE_{T}- or FTF_{T}-valued because of the initial condition. However, we will establish well-posedness and regularity by studying Xxε,uS()xX_{x}^{\varepsilon,u}-S(\cdot)x, which exhibits greater regularity. First, in Theorem 2.5, we assume the parameter conditions of Theorem 0.4 and establish these properties using the space ETE_{T}. Then, in Theorem 2.7, we prove an analogous result using the space FTF_{T} under the conditions of Theorem 0.2.

For our first theorem, the following lemma will be essential to setting up a suitable Picard iteration.

Lemma 2.4.

Let the conditions of Theorem 0.4 be satisfied. Let

p~{(4,2pqλpq),if q<λp,(4,),if qλp.\tilde{p}\in\begin{cases}(4,\frac{2pq}{\lambda p-q}),\,&\text{if }q<\lambda p,\\ (4,\infty),&\text{if }q\geq\lambda p.\end{cases}

Let N[0,)N\in[0,\infty), let uL2(0,T,L2(D))u\in L^{2}(0,T;L^{2}(D)) with uL2(0,T,L2(D))N\|u\|_{L^{2}(0,T;L^{2}(D))}\leq N, and let xLq(D)x\in L^{q}(D). Define mappings Ψxb,Ψxσ,u:ETET\Psi_{x}^{b},\Psi_{x}^{\sigma,u}\colon E_{T}\to E_{T} and Ψxσ:Lp~(Ω,ET)Lp~(Ω,ET)\Psi_{x}^{\sigma}\colon L^{\tilde{p}}(\Omega;E_{T})\to L^{\tilde{p}}(\Omega;E_{T}) by

(2.7) Ψxb(ϕ)(t,ξ)0tDG(ts,ξη)b(S(s)x(η)+ϕ(s,η))𝑑η𝑑sΨxσ,u(ϕ)(t,ξ)0tDG(ts,ξη)σ(S(s)x(η)+ϕ(s,η))u(s,η)𝑑η𝑑sΨxσ(ϕ)(t,ξ)0tDG(ts,ξη)σ(S(s)x(η)+ϕ(s,η))W(dηds).\displaystyle\begin{split}\Psi_{x}^{b}(\phi)(t,\xi)&\coloneqq\int_{0}^{t}\int_{D}G(t-s,\xi-\eta)b(S(s)x(\eta)+\phi(s,\eta))\,\mathrm{d}\eta\,\mathrm{d}s\\ \Psi_{x}^{\sigma,u}(\phi)(t,\xi)&\coloneqq\int_{0}^{t}\int_{D}G(t-s,\xi-\eta)\sigma(S(s)x(\eta)+\phi(s,\eta))u(s,\eta)\,\mathrm{d}\eta\,\mathrm{d}s\\ \Psi_{x}^{\sigma}(\phi)(t,\xi)&\coloneqq\int_{0}^{t}\int_{D}G(t-s,\xi-\eta)\sigma(S(s)x(\eta)+\phi(s,\eta))W(\,\mathrm{d}\eta\,\mathrm{d}s).\end{split}

The mappings above are well-defined, and the following Lipschitz bounds hold for all ϕ,ψET\phi,\psi\in E_{T}, ϕ~,ψ~Lp~(Ω,ET)\tilde{\phi},\tilde{\psi}\in L^{\tilde{p}}(\Omega;E_{T}), and t(0,T]t\in(0,T]:

(2.8) Ψxb(ϕ)Ψxb(ψ)Etp~0tT,p~ϕψEsp~ds,Ψxσ,u(ϕ)Ψxσ,u(ψ)Etp~0tN,T,p,p~ϕψEsp~ds,Ψxσ(ϕ~)Ψxσ(ψ~)Lp~(Ω,Et)p~0tT,p,p~ϕ~ψ~Lp~(Ω,Es)p~ds.\displaystyle\begin{split}\|\Psi_{x}^{b}(\phi)-\Psi_{x}^{b}(\psi)\|_{E_{t}}^{\tilde{p}}&\lesssim_{T,\tilde{p}}\int_{0}^{t}\|\phi-\psi\|_{E_{s}}^{\tilde{p}}\,\mathrm{d}s,\\ \|\Psi_{x}^{\sigma,u}(\phi)-\Psi_{x}^{\sigma,u}(\psi)\|_{E_{t}}^{\tilde{p}}&\lesssim_{N,T,p,\tilde{p}}\int_{0}^{t}\|\phi-\psi\|_{E_{s}}^{\tilde{p}}\,\mathrm{d}s,\\ \|\Psi_{x}^{\sigma}(\tilde{\phi})-\Psi_{x}^{\sigma}(\tilde{\psi})\|_{L^{\tilde{p}}(\Omega;E_{t})}^{\tilde{p}}&\lesssim_{T,p,\tilde{p}}\int_{0}^{t}\|\tilde{\phi}-\tilde{\psi}\|_{L^{\tilde{p}}(\Omega;E_{s})}^{\tilde{p}}\,\mathrm{d}s.\end{split}

Furthermore, Ψxb\Psi_{x}^{b} and Ψxσ,u\Psi_{x}^{\sigma,u} with u𝒜Nu\in{\mathcal{A}}_{N} (see (1.8)) are also well-defined as mappings Lp~(Ω,ET)Lp~(Ω,ET)L^{\tilde{p}}(\Omega;E_{T})\to L^{\tilde{p}}(\Omega;E_{T}), and the first two bounds in (2.8) hold a.s. for ϕ,ψLp~(Ω,ET)\phi,\psi\in L^{\tilde{p}}(\Omega;E_{T}).

Proof.

Step 1a (Ψxb\Psi_{x}^{b} is well-defined): Let xL1(D)Lq(D)x\in L^{1}(D)\supset L^{q}(D) and ϕET\phi\in E_{T}. By contractiveness of SS on Lp(D)L^{p}(D), (1.4), and the linear growth in (0.2) for bb, we have

Ψxb(ϕ)(t,)Lp(D)\displaystyle\|\Psi_{x}^{b}(\phi)(t,\cdot)\|_{L^{p}(D)} 0tb(S(s)x+ϕ(s))Lp(D)𝑑s\displaystyle\leq\int_{0}^{t}\|b(S(s)x+\phi(s))\|_{L^{p}(D)}\,\mathrm{d}s
0t1+|S(s)x|+|ϕ(s)|Lp(D)𝑑s\displaystyle\lesssim\int_{0}^{t}\|1+|S(s)x|+|\phi(s)|\|_{L^{p}(D)}\,\mathrm{d}s
0tT,p,|D|1+s12(11p)xL1(D)+ϕ(s)Lp(D)𝑑s\displaystyle\lesssim_{T,p,|D|}\int_{0}^{t}1+s^{-\frac{1}{2}(1-\frac{1}{p})}\|x\|_{L^{1}(D)}+\|\phi(s)\|_{L^{p}(D)}\,\mathrm{d}s
T,p1+xL1(D)+ϕC([0,t],Lp(D)),\displaystyle\lesssim_{T,p}1+\|x\|_{L^{1}(D)}+\|\phi\|_{C([0,t];L^{p}(D))},

using that 12(11p)>12>1-\frac{1}{2}(1-\frac{1}{p})>-\frac{1}{2}>-1. Moreover, by the estimates above, we have b(S()x+ϕ)L1([0,T],Lp(D))b(S(\cdot)x+\phi)\in L^{1}([0,T];L^{p}(D)). By classical continuous semigroup theory, it follows that Ψxb(ϕ)=Sb(S()x+ϕ)C([0,T],Lp(D))=ET\Psi_{x}^{b}(\phi)=S*b(S(\cdot)x+\phi)\in C([0,T];L^{p}(D))=E_{T}. The estimate above gives

Ψxb(ϕ)ETT,p,|D|1+xL1(D)+ϕET,\|\Psi_{x}^{b}(\phi)\|_{E_{T}}\lesssim_{T,p,|D|}1+\|x\|_{L^{1}(D)}+\|\phi\|_{E_{T}},

proving that Ψxb:ETET\Psi_{x}^{b}\colon E_{T}\to E_{T} is well-defined.

Step 1b (Ψxσ,u\Psi_{x}^{\sigma,u} is well-defined): Let xLq(D)x\in L^{q}(D). First, we show that S()xLλp~(0,T,Lλp(D))S(\cdot)x\in L^{\lambda\tilde{p}}(0,T;L^{\lambda p}(D)). If qλpq\leq\lambda p, then the semigroup bound (1.4) gives

S(t)xLλp(D)λp~T,λ,p,p~,qt12(1q1λp)λp~xLq(D)λp~.\|S(t)x\|_{L^{\lambda p}(D)}^{\lambda\tilde{p}}\lesssim_{T,\lambda,p,\tilde{p},q}t^{-\frac{1}{2}(\frac{1}{q}-\frac{1}{\lambda p})\lambda\tilde{p}}\|x\|_{L^{q}(D)}^{\lambda\tilde{p}}.

If q>λpq>\lambda p, then Lq(D)Lλp(D)L^{q}(D)\subset L^{\lambda{p}}(D) by Hölder’s inequality, and using the contractivity of SS on Lq(D)L^{q}(D),

S(t)xLλp(D)λp~λ,p,p~,q,|D|S(t)xLq(D)λp~xLq(D)λp~.\|S(t)x\|_{L^{\lambda p}(D)}^{\lambda\tilde{p}}\lesssim_{\lambda,p,\tilde{p},q,|D|}\|S(t)x\|_{L^{q}(D)}^{\lambda\tilde{p}}\leq\|x\|_{L^{q}(D)}^{\lambda\tilde{p}}.

Combining both cases, we have for all T1(0,T]T_{1}\in(0,T]:

(2.9) 0T1S(s)xLλp(D)λp~𝑑s\displaystyle\int_{0}^{T_{1}}\|S(s)x\|_{L^{\lambda p}(D)}^{\lambda\tilde{p}}\,\mathrm{d}s T,λ,p,p~,q,|D|0T1s12((1q1λp)0)λp~xLq(D)λp~dsT,λ,p,p~,qxLq(D)λp~,\displaystyle\lesssim_{T,\lambda,p,\tilde{p},q,|D|}\int_{0}^{T_{1}}s^{-\frac{1}{2}((\frac{1}{q}-\frac{1}{\lambda p})\vee 0)\lambda\tilde{p}}\|x\|_{L^{q}(D)}^{\lambda\tilde{p}}\,\mathrm{d}s\lesssim_{T,\lambda,p,\tilde{p},q}\|x\|_{L^{q}(D)}^{\lambda\tilde{p}},

provided that

12(1q1λp)λp~>1 if qλp.-\frac{1}{2}(\frac{1}{q}-\frac{1}{\lambda p})\lambda\tilde{p}>-1\quad\text{ if }q\leq\lambda p.

Assuming p~>4\tilde{p}>4 and qλpq\leq\lambda p, the latter holds if and only if

2λpp+2<q,p~(4,2pqλpq),\frac{2\lambda p}{p+2}<q,\quad\tilde{p}\in(4,\frac{2pq}{\lambda p-q}),

which coincides exactly with our assumptions for the case qλpq\leq\lambda p. Thus indeed, S()xLλp~(0,T,Lλp(D))S(\cdot)x\in L^{\lambda\tilde{p}}(0,T;L^{\lambda p}(D)).

Let ϕET\phi\in E_{T}. To prove that Ψxσ,u(ϕ)ET\Psi_{x}^{\sigma,u}(\phi)\in E_{T}, we can apply Lemma 2.1 with p1=p2=pp_{1}=p_{2}=p and φ=σ(S()x+ϕ)\varphi=\sigma(S(\cdot)x+\phi), provided that φ=σ(S()x+ϕ)Lp~([0,T],Lp(D))\varphi=\sigma(S(\cdot)x+\phi)\in L^{\tilde{p}}([0,T];L^{p}(D)). By the growth bound (0.2) for σ\sigma and by (2.9), we have

0Tσ(S(s)x+ϕ(s,))Lp(D)p~𝑑s\displaystyle\int_{0}^{T}\|\sigma(S(s)x+\phi(s,\cdot))\|_{L^{p}(D)}^{\tilde{p}}\,\mathrm{d}s 0Tλ,p~1Lp(D)p~+S(s)xLλp(D)λp~+ϕ(s,)Lλp(D)λp~𝑑s\displaystyle\lesssim_{\lambda,\tilde{p}}\int_{0}^{T}\|1\|_{L^{p}(D)}^{\tilde{p}}+\|S(s)x\|_{L^{\lambda p}(D)}^{\lambda\tilde{p}}+\|\phi(s,\cdot)\|_{L^{\lambda p}(D)}^{\lambda\tilde{p}}\,\mathrm{d}s
T,λ,p,p~,q,|D|1+xLq(D)λp~+TϕETλp~\displaystyle\lesssim_{T,\lambda,p,\tilde{p},q,|D|}1+\|x\|_{L^{q}(D)}^{\lambda\tilde{p}}+T\|\phi\|_{E_{T}}^{\lambda\tilde{p}}
(2.10) T1+xLq(D)λp~+ϕETp~<,\displaystyle\lesssim_{T}1+\|x\|_{L^{q}(D)}^{\lambda\tilde{p}}+\|\phi\|_{E_{T}}^{\tilde{p}}<\infty,

where we used that λ(0,1]\lambda\in(0,1]. Thus, Lemma 2.1 indeed gives

Ψxσ,u(ϕ)ETp~N,T,λ,p,p~,q,|D|1+xLq(D)λp~+ϕETp~<.\|\Psi_{x}^{\sigma,u}(\phi)\|_{E_{T}}^{\tilde{p}}\lesssim_{N,T,\lambda,p,\tilde{p},q,|D|}1+\|x\|_{L^{q}(D)}^{\lambda\tilde{p}}+\|\phi\|_{E_{T}}^{\tilde{p}}<\infty.

Step 1c (Ψxσ\Psi_{x}^{\sigma} is well-defined): Let ϕ~Lp~(Ω,ET)\tilde{\phi}\in L^{\tilde{p}}(\Omega;E_{T}). We apply Proposition 2.2 with p1=p2=pp_{1}=p_{2}=p and φ=σ(S()x+ϕ~)\varphi=\sigma(S(\cdot)x+\tilde{\phi}). Thanks to this result, it suffices to prove that φLp~(Ω,Lp~([0,T],Lp(D)))\varphi\in L^{\tilde{p}}(\Omega;L^{\tilde{p}}([0,T];L^{p}(D))). Applying (2.2) pointwise in ωΩ\omega\in\Omega and taking the expectation yields

(2.11) 𝔼[0Tσ(S(s)x+ϕ~(s,))Lp(D)p~𝑑s]\displaystyle{\mathbb{E}}[\int_{0}^{T}\|\sigma(S(s)x+\tilde{\phi}(s,\cdot))\|_{L^{p}(D)}^{\tilde{p}}\,\mathrm{d}s] T,λ,p,p~,q,|D|1+xLq(D)λp~+𝔼[ϕ~ETp~]<,\displaystyle\lesssim_{T,\lambda,p,\tilde{p},q,|D|}1+\|x\|_{L^{q}(D)}^{\lambda\tilde{p}}+{\mathbb{E}}[\|\tilde{\phi}\|_{E_{T}}^{\tilde{p}}]<\infty,

thus Ψxσ\Psi_{x}^{\sigma} is well-defined.

Step 2a (Ψxb\Psi_{x}^{b} is Lipschitz): Let ϕ,ψET\phi,\psi\in E_{T}. Since SS is contractive on Lp(D)L^{p}(D) and bb is Lipschitz continuous, we have for all 0rtT0\leq r\leq t\leq T:

Ψxb(ϕ)(r)Ψxb(ψ)(r)Lp(D)\displaystyle\|\Psi_{x}^{b}(\phi)(r)-\Psi_{x}^{b}(\psi)(r)\|_{L^{p}(D)} 0rS(rs)(b(S(s)x+ϕ(s))b(S(s)x+ψ(s)))Lp(D)𝑑s\displaystyle\leq\int_{0}^{r}\big\|S(r-s)\big(b(S(s)x+\phi(s))-b(S(s)x+\psi(s))\big)\big\|_{L^{p}(D)}\,\mathrm{d}s
0rb(S(s)x+ϕ(s))b(S(s)x+ψ(s))Lp(D)𝑑s\displaystyle\leq\int_{0}^{r}\|b(S(s)x+\phi(s))-b(S(s)x+\psi(s))\|_{L^{p}(D)}\,\mathrm{d}s
0tϕ(s)ψ(s)Lp(D)𝑑s\displaystyle\lesssim\int_{0}^{t}\|\phi(s)-\psi(s)\|_{L^{p}(D)}\,\mathrm{d}s
0tϕψEs𝑑s.\displaystyle\leq\int_{0}^{t}\|\phi-\psi\|_{E_{s}}\,\mathrm{d}s.

Taking the supremum over r[0,t]r\in[0,t], raising both sides to the p~\tilde{p}-th power, and applying Hölder’s inequality (recall that p~>41\tilde{p}>4\geq 1) yields

Ψxb(ϕ)Ψxb(ψ)Etp~\displaystyle\|\Psi_{x}^{b}(\phi)-\Psi_{x}^{b}(\psi)\|_{E_{t}}^{\tilde{p}} (0tϕψEs𝑑s)p~0tT,p~ϕψEsp~𝑑s.\displaystyle\lesssim\Big(\int_{0}^{t}\|\phi-\psi\|_{E_{s}}\,\mathrm{d}s\Big)^{\tilde{p}}\lesssim_{T,\tilde{p}}\int_{0}^{t}\|\phi-\psi\|_{E_{s}}^{\tilde{p}}\,\mathrm{d}s.

Step 2b (Ψxσ,u\Psi_{x}^{\sigma,u} is Lipschitz): Lemma 2.1 and the Lipschitz continuity of σ\sigma imply that for all t[0,T]t\in[0,T] and ϕ,ψET\phi,\psi\in E_{T},

Ψxσ,u(ϕ)Ψxσ,u(ψ)Etp~\displaystyle\|\Psi_{x}^{\sigma,u}(\phi)-\Psi_{x}^{\sigma,u}(\psi)\|_{E_{t}}^{\tilde{p}} 0tN,T,p,p~σ(S(s)x+ϕ(s,))σ(S(s)x+ψ(s,))Lp(D)p~𝑑s\displaystyle\lesssim_{N,T,p,\tilde{p}}\int_{0}^{t}\|\sigma(S(s)x+\phi(s,\cdot))-\sigma(S(s)x+\psi(s,\cdot))\|_{L^{p}(D)}^{\tilde{p}}\,\mathrm{d}s
0tϕψEsp~𝑑s.\displaystyle\lesssim\int_{0}^{t}\|\phi-\psi\|_{E_{s}}^{\tilde{p}}\,\mathrm{d}s.

Step 2c (Ψxσ\Psi_{x}^{\sigma} is Lipschitz): Proposition 2.2, the Lipschitz continuity of σ\sigma, and Tonelli’s theorem imply that for all t[0,T]t\in[0,T] and ϕ~,ψ~Lp~(Ω,ET)\tilde{\phi},\tilde{\psi}\in L^{\tilde{p}}(\Omega;E_{T}):

Ψσ(ϕ~)Ψσ(ψ~)Lp~(Ω,Et)p~\displaystyle\|\Psi_{\sigma}(\tilde{\phi})-\Psi_{\sigma}(\tilde{\psi})\|_{L^{\tilde{p}}(\Omega;E_{t})}^{\tilde{p}} T,p,p~𝔼[0tσ(S(s)x+ϕ~(s,))σ(S(s)x+ψ~(s,))Lp(D)p~ds]\displaystyle\lesssim_{T,p,\tilde{p}}{\mathbb{E}}\Big[\int_{0}^{t}\|\sigma(S(s)x+\tilde{\phi}(s,\cdot))-\sigma(S(s)x+\tilde{\psi}(s,\cdot))\|_{L^{p}(D)}^{\tilde{p}}\,\mathrm{d}s\Big]
0tϕ~ψ~Lp~(Ω,Es)p~𝑑s.\displaystyle\lesssim\int_{0}^{t}\|\tilde{\phi}-\tilde{\psi}\|_{L^{\tilde{p}}(\Omega;E_{s})}^{\tilde{p}}\,\mathrm{d}s.

Step 3 (Ψxb\Psi_{x}^{b} and Ψxσ,u\Psi_{x}^{\sigma,u} on Lp~(Ω,ET)L^{\tilde{p}}(\Omega;E_{T})): Let ϕ~,ψ~Lp~(Ω,ET)\tilde{\phi},\tilde{\psi}\in L^{\tilde{p}}(\Omega;E_{T}) and u𝒜Nu\in{\mathcal{A}}_{N}. Applying the estimates of Steps 1a and 1b pointwise in ωΩ\omega\in\Omega gives Ψxb(ϕ~),Ψxσ,u(ϕ~)Lp~(Ω,ET)\Psi_{x}^{b}(\tilde{\phi}),\Psi_{x}^{\sigma,u}(\tilde{\phi})\in L^{\tilde{p}}(\Omega;E_{T}), and the Lipschitz estimates of (2.8) apply a.s. ∎

Next, we establish existence and uniqueness for the controlled stochastic heat eaquation (1.9).

Theorem 2.5.

Let the conditions in Theorem 0.4 be satisfied. Let ε,N[0,)\varepsilon,N\in[0,\infty), u𝒜Nu\in{\mathcal{A}}_{N} (defined in (1.8)), and let xLq(D)x\in L^{q}(D).

Then, there exists a mild solution Xxε,uX_{x}^{\varepsilon,u} to (1.9) in the sense of Definition 1.1, which satisfies

Xxε,uLλp~(0,T,Lλp(D))+Lp~(Ω,C([0,T],Lp(D)))Lp~(Ω,Lλp~(0,T,Lλp(D)))\displaystyle X_{x}^{\varepsilon,u}\in L^{\lambda\tilde{p}}(0,T;L^{\lambda p}(D))+L^{\tilde{p}}(\Omega;C([0,T];L^{p}(D)))\subset L^{\tilde{p}}(\Omega;L^{\lambda\tilde{p}}(0,T;L^{\lambda p}(D)))
 for all p~{(4,2pqλpq),if q<λp,(4,),if qλp.\displaystyle\qquad\text{ for all }\quad\tilde{p}\in\begin{cases}(4,\frac{2pq}{\lambda p-q}),\,&\text{if }q<\lambda p,\\ (4,\infty),&\text{if }q\geq\lambda p.\end{cases}

Furthermore, Vxε,uV^{\varepsilon,u}_{x}, Yxε,uY^{\varepsilon,u}_{x}, Zxε,uZ^{\varepsilon,u}_{x} (see (1.10)), and Xxε,uS()xX^{\varepsilon,u}_{x}-S(\cdot)x all belong to Lp~(Ω,C([0,T],Lp(D)))L^{\tilde{p}}(\Omega;C([0,T];L^{p}(D))). Uniqueness holds amongst mild solutions belonging to {S()x+ϕ:ϕLp~(Ω,C([0,T],Lp(D)))}\{S(\cdot)x+\phi:\phi\in L^{\tilde{p}}(\Omega;C([0,T];L^{p}(D)))\}.

Proof.

Define the space ETE_{T} by (2.6). We use a Picard iteration scheme in the space Lp~(Ω,ET)L^{\tilde{p}}(\Omega;E_{T}). Let ϕ00\phi_{0}\coloneqq 0. Using the mappings from Lemma 2.4, define recursively for nn\in\mathbb{N}:

(2.12) ϕn\displaystyle\phi_{n} Ψxb(ϕn1)+Ψxσ,u(ϕn1)+εΨxσ(ϕn1).\displaystyle\coloneqq\Psi_{x}^{b}(\phi_{n-1})+\Psi_{x}^{\sigma,u}(\phi_{n-1})+\sqrt{\varepsilon}\Psi_{x}^{\sigma}(\phi_{n-1}).

By the Lipschitz bounds of Lemma 2.4, we have for all nn\in\mathbb{N} and t[0,T]t\in[0,T]:

ϕn+1ϕnLp~(Ω,Et)p~\displaystyle\|\phi_{n+1}-\phi_{n}\|_{L^{\tilde{p}}(\Omega;E_{t})}^{\tilde{p}} p~Ψxb(ϕn)Ψxb(ϕn1)Lp~(Ω,Et)p~\displaystyle\lesssim_{\tilde{p}}\|\Psi_{x}^{b}(\phi_{n})-\Psi_{x}^{b}(\phi_{n-1})\|_{L^{\tilde{p}}(\Omega;E_{t})}^{\tilde{p}}
+Ψxσ,u(ϕn)Ψxσ,u(ϕn1)Lp~(Ω,Et)p~+εp~/2Ψxσ(ϕn)Ψxσ(ϕn1)Lp~(Ω,Et)p~\displaystyle\qquad+\|\Psi_{x}^{\sigma,u}(\phi_{n})-\Psi_{x}^{\sigma,u}(\phi_{n-1})\|_{L^{\tilde{p}}(\Omega;E_{t})}^{\tilde{p}}+\varepsilon^{\tilde{p}/2}\|\Psi_{x}^{\sigma}(\phi_{n})-\Psi_{x}^{\sigma}(\phi_{n-1})\|_{L^{\tilde{p}}(\Omega;E_{t})}^{\tilde{p}}
0tN,T,p,p~,εϕnϕn1Lp~(Ω,Es)p~𝑑s.\displaystyle\lesssim_{N,T,p,\tilde{p},\varepsilon}\int_{0}^{t}\|\phi_{n}-\phi_{n-1}\|_{L^{\tilde{p}}(\Omega;E_{s})}^{\tilde{p}}\,\mathrm{d}s.

By induction, using the preceding estimate, we obtain for all n0n\in\mathbb{N}_{0} and t[0,T]t\in[0,T]:

ϕn+1ϕnLp~(Ω,Et)p~ϕ1ϕ0Lp~(Ω,ET)p~Cntnn!,\|\phi_{n+1}-\phi_{n}\|_{L^{\tilde{p}}(\Omega;E_{t})}^{\tilde{p}}\leq\|\phi_{1}-\phi_{0}\|_{L^{\tilde{p}}(\Omega;E_{T})}^{\tilde{p}}\frac{C^{n}t^{n}}{n!},

for a constant CC depending only on N,T,p,p~N,T,p,\tilde{p} and ε\varepsilon. Consequently,

n=0ϕn+1ϕnLp~(Ω,ET)ϕ1ϕ0Lp~(Ω,ET)n=0(CnTnn!)1/p~<.\sum_{n=0}^{\infty}\|\phi_{n+1}-\phi_{n}\|_{L^{\tilde{p}}(\Omega;E_{T})}\leq\|\phi_{1}-\phi_{0}\|_{L^{\tilde{p}}(\Omega;E_{T})}\sum_{n=0}^{\infty}\Big(\frac{C^{n}T^{n}}{n!}\Big)^{1/\tilde{p}}<\infty.

Since ϕmϕkLp~(Ω,ET)n=kϕn+1ϕnLp~(Ω,ET)\|\phi_{m}-\phi_{k}\|_{L^{\tilde{p}}(\Omega;E_{T})}\leq\sum_{n=k}^{\infty}\|\phi_{n+1}-\phi_{n}\|_{L^{\tilde{p}}(\Omega;E_{T})}, it follows that (ϕn)n(\phi_{n})_{n} is Cauchy in Lp~(Ω,ET)L^{\tilde{p}}(\Omega;E_{T}), and therefore, it has a limit ϕ\phi in Lp~(Ω,ET)L^{\tilde{p}}(\Omega;E_{T}). Moreover, Tonelli’s theorem yields

𝔼[n=0ϕn+1ϕnET]=n=0𝔼[ϕn+1ϕnET]n=0ϕn+1ϕnLp~(Ω,ET)<.\displaystyle{\mathbb{E}}\Big[\sum_{n=0}^{\infty}\|\phi_{n+1}-\phi_{n}\|_{E_{T}}\Big]=\sum_{n=0}^{\infty}{\mathbb{E}}\Big[\|\phi_{n+1}-\phi_{n}\|_{E_{T}}\Big]\leq\sum_{n=0}^{\infty}\|\phi_{n+1}-\phi_{n}\|_{L^{\tilde{p}}(\Omega;E_{T})}<\infty.

Hence, n=0ϕn+1ϕnET<\sum_{n=0}^{\infty}\|\phi_{n+1}-\phi_{n}\|_{E_{T}}<\infty a.s. By a telescoping sum, it follows that (ϕn)n(\phi_{n})_{n} is Cauchy in ETE_{T} a.s., hence (ϕn)n(\phi_{n})_{n} converges also a.s. to ϕ\phi in ETE_{T} (by uniqueness of limits in probability).

Define XS()x+ϕX\coloneqq S(\cdot)x+\phi and note that XLλp~(0,T,Lλp(D))+Lp~(Ω,ET)X\in L^{\lambda\tilde{p}}(0,T;L^{\lambda p}(D))+L^{\tilde{p}}(\Omega;E_{T}) by (2.9). We prove that XX is a mild solution to (1.9). Since S()xS(\cdot)x is deterministic and Borel measurable, it is progressively measurable. Moreover, since ϕnϕ\phi_{n}\to\phi a.s. in ETE_{T} and each ϕn\phi_{n} is adapted, completeness of the filtration implies that ϕ\phi has an indistinguishable adapted continuous version, which is progressively measurable. Hence XX is indistinguishable from a progressively measurable process. Next, we verify that XX satisfies the mild solution formula. By (2.8) and Tonelli’s theorem,

Ψxb(ϕn)Ψxb(ϕ)Lp~(Ω,ET)+Ψxσ,u(ϕn)Ψxσ,u(ϕ)Lp~(Ω,ET)+Ψxσ(ϕn)Ψxσ(ϕ)Lp~(Ω,ET)\displaystyle\|\Psi_{x}^{b}(\phi_{n})-\Psi_{x}^{b}(\phi)\|_{L^{\tilde{p}}(\Omega;E_{T})}+\|\Psi_{x}^{\sigma,u}(\phi_{n})-\Psi_{x}^{\sigma,u}(\phi)\|_{L^{\tilde{p}}(\Omega;E_{T})}+\|\Psi_{x}^{\sigma}(\phi_{n})-\Psi_{x}^{\sigma}(\phi)\|_{L^{\tilde{p}}(\Omega;E_{T})}
0TN,T,p,p~ϕnϕLp~(Ω,Es)𝑑sTϕnϕLp~(Ω,ET)0as n.\displaystyle\lesssim_{N,T,p,\tilde{p}}\textstyle{\int_{0}^{T}}\|\phi_{n}-\phi\|_{L^{\tilde{p}}(\Omega;E_{s})}\,\mathrm{d}s\leq T\|\phi_{n}-\phi\|_{L^{\tilde{p}}(\Omega;E_{T})}\to 0\qquad\text{as }n\to\infty.

In particular, we conclude that the following convergences hold in Lp(D)L^{p}(D), a.s. for all t[0,T]t\in[0,T]:

limnΨxb(ϕn)(t)=Ψxb(ϕ)(t),limnΨxσ,u(ϕn)(t)=Ψxσ,u(ϕ)(t),limnΨxσ(ϕn)(t)=Ψxσ(ϕ)(t).\lim_{n\to\infty}\Psi_{x}^{b}(\phi_{n})(t)=\Psi_{x}^{b}(\phi)(t),\quad\lim_{n\to\infty}\Psi_{x}^{\sigma,u}(\phi_{n})(t)=\Psi_{x}^{\sigma,u}(\phi)(t),\quad\lim_{n\to\infty}\Psi_{x}^{\sigma}(\phi_{n})(t)=\Psi_{x}^{\sigma}(\phi)(t).

Consequently, we have a.s. for all t[0,T]t\in[0,T], a.e. on DD:

X(t)=S(t)x+ϕ\displaystyle X(t)=S(t)x+\phi =S(t)x+limnϕn(t)\displaystyle=S(t)x+\lim_{n\to\infty}{\phi}_{n}(t)
=S(t)x+limn(Ψxb(ϕn)(t)+Ψxσ,u(ϕn)(t)+εΨxσ(ϕn)(t))\displaystyle=S(t)x+\lim_{n\to\infty}\big(\Psi_{x}^{b}(\phi_{n})(t)+\Psi_{x}^{\sigma,u}(\phi_{n})(t)+\sqrt{\varepsilon}\Psi_{x}^{\sigma}(\phi_{n})(t)\big)
=S(t)x+Ψxb(ϕ)(t)+Ψxσ,u(ϕ)(t)+εΨxσ(ϕ)(t)\displaystyle=S(t)x+\Psi_{x}^{b}(\phi)(t)+\Psi_{x}^{\sigma,u}(\phi)(t)+\sqrt{\varepsilon}\Psi_{x}^{\sigma}(\phi)(t)
=S(t)x+Ψxb(S()x+X)(t)+Ψxσ,u(S()x+X)(t)+εΨxσ(S()x+X)(t).\displaystyle=S(t)x+\Psi_{x}^{b}(-S(\cdot)x+X)(t)+\Psi_{x}^{\sigma,u}(-S(\cdot)x+X)(t)+\sqrt{\varepsilon}\Psi_{x}^{\sigma}(-S(\cdot)x+X)(t).

Recalling the definition of Ψxb,Ψxb,\Psi_{x}^{b},\Psi_{x}^{b}, and Ψxb\Psi_{x}^{b}, this proves that Xxε,uXX_{x}^{\varepsilon,u}\coloneqq X is a mild solution.

For the regularity claims, note that by construction, Xxε,uS()x=ϕLp~(Ω,ET)X_{x}^{\varepsilon,u}-S(\cdot)x=\phi\in L^{\tilde{p}}(\Omega;E_{T}). Moreover,

Vxε,u=Ψxb(Xxε,uS()x),Yxε,u=Ψxσ,u(Xxε,uS()x),Zxε,u=Ψxσ(Xxε,uS()x).V_{x}^{\varepsilon,u}=\Psi_{x}^{b}(X_{x}^{\varepsilon,u}-S(\cdot)x),\quad Y_{x}^{\varepsilon,u}=\Psi_{x}^{\sigma,u}(X_{x}^{\varepsilon,u}-S(\cdot)x),\quad Z_{x}^{\varepsilon,u}=\Psi_{x}^{\sigma}(X_{x}^{\varepsilon,u}-S(\cdot)x).

Since the Ψ\Psi mappings were proved to map Lp~(Ω,ET)Lp~(Ω,ET)L^{\tilde{p}}(\Omega;E_{T})\to L^{\tilde{p}}(\Omega;E_{T}), we conclude that Vxε,u,Yxε,uV_{x}^{\varepsilon,u},Y_{x}^{\varepsilon,u} and Zxε,uZ_{x}^{\varepsilon,u} belong to Lp~(Ω,ET)L^{\tilde{p}}(\Omega;E_{T}).

Finally, we prove uniqueness within Γ{S()x+ϕ:ϕLp~(Ω,ET)}\Gamma\coloneqq\{S(\cdot)x+\phi:\phi\in L^{\tilde{p}}(\Omega;E_{T})\}. If Xxε,u,1,Xxε,u,2X_{x}^{\varepsilon,u,1},X_{x}^{\varepsilon,u,2} are mild solutions and belong to Γ\Gamma, then for j=1,2j=1,2: Xxε,u,jS()xLp~(Ω,ET)X_{x}^{\varepsilon,u,j}-S(\cdot)x\in L^{\tilde{p}}(\Omega;E_{T}) and

Xxε,u,jS()x=Ψxb(Xxε,u,jS()x)+Ψxσ,u(Xxε,u,jS()x)+εΨxσ(Xxε,u,jS()x).X_{x}^{\varepsilon,u,j}-S(\cdot)x=\Psi_{x}^{b}(X_{x}^{\varepsilon,u,j}-S(\cdot)x)+\Psi_{x}^{\sigma,u}(X_{x}^{\varepsilon,u,j}-S(\cdot)x)+\sqrt{\varepsilon}\Psi_{x}^{\sigma}(X_{x}^{\varepsilon,u,j}-S(\cdot)x).

Hence, (2.8) applied with ϕ=Xxε,u,1S()x\phi=X_{x}^{\varepsilon,u,1}-S(\cdot)x and ψ=Xxε,u,2S()x\psi=X_{x}^{\varepsilon,u,2}-S(\cdot)x implies that for all t[0,T]t\in[0,T]:

Xxε,u,1Xxε,u,2Lp~(Ω,Et)0tN,T,p,p~,εXxε,u,1Xxε,u,2Lp~(Ω,Es)𝑑s,\|X_{x}^{\varepsilon,u,1}-{X}_{x}^{\varepsilon,u,2}\|_{L^{\tilde{p}}(\Omega;E_{t})}\lesssim_{N,T,p,\tilde{p},\varepsilon}\int_{0}^{t}\|X_{x}^{\varepsilon,u,1}-{X}_{x}^{\varepsilon,u,2}\|_{L^{\tilde{p}}(\Omega;E_{s})}\,\mathrm{d}s,

and Grönwall’s inequality yields Xxε,u,1Xxε,u,2=0X_{x}^{\varepsilon,u,1}-{X}_{x}^{\varepsilon,u,2}=0 a.s., proving uniqueness. ∎

Next, we work toward our second well-posedness and regularity result, Theorem 2.7, which treats initial data in Lq(D)L^{q}(D) under the parameter conditions of Theorem 0.2. As a preliminary step, we derive a suitable analog of Lemma 2.4 using the space FTF_{T}.

Lemma 2.6.

Let the conditions of Theorem 0.2 hold and let p~(6,3qλ)\tilde{p}\in(6,\frac{3q}{\lambda}). Let N[0,)N\in[0,\infty), let uL2(0,T,L2(D))u\in L^{2}(0,T;L^{2}(D)) with uL2(0,T,L2(D))N\|u\|_{L^{2}(0,T;L^{2}(D))}\leq N, and let xLq(D)x\in L^{q}(D). Define the space FTF_{T} as in (2.6), and define Ψxb,Ψxσ,u:FTFT\Psi_{x}^{b},\Psi_{x}^{\sigma,u}\colon F_{T}\to F_{T} and Ψxσ:Lp~(Ω,FT)Lp~(Ω,FT)\Psi_{x}^{\sigma}\colon L^{\tilde{p}}(\Omega;F_{T})\to L^{\tilde{p}}(\Omega;F_{T}) by the formulas in (2.7).

Then, these mappings are well-defined and the following Lipschitz bounds hold for all ϕ,ψFT\phi,\psi\in F_{T}, ϕ~,ψ~Lp~(Ω,FT)\tilde{\phi},\tilde{\psi}\in L^{\tilde{p}}(\Omega;F_{T}), and t(0,T]t\in(0,T]:

(2.13) Ψxb(ϕ)Ψxb(ψ)Ftp~0tT,p~ϕψFsp~ds,Ψxσ,u(ϕ)Ψxσ,u(ψ)Ftp~0tN,T,p~ϕψFsp~ds,Ψxσ(ϕ~)Ψxσ(ψ~)Lp~(Ω,Ft)p~0tT,p~ϕ~ψ~Lp~(Ω,Fs)p~ds.\displaystyle\begin{split}\|\Psi_{x}^{b}(\phi)-\Psi_{x}^{b}(\psi)\|_{F_{t}}^{\tilde{p}}&\lesssim_{T,\tilde{p}}\int_{0}^{t}\|\phi-\psi\|_{F_{s}}^{\tilde{p}}\,\mathrm{d}s,\\ \|\Psi_{x}^{\sigma,u}(\phi)-\Psi_{x}^{\sigma,u}(\psi)\|_{F_{t}}^{\tilde{p}}&\lesssim_{N,T,\tilde{p}}\int_{0}^{t}\|\phi-\psi\|_{F_{s}}^{\tilde{p}}\,\mathrm{d}s,\\ \|\Psi_{x}^{\sigma}(\tilde{\phi})-\Psi_{x}^{\sigma}(\tilde{\psi})\|_{L^{\tilde{p}}(\Omega;F_{t})}^{\tilde{p}}&\lesssim_{T,\tilde{p}}\int_{0}^{t}\|\tilde{\phi}-\tilde{\psi}\|_{L^{\tilde{p}}(\Omega;F_{s})}^{\tilde{p}}\,\mathrm{d}s.\end{split}

Furthermore, Ψxb\Psi_{x}^{b} and Ψxσ,u\Psi_{x}^{\sigma,u} with u𝒜Nu\in{\mathcal{A}}_{N} (see (1.8)) are also well-defined as mappings Lp~(Ω,FT)Lp~(Ω,FT)L^{\tilde{p}}(\Omega;F_{T})\to L^{\tilde{p}}(\Omega;F_{T}), and the first two bounds in (2.13) hold a.s. for ϕ,ψLp~(Ω,FT)\phi,\psi\in L^{\tilde{p}}(\Omega;F_{T}).

Proof.

The proof of Lemma 2.4 can be copied with the small adjustments indicated below. Let p~,N,u,\tilde{p},N,u, and xx be as in the statement.

Step 1a: We have by the linear growth of bb and (1.4):

b(S(s)x+ϕ(s))L(D)1+S(s)xL(D)+ϕ(s)C(D)T1+s1/2xL1(D)+ϕ(s)C(D).\|b(S(s)x+\phi(s))\|_{L^{\infty}(D)}\lesssim 1+\|S(s)x\|_{L^{\infty}(D)}+\|\phi(s)\|_{C(D)}\lesssim_{T}1+s^{-1/2}\|x\|_{L^{1}(D)}+\|\phi(s)\|_{C(D)}.

Consequently, b(S()x+ϕ)L1(0,T,L(D))b(S(\cdot)x+\phi)\in L^{1}(0,T;L^{\infty}(D)) and since SS is a contractive semigroup on L(D)L^{\infty}(D),

Ψxb(ϕ)(t)L(D)0tb(S(s)x+ϕ(s))L(D)dsT1+xL1(D)+ϕFT.\|\Psi_{x}^{b}(\phi)(t)\|_{L^{\infty}(D)}\leq\int_{0}^{t}\|b(S(s)x+\phi(s))\|_{L^{\infty}(D)}\,\mathrm{d}s\lesssim_{T}1+\|x\|_{L^{1}(D)}+\|\phi\|_{F_{T}}.

For continuity of Ψxb(ϕ)\Psi_{x}^{b}(\phi) in space and time, note that b(S()x+ϕ)L1(0,T,C(D))b(S(\cdot)x+\phi)\in L^{1}(0,T;C(D)) by the smoothing properties of the heat semigroup and Lipschitz continuity of bb. Moreover, SS is strongly continuous on C(D)C(D), so standard semigroup theory yields Sb(S()x+ϕ)C([0,T],C(D))S*b(S(\cdot)x+\phi)\in C([0,T];C(D)).

Step 1b: We derive an Lλp~L^{\lambda\tilde{p}} bound for the term S()xS(\cdot)x. When qλp~q\leq\lambda\tilde{p}, (1.4) guarantees that

S(t)xLλp~(D)T,λ,p~,qt12(1q1λp~)xLq(D).\|S(t)x\|_{L^{\lambda\tilde{p}}(D)}\lesssim_{T,\lambda,\tilde{p},q}t^{-\frac{1}{2}(\frac{1}{q}-\frac{1}{\lambda\tilde{p}})}\|x\|_{L^{q}(D)}.

When q>λp~q>\lambda\tilde{p}, we have Lq(D)Lλp~(D)L^{q}(D)\subset L^{\lambda\tilde{p}}(D) and

S(t)xLλp~(D)λ,p~,q,|D|S(t)xLq(D)xLq(D).\|S(t)x\|_{L^{\lambda\tilde{p}}(D)}\lesssim_{\lambda,\tilde{p},q,|D|}\|S(t)x\|_{L^{q}(D)}\leq\|x\|_{L^{q}(D)}.

Combining both cases, we obtain

(2.14) S(t)xLλp~(D)T,λ,p~,q,|D|t12((1q1λp~)0)xLq(D).\displaystyle\|S(t)x\|_{L^{\lambda\tilde{p}}(D)}\lesssim_{T,\lambda,\tilde{p},q,|D|}t^{-\frac{1}{2}((\frac{1}{q}-\frac{1}{\lambda\tilde{p}})\vee 0)}\|x\|_{L^{q}(D)}.

If qλp~q\leq\lambda\tilde{p}, we have S()xLλp~([0,T]×D)S(\cdot)x\in L^{\lambda\tilde{p}}([0,T]\times D) if 12(1q1λp~)λp~>1-\frac{1}{2}(\frac{1}{q}-\frac{1}{\lambda\tilde{p}})\lambda\tilde{p}>-1, or equivalently, p~<3qλ.\tilde{p}<\frac{3q}{\lambda}. Our assumption that p~(6,3qλ)\tilde{p}\in(6,\tfrac{3q}{\lambda}) thus guarantees that S()xLλp~([0,T]×D)S(\cdot)x\in L^{\lambda\tilde{p}}([0,T]\times D).

Consequently, for ϕFT\phi\in F_{T}, we have φσ(S()x+ϕ)Lp~([0,T]×D)\varphi\coloneqq\sigma(S(\cdot)x+\phi)\in L^{\tilde{p}}([0,T]\times D) by (2.2) applied with p=p~p=\tilde{p} and noting that FTETF_{T}\hookrightarrow E_{T}. Hence, for well-definedness of Ψxσ,u:FTFT\Psi_{x}^{\sigma,u}\colon F_{T}\to F_{T}, we can apply Lemma 2.1 with p1p~p_{1}\coloneqq\tilde{p} and p2p_{2}\coloneqq\infty. Note that for these p1p_{1} and p2p_{2}, (2.1) holds if and only if p~(6,)\tilde{p}\in(6,\infty), which we assumed.

Step 1c: For well-definedness of Ψxσ:Lp~(Ω,FT)Lp~(Ω,FT)\Psi_{x}^{\sigma}\colon L^{\tilde{p}}(\Omega;F_{T})\to L^{\tilde{p}}(\Omega;F_{T}), we apply Theorem 2.3 instead of Proposition 2.2. Note that for ϕLp~(Ω,FT)\phi\in L^{\tilde{p}}(\Omega;F_{T}), we have φσ(S()x+ϕ)Lp~(Ω,Lp~([0,T]×D))\varphi\coloneqq\sigma(S(\cdot)x+\phi)\in L^{\tilde{p}}(\Omega;L^{\tilde{p}}([0,T]\times D)) by (2.11) applied with p=p~p=\tilde{p} and since FTETF_{T}\hookrightarrow E_{T}.

Steps 2a–2c: In the proofs of these steps, we can replace every EtE_{t} by FtF_{t} and Lp(D)L^{p}(D) by L(D)L^{\infty}(D), applying Lemma 2.1 with p1=p2=p_{1}=p_{2}=\infty in Step 2b and applying Theorem 2.3 in Step 2c.

Step 3: In the proof we can replace ETE_{T} by FTF_{T} and Lp(D)L^{p}(D), Lλp(D)L^{\lambda p}(D) by Lp~(D)L^{\tilde{p}}(D), Lλp~(D)L^{\lambda\tilde{p}}(D). ∎

Theorem 2.7.

Let the conditions in Theorem 0.2 be satisfied. Let ε,N[0,)\varepsilon,N\in[0,\infty), u𝒜Nu\in{\mathcal{A}}_{N} (defined in (1.8)), and let xLq(D)x\in L^{q}(D).

Then, there exists a mild solution Xxε,uX_{x}^{\varepsilon,u} to (1.9) in the sense of Definition 1.1, which satisfies

Xxε,uLλp~(0,T,Lλp~(D))+Lp~(Ω,C([0,T],C(D)))Lp~(Ω,Lλp~(0,T,Lλp~(D)))\displaystyle X_{x}^{\varepsilon,u}\in L^{\lambda\tilde{p}}(0,T;L^{\lambda\tilde{p}}(D))+L^{\tilde{p}}(\Omega;C([0,T];C(D)))\subset L^{\tilde{p}}(\Omega;L^{\lambda\tilde{p}}(0,T;L^{\lambda\tilde{p}}(D)))
 for all p~(6,3qλ).\displaystyle\qquad\text{ for all }\quad\tilde{p}\in(6,\frac{3q}{\lambda}).

Furthermore, Vxε,uV^{\varepsilon,u}_{x}, Yxε,uY^{\varepsilon,u}_{x}, Zxε,uZ^{\varepsilon,u}_{x} (see (1.10)), and Xxε,uS()xX^{\varepsilon,u}_{x}-S(\cdot)x all belong to Lp~(Ω,C([0,T],C(D)))L^{\tilde{p}}(\Omega;C([0,T];C(D))). Uniqueness holds amongst mild solutions belonging to {S()x+ϕ:ϕLp~(Ω,C([0,T],C(D)))}\{S(\cdot)x+\phi:\phi\in L^{\tilde{p}}(\Omega;C([0,T];C(D)))\}.

Proof.

We can copy the Picard iteration scheme from the proof of Theorem 2.5 with ETE_{T} replaced by FTF_{T}. Since Lemma 2.6 fully replaces Lemma 2.4, all arguments remain valid and yield existence and uniqueness of mild solutions belonging to {S()x+ϕ:ϕLp~(Ω,C([0,T],C(D)))}\{S(\cdot)x+\phi:\phi\in L^{\tilde{p}}(\Omega;C([0,T];C(D)))\}. Moreover, the regularity claims can be proved analogously, using the splitting Xxε,u=S()x+Vxε,u+Yxε,u+εZxε,uX_{x}^{\varepsilon,u}=S(\cdot)x+V_{x}^{\varepsilon,u}+Y_{x}^{\varepsilon,u}+\sqrt{\varepsilon}Z_{x}^{\varepsilon,u} with Xxε,uS()xFTX_{x}^{\varepsilon,u}-S(\cdot)x\in F_{T}. The latter then implies that Vxε,u=Ψxb(Xxε,uS()x)FTV_{x}^{\varepsilon,u}=\Psi_{x}^{b}(X_{x}^{\varepsilon,u}-S(\cdot)x)\in F_{T}, Yxε,u=Ψσ,u(Xxε,uS()x)FTY_{x}^{\varepsilon,u}=\Psi_{\sigma,u}(X_{x}^{\varepsilon,u}-S(\cdot)x)\in F_{T}, and Zxε,u=Ψσ(Xxε,uS()x)FTZ_{x}^{\varepsilon,u}=\Psi_{\sigma}(X_{x}^{\varepsilon,u}-S(\cdot)x)\in F_{T}. ∎

As a consequence of Theorems 2.5 and 2.7, we obtain the following results for classical initial data and solution spaces. For the uncontrolled case (u=0u=0), corresponding results are classical, see e.g. [17]. For the controlled problem, part (2) below also follows from [40, Th. 7.1], noting that the boundary conditions do not affect that result.

Corollary 2.8 (Well-posedness, classical cases (q=pq=p)).

Suppose that Assumption 0.1 holds. Let ε,N[0,)\varepsilon,N\in[0,\infty) and u𝒜Nu\in{\mathcal{A}}_{N}.

  1. (1)

    If xLp(D)x\in L^{p}(D) and p~(4,)\tilde{p}\in(4,\infty), then there exists a unique mild solution Xxε,uX_{x}^{\varepsilon,u} to (1.9) that belongs to Lp~(Ω,C([0,T],Lp(D)))L^{\tilde{p}}(\Omega;C([0,T];L^{p}(D))).

  2. (2)

    If xC(D)x\in C(D) and p~(6,)\tilde{p}\in(6,\infty), then there exists a unique mild solution Xxε,uX_{x}^{\varepsilon,u} to (1.9) that belongs to Lp~(Ω,C([0,T],C(D)))L^{\tilde{p}}(\Omega;C([0,T];C(D))).

Proof.

(1): Let xLp(D)x\in L^{p}(D) and p~(4,)\tilde{p}\in(4,\infty). The parameter constraint of Theorem 0.4 holds for any λ(0,1]\lambda\in(0,1] if q=pq=p. Theorem 2.5 applied with qpλpq\coloneqq p\geq\lambda p yields the existence of a mild solution Xxε,uX_{x}^{\varepsilon,u} with Xxε,uS()xLp~(Ω,C([0,T],Lp(D)))X_{x}^{\varepsilon,u}-S(\cdot)x\in L^{\tilde{p}}(\Omega;C([0,T];L^{p}(D))). Since SS is a strongly continuous semigroup on Lp(D)L^{p}(D), we also have S()xC([0,T],Lp(D))S(\cdot)x\in C([0,T];L^{p}(D)), thus Xxε,uLp~(Ω,C([0,T],Lp(D)))X_{x}^{\varepsilon,u}\in L^{\tilde{p}}(\Omega;C([0,T];L^{p}(D))) and uniqueness follows from the uniqueness established in Theorem 2.5.

(2): Let xC(D)x\in C(D) and p~(6,)\tilde{p}\in(6,\infty). Fix a sufficiently large q[1,)q\in[1,\infty) such that p~<3q/λ\tilde{p}<3q/\lambda and 2λ<q2\lambda<q. By Theorem 2.7 we obtain a mild solution Xxε,uX_{x}^{\varepsilon,u} with Xxε,uS()xLp~(Ω,C([0,T],C(D)))X_{x}^{\varepsilon,u}-S(\cdot)x\in L^{\tilde{p}}(\Omega;C([0,T];C(D))). Moreover, S()xC([0,T],C(D))S(\cdot)x\in C([0,T];C(D)), thus Xxε,uX_{x}^{\varepsilon,u} has the stated regularity and uniqueness follows from the uniqueness established in Theorem 2.7. ∎

3. ULDP on C([0,T],C(D))C([0,T];C(D)) over LqL^{q}-bounded subsets of C(D)C(D)

In this section we prove Theorem 0.2, using the sufficient condition stated in Theorem 1.3(1). To verify that condition, we will establish uniform estimates for Xxε,uXx0,uLp~(Ω,C([0,T],C(D)))\|X_{x}^{\varepsilon,u}-X_{x}^{0,u}\|_{L^{\tilde{p}}(\Omega;C([0,T];C(D)))}, for a suitable p~>6\tilde{p}>6. Note that we can write

Xxε,uXx0,u=Vxε,uVx0,u+Yxε,uYx0,u+εZxε,u.X_{x}^{\varepsilon,u}-X_{x}^{0,u}=V_{x}^{\varepsilon,u}-V_{x}^{0,u}+Y_{x}^{\varepsilon,u}-Y_{x}^{0,u}+\sqrt{\varepsilon}Z_{x}^{\varepsilon,u}.

As we will show in the proof of Theorem 0.2, the differences Vxε,uVx0,uV_{x}^{\varepsilon,u}-V_{x}^{0,u} and Yxε,uYx0,uY_{x}^{\varepsilon,u}-Y_{x}^{0,u} can be estimated suitably in terms of Xxε,uXx0,uX_{x}^{\varepsilon,u}-X_{x}^{0,u}, and will disappear after an application of Grönwall’s inequality, thus it does not lead to any xx-dependence. In contrast, Zxε,u(t,ξ)Z_{x}^{\varepsilon,u}(t,\xi) does depend on xx, and we need to derive estimates for this term that are uniform with respect to LqL^{q}-bounded initial data xx.

Let us outline how we will deal with the term Zxε,u(t,ξ)Z_{x}^{\varepsilon,u}(t,\xi). Using Theorem 2.3, it will be shown that for a suitable p~\tilde{p}, we have

𝔼[ε1/2Zxε,uC([0,T],C(D))p~]\displaystyle{\mathbb{E}}[\|\varepsilon^{1/2}Z_{x}^{\varepsilon,u}\|_{C([0,T];C(D))}^{\tilde{p}}] T,p~,λεp~/2𝔼[0TD1+|Xxε,uXx0,u|p~+|Xx0,u|λp~dξdt].\displaystyle\lesssim_{T,\tilde{p},\lambda}\varepsilon^{\tilde{p}/2}{\mathbb{E}}\Big[\int_{0}^{T}\int_{D}1+|X_{x}^{\varepsilon,u}-X_{x}^{0,u}|^{\tilde{p}}+|X_{x}^{0,u}|^{\lambda\tilde{p}}\,\mathrm{d}\xi\,\mathrm{d}t\Big].

Since we study the limit ε0\varepsilon\downarrow 0, we can assume that εε0\varepsilon\leq\varepsilon_{0}, and the second term in the sum above will be absorbed through Grönwall’s inequality. Due to the prefactor εp~/2\varepsilon^{\tilde{p}/2} above, it then remains to prove that

supu𝒜NsupxBRqC(D)𝔼[0TD|Xx0,u(t,ξ)|λp~𝑑ξ𝑑t]<.\sup_{u\in{\mathcal{A}}_{N}}\sup_{x\in B_{R}^{q}\cap C(D)}{\mathbb{E}}\Big[\int_{0}^{T}\int_{D}|X_{x}^{0,u}(t,\xi)|^{\lambda\tilde{p}}\,\mathrm{d}\xi\,\mathrm{d}t\Big]<\infty.

In Subsection 3.1, we will deal with Xx0,uX_{x}^{0,u}, and establish an even stronger, almost sure bound, which is uniform over xBRqx\in B^{q}_{R}. In Subsection 3.2, we provide the proof of Theorem 0.2.

3.1. Uniform Lλp~L^{\lambda\tilde{p}}-bound for Xx0,uX_{x}^{0,u}

In the next lemma, we prove an almost sure, uniform bound for solutions Xx0,uX_{x}^{0,u} to equation (1.9) with ε=0\varepsilon=0. This bound is uniform with respect to L2L^{2}-bounded (forcing) functions uu, and uniform with respect to LqL^{q}-bounded initial data xx.

Lemma 3.1.

Let Assumption 0.1 hold. Let N>0N>0 and q[1,)q\in[1,\infty). Suppose that λ[0,q2)(0,1]\lambda\in[0,\frac{q}{2})\cap(0,1] and p~(6,3qλ)\tilde{p}\in(6,\frac{3q}{\lambda}). Then, there exists a constant M>0M>0 such that for all u𝒜Nu\in{\mathcal{A}}_{N} and xLq(D)x\in L^{q}(D):

(3.1) 0TD|Xx0,u(t,ξ)|λp~𝑑ξ𝑑tM(1+xLq(D)λp~)a.s.\int_{0}^{T}\int_{D}|X_{x}^{0,u}(t,\xi)|^{\lambda\tilde{p}}\,\mathrm{d}\xi\,\mathrm{d}t\leq M(1+\|x\|_{L^{q}(D)}^{\lambda\tilde{p}})\quad\text{a.s.}

The value of MM depends only on T,λ,p~,q,|D|,T,\lambda,\tilde{p},q,|D|, and NN.

Proof of Lemma 3.1.

The condition λ<q/2\lambda<{q}/{2} is there only to ensure that the range for p~\tilde{p} is non-empty. Note that Xx0,u(ω)=Xx0,u(ω)X_{x}^{0,u}(\omega)=X_{x}^{0,u(\omega)} for a.e. ωΩ\omega\in\Omega, i.e. stochasticity only enters into Xx0,uX_{x}^{0,u} through u(ω)u(\omega). Therefore, recalling the definition of 𝒜N{\mathcal{A}}_{N}, it suffices to prove that for all uL2(0,T,L2(D))u\in L^{2}(0,T;L^{2}(D)) with uL2(0,T,L2(D))N\|u\|_{L^{2}(0,T;L^{2}(D))}\leq N, the bound in (3.1) holds.

Let uL2(0,T,L2(D))u\in L^{2}(0,T;L^{2}(D)) with uL2(0,T,L2(D))N\|u\|_{L^{2}(0,T;L^{2}(D))}\leq N. We will apply Grönwall’s inequality to tXx0,uLλp~([0,t]×D)t\mapsto\|X_{x}^{0,u}\|_{L^{\lambda\tilde{p}}([0,t]\times D)}. Using the notation of (1.10), we have

Xx0,u(t,ξ)=[S(t)x](ξ)+Vx0,u(t,ξ)+Yx0,u(t,ξ).X_{x}^{0,u}(t,\xi)=[S(t)x](\xi)+V_{x}^{0,u}(t,\xi)+Y_{x}^{0,u}(t,\xi).

By (2.14), we have for any p~(6,3qλ)\tilde{p}\in(6,\frac{3q}{\lambda}):

(3.2) S()xLλp~([0,T]×D)T,λ,p~,q,|D|xLq(D).\|S(\cdot)x\|_{L^{\lambda\tilde{p}}([0,T]\times D)}\lesssim_{T,\lambda,\tilde{p},q,|D|}\|x\|_{L^{q}(D)}.

For Vx0,uV_{x}^{0,u}, assuming for the moment that λp~1\lambda\tilde{p}\geq 1, we have by Minkowki’s inequality, Hölder’s inequality and the linear growth of bb:

Vx0,u(t,)Lλp~(D)λp~\displaystyle\|V_{x}^{0,u}(t,\cdot)\|_{L^{\lambda\tilde{p}}(D)}^{\lambda\tilde{p}} (0tS(ts)b(Xx0,u(s,))Lλp~(D)𝑑s)λp~\displaystyle\leq\Big(\int_{0}^{t}\|S(t-s)b(X_{x}^{0,u}(s,\cdot))\|_{L^{\lambda\tilde{p}}(D)}\,\mathrm{d}s\Big)^{\lambda\tilde{p}}
0tT,λ,p~S(ts)b(Xx0,u(s,))Lλp~(D)λp~𝑑s\displaystyle\lesssim_{T,\lambda,\tilde{p}}\int_{0}^{t}\|S(t-s)b(X_{x}^{0,u}(s,\cdot))\|_{L^{\lambda\tilde{p}}(D)}^{\lambda\tilde{p}}\,\mathrm{d}s
0tb(Xx0,u(s,))Lλp~(D)λp~𝑑s\displaystyle\leq\int_{0}^{t}\|b(X_{x}^{0,u}(s,\cdot))\|_{L^{\lambda\tilde{p}}(D)}^{\lambda\tilde{p}}\,\mathrm{d}s
T,λ,p~,|D|1+Xx0,uLλp~([0,t]×D)λp~.\displaystyle\lesssim_{T,\lambda,\tilde{p},|D|}1+\|X_{x}^{0,u}\|_{L^{\lambda\tilde{p}}([0,t]\times D)}^{\lambda\tilde{p}}.

For the convolution term Yx0,uY_{x}^{0,u}, we apply Lemma 2.1 with p1=p2=p~>6p_{1}=p_{2}=\tilde{p}>6 and φσ(Xx0,u)\varphi\coloneqq\sigma(X_{x}^{0,u}), and use the growth bound (0.2) for σ\sigma and the fact that λ(0,1]\lambda\in(0,1], giving for all t[0,T]t\in[0,T]:

Yx0,u(t,)Lλp~(D)p~λ,p~,|D|Yx0,u(t,)Lp~(D)p~\displaystyle\|Y_{x}^{0,u}(t,\cdot)\|_{L^{\lambda\tilde{p}}(D)}^{\tilde{p}}\lesssim_{\lambda,\tilde{p},|D|}\|Y_{x}^{0,u}(t,\cdot)\|_{L^{\tilde{p}}(D)}^{\tilde{p}} 0tp~,N,Tσ(Xx0,u(s,))Lp~(D)p~𝑑s\displaystyle\lesssim_{\tilde{p},N,T}\int_{0}^{t}\|\sigma(X_{x}^{0,u}(s,\cdot))\|_{L^{\tilde{p}}(D)}^{\tilde{p}}\,\mathrm{d}s
0tp~,|D|(1+Xx0,u(s,)Lλp~(D)λp~)𝑑s\displaystyle\lesssim_{\tilde{p},|D|}\int_{0}^{t}\big(1+\|X_{x}^{0,u}(s,\cdot)\|_{L^{\lambda\tilde{p}}(D)}^{\lambda\tilde{p}}\big)\,\mathrm{d}s
T1+Xx0,uLλp~([0,t]×D)p~.\displaystyle\lesssim_{T}1+\|X_{x}^{0,u}\|_{L^{\lambda\tilde{p}}([0,t]\times D)}^{\tilde{p}}.

Raising this estimate to the power λ\lambda gives

Yx0,u(t,)Lλp~(D)λp~λ,p~,|D|,N,T1+Xx0,uLλp~([0,t]×D)λp~.\displaystyle\|Y_{x}^{0,u}(t,\cdot)\|_{L^{\lambda\tilde{p}}(D)}^{\lambda\tilde{p}}\lesssim_{\lambda,\tilde{p},|D|,N,T}1+\|X_{x}^{0,u}\|_{L^{\lambda\tilde{p}}([0,t]\times D)}^{\lambda\tilde{p}}.

For all t[0,T]t\in[0,T], we conclude that

Xx0,u(t,)Lλp~(D)λp~\displaystyle\|X_{x}^{0,u}(t,\cdot)\|_{L^{\lambda\tilde{p}}(D)}^{\lambda\tilde{p}} λ,p~S(t)xLλp~(D)λp~+Vx0,u(t,)Lλp~(D)λp~+Yx0,u(t,)Lλp~(D)λp~\displaystyle\lesssim_{\lambda,\tilde{p}}\|S(t)x\|_{L^{\lambda\tilde{p}}(D)}^{\lambda\tilde{p}}+\|V_{x}^{0,u}(t,\cdot)\|_{L^{\lambda\tilde{p}}(D)}^{\lambda\tilde{p}}+\|Y_{x}^{0,u}(t,\cdot)\|_{L^{\lambda\tilde{p}}(D)}^{\lambda\tilde{p}}
λ,p~,|D|,N,TS(t)xLλp~(D)λp~+1+Xx0,uLλp~([0,t]×D)λp~.\displaystyle\lesssim_{\lambda,\tilde{p},|D|,N,T}\|S(t)x\|_{L^{\lambda\tilde{p}}(D)}^{\lambda\tilde{p}}+1+\|X_{x}^{0,u}\|_{L^{\lambda\tilde{p}}([0,t]\times D)}^{\lambda\tilde{p}}.

Hence, using also (3.2) and integrating over tt, we find for all 0T1T0\leq T_{1}\leq T:

Xx0,uLλp~([0,T1]×D)λp~\displaystyle\|X_{x}^{0,u}\|_{L^{\lambda\tilde{p}}([0,T_{1}]\times D)}^{\lambda\tilde{p}} =0T1Xx0,u(t,)Lλp~(D)λp~𝑑t\displaystyle=\int_{0}^{T_{1}}\|X_{x}^{0,u}(t,\cdot)\|_{L^{\lambda\tilde{p}}(D)}^{\lambda\tilde{p}}\,\mathrm{d}t
T,λ,p~,q,|D|,NxLq(D)λp~+1+0T1Xx0,uLλp~([0,t]×D)λp~dt.\displaystyle\lesssim_{T,\lambda,\tilde{p},q,|D|,N}\|x\|_{L^{q}(D)}^{\lambda\tilde{p}}+1+\int_{0}^{T_{1}}\|X_{x}^{0,u}\|_{L^{\lambda\tilde{p}}([0,t]\times D)}^{\lambda\tilde{p}}\,\mathrm{d}t.

Grönwall’s inequality yields

Xx0,uLλp~([0,T]×D)λp~C^(xLq(D)λp~+1)exp(C^),\|X_{x}^{0,u}\|_{L^{\lambda\tilde{p}}([0,T]\times D)}^{\lambda\tilde{p}}\leq\hat{C}(\|x\|_{L^{q}(D)}^{\lambda\tilde{p}}+1)\exp(\hat{C}),

where C^\hat{C} is a constant depending only on T,λ,p~,q,|D|,T,\lambda,\tilde{p},q,|D|, and NN. Setting MC^exp(C^)M\coloneqq\hat{C}\exp(\hat{C}), this concludes the proof if λp~1\lambda\tilde{p}\geq 1.

If λp~<1\lambda\tilde{p}<1, then we put p~1/λ\tilde{p}_{*}\coloneqq 1/\lambda. Note that p~λ=1\tilde{p}_{*}\lambda=1 and p~(6,3qλ)\tilde{p}_{*}\in(6,\frac{3q}{\lambda}) since p~>p~>6\tilde{p}_{*}>\tilde{p}>6 and q1q\geq 1. Thus the proof applies with p~=p~\tilde{p}=\tilde{p}_{*} and gives for all u𝒜Nu\in{\mathcal{A}}_{N}:

0TD|Xx0,u(t,ξ)|𝑑ξ𝑑tM(1+xLq(D)) a.s.\int_{0}^{T}\int_{D}|X_{x}^{0,u}(t,\xi)|\,\mathrm{d}\xi\,\mathrm{d}t\leq M(1+\|x\|_{L^{q}(D)})\quad\text{ a.s.}

Combined with Jensen’s inequality (for the concave function ||λp~|\cdot|^{\lambda\tilde{p}}), we obtain a.s.

0TD|Xx0,u(t,ξ)|λp~dξdt|D|,T,λ,p~(0TD|Xx0,u(t,ξ)|dξdt)λp~\displaystyle\int_{0}^{T}\int_{D}|X_{x}^{0,u}(t,\xi)|^{\lambda\tilde{p}}\,\mathrm{d}\xi\,\mathrm{d}t\lesssim_{|D|,T,\lambda,\tilde{p}}\Big(\int_{0}^{T}\int_{D}|X_{x}^{0,u}(t,\xi)|\,\mathrm{d}\xi\,\mathrm{d}t\Big)^{\lambda\tilde{p}} (M(1+xLq(D)))λp~\displaystyle\leq(M(1+\|x\|_{L^{q}(D)}))^{\lambda\tilde{p}}
Mλp~(1+xLq(D)λp~).\displaystyle\leq M^{\lambda\tilde{p}}(1+\|x\|_{L^{q}(D)}^{\lambda\tilde{p}}).

Redefining MM, we again obtain (3.1). ∎

3.2. Conclusion of the ULDP

Combining Lemma 3.1 with results from Section 2, we can now prove the ULDP of Theorem 0.2.

Proof of Theorem 0.2.

Fix any p~(6,3qλ)\tilde{p}\in(6,\frac{3q}{\lambda}) and N,R[0,)N,R\in[0,\infty). For all u𝒜Nu\in{\mathcal{A}}_{N} and xLq(D)x\in L^{q}(D), Theorem 2.7 provides a unique mild solution Xxε,uLp~(Ω,Lλp~(0,T,Lλp~(D)))X_{x}^{\varepsilon,u}\in L^{\tilde{p}}(\Omega;L^{\lambda\tilde{p}}(0,T;L^{\lambda\tilde{p}}(D))) to (1.9). Using this regularity and the growth condition in (0.2), we have σ(Xxε,u)Lp~(Ω,Lp~(0,T,Lp~(D)))\sigma(X_{x}^{\varepsilon,u})\in L^{\tilde{p}}(\Omega;L^{\tilde{p}}(0,T;L^{\tilde{p}}(D))). Therefore, we may apply Theorem 2.3 with Φσ(Xxε,u)\Phi\coloneqq\sigma(X_{x}^{\varepsilon,u}), giving together with (0.2):

𝔼[ε1/2Zxε,u\displaystyle{\mathbb{E}}[\|\varepsilon^{1/2}Z_{x}^{\varepsilon,u} C([0,T],C(D))p~]p~,Tεp~/2𝔼[0TD1+|Xxε,u|λp~dξdt]\displaystyle\|_{C([0,T];C(D))}^{\tilde{p}}]\lesssim_{\tilde{p},T}\varepsilon^{\tilde{p}/2}{\mathbb{E}}\Big[\int_{0}^{T}\int_{D}1+|X_{x}^{\varepsilon,u}|^{\lambda\tilde{p}}\,\mathrm{d}\xi\,\mathrm{d}t\Big]
λ,p~εp~/2𝔼[0TD1+|Xxε,uXx0,u|p~+|Xx0,u|λp~dξdt]\displaystyle\lesssim_{\lambda,\tilde{p}}\varepsilon^{\tilde{p}/2}{\mathbb{E}}\Big[\int_{0}^{T}\int_{D}1+|X_{x}^{\varepsilon,u}-X_{x}^{0,u}|^{\tilde{p}}+|X_{x}^{0,u}|^{\lambda\tilde{p}}\,\mathrm{d}\xi\,\mathrm{d}t\Big]
T,|D|εp~/2(1+0T𝔼[Xxε,uXx0,uC([0,t],C(D))p~]dt+𝔼[0TD|Xx0,u|λp~dξdt]).\displaystyle\lesssim_{T,|D|}\varepsilon^{\tilde{p}/2}\bigg(1+\int_{0}^{T}{\mathbb{E}}\big[\|X_{x}^{\varepsilon,u}-X_{x}^{0,u}\|_{C([0,t];C(D))}^{\tilde{p}}\big]\,\mathrm{d}t+{\mathbb{E}}\Big[\int_{0}^{T}\int_{D}|X_{x}^{0,u}|^{\lambda\tilde{p}}\,\mathrm{d}\xi\,\mathrm{d}t\Big]\bigg).

In the third line we used that (|z1|+|z2|)λp~2λp~1(|z1|λp~+|z2|λp~)(|z_{1}|+|z_{2}|)^{\lambda\tilde{p}}\leq 2^{\lambda\tilde{p}-1}(|z_{1}|^{\lambda\tilde{p}}+|z_{2}|^{\lambda\tilde{p}}), and we used that |z|λ1+|z||z|^{\lambda}\leq 1+|z| since λ(0,1]\lambda\in(0,1]. For the last term in the upper bound above, we have by Lemma 3.1 (see (1.11)):

Msupu𝒜NsupxBRq𝔼[0TD|Xx0,u(t,ξ)|λp~dξdt]T,λ,p~,q,|D|,N,R1<.M_{*}\coloneqq\sup_{u\in{\mathcal{A}}_{N}}\sup_{x\in B_{R}^{q}}{\mathbb{E}}\Big[\int_{0}^{T}\int_{D}|X_{x}^{0,u}(t,\xi)|^{\lambda\tilde{p}}\,\mathrm{d}\xi\,\mathrm{d}t\Big]\lesssim_{T,\lambda,\tilde{p},q,|D|,N,R}1<\infty.

Furthermore, by (2.13) of Lemma 2.6 (Vxε,u=Ψxb(Xxε,uS()x)V_{x}^{\varepsilon,u}=\Psi_{x}^{b}(X_{x}^{\varepsilon,u}-S(\cdot)x), Yxε,u=Ψxσ,u(Xxε,uS()x)Y_{x}^{\varepsilon,u}=\Psi_{x}^{\sigma,u}(X_{x}^{\varepsilon,u}-S(\cdot)x) for ε0\varepsilon\geq 0) and the Tonelli’s theorem, we have for all u𝒜Nu\in{\mathcal{A}}_{N} and xLq(D)x\in L^{q}(D):

𝔼[\displaystyle{\mathbb{E}}[ Vxε,uVx0,uC([0,T1],C(D))p~]+𝔼[Yxε,uYx0,uC([0,T1],C(D))p~]\displaystyle\|V_{x}^{\varepsilon,u}-V_{x}^{0,u}\|_{C([0,T_{1}];C(D))}^{\tilde{p}}]+{\mathbb{E}}[\|Y_{x}^{\varepsilon,u}-Y_{x}^{0,u}\|_{C([0,T_{1}];C(D))}^{\tilde{p}}]
0T1N,T,p~𝔼[Xxε,uXx0,uC([0,t],C(D))p~]𝑑t.\displaystyle\lesssim_{N,T,\tilde{p}}\int_{0}^{T_{1}}{\mathbb{E}}[\|X_{x}^{\varepsilon,u}-X_{x}^{0,u}\|_{C([0,t];C(D))}^{\tilde{p}}]\,\mathrm{d}t.

Combining all estimates above, we find that for all 0T1T0\leq T_{1}\leq T, ε0(0,)\varepsilon_{0}\in(0,\infty), ε(0,ε0]\varepsilon\in(0,\varepsilon_{0}], u𝒜Nu\in{\mathcal{A}}_{N}, and xBRqx\in B_{R}^{q}, it holds that

𝔼[\displaystyle{\mathbb{E}}[ Xxε,uXx0,uC([0,T1],C(D))p~]\displaystyle\|X_{x}^{\varepsilon,u}-X_{x}^{0,u}\|_{C([0,T_{1}];C(D))}^{\tilde{p}}]
p~𝔼[Vxε,uVx0,uC([0,T1],C(D))p~]+𝔼[Yxε,uYx0,uC([0,T1],C(D))p~]+εp~/2𝔼[Zxε,uC([0,T1],C(D))p~]\displaystyle\lesssim_{\tilde{p}}{\mathbb{E}}[\|V_{x}^{\varepsilon,u}-V_{x}^{0,u}\|_{C([0,T_{1}];C(D))}^{\tilde{p}}]+{\mathbb{E}}[\|Y_{x}^{\varepsilon,u}-Y_{x}^{0,u}\|_{C([0,T_{1}];C(D))}^{\tilde{p}}]+\varepsilon^{\tilde{p}/2}{\mathbb{E}}[\|Z_{x}^{\varepsilon,u}\|_{C([0,T_{1}];C(D))}^{\tilde{p}}]
0T1T,λ,p~,q,|D|,N,R(1+ε0p~/2)𝔼[Xxε,uXx0,uC([0,t],C(D))p~]𝑑t+εp~/2(1+M).\displaystyle\lesssim_{T,\lambda,\tilde{p},q,|D|,N,R}\int_{0}^{T_{1}}(1+\varepsilon_{0}^{\tilde{p}/2}){\mathbb{E}}[\|X_{x}^{\varepsilon,u}-X_{x}^{0,u}\|_{C([0,t];C(D))}^{\tilde{p}}]\,\mathrm{d}t+\varepsilon^{\tilde{p}/2}(1+M_{*}).

Grönwall’s inequality gives that

𝔼[Xxε,uXx0,uC([0,T],C(D))p~]εp~/2C1,{\mathbb{E}}[\|X_{x}^{\varepsilon,u}-X_{x}^{0,u}\|_{C([0,T];C(D))}^{\tilde{p}}]\leq\varepsilon^{\tilde{p}/2}C_{1},

where C1C_{1} is a constant that depends only on T,λ,p~,q,|D|,N,RT,\lambda,\tilde{p},q,|D|,N,R, and ε0\varepsilon_{0}. Finally, Chebychev’s inequality provides the convergence in probability as ε0\varepsilon\to 0 that allows us to apply Theorem 1.3(1), completing the proof of the ULDP. ∎

4. ULDP on C([0,T],Lp(D))C([0,T];L^{p}(D)) over LqL^{q}-bounded subsets of Lp(D)L^{p}(D)

In this section, we prove Theorem 0.4. We will verify the sufficient criterion from Theorem 1.3(2), using an analogous series of estimates as in Section 3, but with Lp(D)L^{p}(D) replacing C(D)C(D). Eventually, we will prove uniform bounds for Xxε,uXx0,uLp~(Ω,C([0,T],Lp(D)))\|X_{x}^{\varepsilon,u}-X_{x}^{0,u}\|_{L^{\tilde{p}}(\Omega;C([0,T];L^{p}(D)))} for a suitable p~>4{\tilde{p}}>4.

Recall the splitting Xxε,uXx0,u=Vxε,uVx0,u+Yxε,uYx0,u+εZxε,uX_{x}^{\varepsilon,u}-X_{x}^{0,u}=V_{x}^{\varepsilon,u}-V_{x}^{0,u}+Y_{x}^{\varepsilon,u}-Y_{x}^{0,u}+\sqrt{\varepsilon}Z_{x}^{\varepsilon,u} (see (1.9)). Again, we will absorb the differences Vxε,uVx0,uV_{x}^{\varepsilon,u}-V_{x}^{0,u} and Yxε,uYx0,uY_{x}^{\varepsilon,u}-Y_{x}^{0,u} in our Grönwall estimate, and are left with bounding Zxε,uZ_{x}^{\varepsilon,u}. As in Section 3, we will reduce this problem to bounding Xx0,uX_{x}^{0,u} uniformly in uu, xx, and ε\varepsilon. The latter will be achieved in Subsection 4.1.

4.1. Uniform Lλp~×LpL^{\lambda\tilde{p}}\times L^{p}-bound for Xx0,uX_{x}^{0,u}

The following result will be crucial for controlling the term Zxε,uZ^{\varepsilon,u}_{x} in the proof of our main result. It can be viewed as an analogue of Lemma 3.1. We note that the stated range for p~\tilde{p} is always non-empty: this follows from (4.1) when q<λpq<\lambda p, and holds trivially when qλpq\geq\lambda p.

Lemma 4.1.

Let Assumption 0.1 hold. Let N>0N>0. Suppose that p[2,),q[1,)p\in[2,\infty),q\in[1,\infty) and λ(0,1]\lambda\in(0,1] satisfy

(4.1) 2λpp+2<q.\frac{2\lambda p}{p+2}<q.

Suppose that p~(4,2pqλpq)\tilde{p}\in(4,\frac{2pq}{\lambda p-q}) if q<λpq<\lambda p, and p~(4,)\tilde{p}\in(4,\infty) if qλpq\geq\lambda p.

Then, there exists a constant M[0,)M\in[0,\infty) such that for all u𝒜Nu\in{\mathcal{A}}_{N} and xLq(D)x\in L^{q}(D):

0TXx0,u(s)Lλp(D)λp~𝑑sM(1+xLq(D)λp~)a.s.\int_{0}^{T}\|X_{x}^{0,u}(s)\|_{L^{\lambda p}(D)}^{\lambda\tilde{p}}\,\mathrm{d}s\leq M(1+\|x\|_{L^{q}(D)}^{\lambda\tilde{p}})\quad\text{a.s.}

The constant MM depends only on T,λ,p,p~,q,|D|,T,\lambda,p,\tilde{p},q,|D|, and NN.

Proof.

Let N[0,)N\in[0,\infty) and xLq(D)x\in L^{q}(D). Since Xx0,u(ω)=Xx0,u(ω)X_{x}^{0,u}(\omega)=X_{x}^{0,u(\omega)} for a.e. ωΩ\omega\in\Omega, and by the definition of 𝒜N{\mathcal{A}}_{N}, it suffices to prove the bound for uL2(0,T,L2(D))u\in L^{2}(0,T;L^{2}(D)) with uL2(0,T,L2(D))N\|u\|_{L^{2}(0,T;L^{2}(D))}\leq N. We will apply Grönwall’s inequality to T1Xx0,uLλp~(0,T1,Lλp(D))λp~T_{1}\mapsto\|X_{x}^{0,u}\|_{L^{\lambda\tilde{p}}(0,T_{1};L^{\lambda p}(D))}^{\lambda\tilde{p}}. Recall that Xx0,u=S()x+Vx0,u+Yx0,uX_{x}^{0,u}=S(\cdot)x+V_{x}^{0,u}+Y_{x}^{0,u}. We estimate these terms separately.

Thanks to the parameter assumptions, we can apply (2.9), giving for all T1(0,T]T_{1}\in(0,T]:

0T1S(s)xLλp(D)λp~dsT,λ,p,p~,q,|D|xLq(D)λp~.\int_{0}^{T_{1}}\|S(s)x\|_{L^{\lambda p}(D)}^{\lambda\tilde{p}}\,\mathrm{d}s\lesssim_{T,\lambda,p,\tilde{p},q,|D|}\|x\|_{L^{q}(D)}^{\lambda\tilde{p}}.

Moreover, assuming for the moment that λp1\lambda p\geq 1 and λp~1\lambda\tilde{p}\geq 1, we have by Minkowki’s inequality, Hölder’s inequality and the linear growth of bb:

Vx0,u(t,)Lλp(D)λp~\displaystyle\|V_{x}^{0,u}(t,\cdot)\|_{L^{\lambda p}(D)}^{\lambda\tilde{p}} (0tS(ts)b(Xx0,u(s,))Lλp(D)𝑑s)λp~\displaystyle\leq\Big(\int_{0}^{t}\|S(t-s)b(X_{x}^{0,u}(s,\cdot))\|_{L^{\lambda p}(D)}\,\mathrm{d}s\Big)^{\lambda\tilde{p}}
0tT,λ,p~S(ts)b(Xx0,u(s,))Lλp(D)λp~𝑑s\displaystyle\lesssim_{T,\lambda,\tilde{p}}\int_{0}^{t}\|S(t-s)b(X_{x}^{0,u}(s,\cdot))\|_{L^{\lambda p}(D)}^{\lambda\tilde{p}}\,\mathrm{d}s
0tb(Xx0,u(s,))Lλp(D)λp~𝑑s\displaystyle\leq\int_{0}^{t}\|b(X_{x}^{0,u}(s,\cdot))\|_{L^{\lambda p}(D)}^{\lambda\tilde{p}}\,\mathrm{d}s
T,λ,p,p~,|D|1+Xx0,uLλp~(0,t,Lλp(D))λp~.\displaystyle\lesssim_{T,\lambda,p,\tilde{p},|D|}1+\|X_{x}^{0,u}\|_{L^{\lambda\tilde{p}}(0,t;L^{\lambda p}(D))}^{\lambda\tilde{p}}.

Integrating over tt, we conclude that for all T1(0,T]T_{1}\in(0,T]:

Vx0,uLλp~(0,T1,Lλp(D))λp~\displaystyle\|V_{x}^{0,u}\|_{L^{\lambda\tilde{p}}(0,T_{1};L^{\lambda p}(D))}^{\lambda\tilde{p}} T,λ,p,p~,|D|1+0T1Xx0,uLλp~(0,t,Lλp(D))λp~dt.\displaystyle\lesssim_{T,\lambda,p,\tilde{p},|D|}1+\int_{0}^{T_{1}}\|X_{x}^{0,u}\|_{L^{\lambda\tilde{p}}(0,t;L^{\lambda p}(D))}^{\lambda\tilde{p}}\,\mathrm{d}t.

By the last part of Lemma 2.1 applied with p1=p2=p2p_{1}=p_{2}=p\geq 2, φσ(Xx0,u)\varphi\coloneqq\sigma(X_{x}^{0,u}), and p~>4\tilde{p}>4, the growth bound (0.2) for σ\sigma, and the fact that λ(0,1]\lambda\in(0,1], we have for all t[0,T]t\in[0,T]:

Yx0,u(t,)Lλp(D)p~λ,p,p~,|D|Yx0,u(t,)Lp(D)p~\displaystyle\|Y_{x}^{0,u}(t,\cdot)\|_{L^{\lambda p}(D)}^{\tilde{p}}\lesssim_{\lambda,p,\tilde{p},|D|}\|Y_{x}^{0,u}(t,\cdot)\|_{L^{p}(D)}^{\tilde{p}} 0tN,T,p~σ(Xx0,u(s,))Lp(D)p~𝑑s\displaystyle\lesssim_{N,T,\tilde{p}}\int_{0}^{t}\|\sigma(X_{x}^{0,u}(s,\cdot))\|_{L^{p}(D)}^{\tilde{p}}\,\mathrm{d}s
0tp~,|D|(1+Xx0,u(s,)Lλp(D)λp~)𝑑s\displaystyle\lesssim_{\tilde{p},|D|}\int_{0}^{t}\big(1+\|X_{x}^{0,u}(s,\cdot)\|_{L^{\lambda p}(D)}^{\lambda\tilde{p}}\big)\,\mathrm{d}s
T1+Xx0,uLλp~(0,t,Lλp(D))p~.\displaystyle\lesssim_{T}1+\|X_{x}^{0,u}\|_{L^{\lambda\tilde{p}}(0,t;L^{\lambda p}(D))}^{\tilde{p}}.

Raising the estimate to the power λ\lambda and integrating over tt, we conclude that for all T1(0,T]T_{1}\in(0,T]:

Yx0,uLλp~(0,T1,Lλp(D))λp~=0T1Yx0,u(t)Lλp(D)λp~𝑑t\displaystyle\|Y_{x}^{0,u}\|_{L^{\lambda\tilde{p}}(0,T_{1};L^{\lambda p}(D))}^{\lambda\tilde{p}}=\int_{0}^{T_{1}}\|Y_{x}^{0,u}(t)\|_{L^{\lambda p}(D)}^{\lambda\tilde{p}}\,\mathrm{d}t λ,p,p~,|D|,N,T1+0T1Xx0,uLλp~(0,t,Lλp(D))λp~dt.\displaystyle\lesssim_{\lambda,p,\tilde{p},|D|,N,T}1+\int_{0}^{T_{1}}\|X_{x}^{0,u}\|_{L^{\lambda\tilde{p}}(0,t;L^{\lambda p}(D))}^{\lambda\tilde{p}}\,\mathrm{d}t.

Combining the estimates above, if λp1\lambda p\geq 1 and λp~1\lambda\tilde{p}\geq 1, we obtain for all T1(0,T]T_{1}\in(0,T]:

\displaystyle\| Xx0,uLλp~(0,T1,Lλp(D))λp~\displaystyle X_{x}^{0,u}\|_{L^{\lambda\tilde{p}}(0,{T_{1}};L^{\lambda p}(D))}^{\lambda\tilde{p}}
λ,p~S()xLλp~(0,T1,Lλp(D))λp~+Vx0,uLλp~(0,T1,Lλp(D))λp~+Yx0,uLλp~(0,T1,Lλp(D))λp~\displaystyle\lesssim_{\lambda,\tilde{p}}\|S(\cdot)x\|_{L^{\lambda\tilde{p}}(0,{T_{1}};L^{\lambda p}(D))}^{\lambda\tilde{p}}+\|V_{x}^{0,u}\|_{L^{\lambda\tilde{p}}(0,{T_{1}};L^{\lambda p}(D))}^{\lambda\tilde{p}}+\|Y_{x}^{0,u}\|_{L^{\lambda\tilde{p}}(0,{T_{1}};L^{\lambda p}(D))}^{\lambda\tilde{p}}
T,λ,p,p~,q,|D|,NxLq(D)λp~+1+0T1Xx0,uLλp~(0,t,Lλp(D))λp~dt.\displaystyle\lesssim_{T,\lambda,p,\tilde{p},q,|D|,N}\|x\|_{L^{q}(D)}^{\lambda\tilde{p}}+1+\int_{0}^{T_{1}}\|X_{x}^{0,u}\|_{L^{\lambda\tilde{p}}(0,t;L^{\lambda p}(D))}^{\lambda\tilde{p}}\,\mathrm{d}t.

Grönwall’s inequality thus gives

Xx0,uLλp~(0,T,Lλp(D))λp~T,λ,p,p~,q,|D|,NxLq(D)λp~+1,\|X_{x}^{0,u}\|_{L^{\lambda\tilde{p}}(0,T;L^{\lambda p}(D))}^{\lambda\tilde{p}}\lesssim_{T,\lambda,p,\tilde{p},q,|D|,N}\|x\|_{L^{q}(D)}^{\lambda\tilde{p}}+1,

concluding the proof if λp1\lambda p\geq 1 and λp~1\lambda\tilde{p}\geq 1.

If λp<1\lambda p<1 or λp~<1\lambda\tilde{p}<1, then we define pp1λp_{*}\coloneqq p\vee\frac{1}{\lambda} and p~p~1λ\tilde{p}_{*}\coloneqq\tilde{p}\vee\frac{1}{\lambda}. We verify that the parameter conditions also hold with (p,p~)(p_{*},\tilde{p}_{*}). Clearly, pp2p_{*}\geq p\geq 2 and p~p~>4\tilde{p}_{*}\geq\tilde{p}>4. Moreover, if q<λpq<\lambda p_{*}, then p1/λp_{*}\neq 1/\lambda (since q1q\geq 1), hence p=pp_{*}=p and p~=p~1λp~2λ<2pqλpq=2pqλpq\tilde{p}_{*}=\tilde{p}\vee\frac{1}{\lambda}\leq\tilde{p}\vee\frac{2}{\lambda}<\frac{2pq}{\lambda p-q}=\frac{2p_{*}q}{\lambda p_{*}-q}. Also, (2.1) is satisfied for pp_{*}: if p=pp_{*}=p, then by assumption, and if p=1/λp_{*}=1/\lambda, then 2λpp+2=2λ1+2λ<1q\frac{2\lambda p_{*}}{p_{*}+2}=\frac{2\lambda}{1+2\lambda}<1\leq q. Consequently, our result can be applied with pp_{*} and p~\tilde{p}_{*}, giving a.s.

0TXx0,u(s)Lλp(D)λp~𝑑sM(1+xLq(D)λp~).\int_{0}^{T}\|X_{x}^{0,u}(s)\|_{L^{\lambda p_{*}}(D)}^{\lambda\tilde{p}_{*}}\,\mathrm{d}s\leq M(1+\|x\|_{L^{q}(D)}^{\lambda\tilde{p}_{*}}).

Applying Jensen’s inequality twice (for ||p/p|\cdot|^{p/p_{*}} on DD and for ||p~/p~|\cdot|^{\tilde{p}/\tilde{p}_{*}} on [0,T][0,T]), we find

0TXx0,u(s)Lλp(D)λp~dsT,p,p,p~,p~,|D|(0TXx0,u(s)Lλp(D)λp~ds)p~/p~\displaystyle\int_{0}^{T}\|X_{x}^{0,u}(s)\|_{L^{\lambda p}(D)}^{\lambda\tilde{p}}\,\mathrm{d}s\lesssim_{T,p,p_{*},\tilde{p},\tilde{p}_{*},|D|}\Big(\int_{0}^{T}\|X_{x}^{0,u}(s)\|_{L^{\lambda p_{*}}(D)}^{\lambda\tilde{p}_{*}}\,\mathrm{d}s\Big)^{\tilde{p}/\tilde{p}_{*}} (M(1+xLq(D)λp~))p~/p~\displaystyle\leq(M(1+\|x\|_{L^{q}(D)}^{\lambda\tilde{p}_{*}}))^{\tilde{p}/\tilde{p}_{*}}
Mp~/p~(1+xLq(D)λp~).\displaystyle\leq M^{\tilde{p}/\tilde{p}_{*}}(1+\|x\|_{L^{q}(D)}^{\lambda\tilde{p}}).

Redefining MM, and noting that p,p~p_{*},\tilde{p}_{*} depend only on λ,p,\lambda,p, and p~\tilde{p}, this proves the result for the general case. ∎

4.2. Conclusion of the ULDP

By employing a Grönwall scheme and applying Lemma 4.1, we now establish Theorem 0.4.

Proof of Theorem 0.4.

First, we observe that it suffices to prove the theorem for q(2λpp+2,λp)q\in(\frac{2\lambda p}{p+2},\lambda p). Indeed, 2λpp+2<λp\frac{2\lambda p}{p+2}<\lambda p, so if qλpq\geq\lambda p, then we can pick q~(2λpp+2,λp)\tilde{q}\in(\frac{2\lambda p}{p+2},\lambda p) and note that Lq(D)L^{q}(D)-bounded sets are also Lq~(D)L^{\tilde{q}}(D) bounded (since DD is bounded). Thus the ULDP follows by applying the stated ULDP with q~\tilde{q}.

From now on, let q(2λpp+2,λp)q\in(\frac{2\lambda p}{p+2},\lambda p) and xLq(D)x\in L^{q}(D). We will apply Grönwall’s inequality to the mapping T1𝔼[Xxε,uXx0,uC([0,T1],Lp(D))p~]T_{1}\mapsto{\mathbb{E}}[\|X_{x}^{\varepsilon,u}-X_{x}^{0,u}\|_{C([0,T_{1}];L^{p}(D))}^{{\tilde{p}}}]. For all t[0,T]t\in[0,T], we have

Xxε,u(t)Xx0,u(t)Lp(D)p~p~Vxε,u(t)Vx0,u(t)Lp(D)p~+Yxε,u(t)Yx0,u(t)Lp(D)p~+εp/2Zxε,u(t)Lp(D)p~.\|X_{x}^{\varepsilon,u}(t)-X_{x}^{0,u}(t)\|_{L^{p}(D)}^{{\tilde{p}}}\lesssim_{{\tilde{p}}}\|V_{x}^{\varepsilon,u}(t)-V_{x}^{0,u}(t)\|_{L^{p}(D)}^{{\tilde{p}}}+\|Y_{x}^{\varepsilon,u}(t)-Y_{x}^{0,u}(t)\|_{L^{p}(D)}^{{\tilde{p}}}+\varepsilon^{p/2}\|Z_{x}^{\varepsilon,u}(t)\|_{L^{p}(D)}^{{\tilde{p}}}.

Note that the difference Vxε,uVx0,uV_{x}^{\varepsilon,u}-V_{x}^{0,u} equals Ψxb(Xxε,uS()x)Ψxb(Xx0,uS()x)\Psi_{x}^{b}(X_{x}^{\varepsilon,u}-S(\cdot)x)-\Psi_{x}^{b}(X_{x}^{0,u}-S(\cdot)x), thus the last part of Lemma 2.4 and Hölder’s inequality yield for all T1(0,T]T_{1}\in(0,T]:

𝔼[Vxε,uVx0,uC([0,T1],Lp(D))p~]\displaystyle{\mathbb{E}}[\|V_{x}^{\varepsilon,u}-V_{x}^{0,u}\|_{C([0,T_{1}];L^{p}(D))}^{\tilde{p}}] T,p~(0T1Xxε,uXx0,uLp~(Ω,C([0,s],Lp(D)))ds)p~\displaystyle\lesssim_{T,\tilde{p}}\Big(\int_{0}^{T_{1}}\|X_{x}^{\varepsilon,u}-X_{x}^{0,u}\|_{L^{\tilde{p}}(\Omega;C([0,s];L^{p}(D)))}\,\mathrm{d}s\Big)^{\tilde{p}}
0T1T,p~𝔼[Xxε,uXx0,uC([0,s],Lp(D))p~]𝑑s.\displaystyle\lesssim_{T,\tilde{p}}\int_{0}^{T_{1}}{\mathbb{E}}\big[\|X_{x}^{\varepsilon,u}-X_{x}^{0,u}\|_{C([0,s];L^{p}(D))}^{{\tilde{p}}}\big]\,\mathrm{d}s.

Similarly, Yxε,uYx0,u=Ψxσ,u(Xxε,uS()x)Ψxσ,u(Xx0,uS()x)Y_{x}^{\varepsilon,u}-Y_{x}^{0,u}=\Psi_{x}^{\sigma,u}(X_{x}^{\varepsilon,u}-S(\cdot)x)-\Psi_{x}^{\sigma,u}(X_{x}^{0,u}-S(\cdot)x), so the last part of Lemma 2.4 and Hölder’s inequality yield

𝔼[Yxε,uYx0,uC([0,T1],Lp(D))p~]\displaystyle{\mathbb{E}}[\|Y_{x}^{\varepsilon,u}-Y_{x}^{0,u}\|_{C([0,T_{1}];L^{p}(D))}^{\tilde{p}}] 0T1N,T,p,p~𝔼[Xxε,uXx0,uC([0,s],Lp(D))p~]𝑑s.\displaystyle\lesssim_{N,T,p,\tilde{p}}\int_{0}^{T_{1}}{\mathbb{E}}\big[\|X_{x}^{\varepsilon,u}-X_{x}^{0,u}\|_{C([0,s];L^{p}(D))}^{{\tilde{p}}}\big]\,\mathrm{d}s.

For Zxε,uZ_{x}^{\varepsilon,u}, we can apply Proposition 2.2 with Φσ(Xxε,u)\Phi\coloneqq\sigma(X_{x}^{\varepsilon,u}) (so Z=Zxε,uZ=Z^{\varepsilon,u}_{x}). The latter, combined with the growth bound (0.2) for σ\sigma and the fact that λ(0,1]\lambda\in(0,1], gives

𝔼[Zxε,u\displaystyle{\mathbb{E}}[\|Z_{x}^{\varepsilon,u} C([0,T1],Lp(D))p~]T,p,p~,|D|𝔼[0T1(1+Xxε,u(s)Lλp(D)λp~)ds]\displaystyle\|_{C([0,T_{1}];L^{p}(D))}^{{\tilde{p}}}]\lesssim_{T,p,\tilde{p},|D|}{\mathbb{E}}\Big[\int_{0}^{T_{1}}(1+\|X_{x}^{\varepsilon,u}(s)\|_{L^{\lambda p}(D)}^{\lambda\tilde{p}})\,\mathrm{d}s\Big]
0T1T,λ,p,p~,|D|𝔼[Xxε,uXx0,uC([0,s],Lp(D))p~]𝑑s+(1+𝔼[0T1Xx0,u(s)Lλp(D)λp~𝑑s]).\displaystyle\lesssim_{T,\lambda,p,\tilde{p},|D|}\int_{0}^{T_{1}}{\mathbb{E}}\big[\|X_{x}^{\varepsilon,u}-X_{x}^{0,u}\|_{C([0,s];L^{p}(D))}^{{\tilde{p}}}\big]\,\mathrm{d}s+\Big(1+{\mathbb{E}}\Big[\int_{0}^{T_{1}}\|X_{x}^{0,u}(s)\|_{L^{\lambda p}(D)}^{{\lambda\tilde{p}}}\,\mathrm{d}s\Big]\Big).

Combining the estimates above, we find that for any fixed ε0>0\varepsilon_{0}>0, for all ε(0,ε0]\varepsilon\in(0,\varepsilon_{0}], and for all T1(0,T]T_{1}\in(0,T],

𝔼[\displaystyle{\mathbb{E}}[ Xxε,uXx0,uC([0,T1],Lp(D))p~]\displaystyle\|X_{x}^{\varepsilon,u}-X_{x}^{0,u}\|_{C([0,T_{1}];L^{p}(D))}^{{\tilde{p}}}]
p~𝔼[Vxε,uVx0,uC([0,T1],Lp(D))p~]+𝔼[Yxε,uYx0,uC([0,T1],Lp(D))p~]+εp~2𝔼[Zxε,uC([0,T1],Lp(D))p~]\displaystyle\lesssim_{{\tilde{p}}}{\mathbb{E}}[\|V_{x}^{\varepsilon,u}-V_{x}^{0,u}\|_{C([0,T_{1}];L^{p}(D))}^{{\tilde{p}}}]+{\mathbb{E}}[\|Y_{x}^{\varepsilon,u}-Y_{x}^{0,u}\|_{C([0,T_{1}];L^{p}(D))}^{{\tilde{p}}}]+\varepsilon^{\frac{{\tilde{p}}}{2}}{\mathbb{E}}[\|Z_{x}^{\varepsilon,u}\|_{C([0,T_{1}];L^{p}(D))}^{{\tilde{p}}}]
0T1N,T,λ,p,p~,|D|(1+ε0p~2)𝔼[Xxε,uXx0,uC([0,s],Lp(D))p~]𝑑s+εp~2(1+𝔼[0T1Xx0,u(s)Lλp(D)λp~𝑑s]).\displaystyle\lesssim_{N,T,\lambda,p,\tilde{p},|D|}\int_{0}^{T_{1}}(1+\varepsilon_{0}^{\frac{\tilde{p}}{2}}){\mathbb{E}}\big[\|X_{x}^{\varepsilon,u}-X_{x}^{0,u}\|_{C([0,s];L^{p}(D))}^{{\tilde{p}}}\big]\,\mathrm{d}s+\varepsilon^{\frac{{\tilde{p}}}{2}}\Big(1+{\mathbb{E}}\Big[\int_{0}^{T_{1}}\|X_{x}^{0,u}(s)\|_{L^{\lambda p}(D)}^{{\lambda\tilde{p}}}\,\mathrm{d}s\Big]\Big).

Now, Grönwall’s inequality gives

(4.2) 𝔼[Xxε,uXx0,uC([0,T],Lp(D))p~]C1εp~2(1+𝔼[0TXx0,u(s)Lλp(D)λp~𝑑s]),{\mathbb{E}}[\|X_{x}^{\varepsilon,u}-X_{x}^{0,u}\|_{C([0,T];L^{p}(D))}^{{\tilde{p}}}]\leq C_{1}\varepsilon^{\frac{{\tilde{p}}}{2}}\Big(1+{\mathbb{E}}\Big[\int_{0}^{T}\|X_{x}^{0,u}(s)\|_{L^{\lambda p}(D)}^{\lambda\tilde{p}}\,\mathrm{d}s\Big]\Big),

for a constant C1C_{1} that depends on N,T,λ,p,p~,|D|,N,T,\lambda,p,\tilde{p},|D|, and ε0\varepsilon_{0}. Moreover, Lemma 4.1 gives for all p~(4,2pqλpq)\tilde{p}\in(4,\frac{2pq}{\lambda p-q}) (see (1.11)):

(4.3) supu𝒜NsupxBRq𝔼[0TXx0,u(s)Lλp(D)λp~𝑑s]<.\sup_{u\in{\mathcal{A}}_{N}}\sup_{x\in B_{R}^{q}}{\mathbb{E}}\Big[\int_{0}^{T}\|X_{x}^{0,u}(s)\|_{L^{\lambda p}(D)}^{\lambda\tilde{p}}\,\mathrm{d}s\Big]<\infty.

Fixing any p~\tilde{p} in the (non-empty) range above, and combining (4.2) and (4.3) with Chebychev’s inequality proves that the condition of Theorem 1.3(2) is satisfied, thus the latter yields the ULDP. ∎

Remark 4.2.

In fact, the proofs of Theorem 0.2 and 0.4 establish that the corresponding condition in Theorem 1.3 holds uniformly over all xBRqx\in B_{R}^{q}. Nevertheless, the ULDP is stated only for xBRqC(D)x\in B_{R}^{q}\cap C(D) and xBRqLp(D)x\in B_{R}^{q}\cap L^{p}(D) respectively, because for xLq(D)C(D)x\in L^{q}(D)\setminus C(D) or xLq(D)Lp(D)x\in L^{q}(D)\setminus L^{p}(D), the solution XxεX_{x}^{\varepsilon} to the SHE may fail to belong to C([0,T],C(D))C([0,T];C(D)) or C([0,T],Lp(D))C([0,T];L^{p}(D)), respectively.

5. Further ULDPs

In this section, we derive further ULDPs using our previous results and the arguments developed in their proofs. Our next corollary establishes ULDPs for initial data that need only belong to Lq(D)L^{q}(D), without the additional requirement of belonging to Lp(D)L^{p}(D), as in Theorem 0.4.

Corollary 5.1 (ULDPs with initial data in Lq(D)L^{q}(D)).

Let the conditions of Theorem 0.4 hold. For ε>0\varepsilon>0 and xLq(D)x\in L^{q}(D), let Xxε=Xxε,0X_{x}^{\varepsilon}=X_{x}^{\varepsilon,0} be the mild solution to (0.1) provided by Theorem 2.5. Then:

  1. (1)

    The family {XxεS()x:ε>0,xLq(D)}\{X_{x}^{\varepsilon}-S(\cdot)x:\varepsilon>0,x\in L^{q}(D)\} satisfies the ULDP on C([0,T],Lp(D))C([0,T];L^{p}(D)) uniformly over bounded subsets of Lq(D)L^{q}(D), with rate functions Jx:C([0,T],Lp(D))[0,+]J_{x}\colon C([0,T];L^{p}(D))\to[0,+\infty] given by

    Jx(φ)12inf{u\displaystyle J_{x}(\varphi)\coloneqq\frac{1}{2}\inf\{\|u L2(0,T,L2(D))2:uL2(0,T;L2(D)),φ=X~x0,u},\displaystyle\|_{L^{2}(0,T;L^{2}(D))}^{2}:u\in L^{2}(0,T;L^{2}(D)),\,\varphi=\tilde{X}_{x}^{0,u}\},

    where inf+\inf\varnothing\coloneqq+\infty and X~x0,u\tilde{X}_{x}^{0,u} is the mild solution to

    {X~t(t,ξ)=ΔX~(t,ξ)+b(X~(t,ξ)+S(t)x(ξ))+σ(X~(t,ξ)+S(t)x(ξ))u(t,ξ),ξD,t(0,T],X~(0,ξ)=0,ξD.\begin{cases}\frac{\partial\tilde{X}}{\partial t}(t,\xi)=\Delta\tilde{X}(t,\xi)+b\big(\tilde{X}(t,\xi)+S(t)x(\xi)\big)\\ \qquad\qquad\quad+\sigma\big(\tilde{X}(t,\xi)+S(t)x(\xi)\big)u(t,\xi),\quad\xi\in D,t\in(0,T],\\ \tilde{X}(0,\xi)=0,\quad\xi\in D.\end{cases}
  2. (2)

    If 𝒥\mathcal{J} is any subset of Lq(D)L^{q}(D), and (,d)(\mathcal{E},d) is any Polish topological vector space such that

    (5.1) {S()x:x𝒥},C([0,T],Lp(D)),\displaystyle\{S(\cdot)x:x\in\mathcal{J}\}\subset\mathcal{E},\quad C([0,T];L^{p}(D))\subset\mathcal{E},

    and there exists a constant KK such that for all f,gf,g\in\mathcal{E} with fgC([0,T],Lp(D))f-g\in C([0,T];L^{p}(D)),

    (5.2) d(f,g)KfgC([0,T],Lp(D)),\displaystyle d(f,g)\leq K\|f-g\|_{C([0,T];L^{p}(D))},

    then the family {Xxε:ε>0,x𝒥}\{X_{x}^{\varepsilon}:\varepsilon>0,x\in\mathcal{J}\} satisfies the ULDP on \mathcal{E} uniformly over Lq(D)L^{q}(D)-bounded subsets of 𝒥\mathcal{J}, with rate functions Ix:[0,+]I_{x}\colon\mathcal{E}\to[0,+\infty] defined by the formula (0.3).

Proof.

Proof of (1): We apply [39, Th. 2.13] with the Polish space C([0,T],Lp(D))C([0,T];L^{p}(D)). We have existence of measurable solution maps 𝒢~xε:C([0,T],)C([0,T],Lp(D))\tilde{\mathcal{G}}_{x}^{\varepsilon}\colon C([0,T];\mathbb{R}^{\infty})\to C([0,T];L^{p}(D)) such that 𝒢~xε(W)=XxεS()x\tilde{\mathcal{G}}_{x}^{\varepsilon}(W)=X_{x}^{\varepsilon}-S(\cdot)x a.s., thanks to Theorem 2.5 and the Yamada–Watanabe theorem. Indeed, for each fixed xLq(D)x\in L^{q}(D), XxεS()xX_{x}^{\varepsilon}-S(\cdot)x is a mild solution to

(5.3) {X~t(t,ξ)=ΔX~(t,ξ)+b(S(t)x(ξ)+X~(t,ξ))+εσ(S(t)x(ξ)+X~(t,ξ))W˙(t,ξ),ξD,t(0,T],X~xε(0,ξ)=0,ξD.\begin{cases}\frac{\partial\tilde{X}}{\partial t}(t,\xi)=\Delta\tilde{X}(t,\xi)+b\big(S(t)x(\xi)+\tilde{X}(t,\xi)\big)\\ \qquad\qquad\quad+\sqrt{\varepsilon}\sigma\big(S(t)x(\xi)+\tilde{X}(t,\xi)\big)\dot{W}(t,\xi),\quad\xi\in D,t\in(0,T],\\ \tilde{X}_{x}^{\varepsilon}(0,\xi)=0,\quad\xi\in D.\end{cases}

Because X~\tilde{X} is a mild solution to (5.3) if and only if X~+S()x\tilde{X}+S(\cdot)x is a mild solution to (0.1), the pathwise uniqueness stated in Theorem 2.5 implies that XxεS()xX_{x}^{\varepsilon}-S(\cdot)x is the unique mild solution to (5.3) within Lp~(Ω,C([0,T],Lp(D)))L^{\tilde{p}}(\Omega;C([0,T];L^{p}(D))). The Yamada–Watanabe theorem (see [33]) yields measurable mappings 𝒢~xε:C([0,T],)C([0,T],Lp(D))\tilde{\mathcal{G}}_{x}^{\varepsilon}\colon C([0,T];\mathbb{R}^{\infty})\to C([0,T];L^{p}(D)) such that 𝒢~xε(W)=X~xεS()x\tilde{\mathcal{G}}_{x}^{\varepsilon}(W)=\tilde{X}_{x}^{\varepsilon}-S(\cdot)x a.s., where we identify W=(Bk)kW=(B_{k})_{k\in\mathbb{N}} as in (1.6).

Taking into account Remark 4.2, the proof of Theorem 0.4 gives immediately

(5.4) limε0supxBRqsupu𝒜N(Xxε,uXx0,uC([0,T],Lp(D))>δ)=0,\lim_{\varepsilon\downarrow 0}\sup_{x\in B_{R}^{q}}\sup_{u\in{\mathcal{A}}_{N}}{\mathbb{P}}(\|X_{x}^{\varepsilon,u}-X_{x}^{0,u}\|_{C([0,T];L^{p}(D))}>\delta)=0,

Since Xxε,uS()x(Xx0,uS()x)=Xxε,uXx0,uX_{x}^{\varepsilon,u}-S(\cdot)x-(X_{x}^{0,u}-S(\cdot)x)=X_{x}^{\varepsilon,u}-X_{x}^{0,u}, the ULDP now follows by [39, Th. 2.13].

Proof of (2): One can combine part (1) either with a contraction argument and the definition of the ULDP, or with another application of [39, Th. 3.12]. Here, we use the latter approach.

Let FC([0,T],Lp(D))F\coloneqq C([0,T];L^{p}(D)). The proof of (1) yielded measurable maps 𝒢~xε:C([0,T],)F\tilde{\mathcal{G}}_{x}^{\varepsilon}\colon C([0,T];\mathbb{R}^{\infty})\to F such that 𝒢~xε(W)=XxεS()x\tilde{\mathcal{G}}_{x}^{\varepsilon}(W)=X_{x}^{\varepsilon}-S(\cdot)x a.s. Now, for any x𝒥x\in\mathcal{J}, we have S()xS(\cdot)x\in\mathcal{E} by assumption, and Φx:F:ϕS()x+ϕ\Phi_{x}\colon F\to\mathcal{E}\colon\phi\mapsto S(\cdot)x+\phi is continuous since FF\hookrightarrow\mathcal{E} and since \mathcal{E} is a topological vector space (so translations are continuous). Thus, the mappings 𝒢xεΦx𝒢~xε:C([0,T],){\mathcal{G}}_{x}^{\varepsilon}\coloneqq\Phi_{x}\circ\tilde{\mathcal{G}}_{x}^{\varepsilon}\colon C([0,T];\mathbb{R}^{\infty})\to\mathcal{E} are Borel measurable and satisfy 𝒢xε(W)=Xxε\mathcal{G}_{x}^{\varepsilon}(W)=X_{x}^{\varepsilon} a.s.

Furthermore, for ε,N[0,)\varepsilon,N\in[0,\infty), and u𝒜Nu\in{\mathcal{A}}_{N}, we have Xxε,u=S()x+(Xxε,uS()x)X_{x}^{\varepsilon,u}=S(\cdot)x+(X_{x}^{\varepsilon,u}-S(\cdot)x)\in\mathcal{E}, using (5.1) and the C([0,T],Lp(D))C([0,T];L^{p}(D))-regularity of Xxε,uS()xX_{x}^{\varepsilon,u}-S(\cdot)x from Theorem 2.5. Moreover, by combining (5.4) with (5.2), we obtain

limε0supxBRq𝒥supu𝒜N(d(Xxε,u,Xx0,u)>δ)=0,\lim_{\varepsilon\downarrow 0}\sup_{x\in B_{R}^{q}\cap\mathcal{J}}\sup_{u\in{\mathcal{A}}_{N}}{\mathbb{P}}(d(X_{x}^{\varepsilon,u},X_{x}^{0,u})>\delta)=0,

thus the stated ULDP thus follows from [39, Th. 3.12]. ∎

Analogous ULDP results can also be established under the conditions of Theorem 0.2.

Remark 5.2.

Corollary 5.1 holds analogously under the conditions of Theorem 0.2, with every instance of Lp(D)L^{p}(D) replaced by C(D)C(D). The proof is completely analogous, noting that well-posedness was provided by Theorem 2.7 and taking again note of Remark 4.2.

The following can be derived as a special case of Corollary 5.1(2).

Corollary 5.3 (ULDP in Lr(0,T,Lp(D))L^{r}(0,T;L^{p}(D))).

Let the conditions of Theorem 0.4 hold. For ε>0\varepsilon>0 and xLq(D)x\in L^{q}(D), let Xxε=Xxε,0X_{x}^{\varepsilon}=X_{x}^{\varepsilon,0} be the mild solution to (0.1) provided by Theorem 2.5. Let r[1,)r\in[1,\infty) be such that r2(1q1p)<1\frac{r}{2}(\frac{1}{q}-\frac{1}{p})<1.

Then, the family {Xxε:ε>0,xLq(D)}\{X_{x}^{\varepsilon}:\varepsilon>0,x\in L^{q}(D)\} satisfies the ULDP on Lr(0,T,Lp(D))L^{r}(0,T;L^{p}(D)) uniformly over bounded subsets of Lq(D)L^{q}(D), with rate functions Ix:Lr(0,T,Lp(D))[0,+]I_{x}\colon L^{r}(0,T;L^{p}(D))\to[0,+\infty] defined by the formula (0.3).

Proof.

We apply Corollary 5.1(2) with 𝒥=Lq(D)\mathcal{J}=L^{q}(D). For any xLq(D)x\in L^{q}(D), we have S()xLr(0,T,Lp(D))S(\cdot)x\in L^{r}(0,T;L^{p}(D)), due to the assumptions on rr and the bound (1.4). Furthermore, we have a continuous embedding C([0,T],Lp(D))Lr(0,T,Lp(D))C([0,T];L^{p}(D))\hookrightarrow L^{r}(0,T;L^{p}(D)), so both (5.1) and (5.2) are satisfied. ∎

Our final result shows that a ULDP over L1(D)L^{1}(D)-bounded sets of initial data always holds in the C([0,T],Lr(D))C([0,T];L^{r}(D)) topology for r2r\leq 2, whenever the noise has sublinear growth (λ(0,1)\lambda\in(0,1)). If the noise has linear growth (λ=1\lambda=1), we obtain the ULDP over Lq(D)L^{q}(D)-bounded sets of initial data for any q>1q>1.

Corollary 5.4 (ULDP on C([0,T],Lr(D))C([0,T];L^{r}(D)) for r[1,2]r\in[1,2]).

Let Assumption 0.1 hold. Let r[1,2]r\in[1,2] and let q=1q=1 if λ<1\lambda<1, and q(1,r]q\in(1,r] if λ=1\lambda=1.

Then, the family {Xxε:ε>0,xLr(D)}\{X_{x}^{\varepsilon}:\varepsilon>0,x\in L^{r}(D)\} of mild solutions to (0.1) satisfies the ULDP on C([0,T],Lr(D))C([0,T];L^{r}(D)), uniformly over initial data in Lq(D)L^{q}(D)-bounded subsets of Lr(D)L^{r}(D), with the rate functions IxI_{x} from Theorem 0.4.

Proof of Corollary 5.4.

We apply Corollary 5.1(2) with p=2p=2, 𝒥Lr(D)Lq(D)\mathcal{J}\coloneqq L^{r}(D)\subset L^{q}(D) and =C([0,T],Lr(D))\mathcal{E}=C([0,T];L^{r}(D)). Because λ<q\lambda<q, p=2p=2 satisfies the parameter condition 2λpp+2=λ<q\frac{2\lambda p}{p+2}=\lambda<q of Theorem 0.4. Furthermore, C([0,T],Lp(D))C([0,T],Lr(D))C([0,T];L^{p}(D))\hookrightarrow C([0,T];L^{r}(D)) since p=2rp=2\geq r, and it holds that S()xLr(D)S(\cdot)x\in L^{r}(D) for all x𝒥x\in\mathcal{J} since SS is a strongly continuous semigroup on Lr(D)L^{r}(D). ∎

References

  • [1] H. Bessaih and A. Millet. Large deviation principle and inviscid shell models. Electron. J. Probab., 14, 2009.
  • [2] H. Bessaih and A. Millet. Large deviations and the zero viscosity limit for 2D stochastic Navier–Stokes equations with free boundary. SIAM J. Math. Anal., 44(3):1861–1893, 2012.
  • [3] A. Biswas and A. Budhiraja. Exit time and invariant measure asymptotics for small noise constrained diffusions. Stoch. Process. Appl., 121(5):899–924, 2011.
  • [4] Z. Brzezniak, Q. Li, and T. Zhang. Large deviation principle of stochastic evolution equations with reflection. J. Evol. Equ., 24(4):91, 2024.
  • [5] A. Budhiraja, P. Dupuis, and V. Maroulas. Large deviations for infinite dimensional stochastic dynamical systems. Ann. Probab., 36(4), 2008.
  • [6] A. Budhiraja, P. Dupuis, and V. Maroulas. Large deviations for stochastic flows of diffeomorphisms. Bernoulli, 16(1), 2010.
  • [7] D. Candil, L. Chen, and C.Y. Lee. Parabolic stochastic PDEs on bounded domains with rough initial conditions: moment and correlation bounds. Stoch. Partial Differ. Equ.: Anal. Comput., 12(3):1507–1573, 2024.
  • [8] C. Cardon-Weber. Large deviations for a Burgers’-type SPDE. Stoch. Process. Appl., 84(1):53–70, 1999.
  • [9] S. Cerrai and M. Röckner. Large deviations for stochastic reaction-diffusion systems with multiplicative noise and non-Lipshitz reaction term. Ann. Probab., 32(1B), 2004.
  • [10] L. Chen and R.C. Dalang. Hölder-continuity for the nonlinear stochastic heat equation with rough initial conditions. Stoch. Partial Differ. Equ.: Anal. Comput., 2(3):316–352, 2014.
  • [11] L. Chen and R.C. Dalang. Moments and growth indices for the nonlinear stochastic heat equation with rough initial conditions. Ann. Probab., 43(6):3006 – 3051, 2015.
  • [12] L. Chen and J. Huang. Comparison principle for stochastic heat equation on d\mathbb{R}^{d}. Ann. Probab., 47(2):989–1035, 2019.
  • [13] L. Chen and J. Huang. Superlinear stochastic heat equation on d\mathbb{R}^{d}. Proc. Am. Math. Soc., 151(09):4063–4078, 2023.
  • [14] L. Chen and K. Kim. Nonlinear stochastic heat equation driven by spatially colored noise: moments and intermittency. Acta Math. Sci., 39(3):645–668, 2019.
  • [15] L. Chen and P. Xia. Asymptotic properties of stochastic partial differential equations in the sublinear regime. Ann. Probab., 54(4):1686–1714, 2026.
  • [16] I. Chueshov and A. Millet. Stochastic 2D hydrodynamical type systems: well posedness and large deviations. Appl. Math. Optim., 61(3):379–420, 2010.
  • [17] G. Da Prato and J. Zabczyk. Stochastic equations in infinite dimensions. Number 152 in Encyclopedia of mathematics and its applications. Cambridge university press, Cambridge, 2nd edition, 2014.
  • [18] A. Dembo and O. Zeitouni. Large Deviations Techniques and Applications, volume 38 of Stochastic Modelling and Applied Probability. Springer, Berlin, Heidelberg, 2010.
  • [19] J. Duan and A. Millet. Large deviations for the Boussinesq equations under random influences. Stoch. Process. Appl., 119(6):2052–2081, 2009.
  • [20] M. Foondun, D. Khoshnevisan, and E. Nualart. On the local well-posedness of randomly forced reaction-diffusion equations with L2L^{2} initial data and a superlinear reaction term. Probability Theory and Related Fields, pages 1–35, 2026.
  • [21] M. Foondun and L. Setayeshgar. Large deviations for a class of semilinear stochastic partial differential equations. Stat. Probab. Lett., 121:143–151, 2017.
  • [22] M.I. Freidlin. Random perturbations of reaction-diffusion equations: The quasi-deterministic approximation. Trans. Am. Math. Soc., 305(2):665–697, 1988.
  • [23] M.I. Freidlin and A.D. Wentzell. Random Perturbations of Dynamical Systems, volume 260 of Grundlehren der mathematischen Wissenschaften. Springer Berlin Heidelberg, Berlin, Heidelberg, 2012.
  • [24] E. Gautier. Uniform large deviations for the nonlinear Schrödinger equation with multiplicative noise. Stoch. Process. Appl., 115(12):1904–1927, 2005.
  • [25] W. Hu, M. Salins, and K. Spiliopoulos. Large deviations and averaging for systems of slow-fast stochastic reaction–diffusion equations. Stoch. PDE: Anal. Comp., 7(4):808–874, 2019.
  • [26] G. Kallianpur and J. Xiong. Large deviations for a class of stochastic partial differential equations. Ann. Probab., 24(1):320–345, 1996.
  • [27] A. Kumar and M.T. Mohan. Small time asymptotics for a class of stochastic partial differential equations with fully monotone coefficients forced by multiplicative Gaussian noise. Stoch. Dyn., 25(05):2550021, 2025.
  • [28] R. Li, R. Wang, and B. Zhang. A large deviation principle for the stochastic heat equation with general rough noise. J. Theor. Probab., 37(1):251–306, 2024.
  • [29] W. Liu. Large deviations for stochastic evolution equations with small multiplicative noise. Appl. Math. Optim., 61(1):27–56, 2010.
  • [30] W. Liu, M. Röckner, and X. Zhu. Large deviation principles for the stochastic quasi-geostrophic equations. Stoch. Process. Appl., 123(8):3299–3327, 2013.
  • [31] Y. Lv and A.J. Roberts. Large deviation principle for singularly perturbed stochastic damped wave equations. Stoch. Anal. Appl., 32(1):50–60, 2014.
  • [32] C. Mo and J. Luo. Large deviations for stochastic differential delay equations. Nonlinear Anal. Theory Methods Appl., 80:202–210, 2013.
  • [33] M. Ondreját. Uniqueness for stochastic evolution equations in Banach spaces. Diss. Math., 426:1–63, 2004.
  • [34] V. Ortiz-López and M. Sanz-Solé. A Laplace principle for a stochastic wave equation in spatial dimension three. In D. Crisan, editor, Stochastic Analysis 2010, pages 31–49. Springer Berlin Heidelberg, Berlin, Heidelberg, 2011.
  • [35] T. Pan, S. Shang, J. Zhai, and T. Zhang. Large deviations of fully local monotone stochastic partial differential equations driven by gradient-dependent noise. Bernoulli, 32(1):249–273, 2026.
  • [36] J. Ren, S. Xu, and X. Zhang. Large deviations for multivalued stochastic differential equations. J. Theor. Probab., 23(4):1142–1156, 2010.
  • [37] J. Ren and X. Zhang. Freidlin–Wentzell’s large deviations for stochastic evolution equations. J. Funct. Anal., 254(12):3148–3172, 2008.
  • [38] M. Röckner, T. Zhang, and X. Zhang. Large deviations for stochastic tamed 3D Navier-Stokes equations. Appl Math Optim, 61(2):267–285, 2010.
  • [39] M. Salins. Equivalences and counterexamples between several definitions of the uniform large deviations principle. Probab. Surveys, 16:99–142, 2019.
  • [40] M. Salins. Systems of small-noise stochastic reaction-diffusion equations satisfy a large deviations principle that is uniform over all initial data. Stoch. Process. Appl., 142:159–194, 2021.
  • [41] M. Salins. Solutions to the stochastic heat equation with polynomially growing multiplicative noise do not explode in the critical regime. Ann. Probab., 53(1):223–238, 2025.
  • [42] S. Shang, P. Wang, and T. Zhang. L2L^{2}-solutions to stochastic reaction-diffusion equations with superlinear drifts driven by space-time white noise. Stoch. Process. Appl., page 104991, 2026.
  • [43] R.B. Sowers. Large deviations for a reaction-diffusion equation with non-Gaussian perturbations. Ann. Probab., 20(1):504–537, 1992.
  • [44] S.S. Sritharan and P. Sundar. Large deviations for the two-dimensional Navier–Stokes equations with multiplicative noise. Stoch. Process. Appl., 116(11):1636–1659, 2006.
  • [45] C. Sun, H. Gao, J. Duan, and B. Schmalfuß. Rare events in the Boussinesq system with fluctuating dynamical boundary conditions. J. Differ. Equ., 248(6):1269–1296, 2010.
  • [46] E. Theewis and M. Veraar. Large deviations for stochastic evolution equations in the critical variational setting. Stoch. Process. Appl., 196:104898, 2026.
  • [47] J. van Neerven, M. Veraar, and L. Weis. Stochastic integration in Banach spaces – a survey. In Stochastic analysis: a series of lectures, volume 68, pages 297–332. Springer, Basel, 2015.
  • [48] T. Xu and T. Zhang. White noise driven SPDEs with reflection: Existence, uniqueness and large deviation principles. Stoch. Process. Appl., 119(10):3453–3470, 2009.
  • [49] J. Zhai and T. Zhang. Large deviations for stochastic models of two-dimensional second grade fluids. Appl. Math. Optim., 75(3):471–498, 2017.