\documentclass[reqno]{amsart} \AtBeginDocument{{\noindent\small {\em Electronic Journal of Differential Equations}, Vol. 2003(2003), No. 111, pp. 1--21.\newline ISSN: 1072-6691. URL: http://ejde.math.txstate.edu or http://ejde.math.unt.edu \newline ftp ejde.math.txstate.edu (login: ftp)} \thanks{\copyright 2003 Texas State University-San Marcos.} \vspace{9mm}} \begin{document} \title[\hfilneg EJDE--2003/111\hfil SDDEs limits solutions] {SDDEs limits solutions to sublinear reaction-diffusion SPDEs} \author[Hassan Allouba\hfil EJDE--2003/111\hfilneg] {Hassan Allouba} \address{Department of Mathematical Sciences, Kent State University, Kent, Ohio 44242, USA} \email{allouba@mcs.kent.edu} \date{} \thanks{Submitted October 3, 2002. Published November 5, 2003.} \thanks{Partially supported by grant MDA904-02-1-0083 from NSA} \subjclass[2000]{60H15, 35R60} \keywords{Reaction-diffusion SPDE, SDDE, SDDE limits solutions, multiscale} \begin{abstract} We start by introducing a new definition of solutions to heat-based SPDEs driven by space-time white noise: SDDEs (stochastic differential-difference equations) limits solutions. In contrast to the standard direct definition of SPDEs solutions; this new notion, which builds on and refines our SDDEs approach to SPDEs from earlier work, is entirely based on the approximating SDDEs. It is applicable to, and gives a multiscale view of, a variety of SPDEs. We extend this approach in related work to other heat-based SPDEs (Burgers, Allen-Cahn, and others) and to the difficult case of SPDEs with multi-dimensional spacial variable. We focus here on one-spacial-dimensional reaction-diffusion SPDEs; and we prove the existence of a SDDEs limit solution to these equations under less-than-Lipschitz conditions on the drift and the diffusion coefficients, thus extending our earlier SDDEs work to the nonzero drift case. The regularity of this solution is obtained as a by-product of the existence estimates. The uniqueness in law of our SPDEs follows, for a large class of such drifts/diffusions, as a simple extension of our recent Allen-Cahn uniqueness result. We also examine briefly, through order parameters $\epsilon_1$ and $\epsilon_2$ multiplied by the Laplacian and the noise, the effect of letting $\epsilon_1,\epsilon_2\to 0$ at different speeds. More precisely, it is shown that the ratio $\epsilon_2/\epsilon_1^{1/4}$ determines the behavior as $\epsilon_1,\epsilon_2\to 0$. \end{abstract} \maketitle \numberwithin{equation}{section} \newtheorem{thm}{Theorem}[section] \newtheorem{cor}[thm]{Corollary} \newtheorem{lem}[thm]{Lemma} \newtheorem{prop}[thm]{Proposition} \newtheorem{defn}[thm]{Definition} \newtheorem{rem}[thm]{Remark} \allowdisplaybreaks %%***Defn of Scrpt****% \font\tenscrpt=eusm10 \font\sevenscrpt=eusm10 scaled 700 \font\fivescrpt=eusm10 scaled 500 \newfam\eusmfam \textfont\eusmfam=\tenscrpt \scriptfont\eusmfam=\sevenscrpt \scriptscriptfont\eusmfam=\fivescrpt \def\eus#1{{\fam\eusmfam\relax#1}} % \section{Introduction and statements of results} We consider the parametrized space-time white noise driven SPDE on $\eus{R}_{T}\overset{\triangle}{=} \mathbb{T}\times\mathbb{R} =[0,T]\times\mathbb{R}$: \begin{equation} \begin{gathered} \frac{\partial U_{\epsilon_1,\epsilon_2}}{\partial t} =\frac{\epsilon_1}2\Delta U_{\epsilon_1,\epsilon_2}+b(U_{\epsilon_1,\epsilon_2}) +\epsilon_2a(U_{\epsilon_1,\epsilon_2}) \frac{\partial^2 W}{\partial t\partial x}; \quad (t,x)\in(0,T]\times\mathbb{R}, \\ U_{\epsilon_1,\epsilon_2}(0,x)=\xi(x) ; \quad x\in \mathbb{R}, \end{gathered} \label{G} \end{equation} where $T>0$ is fixed but arbitrary, $\Delta$ is the Laplace operator in space, and $a,b:\mathbb{R}\to\mathbb{R}$ are Borel measurable. $W(t,x)$ is the Brownian sheet corresponding to the driving space-time white noise--with intensity Lebesgue measure--written formally as $\partial^2W/\partial t\partial x$. As in Walsh \cite{WA}, white noise is regarded as a continuous orthogonal martingale measure, which we denote by $\eus{W}$, with the corresponding Brownian sheet as the random field induced by $\eus{W}$ in the usual way. $\xi(x)$ is taken to be a continuous bounded deterministic function. The parameters $\epsilon_1,\epsilon_2>0$ are order parameters, which allow us to control the competing effects of the Laplacian $\Delta$ and the driving space-time white noise $\partial^2W/\partial t\partial x$. We denote the SPDE in \eqref{G} by $e^{\epsilon_1,\epsilon_2}_{\mbox{\tiny heat}} (a,b,\xi)$ Before going into the statements of our results, let's highlight one of the main features of this article. We introduce and formalize the notion of SDDEs limit solutions (see Definition \ref{limitsolns} and Remark \ref{defnrem} below), and most of our treatment here focuses on this class of solutions to $e^{\epsilon_1,\epsilon_2}_{\mbox{\tiny heat}} (a,b,\xi)$ whose elements are limits of an approximating sequence of stochastic-differential-difference equations (SDDEs) (see \cite{ASDDE1, AD}). SDDEs are obtained from $e^{\epsilon_1,\epsilon_2}_{\mbox{\tiny heat}} (a,b,\xi)$ by discretizing space but leaving time continuous. We have used SDDEs before in \cite{AD,ASDDE1} to give a non-nonstandard proof of Reimers' existence result for $e_{\mbox{\tiny heat}}(a,0,\xi)$ (the case of zero drift) when $a$ is continuous and satisfies a linear growth condition. In addition to extending our existence proof there to the case of nonzero continuous drift (Theorem \ref{ww}) and examining the effects of the order parameters $\epsilon_1,\epsilon_2$ on $e^{\epsilon_1,\epsilon_2}_{\mbox{\tiny heat}} (a,b,\xi)$ (Theorem \ref{thm2}), our new definition (Definition \ref{limitsolns}) of solutions to SPDEs as limits of approximating SDDEs establishes a general approach of SPDEs in which solutions to the SPDE in question are defined {\it entirely }in terms of its approximating SDDEs and their limits (more on this approach and its implications in $d$-dimensional space as well as for other SPDEs is detailed in \cite{Adim}). It is important to note that (even in one dimension spacial variable) this is different from, and has several advantages not shared with, the traditional direct approach (in which solutions to $e^{\epsilon_1,\epsilon_2}_{\mbox{\tiny heat}} (a,b,\xi)$ are understood in the sense of either the standard test function \eqref{TFF} or Green function formulations \eqref{GFF} of $e^{\epsilon_1,\epsilon_2}_{\mbox{\tiny heat}} (a,b,\xi)$). In addition to the obvious numerical advantage; 1.~this is a multiscale approach which allows us to view the model under consideration in two different scales simultaneously (the microscopic-in-space SDDE scale and their limiting SPDE scale), and thus be able to see which properties of the SPDEs being approximated is captured by their SDDEs and which ones are different (e.g.~Proposition \ref{Expbd}, Lemma \ref{boundsonsdifferences}, and Lemma \ref{boundsontdifferences} show how some regularity properties for $e^{\epsilon_1,\epsilon_2}_{\mbox{\tiny heat}} (a,b,\xi)$ are captured by their SDDEs, and the proof of Theorem \ref{thm2} part (ii) is the same for both $e^{\epsilon_1,\epsilon_2}_{\mbox{\tiny heat}} (a,b,\xi)$ and the corresponding SDDEs, while the proof of Theorem \ref{thm2} part (i) is an example of an argument which holds in the continuous setting but not in the SDDE one, possibly pointing to different behaviors in the two different scales), 2.~the role of the heat Green function for $e^{\epsilon_1,\epsilon_2}_{\mbox{\tiny heat}} (a,b,\xi)$ is played by a continuous-time random walk density in the case of SDDEs, allowing us to use powerful and simpler random walk arguments like coupling (see Lemma \ref{5thinequality}) and others to get needed estimates to prove existence and regularity for $e^{\epsilon_1,\epsilon_2}_{\mbox{\tiny heat}} (a,b,\xi)$ under mild conditions on $a$ and $b$ (more on the intimate connection between the Green function and the random walk density is also detailed in \cite{Adim}), and 3.~unlike the usual Green's function formulation in the direct approach to SPDEs, these SDDEs make sense as {\it real-valued} random fields in any spatial dimension $d$, and we use them in \cite{Adim} to extend our Definition \ref{limitsolns} below to a class of solutions for SPDEs in any dimension. In this last regard, SDDEs limit solutions to SPDEs are similar to Lions-Crandal notion of viscosity solutions in that they are defined as limits of more ``regular'' solutions. Consider the sequence of lattices ${(\mathbb{X}_n)\,}_{n=1}^{\infty}$ defined by \begin{equation*} \mathbb{X}_n \overset{\triangle}{=} \{\cdots, -2\delta_n, -\delta_n, 0, \delta_n, 2\delta_n, \cdots \}, \end{equation*} where $\delta_n\to0$ as $n\to\infty$. Then, following \cite{ASDDE1}, $e^{\epsilon_1,\epsilon_2}_{\mbox{\tiny heat}} (a,b,\xi)$ may be approximated by the sequence of SDDEs $\{e_{\mbox{\tiny heat}}^{\mbox{\tiny SDDE}}(a,b,\xi,n)\}_{n\in\mathbb{N}}$: \begin{equation} \begin{gathered} \begin{aligned} d\tilde{U}^x_{n,\epsilon_1,\epsilon_2}(t) &= \big[\frac{\epsilon_1}{2}\Delta_n\tilde{U}^x_{n,\epsilon_1,\epsilon_2}(t) +b(\tilde{U}^x_{n,\epsilon_1,\epsilon_2}(t))\big]dt\\ &\quad +\epsilon_2a(\tilde{U}^x_{n,\epsilon_1,\epsilon_2}(t)) \dfrac{dW_n^x(t)}{\delta_n^{1/2}}; \quad (t,x)\in(0,T]\times\mathbb{X}_n, \end{aligned}\\ \Tilde{U}_{n,\epsilon_1,\epsilon_2}^x(0) = \xi(x); \quad x\in\mathbb{X}_n, \end{gathered} \label{SDDE} \end{equation} where $\Delta_n f(x)$ is the $n$-th discrete Laplacian \begin{equation} \Delta_nf(x) = \frac{f(x+\delta_n)-2f(x)+f(x-\delta_n)}{\delta_n^2}. \label{LG} \end{equation} For each $n\in\mathbb{N}$, we think of $W_n^x(t)$ as a sequence of independent standard Brownian motions indexed by the set $\mathbb{X}_n$ (independence within the same lattice). We also assume that if $m\neq n$ and $x\in\mathbb{X}_m\cap\mathbb{X}_n$ then $W_m^x(t)=W_n^x(t)$, and if $n>m$ and $x\in\mathbb{X}_n\setminus\mathbb{X}_m$ then $W_m^x(t)=0$. \begin{defn} \label{dfn1.1} Fix $\epsilon_1,\epsilon_2$. A solution to the SDDE system $\{e_{\mbox{\tiny heat}}^{\mbox{\tiny SDDE}}(a,b,\xi,n)\}_{n=1}^\infty$; with respect to the Brownian (in $t$) system $\{W_n^x(t)\}_{(n,x)\in\mathbb{N}\times\mathbb{X}_n}$ on the filtered probability space $(\Omega,\eus{F}\{\eus{F}_t\},\mathbb{P})$; is a sequence of real-valued processes $$ \left\{\tilde{U}_{n,\epsilon_1,\epsilon_2} =\{\tilde{U}^x_{n,\epsilon_1,\epsilon_2}(t);(t,x)\in(\mathbb{T}\times\mathbb{X}_n)\} \right\}_{n=1}^\infty $$ with continuous sample paths in $t$ for each fixed $x\in\mathbb{X}_n$ and $n\in\mathbb{N}$ such that, for every $(n,x)\in\mathbb{N}\times\mathbb{X}_n$, $\tilde{U}^x_{n,\epsilon_1,\epsilon_2}(t)$ is $\eus{F}_t$-adapted, and \begin{equation} \begin{aligned} \tilde{U}^x_{n,\epsilon_1,\epsilon_2}(t) &=\int_0^t\left[\frac{\epsilon_1}{2}\Delta_n\tilde{U}^{x}_{n,\epsilon_1, \epsilon_2}(s)+b(\tilde{U}^{x}_{n,\epsilon_1,\epsilon_2}(s))\right]ds +\epsilon_2a(\tilde{U}^{x}_{n,\epsilon_1,\epsilon_2}(s)) \frac{dW_n^x(s)}{\delta_n^{1/2}} \\ &\quad+\xi(x);\quad(t,x)\in\mathbb{T}\times\mathbb{X}_n,\ n\in\mathbb{N},\mbox{ a.s. }\mathbb{P}. \end{aligned} \label{ISDDE} \end{equation} A solution is said to be strong if $\{W_n^x(t)\}_{(n,x)\in\mathbb{N}\times\mathbb{X}_n}$ and $(\Omega,\eus{F}\{\eus{F}_t\},\mathbb{P})$ are fixed a priori$;$ and with \begin{equation} \eus{F}_t=\sigma\left\{\sigma\left(W_n^x(s);0\le s\le t,x\in\mathbb{X}_n,n\in\mathbb{N}\right)\cup \eus{N}\right\};\quad t\in\mathbb{T}, \label{filt} \end{equation} where $\eus{N}$ is the collection of null sets $$ \Big\{O:\exists\,G\in\sigma\Big( \bigcup_{t\ge0}\sigma\left(W_n^x(s);0\le s\le t,x\in\mathbb{X}_n,n\in\mathbb{N}\right)\Big),\ O\subseteq G\ \mbox{and}\ \mathbb{P}(G)=0\Big\} $$ A solution is termed weak if we are free to choose $(\Omega,\eus{F}\{\eus{F}_t\},\mathbb{P})$ and the Brownian system on it and without requiring $\eus{F}_t$ to satisfy $\eqref{filt}$. \end{defn} Unless otherwise stated all filtrations are assumed to satisfy the usual conditions, and any filtered probability space with such filtration is called a usual probability space. Now, as in Lemma 2.1 in \cite{ASDDE1}, we easily have the following representation and existence result for our approximating SDDEs \begin{lem} \label{GSDDE} Under the conditions \begin{equation} \begin{aligned} &(a)\quad a(u) \mbox{ and } b(u)\mbox{ are continuous in }u; \quad u\in\mathbb{R},\\ &(b)\quad a^2(u)\le K(1+u^{2})\mbox{ and }b^2(u)\le K(1+u^{2});\quad u\in\mathbb{R}, \\ &(c)\quad \xi \mbox{ is continuous, nonrandom, and bounded on }\mathbb{R}, \end{aligned} \label{acond} \end{equation} for some constant $K>0$, the SDDE system $\eqref{SDDE}$ is equivalent to the discrete-space continuous-time Green function formulation \begin{equation} \begin{split} \tilde{U}^x_{n,\epsilon_1,\epsilon_2}(t) &= \sum_{y\in\mathbb{X}_n}\int_0^t {Q_{\delta_n,\epsilon_1}^{t-s;x-y}} \Big[\epsilon_2{a(\tilde{U}_{n,\epsilon_1,\epsilon_2}^y(s))} \frac{dW_n^y(s)} {\delta_n^{1/2}}+{b(\tilde{U}_{n,\epsilon_1,\epsilon_2}^y(s))} ds\Big]\\ &\quad +\sum_{y\in\mathbb{X}_n} {Q_{\delta_n,\epsilon_1}^{t;x-y}}\xi(y); \quad (t,x)\in\mathbb{T}\times\mathbb{X}_n, \end{split} \label{GSDDEeq} \end{equation} where $Q_{\delta_n,\epsilon_1}^{t:x}$ is the fundamental solution to the parametrized deterministic heat equation on the lattice $\mathbb{X}_n$: $$ \frac{du_{n,\epsilon_1}^x(t)}{dt} = \frac{\epsilon_1}{2}\Delta_nu_{n,\epsilon_1}^x(t); \quad (t,x)\in(0,T]\times\mathbb{X}_n. $$ Furthermore, there is at least one weak solution to $\eqref{SDDE}$ $($and hence $\eqref{GSDDEeq})$. \end{lem} Similar to the zero drift case (Lemma 2.1 in \cite{ASDDE1}), the equivalence assertion in Lemma \ref{GSDDE} follows as in the continuous time-space case (e.g., Walsh \cite{WA}) from an equivalence of test function and Green function formulations argument, and the existence is a straightforward generalization of standard SDEs arguments and the details will be omitted. \begin{rem} \rm Just as in the continuous time-case, where we can look probabilistically at the fundamental solution of the deterministic heat equation as the density of Brownian motion, we note that $Q_{\delta_n,\epsilon_1}^{t;x}$ is the density of a symmetric $1$-dimensional random walk on $\mathbb{X}_n$, in which the times between transitions are exponentially distributed with mean $\delta_n^2/\epsilon_1$, for all $n$. To simplify notations, we will suppress the dependence on the parameters $\epsilon_1,\epsilon_2$ unless we want to expressly consider the effect of variations in them on the SPDE or SDDE (the subscript $\delta_n$ in $Q^{t;x}_{\delta_n}$ is to remind us that the lattice points are $\delta_n$ apart). Of course, by enlarging the filtration $\eus{F}_t$, we can accomodate random initial $\xi$. Also, if the space $\mathbb{R}$ is replaced by a closed bounded interval $\mathbb{L}=[a,b]$, $a,b\in\mathbb{R}$ then $\mathbb{X}_n$ is replaced by $\mathbb{X}_n\cap\mathbb{L}$; and the random walk will be either reflected or absorbed at $a\mbox{ and }b$, with corresponding densities $Q_{\delta_n}$, depending on whether we have Neumann or Dirichlet boundary conditions. \label{Qasdensity} \end{rem} Using linear interpolation, we extend the definition of the already continuous-in-time process $\tilde{U}^{x}_n(t)$ on $\mathbb{T}\times\mathbb{X}_n$, so as to obtain a continuous process on $\mathbb{T}\times\mathbb{R}$, for each $n\in\mathbb{N}$, which we will also denote by $\tilde{U}^{x}_n(t)$. Henceforth, any such sequence $\{\tilde U_n\}$ of interpolated $\tilde U_n$'s will be called a continuous solution to $e_{\mbox{\tiny heat}}^{\mbox{\tiny SDDE}}(a,b,\xi,n)$. We now give our definitions of SDDEs limit solutions to $e^{\epsilon_1,\epsilon_2}_{\mbox{\tiny heat}} (a,b,\xi)$ (of course solutions on $\mathbb{R}_+\times\mathbb{R}$ are defined in the same way, replacing $\mathbb{T}$ with $\mathbb{R}_+$). \begin{defn} \rm [SDDEs limits solutions to $e^{\epsilon_1,\epsilon_2}_{\mbox{\tiny heat}} (a,b,\xi)$] We say that the random field $U(t,x)$ is a continuous SDDE limit solution to $e^{\epsilon_1,\epsilon_2}_{\mbox{\tiny heat}} (a,b,\xi)$ on $\mathbb{T}\times\mathbb{R}$ iff there is a continuous solution $\{\tilde{U}^{x}_n(t)\}$ to the SDDE system $e_{\mbox{\tiny heat}}^{\mbox{\tiny SDDE}}(a,b,\xi,n)$ on a usual probability space $(\Omega,\eus{F}\{\eus{F}_t\},\mathbb{P})$ and with respect to a Brownian system $\{W_n^x(t)\}_{(n,x)\in\mathbb{N}\times\mathbb{X}_n}$ such that $U$ has $\mathbb{P}$-a.s. continuous paths on $\mathbb{T}\times\mathbb{R}$, and $U$ is the limit or a modification of the limit of $\tilde U_n$ (or of a subsequence $\tilde{U}_{n_k}$) as $n\nearrow\infty$ ($k\nearrow\infty$). When desired; the types of the solution and the limit are explicitly stated (e.g., we say strong (weak) SDDEs weak, in probability, $L^p$, or a.s.~limit solution to indicate that the solution to the approximating SDDEs system is strong (weak) and that the limit of the SDDEs is in the weak, in probability, $L^p$, or a.s.~sense, respectively). We say that uniqueness in law holds if whenever $U$ and $V$ are SDDEs limit solutions, $U$ and $V$ have the same law. \label{limitsolns} \end{defn} \begin{rem} \rm Although in this article we restrict our treatment to the weak SDDEs weak limit solutions and weak uniqueness (in law), Definition \ref{limitsolns} easily admits limits in any sense, not only those mentioned above, as well as stronger uniqueness (pathwise). \label{defnrem} \end{rem} Next, we use the existence in Lemma \ref{GSDDE} to show the existence of a SDDE limit solution to $e^{\epsilon_1,\epsilon_2}_{\mbox{\tiny heat}} (a,b,\xi)$, and that this solution is a solution in the standard sense of satisfying the test function formulation of $e^{\epsilon_1,\epsilon_2}_{\mbox{\tiny heat}} (a,b,\xi)$. This complements the results in \cite{ASDDE1} by treating the case of continuous drift which grows no more than linearly. \begin{thm} [Weak existence, uniqueness, and regularity] Fix $\epsilon_1$ and $\epsilon_2$. If the conditions in \eqref{acond} hold for some constant $K>0$, then every sequence of continuous SDDEs solutions is tight in $C(\mathbb{T}\times\mathbb{R};\mathbb{R})$ and we have a weak SDDE weak limit solution $U$ to $e^{\epsilon_1,\epsilon_2}_{\mbox{\tiny heat}} (a,b,\xi)$. Moreover, there is a continuous random field $Y,$ with the same law as $U,$ that satisfies the test function formulation \ of $e^{\epsilon_1,\epsilon_2}_{\mbox{\tiny heat}} (a,b,\xi)$: \begin{equation} \begin{split} &(Y(t)-\xi,\varphi)-\frac{\epsilon_1}{2}\int_0^t(Y(s),\varphi'')ds-\int_0^t(b(Y(s)),\varphi)ds\\ &= \epsilon_2\int_0^t \int_{\mathbb{R}}a(Y(s,x))\varphi(x)\eus{W}(dx,ds) \end{split} \label{TFF} \end{equation} for every $\varphi\in C_c^{\infty}(\mathbb{R};\mathbb{R})$, where $(\cdot,\cdot)$ denotes the scalar product on $L^2(\mathbb{R})$. The continuous paths of $Y$ are H\"older $\gamma_s\in(0,\frac12)$ in space and H\"older $\gamma_t\in(0,\frac14)$ in time and are $L^p$ bounded for every $p\ge2$. If $a(u)=u^\gamma$, with $1/2\le\gamma\le1$ and $b(u)=\sum_{i=1}^N c_iu^{\alpha_i}$ for constants $c_i\in\mathbb{R}$, $N\in\mathbb{N}$, and $1\ge\alpha_i\ge\gamma$, $i=1,\ldots,N;$ then uniqueness in law holds for $e^{\epsilon_1,\epsilon_2}_{\mbox{\tiny heat}} (a,b,\xi)$ on $[0,T]\times[0,L]$, for any $T,L\in\mathbb{R}_+^2$, and hence the convergence to $U$ is along the whole SDDEs sequence. \label{ww} \end{thm} \begin{rem} \rm In Theorem \ref{ww} we suppressed the dependence on the epsilons, since the results are given for fixed $\epsilon_1,\epsilon_2$; i.e., $U=U_{\epsilon_1,\epsilon_2}$ and $Y=Y_{\epsilon_1,\epsilon_2}$. \label{fixedepsilon} \end{rem} The next result reveals that the behavior of $e^{\epsilon_1,\epsilon_2}_{\mbox{\tiny heat}} (a,b,\xi)$ as $\epsilon_1,\epsilon_2\searrow0$ is controlled by the ratio ${\epsilon_2}/{{ \epsilon_1}^{1/4}}$: the solutions blow up in $L^2$ if ${\epsilon_2}/{{ \epsilon_1}^{1/4}}\nearrow\infty$ and they converge to the deterministic $\epsilon_1\searrow0$ limit in $L^{2q}$ ($q\ge1$) if ${\epsilon_2}/{{ \epsilon_1}^{1/4}}\searrow0$. \begin{thm} [Limits of $e^{\epsilon_1,\epsilon_2}_{\mbox{\tiny heat}} (a,b,\xi)$ as $\epsilon_1,\epsilon_2\searrow0$] \label{thm2} $\mbox{i})$ Suppose that the conditions in \eqref{acond} hold, and that there are constants $K_l,K_L>0$ such that $K_l\le a(u)\le K_L$ for all $u\in\mathbb{R}$. If $\epsilon_1,\epsilon_2\searrow0$ (or $\epsilon_1,\epsilon_2\nearrow\infty$) such that the ratio ${\epsilon_2}/{{ \epsilon_1}^{1/4}}\to\infty;$ then, $\sup_{0\le s\le T}\sup_{{x\in\mathbb{R}}}\mathbb{E} U_{\epsilon_1,\epsilon_2}^{2}(s,x)\nearrow\infty$, for any $T>0$ and for any SDDEs weak limit solution $U_{\epsilon_1,\epsilon_2}$ to $e^{\epsilon_1,\epsilon_2}_{\mbox{\tiny heat}} (a,b,\xi)$ $\mbox{ii})$ If, in addition to the conditions in \eqref{acond}, $b$ is Lipschitz and $a$ is bounded; and if $U_{\epsilon_1}(t,x)$ is the solution to the deterministic PDE obtained from $e^{\epsilon_1,\epsilon_2}_{\mbox{\tiny heat}} (a,b,\xi)$ by setting $a\equiv0$, and $U_{\epsilon_1,\epsilon_2}$ is a SDDEs weak limit solution to $e^{\epsilon_1,\epsilon_2}_{\mbox{\tiny heat}} (a,b,\xi)$, then for every $q\ge1$ $$\sup_{ 0\le t\le T}\sup_{x\in\mathbb{R} }\mathbb{E}\left|U_{\epsilon_1,\epsilon_2}(t,x)-U_{\epsilon_1}(t,x)\right|^{2q}\longrightarrow0$$ as $\epsilon_1,\epsilon_2, \mbox{ and }\epsilon_2/\epsilon_1^{1/4}\to 0$. Also, if $\left\{\tilde{U}^x_{n,\epsilon_1,\epsilon_2}(t)\right\}$ is a solution to the SDDEs system $\left\{e_{\mbox{\tiny heat}}^{\mbox{\tiny SDDE}}(a,b,\xi,n)\right\}$ and if $\left\{\tilde{U}_{n,\epsilon_1}\right\}$ is a solution to the deterministic system obtained from $e_{\mbox{\tiny heat}}^{\mbox{\tiny SDDE}}(a,b,\xi,n)$ by setting $a\equiv0$, then for every $q,n\ge1$ $$\sup_{ 0\le t\le T}\sup_{x\in\mathbb{R} }\mathbb{E}\left|\tilde{U}^x_{n,\epsilon_1,\epsilon_2}(t)-\tilde{U}_{n,\epsilon_1}(t,x)\right|^{2q}\longrightarrow0$$ as $\epsilon_1,\epsilon_2, \mbox{ and }\epsilon_2/\epsilon_1^{1/4}\to 0$. \end{thm} \begin{rem} \rm Taking note of Remark \ref{fixedepsilon}, it follows from well known facts in Walsh \cite{WA} that, under \eqref{acond}, \eqref{TFF} is equivalent to the Green function formulation \eqref{GFF} below \begin{equation} \begin{split} &Y_{\epsilon_1,\epsilon_2}(t,x)-\int_{\mathbb{R}}G_{\epsilon_1}(t;x,y)\xi(y)dy\\&=\int_{\mathbb{R}}\int_0^t G_{\epsilon_1}(s,t;x,y)\Big\lbrack{\epsilon_2 a(Y_{\epsilon_1,\epsilon_2}(s,y))} \eus{W}(ds,dy)+ b(Y_{\epsilon_1,\epsilon_2}(s,y))dsdy\Big\rbrack, \end{split} \label{GFF} \end{equation} where $$ G_{\epsilon_1}(t;x,y)=\frac{1}{\sqrt{2\epsilon_1\pi t}}e^{-(x-y)^2/2\epsilon_1t}. $$ Also, note that when $\epsilon_1,\epsilon_2$ are fixed, $\sup_{0\le s\le T}\sup_{{x\in\mathbb{R}}}\mathbb{E} U_{\epsilon_1,\epsilon_2}^{2}(s,x)<\infty $ for all $T>0$ (see Proposition \ref{Expbd}, \eqref{Lp}, and note that $Y$ has the same law as $U$). \label{rm2} \end{rem} \section{Existence, uniqueness, and regularity} The proof of Theorem \ref{ww} proceeds in several steps as in the heat SPDE in \cite{ASDDE1}, with the extra difficulty caused by the extra term $b(U)$: we first get Kolmogorov type estimates on the spatial and temporal differences of the continuous $\tilde{U}^{x}_n(t)$'s establishing tightness, and so by Lemma \ref{GSDDE} and Definition \ref{limitsolns} this implies the existence of a weak SDDE weak limit solution to $e^{\epsilon_1,\epsilon_2}_{\mbox{\tiny heat}} (a,b,\xi)$. Then we show that the limit satisfies the test function formulation of $e^{\epsilon_1,\epsilon_2}_{\mbox{\tiny heat}} (a,b,\xi)$. Since Theorem \ref{ww} is stated for fixed $\epsilon_1,\epsilon_2$, we suppress the dependence on these parameters (except in Lemma \ref{2ndQinequality} below which we use in Remark \ref{SDDEno} in the third section), and we assume without loss of generality that they are both $1$. Throughout the article, $K$ will denote a constant that may change its value from one step to the next. \subsection{Random walk estimates} The first set of estimates we need are bounds on the random walk density $Q^{t;x}_{\delta_n}$. Since all the results in this section hold for all $n$, we will suppress the dependence on $n$, except in \eqref{determ}, to simplify the notation. The first three lemmas are taken directly from \cite{ASDDE1} p.~32 and are reproduced below for convenience: \begin{lem} There is a constant $K$ such that \[\sum_{x\in\mathbb{X}} {(Q^{t:x}_\delta)}^2 \leq K\delta/\sqrt t\quad\mbox{ and }\quad\sum_{x\in\mathbb{X}} {(Q_{\delta,\epsilon_1}^{t;x})}^2 \leq K\delta/\sqrt{\epsilon_1 t}, \] and hence \[\int_0^t\sum_{x\in\mathbb{X}} {(Q^{s;x}_\delta)}^2 \, ds \leq K\delta\sqrt t \quad\mbox{ and }\quad\int_0^t\sum_{x\in\mathbb{X}} {(Q_{\delta,\epsilon_1}^{s;x})}^2 \, ds \leq K\delta\sqrt{\frac{t}{\epsilon_1}}.\] \label{2ndQinequality} \end{lem} \begin{lem} There is a constant $K$ such that \[\int_0^t\sum_{x\in\mathbb{X}} {(Q^{s;x}_\delta - Q^{s;x+z}_\delta)}^2 ds \leq K\delta |z|.\] \label{3rdQinequality} \end{lem} \begin{lem} There is a constant $K$ such that \[\int_0^t\sum_{x\in\mathbb{X}} {(Q^{t-s;x}_\delta - Q^{r-s;x}_\delta)}^2 ds \leq K\delta\sqrt {t-r},\] for $r0$ such that, whenever $\delta<\delta^*$, the total variation distance in \eqref{ttlvdist} is bounded by a constant multiple of $(t^{1/2} - s^{1/2}) / t^{1/2}$ as well, with a possibly different constant. So, for small enough $\delta$, we get the first inequality. The second inequality follows upon using the fact that $(t^{1/2} - s^{1/2}) / t^{1/2} \leq (t-s) / t$ and integrating, using the convention that $Q_{\delta}^{t;x}=0$ if $t<0$. \end{proof} From this point on, and without explicitly stating it, we will assume $\delta<\delta^*$ whenever needed. Consequently, we have the following statement. \begin{lem} There is a constant $K$, depending only on $T$, such that $$ \int_0^t\sum_{x\in\mathbb{X}} \left|{Q^{t-s;x}_\delta - Q^{r-s;x}_\delta}\right|ds \le K\big[(t-r)+ (t-r)^{1-e^{-1}}\big], $$ for $0\le r0$ independent of $k$, where the last two inequalities follow from Burkholder's inequality, the boundedness and compact supportedness of $\varphi$, the growth condition on $a$ ((b) in \eqref{acond}), and Proposition \ref{Expbd}. Thus, uniform integrability of the sequence $\{S^\varphi(Y_k,t)\}_k$ follows for each $\varphi$ and each $t$. So, If $s < t$ \begin{equation*} \begin{split} &\mathbb{E}\left[ M^\varphi(t) - M^\varphi(s)\left| \eus{G}_s\right.\right] = \lim_{k\to\infty}\mathbb{E}\left[S^\varphi(Y_k,t) - S^\varphi(Y_k,s)\left| \eus{G}_s\right.\right] = 0. \end{split} \end{equation*} This proves (i). \noindent (ii)\quad From \eqref{SDDE} it follows that \begin{equation*} \begin{split} & d \Big[\sum_{x\in {\mathbb{X}}_{n_k}}\tilde{U}^x_{n_k}(t)\varphi(x)\delta_{n_k} \Big]\\ &= \sum_{x\in {\mathbb{X}}_{n_k}} a\left(\tilde{U}^x_{n_k}(t)\right)\varphi(x) \sqrt{\delta_{n_k}}dW^x_{n_k}(t) \\ &\quad+\Big[\sum_{x\in {\mathbb{X}}_{n_k}}\big(\frac12 \Delta_{n_k}\tilde{U}^x_{n_k} (t)+b(\tilde{U}^x_{n_k}(t))\big)\varphi(x)\delta_{n_k}\Big]dt\\ &= \Big[\frac12\sum_{x\in {\mathbb{X}}_{n_k}}\tilde{U}^x_{n_k}(t) \Delta_{n_k}\varphi(x)\delta_{n_k} + \sum_{x\in {\mathbb{X}}_{n_k}}b(\tilde{U}^x_{n_k}(t))\varphi(x)\delta_{n_k}\Big]dt\\ &+\sum_{x\in {\mathbb{X}}_{n_k}} a\left(\tilde{U}^x_{n_k}(t)\right)\varphi(x) \sqrt{\delta_{n_k}}dW^x_{n_k}(t). \end{split} \end{equation*} Observing that the first two terms on the right hand side of the last equality in the above equation are of bounded variation, and that the $\left(W^x_{n_k}(t)\right)_{x\in {\mathbb{X}}_{n_k}}$ is a sequence of independent Brownian motions, we obtain, after inspecting \eqref{sumofsubseq}, that \begin{equation} \Big\langle S^\varphi(\tilde{U}_{n_k},\cdot)\Big\rangle_t =\Big\langle{\sum_{x\in {\mathbb{X}}_{n_k}}\tilde{U}^x_{n_k}(\cdot) \varphi(x)\delta_{n_k}}\Big\rangle_t = \int_0^t \Big[\sum_{x\in {\mathbb{X}}_{n_k}}a^2\left(\tilde{U}^x_{n_k}(s)\right) \varphi^2(x)\delta_{n_k}\Big]ds \label{QVoftheutilde} \end{equation} Again, replacing the $\tilde{U}^x_{n_k}(t)$ in \eqref{QVoftheutilde} by the $Y_k(t,x)$, we get, for $0\le r\le t\le T$, \begin{equation} %\begin{split} \mathbb{E}\left[\left(S^\varphi(Y_k,t)-S^\varphi(Y_k,r)\right)^2 \big|\eus{G}_r\right] =\mathbb{E}\Big[\int_r^t\sum_{x\in {\mathbb{X}}_{n_k}}a^2\left(Y_k(s,x)\right) \varphi^2(x)\delta_{n_k}ds\Big|\eus{G}_r\Big]. %\end{split} \label{expl} \end{equation} Again, we observe that $\left(S^\varphi(Y_k,t)-S^\varphi(Y_k,r)\right)^2$ are uniformly integrable, for each $r$ and $t$ and each $\varphi$. To see that, fix $p\ge1$, $0\le r\le t\le T$, and $\varphi\in C_c^{\infty}(\mathbb{R};\mathbb{R})$; and apply Burkholder's inequality to obtain \begin{equation} \begin{split} \mathbb{E}\left|S^\varphi(Y_k,t)-S^\varphi(Y_k,r)\right|^{2p} &= \mathbb{E}\Big|\int_r^t\sum_{x\in {\mathbb{X}}_{n_k}}a\left(Y_k(s,x)\right) \varphi(x)\sqrt{\delta_{n_k}}dW^x_{n_k}(s)\Big|^{2p}\\ &\le K\mathbb{E} \Big|\sum_{x\in {\mathbb{X}}_{n_k}}\int_0^ta^2\left(Y_k(s,x) \right)\varphi^2(x)\delta_{n_k}ds\Big|^{p} \end{split} \label{ui2} \end{equation} for some constant $K>0$ independent of $k$. Now, let ${\eta}_{_k}^t$ be the measure defined on $[0,t]\times{\mathbb{X}}_{n_k}$ by $d{\eta}_{_k}^t(s,x)=\varphi^2(x)\delta_{n_k}ds$ and let $|{\eta}_{_k}^t| = {\eta}_{_k}^t([0,t] \times{\mathbb{X}}_{n_k})$. Clearly, for a fixed $\varphi$, $$ \sup_{\substack{k\in\mathbb{N}\\0\le t\le T}}|{\eta}_{_k}^t|\le K $$ for some constant $K>0$ independent of $k$ ($K$ depends only on $T$, $\sup_x\varphi^2(x)$, and the Lebesgue measure of the support of $\varphi$). Then, rewriting \eqref{ui2} and---observing that ${\eta}_{_k}^t/|{\eta}_{_k}^t|$ is a probability measure---applying Jensen's inequality yields \begin{equation} \begin{split} &\mathbb{E}\left|S^\varphi(Y_k,t)-S^\varphi(Y_k,r)\right|^{2p}\\ &\le K\mathbb{E} \Big|\int_{[0,t]\times{\mathbb{X}}_{n_k}} a^2\left(Y_k(s,x)\right) \frac{d{\eta}_{_k}^t(s,x)}{|{\eta}_{_k}^t|}\Big|^{p}|{\eta}_{_k}^t|^p \\ &\le {|{\eta}_{_k}^t|}^{p-1}K\int_{[0,t]\times{\mathbb{X}}_{n_k}}\mathbb{E}\left[a^{2p}\left(Y_k(s,x)\right)\right]d{\eta}_{_k}^t(s,x)\\ &\le K\sum_{x\in {\mathbb{X}}_{n_k}}\int_0^T\mathbb{E}\left[a^{2p}\left(Y_k(s,x)\right)\right]\varphi^{2}(x){\delta_{n_k}}ds \le K <\infty, \end{split} \label{ui70} \end{equation} for some constant $K>0$ independent of $k$, where in the next to last inequality we also used the growth condition on $a$ ((b) in \eqref{acond}) and Proposition \ref{Expbd} ($Y_k \overset{d}{=} \Tilde{U}_{n_k}$), along with the compact supportedness and boundedness of $\varphi$. Thus, \begin{equation} \lim_{k\to\infty}\mathbb{E}\left[\left.\left(S^\varphi(Y_k,t)-S^\varphi(Y_k,r)\right)^2\right|\eus{G}_r\right]\\ =\mathbb{E}\left[\left.\left(M^\varphi(t)-M^\varphi(r)\right)^2\right|\eus{G}_r\right]. \label{expl1} \end{equation} Also, for the same reasons as in the next to last inequality in \eqref{ui70}, we see that $$ \mathbb{E}\int_r^t\sum_{x\in {\mathbb{X}}_{n_k}}a^2\,\left(Y_k(s,x)\right) \varphi^2(x)\delta_{n_k}ds\le K<\infty, $$ for some constant $K>0$ independent of $k$. Therefore, for each $r,t$ and each $\varphi$, $$ \Big\{\int_r^t\sum_{x\in {\mathbb{X}}_{n_k}}a^2\,\left(Y_k(s,x)\right) \varphi^2(x)\delta_{n_k}ds\Big\}_k $$ is a uniformly integrable sequence and thus \begin{equation} \begin{split} &\lim_{k\to\infty}\mathbb{E}\Big[\int_r^t\sum_{x\in {\mathbb{X}}_{n_k}}a^2\, \left(Y_k(s,x)\right)\varphi^2(x)\delta_{n_k}ds\Big|\eus{G}_r\Big]\\ &=\mathbb{E}\Big[\int_r^t\int_{\mathbb{R}}a^2\left(Y^x(s)\right)\varphi^2(x)dx ds \Big|\eus{G}_r\Big]. \end{split} \label{expl2} \end{equation} Now, equations \eqref{expl}, \eqref{expl1}, and \eqref{expl2} yield \begin{equation} \left\langle{M^\varphi(\cdot)}\right\rangle_t = \int_0^t \int_{{\mathbb{R}}}a^2(Y^x(s))\varphi^2(x)dx ds, \label{QVoftheY} \end{equation} and (ii) is proved. \end{proof} \subsection{Regularity and Uniqueness} Having established existence for the SPDE $e^{\epsilon_1,\epsilon_2}_{\mbox{\tiny heat}} (a,b,\xi)$ under our conditions \eqref{acond}, we turn to the proof of some properties of our solution $Y$. \begin{proof}[Proof of the regularity part of Theorem \ref{ww}] We divide the proof in two steps: \noindent (1) $Y$ is $L^p$ bounded for all $p\ge2$:\quad First, note that $Y_k\overset{d}{=}\tilde{U}_{n_k} $ and Proposition \ref{Expbd} give us, for each $q\ge1$: \begin{equation} \mathbb{E}\left|Y_k(t,x)\right|^{2q} =\mathbb{E}\left|\tilde{U}^x_{n_k}(t)\right|^{2q} \le K\exp{(KT)}<\infty;\quad \forall (k,t,x)\in\mathbb{N}\times\mathbb{T}\times\mathbb{R}, \label{ui7} \end{equation} for some constant $K$ (independent of $k,t,x$). It follows that, for each $(t,x)\in\mathbb{T}\times\mathbb{R}$ the sequence $\{|Y_k(t,x)|^p\}_k$ is uniformly integrable for each $p\ge2$. Thus, \begin{equation} \mathbb{E}|Y(t,x)|^{p}\le\lim_{k\to\infty}\mathbb{E}\left|Y_k(t,x)\right|^{p} \le K<\infty;\ \forall (t,x)\in\mathbb{T}\times\mathbb{R},\ \forall p\ge2, \label{Lp} \end{equation} and the desired conclusion follows. \noindent (2) The continuous paths of $Y$ are H\"older $\gamma_s\in(0,\frac12)$ in space and H\"older $\gamma_t\in(0,\frac14)$ in time:\quad Using Proposition \ref{Expbd}, we get, for each $q\ge1$, that \begin{equation} \begin{split} &\mathbb{E}\left|Y_k(t,x)-Y_k(t,y)\right|^{2q}+\mathbb{E}\left|Y_k(t,x)-Y_k(r,x)\right|^{2q}\\&\le K\left(\mathbb{E}|Y_k(t,x)|^{2q}+\mathbb{E}|Y_k(t,y)|^{2q}+\mathbb{E}|Y_k(r,x)|^{2q}\right)\\ &\le K;\quad \forall (k,r,t,x,y)\in\mathbb{N}\times\mathbb{T}^2\times\mathbb{R}^2. \end{split} \label{ui14} \end{equation} So, for each $(r,t,x,y)\in\mathbb{T}^2\times\mathbb{R}^2$, the sequences $\big\{\left|Y_k(t,x)-Y_k(t,y)\right|^{2q}\big\}_k$ and $\big\{\left|Y_k(t,x)-Y_k(r,x)\right|^{2q}\big\}_k$ are uniformly integrable, for each $q\ge1$. Therefore, using Lemma \ref{boundsonsdifferences} and Lemma \ref{boundsontdifferences}, we obtain \begin{equation} \begin{split} (i)\quad&\mathbb{E}\left|Y(t,x)-Y(t,y)\right|^{2q}\\ &=\lim_{k\to\infty}\mathbb{E}\left|Y_k(t,x)-Y_k(t,y)\right|^{2q}\\ &=\lim_{k\to\infty}\mathbb{E} \left|\tilde{U}_{n_k}^x(t)-\tilde{U}_{n_k}^y(t)\right|^{2q} \le K|x-y|^q; \quad \mbox{whenever } |x-y|<1,\\ (ii)\quad&\mathbb{E}\left|Y(t,x)-Y(r,x)\right|^{2q}\\ &=\lim_{k\to\infty}\mathbb{E}\left|Y_k(t,x)-Y_k(r,x)\right|^{2q}\\ &=\lim_{k\to\infty}\mathbb{E}\left|\tilde{U}_{n_k}^x(t)-\tilde{U}_{n_k}^x(r) \right|^{2q} \le K|t-r|^{q/2}; \quad\mbox{whenever } |t-r|<1. \end{split} \label{timespace} \end{equation} Now, letting $q_n=n+1$ for $n\in\{0,1,\ldots\}$ and let $n=m+1$ for $m\in\{0,1,\ldots\}$, we then have from \eqref{timespace} that \begin{equation} \begin{split} (i)\quad &\mathbb{E}\left|Y(t,x)-Y(t,y)\right|^{2+2n} \le K|x-y|^{1+n};\quad\mbox{whenever } |x-y|<1,\\ (ii)\quad &\mathbb{E}\left|Y(t,x)-Y(r,x)\right|^{4+2m} \le K|t-r|^{1+\frac{m}{2}}; \quad\mbox{whenever } |t-r|<1. \end{split} \label{timespace2} \end{equation} By Theorem 2.8 p.~53 \cite{KS} we get that $\gamma_s\in(0,\frac{n}{2n+2}) $ and $\gamma_t\in(0,\frac{m/2}{2m+4})$ $ \forall m,n$, from which the proof follows upon taking the limits as $m,n\to\infty$. \end{proof} \begin{proof}[Proof of the uniqueness part of Theorem \ref{ww}] Consider $e^{\epsilon_1,\epsilon_2}_{\mbox{\tiny heat}} (a,b,\xi)$ on the rectangle $\eus{R}_{T,L}\overset{\triangle}{=} [0,T]\times[0,L]$ for some $T,L>0$, and assume that $a(u)$ and $b(u)$ are as given in Theorem \ref{ww}. Then as in the proof of Theorem 1.2 in \cite{ACR1} (see also the comment after Remark 1.1 in \cite{ACR2}), we only need to show that, if $\lambda$ is Lebesgue measure on $\eus{R}_{T,L}$, then the ratios $b(U)/a(U)$ and $b(V)/a(V)$ are in $L^2(\eus{R}_{T,L},\lambda)$ almost surely whenever $U$ solves $e^{\epsilon_1,\epsilon_2}_{\mbox{\tiny heat}}(a,0,\xi)$ and $V$ solves $e^{\epsilon_1,\epsilon_2}_{\mbox{\tiny heat}}(a,b,\xi)$. But this easily follows as in the proof of Theorem 1.2 \cite{ACR1} under our conditions, since we always assume that solutions to $e^{\epsilon_1,\epsilon_2}_{\mbox{\tiny heat}} (a,b,\xi)$, and hence $U$ and $V$ are continuous. \end{proof} \section{The vanishing of the Laplacian vs.~noise} We now prove of Theorem \ref{thm2}, which asserts that ${\epsilon_2}/{{ \epsilon_1}^{1/4}}$ is the correct scaling of $\epsilon_1$ and $\epsilon_2$ when we investigate the asymptotic behavior as $\epsilon_1,\epsilon_2\to0$. \begin{proof}[Proof of Theorem \ref{thm2}] Throughout this proof we use the process $Y$ of Theorem \ref{ww}, Remark \ref{fixedepsilon}, and Remark \ref{rm2} to get to the desired conclusions. This is justified by the fact that $Y$ has the same law as $U$. \noindent (i)\quad We prove it by contradiction. So, assume there is a $T>0$ such that $$ \lim_{\substack{\epsilon_1,\epsilon_2\downarrow0\\ {\epsilon_2}/{{ \epsilon_1}^{1/4}}\to\infty}} \sup_{0\le s\le T}\sup_{{x\in\mathbb{R}}}\mathbb{E} Y_{\epsilon_1,\epsilon_2}^{2}(s,x) <\infty $$ and assume without loss of generality that $\xi\equiv0$. Observe that \begin{equation} \begin{split} &\mathbb{E}\left|Y_{\epsilon_1,\epsilon_2}(t,x)\right|^2\\ &=\mathbb{E} \Big| \int_{\mathbb{R}}\int_0^t G_{\epsilon_1}(s,t;x,y)\left[{\epsilon_2a(Y_{\epsilon_1,\epsilon_2}(s,y))} \eus{W}(ds,dy)+ b(Y_{\epsilon_1,\epsilon_2}(s,y))ds\,dy\right]\Big|^{2}\\ &= \epsilon_2^2\int_{\mathbb{R}}\int_0^t G_{\epsilon_1}^2(s,t;x,y) {\mathbb{E} a^2(Y_{\epsilon_1,\epsilon_2}(s,y))} ds\,dy \\ &\quad + \mathbb{E}\Big(\int_{\mathbb{R}}\int_0^t G_{\epsilon_1}(s,t;x,y) b(Y_{\epsilon_1,\epsilon_2}(s,y))dsdy\Big)^2\\ &\quad + 2\epsilon_2\mathbb{E}\Big(\int_{\mathbb{R}}\int_0^t G_{\epsilon_1} (s,t;x,y)a(Y_{\epsilon_1,\epsilon_2}(s,y)) \eus{W}(ds,dy)\\ &\quad \times\int_{\mathbb{R}}\int_0^t G_{\epsilon_1}(s,t;x,y) b(Y_{\epsilon_1,\epsilon_2}(s,y))dsdy\Big)\\ &\ge K^2_{l}\epsilon_2^2\int_{\mathbb{R}}\int_0^t G_{\epsilon_1}^2(s,t;x,y) ds\,dy +\mathbb{E}\Big(\int_{\mathbb{R}}\int_0^t G_{\epsilon_1}(s,t;x,y) b(Y_{\epsilon_1,\epsilon_2}(s,y))dsdy\Big)^2\\ &\quad 2\epsilon_2\mathbb{E}\Big(\int_{\mathbb{R}}\int_0^t G_{\epsilon_1} (s,t;x,y)a(Y_{\epsilon_1,\epsilon_2}(s,y)) \eus{W}(ds,dy)\\ &\quad\times\int_{\mathbb{R}}\int_0^t G_{\epsilon_1}(s,t;x,y) b(Y_{\epsilon_1,\epsilon_2}(s,y))dsdy\Big), \end{split} \label{as00} \end{equation} where we used the assumption $01$, case; and we will prove it in this generality. \begin{thm} Assume that $a,b,$ and $\xi$ satisfy the conditions in $\eqref{acond}$. Then, the (local) martingale problem in Theorem \ref{1stmartprob} is equivalent to the test function formulation of $e^{\epsilon_1,\epsilon_2}_{\mbox{\tiny heat}} (a,b,\xi)$. \label{marttff} \end{thm} \begin{proof} If the test function formulation of $e^{\epsilon_1,\epsilon_2}_{\mbox{\tiny heat}} (a,b,\xi)$ holds on $(\Omega,\eus{F}\{\eus{F}_t\},\mathbb{P})$, then \begin{equation} \begin{split} M^\varphi(t)&\overset{\triangle}{=} (U(t)-\xi,\varphi)-\frac12 \int_0^t(U(s), \varphi'')ds-\int_0^t(b(U(s)),\varphi)ds\\ &=\int_0^t \int_{\mathbb{R}^d} a(U(s,x))\varphi(x)\eus{W}(dx,ds)\\ &=\int_0^t \int_{{{\mathbb S}}^\varphi}a(U(s,x)) \varphi(x)\eus{W}(dx,ds), \end{split} \label{TFFa} \end{equation} where ${{\mathbb S}}^\varphi\subset\mathbb{R}^d$ is the compact support of $\varphi$. It follows from the assumptions on $a$ and the boundedness of $\varphi$ that $M^\varphi(t)$ is an $\eus{F}_t$-local martingale under $\mathbb{P}$ and that \begin{equation} \langle M^\varphi(\cdot)\rangle_t=\int_0^t \int_{S^\varphi}a^2(U(s,x))\varphi^2(x)dx\,ds= \int_0^t \int_{\mathbb{R}^d}a^2(U(s,x))\varphi^2(x)dxds. \label{QVa} \end{equation} For the other direction, assume that $M^\varphi(t)$, as defined in \eqref{TFFa}, is a local martingale on $(\Omega,\eus{F}\{\eus{F}_t\},\mathbb{P})$ with quadratic variation given by \eqref{QVa}. Suppose also that $a$ vanishes almost nowhere in $(u,\omega)\in\mathbb{R}\times\Omega$ (if this fails we can always do the same as in the finite dimensional case cf.~Ikeda and Watanabe \cite{IW} or Doob \cite{Db}). Now let $\lambda$ denote Lebesgue measure on $\eus{B}({\mathbb{R}_+\times\mathbb{R}^d})$ and on $\eus{B}({\mathbb{R}^d})$; and, for each $t\ge 0$, define the random measure $\eus{M}_t(A)=\eus{M}([0,t]\times A)$ on the ring $\eus{R}\overset{\triangle}{=} \{A\in\eus{B}(\mathbb{R}^d);\lambda([0,t] \times A)<\infty,\ \forall t>0\}$ by the recipe \begin{equation} \begin{gathered} \int_0^t\int_{A}\varphi(x)\eus{M}(dx,ds) =\int_0^t\int_{\mathbb{R}^d}\varphi(x)\eus{M}(dx,ds) \overset{\triangle}{=} M^\varphi(t); \\ \forall A\in\eus{R}\mbox{ with } \lambda(A\triangle{{\mathbb S}}^\varphi)=0 \mbox{ or such that } A \supset{{\mathbb S}}^\varphi,\ \forall\varphi\in C_c^\infty(\mathbb{R}^d;\mathbb{R}). \end{gathered} \label{martdef} \end{equation} By assumption, we have that $M^\varphi(t)$ is a continuous local martingale for each $\varphi\in C_c^\infty(\mathbb{R}^d;\mathbb{R})$. Furthermore, if $\varphi_1,\varphi_2\in C_c^\infty(\mathbb{R}^d;\mathbb{R})$ have disjoint supports (${{\mathbb S}}^{\varphi_1}\cap{{\mathbb S}}^{\varphi_2}=\phi$), then for any disjoint $A,B\in\eus{R}$ with $\lambda(A\triangle{{\mathbb S}}^\varphi_1)=0$ and $\lambda(B\triangle{{\mathbb S}}^\varphi_2)=0$ we have by the definition of $M^\varphi(t)$, the fact that the second and third terms in $M^\varphi(t)$'s definition in \eqref{TFFa} are of bounded variation, \eqref{QVa}, and \eqref{martdef} \begin{equation} \begin{split} &\Big\langle\int_0^{\cdot}\int_{A}\varphi_1(x)\eus{M}(dx,ds), \int_0^{\cdot}\int_{B}\varphi_2(x)\eus{M}(dx,ds)\Big\rangle_t\\ &=\langle(U(t)-\xi,\varphi_1),(U(t)-\xi,\varphi_2)\rangle_t\\ &=\frac14\left[\langle(U(t)-\xi,\varphi_1+\varphi_2)\rangle_t - \langle(U(t)-\xi,\varphi_1-\varphi_2)\rangle_t\right]\\ &=\int_0^t\int_{\mathbb{R}^d}a^2(U(s,x)) \varphi_1(x)\varphi_2(x)dx\,ds=0\,. \end{split} \label{ortho} \end{equation} Thus, $\eus{M}$ is a continuous orthogonal local martingale measure \cite{A298}. By the quadratic variation assumption on $M^\varphi(t)$, we also have that for each $\varphi\in C_c^\infty(\mathbb{R}^d;\mathbb{R})$ \begin{equation} \begin{split} \int_0^t \int_{\mathbb{R}^d}\varphi^2(x)a^2(U(s,x))dxds&= \Big\langle\int_0^{\cdot}\int_{\mathbb{R}^d}\varphi(x)\eus{M}(dx,ds)\Big\rangle_t \\ &=\int_0^{t}\int_{\mathbb{R}^d}\varphi^2(x)\nu_{\eus{M}}(dx,ds). \end{split} \label{QVB} \end{equation} So that the intensity measure $\nu_{\eus{M}}$ of $\eus{M}$ is given by \begin{equation} \nu_{\eus{M}}(dx,ds)=a^2(U(s,x))dxds,\mbox{ on sets of the form } [0,t]\times A,\ A\in\eus{R}. \label{intensity} \end{equation} We now show that there is a space-time white noise $\eus{W}$ such that \begin{equation} \int_0^t\int_{\mathbb{R}^d}\varphi(x)\eus{M}(dx,ds) =\int_0^t \int_{\mathbb{R}^d}a(U(s,x))\varphi(x)\eus{W}(dx,ds),\ \forall\varphi\in C_c^\infty(\mathbb{R}^d;\mathbb{R}). \label{Kalas} \end{equation} For each $A\in\eus{R}$, let $$ \eus{W}_t(A)\overset{\triangle}{=} \int_0^t\int_{A}\frac{\eus{M}(dx,ds)}{a(U(s,x))}. $$ $\eus{W}=\{\eus{W}_t(A);t\in\mathbb{R}_+, A\in\eus{R}\}$ is clearly a continuous orthogonal local martingale measure with intensity $\nu_{\eus{W}}=\lambda([0,t]\times A),$ where $\lambda$ is Lebesgue measure, so it is a white noise and clearly \eqref{Kalas} holds, completing the proof. \end{proof} \begin{thebibliography}{00} \bibitem{ACR1} Allouba, H. Uniqueness in law for the Allen-Cahn SPDE via change of measure. {\it C.~R.~Acad.~Sci.} {\bf 330} (2000), 371--376. \bibitem{ACR2} Allouba, H. SPDE law equivalence and the compact support property: applications to the Allen-Cahn SPDE. {\it C.~R.~Acad.~Sci.} {\bf 331} (2000), 245--250. \bibitem{ASDDE1} Allouba, H. A non-nonstandard proof of Reimers' existence result for heat SPDEs. {\it J. Appl. Math. Stochastic Anal.} {\bf 11} no. 1, (1998), 29--41. \bibitem{Adim} Allouba, H. SDDEs Limits solutions to SPDEs: the $d>1$ spatial dimension case. {\it In preparation }. \bibitem{A298} Allouba, H. Different types of SPDEs in the eyes of Girsanov's theorem. Stochastic Anal. Appl. 16, no. 5, (1998), 787--810. \bibitem{AD} {Allouba, H.} Different types of SPDEs: existence, uniqueness, and Girsanov theorem. Ph.D. Dissertation, Cornell University (1996). \bibitem{B1} {Bass, R.~F.} Probabilistic techniques in analysis. Springer-Verlag, New York (1995). \bibitem{B2} {Bass, R.~F.} Diffusions and elliptic operators. Springer-Verlag, New York (1997). \bibitem{BC} {Bass, R.~F.~and Chen, Z.} Stochastic differential equations for Dirichlet processes. Probab. Theory Related Fields 121, no. 3, (2001), 422--446. \bibitem{Bl1} {Blount, D.} Comparison of stochastic and deterministic models of a linear chemical reaction with diffusion. Ann. Probab. 19, no. 4, (1991), 1440--1462. \bibitem{Bl2} {Blount, D.} Limit theorems for a sequence of nonlinear reaction-diffusion systems. Stochastic Process. Appl. 45, no. 2, (1993), 193--207. \bibitem{Db} {Doob, J.} Stochastic processes. John Wiley and Sons (1953). \bibitem{IW} {Ikeda, N.~and Watanabe, S.} Stochastic differential equations and diffusions, North-Holland Publishing Company (1989). \bibitem{KS} {Karatzas, I.~and Shreve, S.} Brownian motion and stochastic calculus. Springer-Verlag (1988). \bibitem{S1} Sowers, R. B. Multidimensional reaction-diffusion equations with white noise boundary perturbations. Ann. Probab. 22 no. 4, (1994), 2071--2121. \bibitem{S2} Sowers, R. B. Large deviations for a reaction-diffusion equation with non-Gaussian perturbations. Ann. Probab. 20 no. 1, (1992), 504--537. \bibitem{SZ} Stroock, D. and Zheng, W. Markov chain approximations to symmetric diffusions {\it Ann. Inst. H. Poincar\'e Probab. Statist.} {\bf 33} no. 5, (1997), 619--649. \bibitem{HT} Thorisson,~H. Coupling, stationarity, and regeneration. {\it Probability and its Applications. Springer-Verlag, New York, xiv+517 pp.} (2000). \bibitem{WA} {Walsh, J. B.} An introduction to stochastic partial differential equations. Ecole d'Et\'e de Probabilit\'es de Saint-Flour XIV. Lecture Notes in Math. {1180}. Springer, New York (1986). \end{thebibliography} \end{document}