The Random Heat Equation in Dimensions Three and Higher: The Homogenization Viewpoint

We consider the stochastic heat equation $\partial_{s}u =\frac{1}{2}\Delta u +(\beta V(s,y)-\lambda)u$, driven by a smooth space-time stationary Gaussian random field $V(s,y)$, in dimensions $d\geq 3$, with an initial condition $u(0,x)=u_0(\varepsilon x)$. It is known that the diffusively rescaled solution $u^{\varepsilon}(t,x)=u(\varepsilon^{-2}t,\varepsilon^{-1}x)$ converges weakly to a scalar multiple of the solution $\bar u$ of a homogenized heat equation with an effective diffusivity $a$, and that fluctuations converge (again, in a weak sense) to the solution of the Edwards-Wilkinson equation with an effective noise strength $\nu$. In this paper, we derive a pointwise approximation $w^\varepsilon(t,x)=\bar u(t,x)\Psi^\varepsilon(t,x)+\varepsilon u_1^\varepsilon(t,x)$, where $\Psi^\varepsilon(t,x)=\Psi(t/\varepsilon^2,x/\varepsilon)$, $\Psi$ is a solution of the SHE with constant initial conditions, and $u_1$ is an explicit corrector. We show that $\Psi(t,x)$ converges to a stationary process $\tilde \Psi(t,x)$ as $t\to\infty$, that $w^\varepsilon(t,x)$ converges pointwise (in $L^1$) to $u^\varepsilon(t,x)$ as $\varepsilon\to 0$, and that $\varepsilon^{-d/2+1}(u^\varepsilon-w^\varepsilon)$ converges weakly to $0$ for fixed $t$. As a consequence, we derive new representations of the diffusivity $a$ and effective noise strength~$\nu$. Our approach uses a Markov chain in the space of trajectories introduced in Gu, Ryzhik, and Zeitouni, "The Edwards-Wilkinson limit of the random heat equation in dimensions three and higher", as well as tools from homogenization theory. The corrector $u_1^\varepsilon(t,x)$ is constructed using a seemingly new approximation scheme on "long but not too long time intervals".