Another look into the Wong-Zakai Theorem for Stochastic Heat Equation

Yu Gu, Li‐Cheng TsaiarXiv 2018 · DOI · arXiv

Abstract

Consider the heat equation driven by a smooth, Gaussian random potential: \begin{align} \partial_t u_{\varepsilon}=\tfrac12Δu_{\varepsilon}+u_{\varepsilon}(ξ_{\varepsilon}-c_{\varepsilon}), \ \ t>0, x\in\mathbb{R}, \end{align} where $ξ_{\varepsilon}$ converges to a spacetime white noise, and $c_{\varepsilon} $ is a diverging constant chosen properly. For any $ n\geq 1 $, we prove that $ u_{\varepsilon} $ converges in $ L^n $ to the solution of the stochastic heat equation. Our proof is probabilistic, hence provides another perspective of the general result of Hairer and Pardoux, for the special case of the stochastic heat equation. We also discuss the transition from homogenization to stochasticity.

Introduction

The Stochastic Heat Equation (SHE) is a fundamental model in probability and mathematical physics, describing phenomena such as directed polymers in random media and the evolution of interfaces. In one dimension, the SHE is formally written as $\partial_t \mathcal{U} = \frac{1}{2}\Delta \mathcal{U} + \mathcal{U}\xi$, where $\xi$ is spacetime white noise. Because white noise is a distribution rather than a function, the equation is "singular," and the product $\mathcal{U}\xi$ cannot be defined using standard calculus.

The Wong–Zakai theorem provides a way to understand these singular objects by looking at the limit of smooth approximations. By replacing the rough white noise with a mollified version $\xi_\epsilon$, one asks if the solution $u_\epsilon$ to the smoothed equation converges to the solution of the original SPDE as the smoothing parameter $\epsilon \to 0$. In this paper, Yu Gu and Li-Cheng Tsai provide a fresh, purely probabilistic proof of this convergence, bypassing the complex analytic machinery of Regularity Structures.

Methodology

The authors' approach relies on three mathematical pillars that transform the analytic problem into a study of stochastic processes:

  1. Feynman–Kac Representation: Instead of treating the SPDE as a purely analytic object, the solution $u_\epsilon$ is expressed as an expectation over the paths of an auxiliary Brownian motion. This allows the researchers to analyze the solution through the lens of Brownian functionals.
  2. Renormalization: To prevent the solution from blowing up as the noise becomes rougher, a specific diverging constant $c_\epsilon$ must be subtracted. The authors identify this constant as $c_\epsilon = c_*\epsilon^{-1} + \frac{1}{2}\sigma_*^2$, derived from the covariance structure of the mollified noise.
  3. Wiener Chaos Expansion: The solutions are decomposed into a series of multiple stochastic integrals (chaos). The proof of convergence in $L^n$ spaces is achieved by demonstrating that the kernels of the mollified solution's expansion converge to the kernels of the SHE's expansion.

Key Insights

The Role of Local Time

A central insight of the proof is the connection to mutual intersection local time. The authors show that the interaction between the smoothed noise and the Brownian paths in the Feynman–Kac formula converges to the local time of two independent Brownian motions. This provides a physical intuition for the limit: the singular noise effectively "measures" how often Brownian paths intersect.

Scaling and Homogenization

The paper explores the transition between different regimes of noise. At small scales, the noise tends to average out—a process known as homogenization—leading to a deterministic heat equation with Gaussian fluctuations (the Edwards-Wilkinson model). The authors demonstrate that the SHE emerges specifically at the critical scaling $\alpha=2$, where fluctuations are strong enough to remain a driving force in the limit.

Results

The primary result, Theorem 1.1, rigorously proves that the mollified solution $u_\epsilon$ converges pointwisely in $L^n(\Omega)$ to the solution of the SHE for any $n \ge 1$. This holds provided the renormalization constant $c_\epsilon$ is correctly calibrated to the specific mollifier used. Unlike previous proofs that rely on the heavy machinery of paracontrolled distributions or regularity structures, this work shows that classical tools like Brownian local time and Itô isometry are sufficient for the 1D case.

Limitations and Future Directions

While this probabilistic approach is elegant, it is currently restricted to one spatial dimension ($d=1$). In higher dimensions, the SHE becomes significantly more singular (log-critically singular in $d=2$ and supercritical in $d \ge 3$), requiring different techniques. Additionally, the proof relies on the Gaussian nature of the noise to utilize chaos expansions; extending these simplified methods to non-Gaussian noise remains an open area for further research.

Summary updated

Abstract

Consider the heat equation driven by a smooth, Gaussian random potential: \begin{align*} \partial_t u_{\varepsilon}=\tfrac12Δu_{\varepsilon}+u_{\varepsilon}(ξ_{\varepsilon}-c_{\varepsilon}), \ \ t>0, x\in\mathbb{R}, \end{align*} where $ξ_{\varepsilon}$ converges to a spacetime white noise, and $c_{\varepsilon} $ is a diverging constant chosen properly. For any $ n\geq 1 $, we prove that $ u_{\varepsilon} $ converges in $ L^n $ to the solution of the stochastic heat equation. Our proof is probabilistic, hence provides another perspective of the general result of Hairer and Pardoux \cite{Hairer15a}, for the special case of the stochastic heat equation. We also discuss the transition from homogenization to stochasticity.
BibTeX
@misc{gu2018another,
  author = {Yu Gu and Li‐Cheng Tsai},
  title = {{Another look into the Wong-Zakai Theorem for Stochastic Heat Equation}},
  year = {2018},
  eprint = {1801.09164},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1801.09164},
}
Figure