Moments of 2D Parabolic Anderson Model

Yu Gu, Weijun Xu

Introduction

The Parabolic Anderson Model (PAM) is a fundamental stochastic partial differential equation (SPDE) that describes the evolution of a field $u(t, x)$ in a random environment. Formally, the equation is given by: $$\partial_t u = \frac{1}{2} \Delta u + u \cdot \xi$$ where $\xi$ represents spatial white noise. While the one-dimensional case is well-understood, the two-dimensional (2D) case presents significant mathematical challenges due to the singular nature of the noise. This note explores the work of Yu Gu and Weijun Xu, who provide a rigorous treatment of the moments of the 2D PAM using the Feynman-Kac formula.

The Challenge of Two Dimensions

In two spatial dimensions, the white noise $\xi$ is a distribution of negative regularity. Consequently, the product $u \cdot \xi$ is not well-defined in the classical sense because the solution $u$ is not sufficiently smooth to be multiplied by such a singular distribution.

To address this, researchers use regularization and renormalization. This involves smoothing the noise $\xi$ to $\xi_\epsilon$, solving the resulting equation for $u_\epsilon$, and then taking the limit as $\epsilon \to 0$. However, to obtain a non-trivial limit, one must subtract a divergent constant $C_\epsilon$ from the noise. While existence and uniqueness of the solution $u$ have been established through theories like Regularity Structures, characterizing the statistical moments $E[u(t, x)^n]$ remained a difficult task. This paper fills that gap by showing when these moments exist and providing an explicit representation.

Methodology: The Feynman-Kac Formula

The authors utilize the Feynman-Kac formula, which represents the solution to the heat equation with a potential as an expectation over Brownian motion paths. For the regularized equation, the solution can be expressed as: $$u_\epsilon(t, x) = E_B \left[ \exp \left( \int_0^t \xi_\epsilon(B_s) ds - C_\epsilon t \right) \right]$$ where $B_s$ is a 2D Brownian motion starting at $x$.

To find the $n$-th moment, the authors consider $n$ independent Brownian motions $B^1, \dots, B^n$. The $n$-th moment $E[u_\epsilon(t, x)^n]$ then involves the expectation over these $n$ paths of the exponential of the sum of their interactions with the noise.

Intersection Local Times

As the smoothing parameter $\epsilon$ goes to zero, the interaction terms converge to intersection local times. These are measures that quantify how much time the Brownian paths spend intersecting themselves or each other:

  • Self-intersection local time: The time a single path $B^i$ spends hitting its own previous locations.
  • Mutual-intersection local time: The time two different paths $B^i$ and $B^j$ spend at the same location.

A key insight of the paper is that the renormalization constant $C_\epsilon$ required for the SPDE is precisely the same constant needed to "center" the self-intersection local time of a planar Brownian motion. The core technical challenge lies in proving uniform integrability—ensuring the exponential of these renormalized local times remains finite in the limit.

Key Results and Existence

The primary result of the study is the existence of the $n$-th moment for a restricted time interval.

Theorem: Moment Existence

The $n$-th moment $E[u(t, x)^n]$ exists and is finite if the time $t$ satisfies: $$t < \frac{\delta}{n^2}$$ for some constant $\delta > 0$.

This "small time" constraint is a distinctive feature of the 2D model. It suggests that the solution $u$ possesses "heavy tails"—the randomness is so intense that for sufficiently large times, the higher-order moments may become infinite. The authors provide an explicit formula (1.10) for $E[u(t, x)^n]$ involving the exponential of a linear combination of self- and mutual-intersection local times.

Comparison Across Dimensions

The paper highlights how dimensionality fundamentally changes the behavior of the PAM:

  • 1D: The noise is smooth enough that moments exist for all time $t > 0$.
  • 2D: The noise is singular, and moments only exist for small time $t < \delta/n^2$.
  • 3D: The noise is so singular that even the first moment (the mean) of the regularized solution blows up as $\epsilon \to 0$, suggesting that the 3D PAM may not have well-defined moments in the limit at all.

Conclusion

This work provides a crucial link between the analysis of SPDEs and the sample path properties of Brownian motion. By leveraging the Feynman-Kac formula, the authors demonstrate that the renormalization required to solve the 2D PAM is intrinsically tied to the self-intersection properties of planar paths. While it establishes existence for small times, the behavior at large times remains a conjecture, pointing toward the highly intermittent and heavy-tailed nature of the 2D Parabolic Anderson Model.

Summary updated

Abstract

In this note, we use the Feynman-Kac formula to derive a moment representation for the 2D parabolic Anderson model in small time, which is related to the intersection local time of planar Brownian motions.
BibTeX
@inproceedings{gu2018moments,
  author = {Yu Gu and Weijun Xu},
  title = {{Moments of 2D Parabolic Anderson Model}},
  booktitle = {Asymptotic},
  year = {2018},
  doi = {10.3233/asy-171460},
}
Figure