Weak convergence approach for parabolic equations with large, highly oscillatory, random potential

Yu Gu, Guillaume Bal

Abstract

This research investigates the macroscopic behavior of solutions to parabolic equations featuring large-intensity, rapidly oscillating random potentials. By employing a Feynman–Kac probabilistic representation and the Kipnis–Varadhan methodology, the study demonstrates that solutions converge to a deterministic homogenized state in dimensions $d \geq 3$, provided the potential's correlation function meets specific integrability criteria. Conversely, when these criteria are not met, the system retains stochasticity, converging toward a stochastic partial differential equation (SPDE). This work provides a precise description of the transition between deterministic and stochastic limits based on the statistical properties of the random environment.

Introduction

In mathematical physics, understanding how microscopic randomness translates to macroscopic order is a fundamental challenge. This research, published in Annales de l’Institut Henri Poincaré, addresses partial differential equations (PDEs) where the potential term is both large and highly oscillatory.

The research was published in Annales de l’Institut Henri Poincaré, a leading journal for probability and statistics.

The central question is whether such systems "average out" into a predictable state (homogenization) or if the underlying noise persists at larger scales. This paper provides a rigorous mathematical framework to answer these questions for dimensions $d \geq 3$, specifically focusing on the transition between deterministic and stochastic limits.

Methodology

The authors move away from purely analytical PDE methods, instead adopting a probabilistic approach:

  1. Feynman–Kac Representation: They transform the parabolic PDE into a problem of Brownian motion. Instead of solving the equation directly, they analyze the "expected" path of a particle moving through a random landscape.
  2. Kipnis–Varadhan Method: To handle the randomness, they use the "medium seen from the observer" perspective. This allows them to decompose the particle's motion into a martingale (a mathematical model of a fair game) plus a negligible error term.
  3. Quantitative Martingale Central Limit Theorem (CLT): To find the speed at which the random solution approaches the average solution, they apply a quantitative version of the CLT, providing optimal rates of convergence.
  4. Hermite Expansions: For cases where the randomness is long-range, they use Hermite polynomials to expand the potential, showing how it converges to a specific type of multiplicative noise in an SPDE.

Key Insights: The Integrability Threshold

The research identifies a critical "tipping point" based on the integrability of the potential's correlation function.

  • Short-range correlations: If fluctuations decay quickly (satisfying the integrability condition), the system averages out to a deterministic limit.
  • Long-range correlations: If fluctuations persist over long distances, the randomness survives into the macroscopic limit, resulting in an SPDE.

Furthermore, the "speed" of this convergence is shown to be dimension-dependent. The researchers show that the error vanishes at different rates depending on the spatial dimension, with 3D systems converging more slowly than those in 5D and above.

Key Results

The study establishes rigorous error bounds for the homogenization process in dimensions $d \geq 3$:

  • $d = 3$: Error $\leq C(1+t)\sqrt{\epsilon}$
  • $d = 4$: Error $\leq C(1+t)\epsilon\sqrt{|\log \epsilon|}$
  • $d > 4$: Error $\leq C(1+t)\epsilon$

For long-range Gaussian potentials, the authors prove that the solution converges to an SPDE driven by multiplicative noise, effectively deriving the Stratonovich form of the limiting equation.

Conclusion

By bridging the gap between homogenization theory and SPDEs, this work provides a unified view of how the "memory" of a random environment dictates the macroscopic physics of the system. While the results are robust for $d \geq 3$, future work remains to fully describe the distribution of fluctuations and to extend these findings to lower dimensions ($d=1, 2$) where different scaling behaviors are expected.

Summary updated

Abstract

Nous considérons le comportement macroscopique de solutions d’équations paraboliques présentant un terme potentiel aléatoire de grande intensité et oscillant rapidement. Lorsque la fonction de corrélation du potentiel aléatoire satisfait une condition précise d’intégrabilité, nous démontrons que la solution aléatoire converge vers la solution déterministe d’une équation homogénéisée quand la longueur de corrélation du milieu tend vers zéro en toute dimension $d\geq3$. Notre preuve s’appuie sur une représentation probabiliste de type Feynman–Kac et sur la methodologie introduite par Kipnis et Varadhan permettant de montrer la convergence de mouvements browniens dans des milieux aléatoires. Lorsque les coefficients aléatoires sont suffisamment mélangeants, nous présentons un taux optimal de convergence grâce à une approche quantitative du théorème de la limite centrale pour les martingales. Dès que la condition d’intégrabilité mentionée ci-dessus n’est plus satisfaite, nous pensons que la solution restera fortement stochastique pour toute longueur de corrélation du milieu. Nous montrons, pour une classe de potentiels décrits comme des fonctionnelles de champs gaussiens, que la solution converge vers celle d’une équation différentielle stochastique. La représentation de Feynman–Kac et la convergence faible de mouvements browniens nous permet d’obtenir une description précise de la transition d’une limite déterministe vers une limite stochastique en fonction des propriétés de la fonction de corrélation du potentiel aléatoire.
BibTeX
@inproceedings{gu2016weak,
  author = {Yu Gu and Guillaume Bal},
  title = {{Weak convergence approach for parabolic equations with large, highly oscillatory, random potential}},
  booktitle = {Annales},
  year = {2016},
  doi = {10.1214/14-aihp637},
}
Figure