Weak Convergence Approach for Parabolic Equations with Large, Highly Oscillatory, Random Potential

Yu Gu, Guillaume Bal

Abstract

This paper concerns the macroscopic behavior of solutions to parabolic equations with large, highly oscillatory, random potential. When the correlation function of the random potential satisfies a specific integrability condition, we show that the random solution converges, as the correlation length of the medium tends to zero, to the deterministic solution of a homogenized equation in dimension $d\geq3$. Our derivation is based on a Feynman-Kac probabilistic representation and the Kipnis-Varadhan method applied to weak convergence of Brownian motions in random sceneries. For sufficiently mixing coefficients, we also provide an optimal rate of convergence to the homogenized limit using a quantitative martingale central limit theorem. As soon as the above integrability condition fails, the solution is expected to remain stochastic in the limit of a vanishing correlation length. For a large class of potentials given as functionals of Gaussian fields, we show the convergence of solutions to stochastic partial differential equations (SPDE) with multiplicative noise. The Feynman-Kac representation and the corresponding weak convergence of Brownian motions in random sceneries allows us to explain the transition from deterministic to stochastic limits as a function of the correlation function of the random potential.

Introduction

In the fields of physics and engineering, modeling systems with parameters that vary randomly at microscopic scales is a common challenge. For instance, heat diffusion through a highly heterogeneous material requires accounting for small-scale oscillations in the medium. Traditionally, homogenization theory suggests that as these oscillations become infinitely fine (i.e., the correlation length $\epsilon \to 0$), the system's overall behavior converges to a simpler, deterministic "effective" equation.

However, this deterministic limit is not universal. This research investigates parabolic partial differential equations (PDEs) subjected to large, highly oscillatory random potentials. The central question is determining the precise conditions under which the randomness "averages out" into a deterministic state versus when it persists at the macroscopic level, leading to a Stochastic PDE (SPDE) limit. The study focuses on dimensions $d \geq 3$ and identifies the correlation function of the random medium as the primary driver of this transition.

Methodology

To analyze the behavior of the solution $u_\epsilon$, the authors depart from standard analytical PDE techniques in favor of a probabilistic approach. This framework provides a more intuitive understanding of how particles interact with the random environment.

1. Feynman-Kac Representation

The PDE is first transformed into an expectation over Brownian motion paths. This representation allows the solution to be viewed as a particle moving through a "random scenery," where the potential is accumulated along the path of the Brownian motion.

2. Kipnis-Varadhan Method

A classic tool in probability, the Kipnis-Varadhan method is adapted here to prove the weak convergence of Brownian motions within these random environments. This is essential for establishing the central limit theorems required for homogenization.

3. Quantitative Martingale Central Limit Theorem (CLT)

To establish the "speed" or rate of convergence, the authors utilize a quantitative version of the CLT. This allows for precise error estimates between the random solution $u_\epsilon$ and its deterministic homogenized limit $u_{hom}$.

4. Hermite Polynomial Expansion

In cases where the potential is defined as a functional of a Gaussian field, the authors employ Hermite expansions. This technique is particularly useful for analyzing the effects of long-range correlations and proving the convergence to an SPDE when the standard integrability conditions fail.

Key Insights: The Integrability Pivot

The research identifies a sharp transition in macroscopic behavior governed by the integrability of the potential's correlation function, specifically the condition $\langle V, -L^{-1}V \rangle < \infty$.

  • Short-Range Correlations: If this condition is met, the noise is effectively short-ranged. The path-integral of the noise obeys a Law of Large Numbers, and the system homogenizes into a deterministic state.
  • Long-Range Correlations: When the correlations decay slowly enough to break this integrability condition, the randomness does not wash out. Instead, the limit remains stochastic, converging to an SPDE with multiplicative noise.
  • Dimensionality: The results are established for $d \geq 3$. In lower dimensions ($d=1, 2$), the recurrence of Brownian motion—the tendency of a particle to return to its starting point—makes the averaging process significantly more complex and requires different mathematical treatments.

Principal Results

Homogenization and Convergence Rates

The paper proves that under the necessary integrability conditions, the random solution $u_\epsilon$ converges in probability to a deterministic solution $u_{hom}$ (Theorem 2.2). For "strongly mixing" potentials (those with very short-range correlations), the authors provide optimal convergence rates (Theorem 2.6):

  • In 3D, the error scales as $O(\sqrt{\epsilon})$.
  • In dimensions $d > 4$, the error scales as $O(\epsilon)$.

Long-Range Decay and SPDE Limits

For Gaussian potentials with power-law correlation decay $|x|^{-\beta}$, Theorem 2.9 demonstrates that the error rate becomes a function of $\beta$, showing how lingering correlations slow the homogenization process. Most significantly, Theorem 2.11 proves that for a large class of long-range correlated potentials, the solution converges to an SPDE of the form: $$\partial_t u = \frac{1}{2}\Delta u + i \xi u$$ where $\xi$ represents a Gaussian noise term. This result explains why certain physical systems remain inherently noisy even at macroscopic scales.

Discussion and Limitations

While the paper provides a robust framework for understanding the transition from deterministic to stochastic limits, several areas remain for future exploration:

  • Imaginary Potentials: The study focuses on imaginary potentials ($iV_\epsilon$), which are relevant for Schrödinger-type models. This choice provides a uniform bound on the solution's energy. Real-valued potentials, such as those in the Parabolic Anderson Model, are more challenging because the solution can exhibit local "explosions."
  • Low Dimensions: The behavior in $d=1$ and $d=2$ remains a distinct challenge due to the different nature of Brownian motion in those spaces.
  • Fluctuation Theory: While the authors have identified the limit and the rate of convergence, a complete Central Limit Theorem for the fluctuations around the homogenized limit is identified as a subject for future research.

Summary updated

Abstract

This paper concerns the macroscopic behavior of solutions to parabolic equations with large, highly oscillatory, random potential. When the correlation function of the random potential satisfies a specific integrability condition, we show that the random solution converges, as the correlation length of the medium tends to zero, to the deterministic solution of a homogenized equation in dimension $d\geq3$. Our derivation is based on a Feynman-Kac probabilistic representation and the Kipnis-Varadhan method applied to weak convergence of Brownian motions in random sceneries. For sufficiently mixing coefficients, we also provide an optimal rate of convergence to the homogenized limit using a quantitative martingale central limit theorem. As soon as the above integrability condition fails, the solution is expected to remain stochastic in the limit of a vanishing correlation length. For a large class of potentials given as functionals of Gaussian fields, we show the convergence of solutions to stochastic partial differential equations (SPDE) with multiplicative noise. The Feynman-Kac representation and the corresponding weak convergence of Brownian motions in random sceneries allows us to explain the transition from deterministic to stochastic limits as a function of the correlation function of the random potential.
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