Pointwise two-scale expansion for parabolic equations with random coefficients

Yu Gu, Jean-Christophe Mourrat

TL;DR

This paper proves a pointwise version of the "two-scale expansion" for linear parabolic and elliptic equations with random coefficients in dimensions $d \ge 3$. The authors demonstrate that the first-order correction to the homogenized solution holds without needing the extra symmetry assumptions previously thought necessary, revealing a surprising difference between continuous and discrete models.

What Problem Does This Paper Address?

In the study of homogenization, researchers look at how differential equations with highly oscillatory coefficients (representing complex, heterogeneous materials) behave at a large scale. Usually, we can find a "homogenized" equation with constant coefficients that describes the average behavior.

However, the "average" doesn't tell the whole story. To understand the local fluctuations, mathematicians use a two-scale expansion: $$u_\epsilon \approx u_{hom} + \epsilon \cdot (\text{corrector term})$$ While this expansion is well-understood in periodic settings, the stochastic (random) setting is much harder. Specifically, it was widely believed that for this expansion to hold in a pointwise sense (at a specific point $x$ rather than just as an average over a region), the law of the random coefficients had to be symmetric (e.g., invariant under $z \to -z$). Without this symmetry, a "bias" of order $\epsilon$ was expected to persist. This paper tackles the challenge of proving that this expansion holds pointwise even without those symmetry assumptions.

Methods

The authors employ a probabilistic approach to analyze the partial differential equations (PDEs). Key components of their methodology include:

  • Diffusion in Random Environments: They represent the solution to the parabolic equation as an expectation of a random motion (a "particle") moving through the random field.
  • Martingale Decomposition: They use the Kipnis-Varadhan method to decompose the particle's drift into a martingale part and a negligible remainder.
  • Quantitative Martingale Central Limit Theorem (CLT): To get precise rates, they use a quantitative version of the CLT to show how closely the random motion resembles a standard Brownian motion.
  • Spectral Gap Inequalities: To handle the randomness of the coefficients (modeled as a local function of a Poisson point process), they use spectral gap tools to estimate how quickly correlations decay over distance.
  • Green's Function Estimates: They rely on recent, sophisticated bounds on the gradients of the Green's function for elliptic operators with random coefficients.

Datasets

Not discussed in the paper. As a theoretical mathematics paper, it relies on rigorous proofs and analytical derivations rather than empirical datasets or benchmarks.

Key Insights and Ideas

  • The "Symmetry Myth": The most significant insight is that the pointwise two-scale expansion is true in general for continuous equations. The authors point out that previous counter-examples and "beliefs" were based on discrete models (on a lattice $\mathbb{Z}^d$). In the continuous world, the first-order deterministic bias actually vanishes.
  • Vanishing Bias: The authors identify a potential deterministic error term at order $\epsilon$ (defined by constants $c_{ijk}$). Through a clever integration-by-parts argument involving the "corrector" equations, they prove these constants are all zero.
  • Dimension Matters: The result specifically requires $d \ge 3$. This is because the "corrector" (the function that accounts for local variations) is stationary and has better integrability properties in higher dimensions, which is crucial for the decay of the remainder terms.

Key Results

  • Theorem 2.1 (Parabolic Case): For a fixed point $(t, x)$, the difference between the actual solution $u_\epsilon$ and the two-scale expansion is $o(\epsilon)$ in $L^1(\Omega)$. Specifically: $$u_\epsilon(t, x, \omega) - u_{hom}(t, x) = \epsilon \nabla u_{hom}(t, x) \cdot \phi(x/\epsilon, \omega) + o(\epsilon)$$
  • Theorem 2.5 (Elliptic Case): A similar result is derived for elliptic equations, showing that the parabolic techniques can be successfully transferred to the steady-state setting.
  • Quantitative Error Bounds: While the primary result is $o(\epsilon)$, the authors provide specific decay rates for the error terms (e.g., related to $\epsilon^{4/3}$ in $d=3$) depending on the dimension.

Limitations and Future Work

  • Dimension Constraint: The proof relies heavily on the properties of correctors in $d \ge 3$. The behavior in $d=1$ and $d=2$ remains a different, likely more complex, challenge.
  • Range of Dependence: The results assume a "finite range of dependence" (coefficients are determined by a Poisson process in a local neighborhood). Extending this to coefficients with "long-range" correlations (heavy tails) would require different tools.
  • The Discrete-Continuous Gap: The authors acknowledge that their result does not apply to discrete equations. Understanding exactly why the bias vanishes in continuous space but persists in discrete space is an open conceptual question.

Summary updated

Abstract

We investigate the first-order correction in the homogenization of linear parabolic equations with random coefficients. In dimension $3$ and higher and for coefficients having a finite range of dependence, we prove a pointwise version of the two-scale expansion. A similar expansion is derived for elliptic equations in divergence form. The result is surprising, since it was not expected to be true without further symmetry assumptions on the law of the coefficients.
BibTeX
@inproceedings{gu2015pointwise,
  author = {Yu Gu and Jean-Christophe Mourrat},
  title = {{Pointwise two-scale expansion for parabolic equations with random coefficients}},
  booktitle = {Probability},
  year = {2015},
  doi = {10.1007/s00440-015-0667-z},
}

Loading related content...

Figure