[Paper Review] Central limits and homogenization in random media

Guillaume Bal|ArXiv.org|Oct 1, 2007

Advanced Mathematical Modeling in Engineering27 references3 citations

TL;DR

This paper establishes central limit corrections to homogenization in random media by analyzing the asymptotic behavior of solutions to elliptic PDEs with rapidly oscillating random potentials. Using integral equation formulations and mixing conditions, it shows that the difference between perturbed and homogenized solutions converges to a Gaussian process of order $\varepsilon^{d/2}$ for short-range correlations, providing a statistical refinement beyond the deterministic homogenization limit.

ABSTRACT

We consider the perturbation of elliptic operators of the form $P(\bx,\bD)$ by random, rapidly varying, sufficiently mixing, potentials of the form $q(\frac{\bx}\eps,ω)$. We analyze the source and spectral problems associated to such operators and show that the properly renormalized difference between the perturbed and unperturbed solutions may be written asymptotically as $\eps o0$ as explicit Gaussian processes. Such results may be seen as central limit corrections to the homogenization (law of large numbers) process. Similar results are derived for more general elliptic equations in one dimension of space. The results are based on the availability of a rapidly converging integral formulation for the perturbed solutions and on the use of classical central limit results for random processes with appropriate mixing conditions.

Motivation & Objective

• To characterize the statistical structure of the corrector to homogenization in random media, beyond the deterministic homogenized limit.
• To address the gap between full stochastic simulations and deterministic homogenization by deriving central limit-type corrections.
• To analyze source and spectral problems for elliptic operators perturbed by rapidly varying, mixing random potentials.
• To extend known results on correctors in one-dimensional settings to higher dimensions and more general correlation structures.
• To provide a rigorous asymptotic expansion of the solution difference as a Gaussian process under appropriate mixing and decay conditions.

Proposed method

• Formulate the perturbed solution using an integral equation involving the Green’s function of the homogenized operator.
• Apply a rapidly converging integral representation to express the solution difference as a functional of the random potential.
• Use classical central limit theorems for mixing processes to derive the limiting Gaussian distribution of the corrector.
• Analyze the variance of the corrector explicitly using correlation structure and mixing coefficients.
• Generalize results from short-range to long-range correlated random potentials by extending the framework to power-law decaying correlations.
• Apply the method to both steady-state (Helmholtz) and time-dependent problems, including wave and evolution equations.

Experimental results

Research questions

• RQ1 How does the solution to an elliptic PDE with rapidly oscillating random coefficients behave asymptotically as the scale $\varepsilon \to 0$?
• RQ2 What is the statistical structure of the corrector term that corrects the homogenized solution in random media?
• RQ3 Can the central limit theorem be applied to the solution difference in the presence of mixing random potentials?
• RQ4 How does the amplitude of the corrector depend on the correlation structure of the random potential?
• RQ5 Can the method be extended to time-dependent equations such as the wave or evolution equations?

Key findings

• The properly renormalized difference between the perturbed and unperturbed solutions converges to a Gaussian process as $\varepsilon \to 0$.
• For integrable correlation functions, the corrector is a mean-zero Gaussian process of order $\varepsilon^{d/2}$ in dimension $d$.
• When correlations decay as $|\mathbf{x}|^{-\alpha d}$ with $0 < \alpha < 1$, the corrector amplitude scales as $\varepsilon^{\alpha d/2}$.
• The method relies on a rapidly converging integral formulation of the solution difference, enabling the application of central limit theorems.
• The corrector for low-frequency components of the solution can be explicitly estimated and shown to be Gaussian with calculable variance.
• The framework applies to both steady-state and time-dependent problems, including the wave equation, when the Green’s function is sufficiently regular.