Quantitative homogenization of elliptic equations with infinitely many scales

Technical Summary: Quantitative Homogenization of Elliptic Equations with Infinitely Many Scales

1. Problem Statement and Motivation
This paper addresses the multiscale homogenization of elliptic equations where the coefficient matrix oscillates periodically at infinitely many scales. The authors consider the boundary value problem:
$$\begin{cases} -\text{div}(A_\varepsilon(x)\nabla u_\varepsilon) = f & \text{in } \Omega, \\ u_\varepsilon = g & \text{on } \partial\Omega, \end{cases}$$
where $\Omega \subset \mathbb{R}^d$ is a bounded Lipschitz domain. The coefficient matrix $A_\varepsilon(x)$ is defined as the $L^\infty$-limit of matrices with finitely many scales, $A_\varepsilon(x) = \lim_{n \to \infty} A_\varepsilon^n(x)$, where $A_\varepsilon^n(x) = A_n(x, x/\varepsilon_1, \dots, x/\varepsilon_n)$. The scales satisfy $1 = \varepsilon_0 > \varepsilon_1 > \varepsilon_2 > \dots \to 0$.

The motivation stems from physical models involving fractal-like materials and fluid dynamics, particularly turbulence, where vortices exist across a wide range of scales. While classical reiterated homogenization handles a fixed finite number of scales $n$, existing methods rely on inductive arguments where constants typically blow up as $n \to \infty$. This paper aims to develop a general theory that remains uniform as the number of scales approaches infinity, addressing the "perpetual" homogenization problem.

2. Mathematical Framework and Assumptions
The coefficient matrix is constructed via a series expansion:
$$A_\varepsilon(x) = \sum_{\ell=0}^\infty B_\ell\left(x, \frac{x}{\varepsilon_1}, \dots, \frac{x}{\varepsilon_\ell}\right),$$
where $B_\ell$ are continuous and periodic in their fast variables. The authors impose three main structural assumptions:

• Ellipticity and Boundedness: Uniform ellipticity ($\mu |\xi|^2 \le \xi \cdot A_\varepsilon \xi$) and boundedness, ensured by the condition $\sum_{\ell=0}^\infty \|B_\ell\|_{L^\infty} < \infty$.
• Periodicity: Each $B_\ell$ is periodic in its fast variables.
• Decay of Oscillations: To obtain quantitative rates, the authors assume the coefficients $B_\ell$ are Lipschitz continuous and decay at a rate controlled by a sequence $\delta = (\delta_0, \delta_1, \dots)$. Specifically, they require $\sum_{\ell=0}^\infty \delta_\ell < \infty$ and $\sum_{\ell=1}^\infty \ell \delta_\ell < \infty$ (denoted as $[\delta]_0 < \infty$ and $[\delta]_1 < \infty$).

3. Methodology
The proofs rely on a refined reiterated homogenization argument, viewed as a deterministic renormalization group or coarse-graining process. The methodology proceeds in two main directions:

• Qualitative Homogenization: The authors start with the approximate problem involving $n$ scales. Using the stability of $H$-convergence, they show that the homogenized matrices $\bar{A}_n$ of the $n$-scale problems form a Cauchy sequence in $L^\infty$ as $n \to \infty$. This allows them to define a limiting homogenized matrix $\bar{A}$ and prove that $u_\varepsilon$ converges weakly to the solution $u_0$ of the homogenized equation with coefficient $\bar{A}$.

• Quantitative Convergence Rates: To establish optimal $L^2$ convergence rates independent of $n$, the authors decompose the coefficient matrix into a "principal part" (slow oscillation, large amplitude) and a "tail part" (fast oscillation, small amplitude). They prove an improved one-step error estimate for a two-scale problem where the perturbation is small. By iterating this estimate across scales and utilizing the decay of $\delta_\ell$, they control the accumulation of errors. A key technical innovation is showing that the error terms involving the tail parts decay sufficiently fast to allow the limit $n \to \infty$ without the constants blowing up.

• Uniform Lipschitz Estimates: The proof of uniform Lipschitz regularity ($\|\nabla u_\varepsilon\|_{L^\infty} \le C$) employs a large-scale iteration argument (Campanato-type). Unlike standard methods where constants accumulate as $C^n$, the authors utilize the specific structure of the coefficients to derive an iterative formula for the flatness quantity $H(r)$ and the gradient average $h(r)$. They introduce a sequence of error terms $M_m$ which, under specific scale-separation conditions and decay assumptions on $\delta$, are shown to be summable ($\sum M_m < \infty$). This summability prevents the blow-up of constants as the scale $m \to \infty$.

4. Key Results

• Qualitative Homogenization (Theorem 1.1): Under the scale-separation condition $\varepsilon_k / \varepsilon_{k-1} \to 0$, the solution $u_\varepsilon$ converges weakly in $H^1(\Omega)$ to $u_0$, and $A_\varepsilon$ $H$-converges to a homogenized matrix $\bar{A}$.
• Optimal $L^2$ Convergence Rates (Theorem 1.2): Assuming $[\delta]_1 < \infty$ and a bounded Lipschitz domain, the convergence rate is:
  $$\|u_\varepsilon - u_0\|_{L^2(\Omega)} \le C \Lambda [\delta]_1 \sup_{k \ge 1} \frac{\varepsilon_k}{\varepsilon_{k-1}} (\|f\|_{L^2} + \|g\|_{H^{3/2}}).$$
  This recovers the optimal rate for finite scales and remains valid for infinite scales.
• Uniform Lipschitz Estimates (Theorems 1.5 & 1.6): Under a quantitative scale-separation condition (e.g., $(\varepsilon_{k+1}/\varepsilon_k)^N \le \varepsilon_{i+1}/\varepsilon_i$) and assuming $[\delta]_{1+\rho} < \infty$, the authors establish uniform interior and boundary Lipschitz estimates:
  $$\|\nabla u_\varepsilon\|_{L^\infty} \le C \left( \|\nabla u_\varepsilon\|_{L^2} + \|f\|_{L^p} + \dots \right).$$
  These estimates are uniform in $\varepsilon$ provided the smallest scale $\varepsilon_1$ is sufficiently small (or under specific geometric decay conditions on the scales).

5. Significance and Claims
The paper claims to provide the first general homogenization theory for elliptic equations with infinitely many oscillating scales that yields optimal convergence rates and uniform regularity estimates.

• Overcoming Inductive Blow-up: The authors explicitly state that previous results for $n$ scales rely on induction where constants generally blow up as $n \to \infty$. This work overcomes that limitation by exploiting the "small perturbation" structure of the tail terms in the coefficient expansion, allowing for uniform control.
• Fractal and Turbulence Applications: The theory is motivated by and applicable to fractal materials and simplified models of turbulence (referencing Richardson's cascade), where infinite scales are intrinsic.
• Robustness: The results hold under natural scale-separation assumptions and decay conditions on the coefficients, bridging the gap between finite-scale reiterated homogenization and the continuous spectrum of scales found in physical phenomena.

The paper does not propose new experimental setups or future applications beyond the theoretical framework of elliptic homogenization, fluid dynamics, and fractal materials as described in the introduction. The focus remains strictly on the mathematical analysis of the convergence and regularity of these specific multiscale problems.