Stochastic homogenization of fractional obstacle problems

1. Problem Statement

The paper investigates the asymptotic behavior of a class of nonlinear and nonlocal variational problems defined on domains containing a large number of small, randomly distributed obstacles.

• The Model: The prototypical problem is a Dirichlet problem for the fractional $p$-Laplacian ($(-\Delta)^s_p$) with $p \in (1, \infty)$ and $s \in (0, 1)$.
  $$\begin{cases} (-\Delta)^s_p u = f & \text{in } U \setminus T^\omega_\varepsilon, \\ u = 0 & \text{on } \partial U \cup T^\omega_\varepsilon, \end{cases}$$
  where $U \subset \mathbb{R}^n$ is a bounded Lipschitz domain, and $T^\omega_\varepsilon$ is a random set of obstacles.
• Obstacle Configuration: The obstacles are defined via a stationary and ergodic marked point process $N$. Specifically, $T^\omega_\varepsilon$ is a union of balls $B_{\lambda_\varepsilon \rho_i}(\varepsilon x_i)$, where:
  • $\{x_i\}$ are the centers (points of the process).
  • $\{\rho_i\}$ are the marks (radii).
  • $\varepsilon$ is the scaling parameter for the domain (obstacles become dense).
  • $\lambda_\varepsilon$ is a secondary scaling parameter for the obstacle radii.
• Key Challenge: Unlike previous works that assumed "well-separated" obstacles, this model allows for clustering effects (obstacles arbitrarily close to each other) with positive probability. The goal is to determine the critical scaling $\lambda_\varepsilon$ and the resulting effective (homogenized) equation.

2. Methodology

The authors employ a purely variational approach based on $\Gamma$-convergence, avoiding the delicate PDE techniques (like oscillating test functions or extension methods) used in earlier literature for the fractional Laplacian.

A. Variational Framework

The problem is reformulated as the minimization of a functional $F^\omega_\varepsilon$ in the fractional Sobolev space $W^{s,p}_0(U)$:
$$F^\omega_\varepsilon(u) = \int_{U \times U} \frac{|u(x) - u(y)|^p}{|x-y|^{n+sp}} \, dx dy + \chi^\omega_\varepsilon(u),$$
where $\chi^\omega_\varepsilon$ is the indicator function enforcing the zero boundary condition on the obstacles. The convergence of solutions follows from the $\Gamma$-convergence of these functionals.

B. Decomposition of Obstacles

A central technical innovation is the decomposition of the random obstacle set $T^\omega_\varepsilon$ into two disjoint sets based on a truncation parameter $R$:

1. Good Obstacles ($T^g_\varepsilon$): Small, $\varepsilon$-separated obstacles. These are well-behaved and allow for local constructions.
2. Bad Obstacles ($T^b_\varepsilon$): Large obstacles or clusters.
   • The authors prove that under suitable moment conditions, the fractional capacity of the "bad" set vanishes asymptotically ($\text{cap}_{s,p}(T^b_\varepsilon) \to 0$).
   • This implies that clusters do not contribute to the effective limit term, a crucial insight that removes the need for "well-separation" assumptions.

C. Probabilistic Tools

• Ergodic Theorems: Used to establish the almost sure convergence of random sums to their expected values.
• Palm Measures: The authors utilize Palm distributions to compute the expectation measure of a thinned process (the selected "good" obstacles). This allows them to rigorously handle the selection of obstacles and prove the additivity of the capacity density.
• Joining Lemma: A nonlocal version of the "Joining Lemma" (originally by Ansini and Braides for local problems) is adapted. This lemma allows the modification of a sequence with equibounded energy to be locally constant around the "good" obstacles without generating extra energy concentration, facilitating the reconstruction of the limit functional.

3. Key Contributions

1. Removal of Separation Assumptions: This is the first homogenization result for nonlinear fractional obstacle problems that does not require obstacles to be well-separated. It accommodates clustering configurations, extending previous results by Giunti, H"ofer, and Vel'azquez (local case) and Scardia, Zemas, and Zeppieri (nonlinear local case) to the nonlocal setting.
2. General Nonlocal Kernels: The analysis extends beyond the standard fractional Laplacian to a broad class of anisotropic, translation-invariant, homogeneous kernels and randomly shaped obstacles.
3. Methodological Streamlining: The paper introduces a simplified proof strategy compared to prior works, utilizing:
   • A simplified "good/bad" decomposition.
   • Direct application of ergodic theorems to thinned processes via Palm measures.
   • Avoidance of the Caffarelli-Silvestre extension, making the results applicable to general nonlinear nonlocal functionals.

4. Main Results

The Critical Scaling

The authors identify a critical scaling regime for the obstacle radii:
$$\lambda_\varepsilon = \varepsilon^{\frac{n}{n-sp}}$$
This scaling ensures that the capacity of the obstacles remains finite and non-zero in the limit.

The Homogenized Limit

Under the moment condition $E[\sum_{x_i \in Q} \rho_i^{n-sp}] < \infty$, the random functionals $\Gamma$-converge almost surely to a deterministic functional $F$:
$$F(u) = \int_{U \times U} \frac{|u(x) - u(y)|^p}{|x-y|^{n+sp}} \, dx dy + \gamma \int_U |u|^p \, dx$$
where the capacitary constant $\gamma$ is given by:
$$\gamma = \text{cap}_{s,p}(B_1) \cdot E\left[ \sum_{x_i \in Q} \rho_i^{n-sp} \right]$$
Here, $\text{cap}_{s,p}(B_1)$ is the fractional $p$-capacity of the unit ball.

Interpretation

The limit equation is formally:
$$(-\Delta)^s_p u + \gamma |u|^{p-2}u = f \quad \text{in } U, \quad u=0 \text{ on } \partial U.$$
The term $\gamma |u|^{p-2}u$ represents the "strange term" arising from the obstacles. The result shows that even with clustering, the effective behavior is additive and determined by the average density of the fractional capacity of the obstacles.

5. Significance

• Mathematical Rigor: The paper resolves the technical difficulty of handling clusters in nonlocal settings, proving that their contribution to the capacity density is negligible under the derived moment condition.
• Unified Framework: It provides a robust framework for stochastic homogenization of nonlinear nonlocal problems, unifying the treatment of periodic, random, and clustered geometries without relying on specific PDE regularity results.
• Applicability: The results are applicable to a wide range of physical models involving nonlocal interactions (e.g., anomalous diffusion, image processing, material science) where defects or obstacles may be randomly distributed and clustered.

In summary, the paper establishes that the "strange term" in fractional obstacle problems is robust against clustering, provided the obstacle sizes satisfy a specific moment condition, and derives the explicit form of the homogenized operator using advanced probabilistic and variational techniques.