Quenched invariance principle for random walks in random environments admitting a cycle decomposition

Jean‐Dominique Deuschel, Martin Slowik, Wenwen WengarXiv 2024 · DOI · arXiv

Abstract

We study a class of non-reversible, continuous-time random walks in random environments on $\mathbb{Z}^d$ that admit a cycle representation with finite cycle length. The law of the transition rates, taking values in $[0, \infty)$, is assumed to be stationary and ergodic with respect to space shifts. Moreover, the transition rate from $x$ to $y$, denoted by $c^\omega(x,y)$, is a superposition of non-negative random weights on oriented cycles that contain the edge $(x,y)$. We prove a quenched invariance principle under moment conditions that are comparable to the well-known p-q moment condition of Andres, Deuschel, and Slowik [2] for the random conductance model. A key ingredient in proving the sublinearity is an energy estimate for the non-symmetric generator. Our result extends that of Deuschel and Kösters [12] beyond strong ellipticity and bounded cycle lengths.

Introduction

Random walks in random environments (RWRE) are fundamental models for understanding diffusion in disordered media, such as pollutants moving through soil or heat through a composite material. A major goal in this field is to prove an Invariance Principle (or Functional Central Limit Theorem), which shows that at large scales, the path of the random walk converges to the smooth, predictable path of a Brownian motion.

Most existing proofs for this behavior—specifically the Quenched version, which holds for almost every possible realization of the random environment—rely on the walk being reversible. However, many real-world systems are non-reversible. Previous work on non-reversible walks required "strong ellipticity" (transition rates strictly bounded away from zero and infinity) and "bounded cycle lengths." This paper tackles the challenge of proving the invariance principle when these strict bounds are removed, allowing for much more irregular and "degenerate" environments.

The Cyclic Model and Methods

The authors focus on continuous-time random walks on the integer lattice $\mathbb{Z}^d$ where the transition rates are built from a cycle decomposition. In this model, the environment is a collection of oriented cycles, each with a random weight. The rate of jumping from $x$ to $y$ is the sum of the weights of all cycles containing that specific edge.

Key technical components include:

  • Corrector Construction: The walk is decomposed into a martingale and a "corrector" term. The primary challenge is proving that this corrector is sublinear, meaning it grows slower than the walk itself and vanishes when rescaled.
  • Analytical Techniques: The authors use the Lax-Milgram Theorem to construct the corrector and adapt De Giorgi Iteration (typically used in elliptic PDEs) to prove its sublinearity. This method is more robust for handling the non-symmetric nature of the walk's generator than the more common Moser iteration.
  • Energy Estimates: They derive a novel energy estimate for non-symmetric generators that leverages the cycle structure to control the "drift" of the walk without needing a standard "sector condition."

Key Results

  • Theorem 2.6 (QFCLT): The main result establishes that for dimensions $d \ge 2$, the rescaled random walk converges in law to a Brownian motion with a deterministic, non-degenerate covariance matrix.
  • Relaxing Ellipticity: The paper shows that transition rates do not need to be uniformly bounded. Instead, they use p-q moment conditions (integrability requirements on the weights and their inverses), mirroring the state-of-the-art conditions used for symmetric "random conductance" models. The condition $1/p + 1/q < 2/d$ is implemented, which is a well-known benchmark for optimality.
  • Unbounded Cycles: Unlike previous studies that required cycles to have a maximum length, this work allows for cycles of any finite length, significantly broadening the class of applicable environments.

Limitations and Future Work

While this work significantly extends the reach of the Quenched Invariance Principle, several questions remain. The results are specifically for $d \ge 2$; the one-dimensional case often requires different mathematical approaches. Additionally, while the $2/d$ moment condition is prevalent, some symmetric models allow for an improved bound of $2/(d-1)$; whether this improvement is possible for the non-reversible cyclic model is an open question. Finally, extending these results to dynamic random environments, where the environment itself evolves over time, remains a natural next step.

Summary updated

Abstract

We study a class of non-reversible, continuous-time random walks in random environments on $\mathbb{Z}^d$ that admit a cycle representation with finite cycle length. The law of the transition rates, taking values in $[0, \infty)$, is assumed to be stationary and ergodic with respect to space shifts. Moreover, the transition rate from $x$ to $y$, denoted by $c^ω(x,y)$, is a superposition of non-negative random weights on oriented cycles that contain the edge $(x,y)$. We prove a quenched invariance principle under moment conditions that are comparable to the well-known p-q moment condition of Andres, Deuschel, and Slowik [2] for the random conductance model. A key ingredient in proving the sublinearity is an energy estimate for the non-symmetric generator. Our result extends that of Deuschel and Kösters [12] beyond strong ellipticity and bounded cycle lengths.
BibTeX
@misc{deuschel2024quenched,
  author = {Jean‐Dominique Deuschel and Martin Slowik and Wenwen Weng},
  title = {{Quenched invariance principle for random walks in random environments admitting a cycle decomposition}},
  year = {2024},
  eprint = {2411.06861},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2411.06861},
}
Figure