Harnack inequalities on weighted graphs and some applications to the random conductance model

Sebastián Andrés, Jean‐Dominique Deuschel, Martin SlowikProbability 2015 · DOI

TL;DR

This paper establishes new Harnack inequalities for random walks on graphs with unbounded weights, extending classical results to a more general setting. These inequalities are crucial for understanding the long-term behavior of random processes and have direct applications in proving local limit theorems for random conductance models, which are fundamental in statistical physics.

What Problem Does This Paper Address?

The paper tackles the challenge of understanding random walks on graphs where the "conductivity" of edges can vary significantly, even becoming infinitely large or infinitesimally small. Classical results on Harnack inequalities, which describe how "smooth" harmonic functions are, typically assume bounded conductances. This paper extends these powerful tools to graphs with unbounded weights, which are more realistic for modeling phenomena like disordered materials. A key motivation is to provide a rigorous mathematical framework for analyzing the behavior of random conductance models, which are used to describe diffusion in complex media. Specifically, the authors aim to prove a local limit theorem, which describes how the probability distribution of a random walk behaves at large scales.

Methods

The core of the paper lies in developing and proving elliptic and parabolic Harnack inequalities on weighted graphs. The authors adapt and extend Moser's iteration technique, a standard method for proving Harnack inequalities. A key innovation is the use of weighted Sobolev and Poincaré inequalities tailored to the graph setting with varying weights. They also leverage an abstract lemma by Bombieri and Giusti to bridge the gap between controlling local norms and establishing the desired inequalities. For the application to the random conductance model, the authors utilize ergodic theorems to handle the randomness of the edge weights and combine these with the established Harnack inequalities to prove a quenched local limit theorem.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • Generalizing Harnack Inequalities: The paper demonstrates that Harnack inequalities, fundamental tools in PDE theory, can be extended to graphs with unbounded edge weights. This opens up the analysis of a wider range of random processes.
  • Lp Bounds for Harnack Constants: A significant contribution is showing that the constants in the Harnack inequalities can be controlled by Lp norms of the graph's measures (related to conductances), rather than just uniform bounds. This provides a more refined understanding of the inequalities' dependence on the graph's structure.
  • Bridging Theory and Application: The paper successfully connects abstract mathematical tools (Harnack inequalities) to concrete physical models (random conductance model) by proving a quenched local limit theorem. This highlights the practical relevance of the theoretical developments.
  • Optimality of Moment Conditions: The authors provide evidence that the moment conditions they impose on the edge weights are likely optimal for the existence of local limit theorems, suggesting a fundamental limit to how "heavy-tailed" the edge weights can be.

Key Results

  • Elliptic and Parabolic Harnack Inequalities: The paper establishes new elliptic and parabolic Harnack inequalities for random walks on weighted graphs. These inequalities provide bounds on the ratio of maximum to minimum values of harmonic/caloric functions over different regions of the graph. The constants in these inequalities depend on Lp norms of the graph measures.
  • Quenched Local Limit Theorem: As a direct application, the authors prove a quenched local limit theorem for the random conductance model. This theorem describes the convergence of the rescaled transition probabilities of the random walk to a Gaussian distribution, under specific moment conditions on the edge weights.
  • Near-Diagonal Heat Kernel Bounds: The paper also derives near-diagonal bounds on the heat kernel, which are important consequences of the parabolic Harnack inequality.

Limitations and Future Work

  • Moment Conditions: The paper discusses the optimality of the moment conditions (Assumption 1.9) for the local limit theorem, suggesting that deviations from these conditions can lead to the failure of the theorem. However, exploring exact conditions for the failure of the local limit theorem remains an area for future work.
  • Gaussian Off-Diagonal Bounds: The authors note that directly deducing off-diagonal Gaussian bounds from the parabolic Harnack inequality is challenging due to the dependence of the Harnack constant on the underlying ball. They suggest that further work is needed to obtain these bounds in their non-elliptic setting.
  • Examples and Specific Cases: While the paper provides general results, exploring specific examples of weighted graphs and conductances could offer deeper insights and potentially relax some of the derived conditions.

Summary updated

BibTeX
@inproceedings{andres2015harnack,
  author = {Sebastián Andrés and Jean‐Dominique Deuschel and Martin Slowik},
  title = {{Harnack inequalities on weighted graphs and some applications to the random conductance model}},
  booktitle = {Probability},
  year = {2015},
  doi = {10.1007/s00440-015-0623-y},
}
Figure