Metastability of Glauber dynamics with inhomogeneous coupling disorder

Anton Bovier, Frank den Hollander, Saeda Marello, Elena Pulvirenti, Martin SlowikLatin 2022

TL;DR

This paper investigates the phenomenon of metastability in complex spin systems with random interactions, specifically extending previous work on random graphs to more general settings. The authors develop a theoretical framework to show that the "annealed" version of these systems (where random couplings are replaced by their average) can predict the metastability of the "quenched" system (with actual random couplings), a crucial result for understanding the long-term behavior of these systems.

What Problem Does This Paper Address?

The paper tackles the problem of understanding metastability in complex magnetic systems, particularly the Ising model, when the interactions between spins are not uniform but are instead random. This randomness can arise in various physical scenarios, such as in disordered materials or networks. A key challenge is to predict how long such systems remain in a metastable state (a temporary stable state before transitioning to a more stable one) before escaping it. Previous work focused on specific types of random graphs, but this paper aims to generalize these findings to a broader class of systems with "inhomogeneous coupling disorder" and "random interactions." The authors are interested in whether the behavior of a system with averaged interactions (the "annealed" model) can reliably predict the metastability of the system with actual random interactions (the "quenched" model).

Methods

The core of the paper's methodology lies in the potential-theoretic approach to metastability. This approach leverages concepts from probability theory and analysis to study the escape times from metastable states. Key techniques include:

  • Harmonic sums and capacities: These are analytical tools used to quantify the "difficulty" of escaping a metastable state. The paper estimates these quantities for both the quenched and annealed models.
  • Dirichlet principle and Thomson principle: These variational principles are used to obtain upper and lower bounds on capacities, which are crucial for estimating escape times.
  • McDiarmid's concentration inequality: This inequality is used to prove that certain quantities (like the difference between quenched and annealed Hamiltonians) are highly concentrated around their expected values with high probability. This is a key tool for bridging the gap between the quenched and annealed models.
  • Comparison of Hamiltonians: The authors carefully compare the Hamiltonians of the quenched and annealed models, particularly on a high-probability event, to relate their dynamics.

A notable aspect is the use of McDiarmid's inequality, which provides exact constants and is more flexible than previous methods like Talagrand's inequality used in related work.

Datasets

Not discussed in the paper. The paper deals with theoretical models and does not use empirical datasets.

Key Insights and Ideas

  • Annealed vs. Quenched Predictability: The central insight is that, under certain conditions and with high probability, the metastability of the "annealed" model (where random couplings are averaged) implies the metastability of the "quenched" model (with actual random couplings). This is a significant theoretical result that simplifies the analysis of complex disordered systems.
  • Bridging the Gap: The paper provides a rigorous framework for comparing the dynamics of quenched and annealed systems, particularly focusing on how the randomness of couplings affects escape times from metastable states.
  • Generalization of Results: The work extends previous findings on metastability for Ising models on specific random graphs (like Erdős-Rényi) to a more general class of inhomogeneous random graphs and random interactions.
  • Quantitative Estimates: Beyond proving metastability, the paper provides quantitative estimates for the ratio of mean hitting times between the quenched and annealed models, characterizing their tail behavior and moments.

Key Results

The paper presents two main theorems:

  • Theorem 2.10 (Metastability): Under a specific assumption of metastability for the annealed model, the paper proves that the quenched model also exhibits metastability with high probability. This means that the random interactions do not fundamentally destroy the metastable behavior predicted by the averaged model.
  • Theorem 2.12 (Hitting Time Estimates): This theorem provides estimates for the ratio of mean hitting times between the quenched and annealed models. It shows that this ratio is bounded and its tail behavior and moments can be controlled. Specifically, it establishes bounds on the ratio of expected hitting times, demonstrating that they are asymptotically related.

Limitations and Future Work

  • Assumptions on Annealed Model: The main results rely on an assumption of metastability for the annealed model (Assumption 2.7). Proving this assumption for specific models is a separate, often challenging, task.
  • Generalization to Other Dynamics: While the paper focuses on Glauber dynamics, it notes that the results might hold for other reversible, single spin-update dynamics.
  • Unbounded Random Variables: The current work assumes uniformly bounded random variables for the couplings. The authors mention that future work will address the case of unbounded random variables with finite exponential moments.
  • Hopfield Model: The paper briefly discusses the Hopfield model as an example and suggests that addressing its metastable behavior in a more general regime is a topic for future research.
  • Coupling Methods: The paper acknowledges that constructing suitable couplings to prove certain regularity statements for harmonic sums in the quenched setting would require highly non-trivial adaptations of existing strategies due to the random interactions.

Summary updated

Abstract

We introduce a general class of mean-field-like spin systems with random couplings that comprises both the Ising model on inhomogeneous dense random graphs and the randomly diluted Hopfield model. We are interested in quantitative estimates of metastability in large volumes at fixed temperatures when these systems evolve according to a Glauber dynamics, i.e.\ where spins flip with Metropolis transition probabilities at inverse temperature $β$. We identify conditions ensuring that with high probability the system behaves like the corresponding system where the random couplings are replaced by their averages. More precisely, we prove that the metastability of the former system is implied with high probability by the metastability of the latter. Moreover, we consider relevant metastable hitting times of the two systems and find the asymptotic tail behaviour and the moments of their ratio. This work provides an extension of the results known for the Ising model on the Erdős--R{é}nyi random graph. The proofs use the potential-theoretic approach to metastability in combination with concentration inequalities.
BibTeX
@inproceedings{bovier2022metastability,
  author = {Anton Bovier and Frank den Hollander and Saeda Marello and Elena Pulvirenti and Martin Slowik},
  title = {{Metastability of Glauber dynamics with inhomogeneous coupling disorder}},
  booktitle = {Latin},
  year = {2022},
}
Figure