Parallelising Glauber dynamics

TL;DR

This paper demonstrates that parallelizing Glauber dynamics by resampling multiple coordinates simultaneously can significantly accelerate mixing times, leading to efficient parallel algorithms for models like the Ising and p-spin at high temperatures.

Contribution

It introduces a method to parallelize Glauber dynamics, achieving faster mixing and efficient parallel algorithms for specific probabilistic models.

Findings

k-Glauber mixes k times faster in χ²-divergence

Parallel algorithms run in b5O(\u221a n) time for Ising model

High-temperature p-spin model also benefits from parallelization

Abstract

For distributions over discrete product spaces $\prod_{i=1}^n \Omega_i'$, Glauber dynamics is a Markov chain that at each step, resamples a random coordinate conditioned on the other coordinates. We show that $k$-Glauber dynamics, which resamples a random subset of $k$ coordinates, mixes $k$ times faster in $\chi^2$-divergence, and assuming approximate tensorization of entropy, mixes $k$ times faster in KL-divergence. We apply this to obtain parallel algorithms in two settings: (1) For the Ising model $\mu_{J,h}(x)\propto \exp(\frac1 2\left\langle x,Jx \right\rangle + \langle h,x\rangle)$ with $\|J\|<1-c$ (the regime where fast mixing is known), we show that we can implement each step of $\widetilde \Theta(n/\|J\|_F)$-Glauber dynamics efficiently with a parallel algorithm, resulting in a parallel algorithm with running time $\widetilde O(\|J\|_F) = \widetilde O(\sqrt n)$. (2) For the mixed $p$-spin model at high enough temperature, we show that with high probability we can implement each step of $\widetilde \Theta(\sqrt n)$-Glauber dynamics efficiently and obtain running time $\widetilde O(\sqrt n)$.