Parallel kinetic Monte Carlo simulation of Coulomb glasses

TL;DR

This paper introduces a parallel rejection algorithm and GPU implementation for simulating Coulomb glasses, significantly accelerating computations and revealing a scaling law for conductivity relaxation.

Contribution

The paper presents a novel parallel rejection algorithm and GPU-based simulation method for Coulomb glasses, achieving substantial speedups and new insights into conductivity relaxation.

Findings

Speedups of up to 100x over serial code

Numerical evidence of a scaling relation for conductivity relaxation

Effective parallelization of local energy updates

Abstract

We develop a parallel rejection algorithm to tackle the problem of low acceptance in Monte Carlo methods, and apply it to the simulation of the hopping conduction in Coulomb glasses using Graphics Processing Units, for which we also parallelize the update of local energies. In two dimensions, our parallel code achieves speedups of up to two orders of magnitude in computing time over an equivalent serial code. We find numerical evidence of a scaling relation for the relaxation of the conductivity at different temperatures.