Replica exchange nested sampling

TL;DR

This paper introduces replica-exchange nested sampling (RENS), a new method that enhances nested sampling efficiency by integrating replica-exchange moves, enabling better exploration of complex energy landscapes in materials science.

Contribution

The paper presents RENS, a novel extension of nested sampling that incorporates replica exchange to improve sampling efficiency and applicability to complex, multimodal systems.

Findings

RENS accelerates convergence compared to traditional NS.

RENS effectively handles challenging multimodal energy landscapes.

RENS expands the applicability of nested sampling to realistic material models.

Abstract

Nested sampling (NS) has emerged as a powerful tool for exploring thermodynamic properties in materials science. However, its efficiency is often hindered by the limitations of Markov chain Monte Carlo (MCMC) sampling. In strongly multimodal landscapes, MCMC struggles to traverse energy barriers, leading to biased sampling and reduced accuracy. To address this issue, we introduce replica-exchange nested sampling (RENS), a novel enhancement that integrates replica-exchange moves into the NS framework. Inspired by Hamiltonian replica exchange methods, RENS connects independent NS simulations performed under different external conditions, facilitating ergodic sampling and significantly improving computational efficiency. We demonstrate the effectiveness of RENS using four test systems of increasing complexity: a one-dimensional toy system, periodic Lennard-Jones, the two-scale core-softened Jagla model and a machine learned interatomic potential for silicon. Our results show that RENS not only accelerates convergence but also allows the effective handling of challenging cases where independent NS fails, thereby expanding the applicability of NS to more realistic material models.