Extended Ensemble Molecular Dynamics for Thermodynamics of Phases

TL;DR

This paper explores extended ensemble molecular dynamics methods to improve sampling of phase transitions in Lennard-Jones systems, providing insights into thermodynamics and phase behavior of materials.

Contribution

It introduces and compares multibaric-multithermal and multiorder-multithermal ensemble methods for phase transition simulations, highlighting their strengths and limitations.

Findings

MBMT agrees with known EOSs for fluid and gas-liquid transitions

MBMT struggles to predict liquid-solid transition accurately

MOMT enhances sampling and accurately estimates transition temperatures

Abstract

The first-order phase transitions and related thermodynamics properties are primary concerns of materials sciences and engineering. In traditional atomistic simulations, the phase transitions and the estimation of their thermodynamic properties are challenging tasks because the trajectories get trapped in local minima close to the initial states. In this study, we investigate various extended ensemble molecular dynamics (MD) methods based on the multicanonical ensemble method using the Wang-Landau (WL) approach. We performed multibaric-multithermal (MBMT) method to fluid phase, gas-liquid transition, and liquid-solid transition of the Lennard-Jones (LJ) system. The derived thermodynamic properties of the fluid phase and the gas-liquid transition from the MBMT agree well with the previously reported equation of states (EOSs). However, the MBMT cannot correctly predict the liquid-solid transition. The multiorder-multithermal (MOMT) ensemble shows significantly enhanced sampling between liquid and solid states with an accurate estimation of transition temperatures. We further investigated the dynamics of each system based on their free energy shapes, providing fundamental insights for their sampling behaviors. This study guides the prediction of broader crystalline materials, e.g., alloys, for their phases and thermodynamic properties from atomistic modeling.