Overestimation of melting temperatures calculated by first-principles molecular dynamics simulations

TL;DR

This study evaluates the accuracy of melting temperature calculations from first-principles molecular dynamics, revealing systematic overestimations mainly due to cell size limitations and energy dissipation effects, with implications for semiconductors, metals, and oxides.

Contribution

The paper identifies key factors causing overestimation of melting temperatures in FP-MD simulations and demonstrates methods to improve accuracy, including larger cell sizes and adiabatic MD simulations.

Findings

Overestimation of $T_m$ in most cases except Si.

Increasing cell size reduces overestimation but converges slowly.

Adiabatic MD captures energy dissipation effects in liquids.

Abstract

Although the melting temperature, $T_{m}$, of a solid can be calculated based on first-principles molecular dynamics (FP-MD) simulations, systematic assessments of the accuracy of the resulting values have not yet been reported. FP-MD simulations require significant computational resources and hence an examination of the effect of cell size on convergence is difficult. In addition, calculation of the energy of a liquid is not a trivial problem because of energy dissipation effects. The present work attempts to resolve these problems, and thus allow the accuracy of $T_{m}$ values obtained from FP-MD simulations to be assessed for typical semiconductors, metals, and oxides. With the exception of Si, the $T_{m}$ value was overestimated in all cases. This overestimation can be reduced by increasing the cell size, although the convergence is slow unless the potential is very shallow. For oxides, this overestimation may not be removed by increasing the cell size. The LDA/GGA error of overbinding affects the melting enthalpy and thereby $T_{m}$. In order to fully capture the energy dissipation nature of liquids, adiabatic MD simulations are required, and such simulations have been performed in the present study.