A Comparative Study of Molecular Dynamics Approaches for Simulating Ionic Conductivity in Solid Lithium Electrolytes

TL;DR

This study benchmarks molecular dynamics methods, including DFT and machine-learning potentials, for predicting ionic conductivity in lithium electrolytes, highlighting comparable accuracy and significant speed advantages of ML approaches.

Contribution

It provides a systematic comparison of DFT and ML-based MD approaches for ionic conductivity prediction in lithium electrolytes, demonstrating ML's efficiency and potential.

Findings

ML potentials achieve comparable accuracy to DFT in conductivity prediction.

ML simulations run over 350 times faster than DFT on comparable hardware.

The framework facilitates future comparisons of different ML interatomic potentials.

Abstract

Accurate prediction of ionic conductivity is critical for the design of high-performance solid-state electrolytes in next-generation batteries. We benchmark molecular dynamics (MD) approaches for computing ionic conductivity in 21 lithium solid electrolytes for which experimental ionic conductivity has been previously reported in the literature. In particular, we compare simulations driven by density functional theory (DFT) and by universal machine-learning interatomic potentials (uMLIPs), namely a MACE foundation model. We find comparable performance between DFT and MACE, despite MACE on one GPU more than 350 times faster than DFT on a 64-CPU node. The framework developed here is designed to enable systematic comparisons with additional uMLIPs and fine-tuned models in future work.