Benchmarking Machine Learning Models for Predicting Lithium Ion Migration

Abstract

The development of fast ionic conductors to improve the performance of electrochemical devicesrelies on expensive high-throughput (HT) density functional theory (DFT) calculations of transportproperties. Machine learning (ML) can accelerate HT workflows but requires high-quality data toensure accurate predictions from trained models. In this study, we introduce the LiTraj dataset,which comprises 13,000 percolation and 122,000 migration barriers, and 1,700 migration trajectories,calculated for Li-ion in diverse crystal structures using empirical force fields and DFT, respectively.Using LiTraj, we demonstrate that classical ML models and graph neural networks (GNNs) forstructure-to-property prediction of percolation and migration barriers can distinguish between ”fast”and ”poor” ionic conductors. Furthermore, we evaluate the capability of GNN-based universalML interatomic potentials (uMLIPs) to identify optimal Li-ion migration trajectories. Fine-tuneduMLIPs achieve near-DFT accuracy in predicting migration barriers, significantly accelerating HTscreenings of new ionic conductors.