Atomistic Modeling of Bulk and Grain Boundary Diffusion for Solid Electrolyte Li₆PS₅Cl Accelerated by Machine-Learning Interatomic Potentials

Li₆PS₅Cl is a promising candidate for the solid electrolyte in all-solid-state lithium-ion batteries. For applications, this material is in a polycrystalline state with many grain boundaries (GBs) rather than an idealized single-crystalline state. Atomistic simulations of Li₆PS₅Cl with GBs, however, remain rare due to high computational cost. In this study, machine-learning interatomic potentials, specifically moment tensor potentials (MTPs) [1], are employed to accelerate the simulations while preserving the ab initio accuracy. In the initial stage, energies and forces of a small number of configurations are generated under the <i>ab initio</i> framework. MTPs are then fitted to the <i>ab initio</i> data, and active learning techniques are used to further stabilize the MTPs. The usage of MTPs enables molecular dynamics (MD) simulations in larger system sizes (up to 20 000 atoms) and longer time scales (several ns). Two tilt GBs Σ3[110]/(1-12), Σ3[110]/(-111) and one twist GB Σ5[001]/(001) are focused on, all of which show relatively low GB energies and an enhanced ionic conductivity for Li compared to bulk. Diffusion mechanisms specific to each GB are analyzed. This research offers new insights into the design of solid electrolytes through GB engineering and emphasizes the importance of considering GBs for materials modeling.<br/><br/>[1] A.V. Shapeev, Multiscale Modeling & Simulation <b>14</b>, 1153 (2016).