Modelling Phase Transformations and Diffusion in Na-Ion Conductors Using Machine Learned Interatomic Potentials

Machine learned interatomic potentials (MLIP) have become an indispensable tool in computational materials science, promising to extend the accessible length and timescales of ab initio molecular dynamics (AIMD), while retaining an accurate description of the relevant interactions. The last decade has seen a tremendous increase in the accuracy, data efficiency and scalability of the MLIPs. Consequently, their applications are being pushed towards increasingly complex materials and phenomena.<br/><br/>Here, we present an application of the Allegro [1] MLIP, a recently developed local E(3)-equivariant graph neural network potential, to a set of promising Na-ion conductors. Specifically, we study the Na_3-xSb_1-xWxS₄ system, where aliovalent W-dopants are introduced along with charge-compensating Na-vacancies in order to facilitate Na-ion diffusion. Describing this system and the relevant physical phenomena using an MLIP requires it to describe ionic diffusion and structural phase-transformation and their dependence on W-dopant concentration.<br/><br/>In the talk, we showcase how our trained MLIP performs on this demanding set of tasks [2]. In particular, we use large scale MD simulations to shows that an increased Na vacancy population stabilises the global cubic phase at lower temperatures with enhanced Na ion diffusion, and that the explicit role of the substitutional W dopants is limited. In the global cubic phase we observe large and long-lived deviations of atoms from the averaged symmetry, echoing recent experimental suggestions. Evidence of correlated Na ion diffusion is also presented that underpins the suggested superionic nature of these materials.<br/><br/>[1] Nature Communications,14, 579 (2023)<br/>[2] Klarbring and Walsh, arXiv preprint arXiv:2403.20138 (2024)