Regensburg 2022 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
MM: Fachverband Metall- und Materialphysik
MM 14: Materials for Storage and Conversion of Energy (joint session MM/KFM)
MM 14.1: Talk
Tuesday, September 6, 2022, 10:15–10:30, H46
How Important are Long-Range Electrostatics in Machine-Learning Potentials for Battery Materials? — •Carsten Staacke1, Hendrik Heenen1, Christoph Scheurer1, Gabor Csanyi2, Karsten Reuter1, and Johannes Margraf1 — 1Fritz Haber Institut, Berlin, Germany — 2Engineering Department, Cambridge University, UK
All-solid-state Li-ion batteries promise gains in safety and durability by combining high Li-ion conductivity and mechanical ductility. In this respect, solid-state electrolytes (SSE) such as the Li7P3S11 glass-ceramic have gained much attention. Modern machine learning (ML) potentials are increasingly being adopted as a tool for modeling SSEs at the atomistic level. However, the local nature of these ML potentials typically means that long-range contributions arising, e.g., from electrostatic interactions are neglected. To this end, we have combined short-ranged machine-learning potentials based on the Gaussian Approximation Potential (GAP) approach with a classical electrostatic model in the long-range (ES-GAP). We will present a first-principles validation of both, the pure GAP potential and the new ES-GAP for the LPS SSE. In particular, the role of Coulomb interactions in isotropic vs. non-isotropic system simulations will be evaluated. In standard isotropic simulation tasks, such as determining ionic conductivities, both GAP and ES-GAP yield similar results. In contrast, simulations on non-isotropic systems show the importance of ES contributions and provide new insights into interface stability of Li7P3S11.