Speakers

Foundation potentials (FPs), i.e., machine learning interatomic potentials with near universal coverage of the entire periodic table, have emerged as a promising new tool in the study of materials. In this talk, I will provide a general overview of the capabilities of FPs, especially in the context of design of solid electrolytes for all-solid-state batteries. I will demonstrate how modern FPs trained on robust, diverse datasets such as MatPES, are a highly effective surrogate for expensive density functional theory (DFT) calculations to access a wide range of PES-derived properties, such as the ionic conductivity, phase stability, electrochemical stability, phonon dispersion, elastic moduli, etc. I will also discuss approaches to fine-tune such FPs with a minimal amount of data, resulting in significantly improved accuracies and efficiency for simulations at scale. I will share new insights obtained from FP-accelerated simulations with regards the differences in ionic conductivities between different structures and chemistries.