Despite the emergence of powerful machine learning interatomic potentials, their use for mixtures is rarely explore due to the lack of transferability between different mixture compositions. In molten salt systems, where the use of ML potentials is crucial for accurately accounting for strong polarizability, discovery of experimentally relevant eutectic mixtures is inhibited by this lack of transferability. We present a composition transferable Gaussian Approximation Potential for LiCl-KCl melts, achieving strong agreement with High Energy X-ray Diffraction (HEXRD) experiments across a broad composition range, including a eutectic not explicitly included in model training set. Our approach will find broad applicability across condensed phase systems.

Composition-transferable machine learning potential for binary molten salts melts validated by high energy X-ray diffraction

Fall 2022 abstracts, posters, and presentations will launch in mid-September.

ACS Fall 2022

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.