Skip to main content
Photon Sciences

Machine-learning Predicts Potentials for Molten Salts from X-ray Pair Distribution Function (PDF) Data

A new machine learning approach significantly lowers the barrier to understanding and designing molten salts across the periodic table.

Molten salt reactors use liquid fluoride or chloride salt as a fuel, but the extreme chemical reactivity of molten alkali chlorides at high temperatures has presented a significant challenge in characterizing atomic structures and physical properties experimentally. Modeling is inhibited by the cost of quantum-mechanically treating the high polarizabilities of molten salts. Our methodology reduces the number of expensive ab initio evaluations by several orders of magnitude. Here, we integrate configurational sampling of density functional theory calculations using classical force fields with active learning to automate and accelerate the generation of Gaussian approximation potentials (GAP) for molten salts.

The developed models provide new physical insights into the temperature-dependent coordination environment of the molten state as well as densities, self-diffusion constants, and ionic conductivities. The ML methodology indicates that GAP models are able to capture the many-body interactions required to accurately model ionic systems. This significantly lowers the barrier to the understanding and design of molten salts across the periodic table.