The computational modeling of molten salts is of crucial importance to emerging "carbon free" energy applications such as molten salt reactors and concentrated solar power. A realistic atomistic description of molten salt melts is inhibited by the computational cost of quantum mechanically treating the high polarizabilities characteristic of molten salts. In this talk, we will present a data efficient active learning on configurations sampled from classical forcefield to automate the generation of near-DFT accurate machine learning Gaussian Approximation Potentials (GAP). We will use molten LiCl as a proof of concept to show that a GAP model can be generated using fewer than 600 active learned training configurations and exhibits a 19,000x speedup relative to conventional ab initio molecular dynamics. We will also discuss the quality of the GAP model simulation results by its ability to reproduce experimental structure factors, densities, self-diffusion coefficients, and ionic conductivities. This hybrid simulation strategy significantly accelerates the generation of machine learning potentials for molten salts by reducing the expensive ab initio calculations required for parameterization to O(100) evaluations, enabling the facile generation of first-principles quality predictions of structural and dynamical properties of molten salts. We end the talk with a discussion on modeling binary molten salt mixture melts.

Active learning driven automated parametrization of machine learning forcefields for molten salt melts

ACS Fall 2021

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