Salty ices modeling using data-driven approaches and atomistic simulations – SIMODAS

Ice phases incorporating a non-negligible amount of salt have been recently experimentally demonstrated. In particular, both high- and low-pressure phases have been reported, showing interest respectively in models for planetary interior compositions, and solid aqueous electrolytes for battery devices. Investigating these phases in greater detail and potentially identifying new ones is therefore interesting both for fundamental and applied research perspectives. Since high-pressure characterization experiments are challenging, atomic-scale numerical simulations such as molecular dynamics can help providing valuable insights in the formation and stability of materials. Four challenges prevent reaching a quantitative description of salty ice phases through atom scale modeling: the computational cost of ab initio approaches, the subtle balance of interactions in water, the low accuracy and transferability of empirical potentials, and the selection of an appropriate collective variable to follow transformations between phases. The SIMODAS project aims at solving all four challenges, by combining two data-driven approaches: one to derive optimal collective variables, and one to extract accurate and transferable interaction potentials. With this methodology in hand, the formation processes and stability field of a range of salty ices will be investigated. In particular, several halides will be considered, to extract knowledge regarding the necessary conditions for the favorable formation of crystallines phases. In addition, work will be devoted to characterize the ion transport properties of these phases to provide valuable information for planetary models and materials design for electrolytes. Finally, several methodological improvements related to the committor probability will be investigated, ranging from accelerating its estimation from molecular dynamics simulations, to investigating local approximations enabling significant dimensionality reduction in optimizing collective variables.