Improving the performance of liquid electrolytes is one of today's major challenges in the field of batteries, with the aim of improving efficiency, safety and economy. Recent advances include superconcentrated media such as WIS (“Water-In-Salts”) solutions. Their properties depend crucially on the chemistry and physics of the interfaces between water and ions (Li+ for lithium-ion batteries, but also Na+, K+, Zn2+), both at a distance and close to the electrodes.

Atomic-scale modeling of these superconcentrated liquid electrolytes requires the study of nanoscopic structures and phenomena taking place over long timescales. One relevant solution is to build potentials by machine learning, based on ab initio molecular dynamics (AIMD) trajectories. This method combines an accurate description of the interactions between ions and water molecules, including the breaking and forming of chemical bonds, with fast calculation speed. In particular, the DeePMD kit has recently been successfully ported to GPU architectures, paving the way for calculations on exascale supercomputers (whose power exceeds 10^18 floating-point operations per second).
This theoretical study will be supported by an experimental counterpart, thanks to direct collaboration with a team in the unit specializing in electrochemistry.