Data-driven design of advanced magnesium-battery electrolyte via dynamic solvation models

Abstract

Artificial intelligence (AI) facilitates electrolyte screening by correlating the complex physicochemical properties of solvent/clusters with battery performance. However, modeling and interpreting the high-dimensional relationships between the dynamic evolution of ion-solvent clusters and their electrochemical performance with machine learning remains challenging by using the traditional static model. In this work, we developed a dynamic solvation model by precisely extracting descriptors of the composition, solvation, and migration stages for solvated ions. Taking rechargeable magnesium batteries (RMBs) as the sample, the model reveals that the optimal anion-coordinated solvation structure for RMBs features ligand coordination numbers (CNs) of 2/3/4 and an atomic CN of 5, enhancing desolvation and solid electrolyte interphase formation. Additionally, the diffusion coefficient, crucial for ionic conductivity, is influenced by dielectric constants and solvent properties. An intelligent screening process based on this model identifies electrolytes that demonstrate a low overpotential and long cycle life in experimental validation, offering new perspectives on designing high-performance batteries using artificial intelligence.