Machine Learning-Accelerated Molecular Dynamics for Electrolyte R&D

This document presents a comprehensive approach to accelerating electrolyte research and development using machine learning-accelerated molecular dynamics. It begins by highlighting the critical role of electrolyte materials in energy storage systems, where safety, conductivity, viscosity, and redox potential limit performance. The talk outlines the development of computational electrochemical methods for electrolytes, emphasizing the need to accurately describe complex solvation structures in real electrolytes beyond simple dilute solutions. A key focus is constructing a universal potential for electrolytes using machine learning, enabling efficient simulation of electrode/electrolyte interfaces. The document discusses the electrochemical stability window, contrasting band gaps with actual stability, and analyzes concentration effects via projected density of states. Free energy calculation methods, including particle insertion and thermodynamic integration, are introduced to compute redox potentials and acidity constants. The development of computational reference electrodes based on ab initio molecular dynamics allows for accurate potential calculations benchmarked against experiments. Overall, the work aims to bridge the gap between computational chemistry and practical electrolyte design, leveraging artificial intelligence to explore the vast material search space for next-generation energy storage.