Ion Dynamics in Amorphous Solid Electrolytes Studied Using Neural Network Potentials

Understanding ion behavior at the atomic scale within all-solid-state lithium (Li) batteries is essential for advancing their performance. This chapter introduces the neural network potential (NNP), a machine-learning interatomic potential method, which is expected to exhibit accuracy comparable to first-principles calculations based on density functional theory while significantly reducing computational costs. Additionally, a neural network (NN) model for predicting Born effective charges is briefly explained. Utilizing these NN-based models, ion behavior in amorphous-Li3PO4 under an applied electric field is investigated. Results reveal enhanced Li motion along the electric field direction, which is a physically reasonable observation. Moreover, ionic mobility in Li3PO4 is notably increased in its amorphous state compared to crystalline, indicating ion susceptibility to electric fields in metastable configurations. Furthermore, extensive molecular dynamics simulations employing an NNP successfully reproduce the crystallization process of Li3PS4 glass through heat treatment.