Density functional and neural-network potential simulations of Ag migration in disordered GeS₂ electrolytes

Abstract

Density functional/molecular dynamics (DF/MD) simulations on GeS2 with four compositions of silver (2.5%–20%) provide valuable information on the structures and energetics of the amorphous material. Ag fills the available empty volume (cavities, voids) and interacts mainly with sulfur. The number of Ag-Ag contacts increases with increasing Ag composition, leading mainly to branched chains that partially disrupt the covalent Ge-S network. We use DF/MD simulations to study the energetics of Ag dissolution, the structures of Ag-GeS2 alloys for differing Ag content, migration paths in the energy surfaces at 0 K, and liquid dynamics at 1100 K, close to the experimental melting point. We discuss the electronic structures of the optimized glass structures. DF/MD simulations can describe the dynamics of these systems at liquid temperatures, but are too short to allow the study of Ag migration at lower temperatures. We have used DF/MD trajectories and generated snapshots of high-T liquids to develop a neural network potential (NNP) that allows simulations of these systems for tens of nanoseconds. These simulations are used for migration dynamics at 500–800K, leading to a linear behavior of the mean-square displacement on a logarithmic scale; the effective migration energy of Ag (0.33−0.43eV, with little composition dependence) agrees well with the static DF migration barriers.