Exploring diffusion behavior of superionic materials using machine-learning interatomic potentials

Abstract

Superionic materials possess mobile atoms with liquidlike behavior in the rigid frameworks of other atoms. Theoretically, the diffusion behavior of the mobile atoms is usually probed by ab initio molecular dynamics simulations where enormous computing resources are requested for a complete thorough study. Thus, only limited cases are investigated without providing the most critical quantity, such as the diffusion barrier. To address this shortcoming, we perform molecular dynamics simulations based on machine-learning interatomic potentials, fitted from ab initio molecular dynamics simulations, to have complete studies for ${\mathrm{Ag}}_2\mathrm{S}, {\mathrm{Ag}}_8{\mathrm{SiTe}}_6, {\mathrm{Cu}}_2\mathrm{S}$, and ${\mathrm{Zn}}_{3.6+x}{\mathrm{Sb}}_3$ systems. Our results indicate that the Arrhenius equation can describe very well the diffusion behaviors of the studied superionic systems where the activation barriers range from 0.09–0.22 eV. The small diffusion barrier provides the fundamental origin for the liquid behaviors of superionic materials.