Machine-learning molecular dynamics study of thermal properties of CaF

Nakamura, Hiroki  ; Machida, Masahiko  

It is not easy to measure high-temperature properties of actinide dioxides, which are the main component of nuclear fuels, due to their high melting points. CaF is alternative materials of actinide dioxides since it has the same crystal structure and a lower melting point. In this paper, we perform machine-learning molecular dynamics with the Behler-Parrinello neural-network potential which is optimized by first-principles calculation results. We confirm the Bredig transition peak in the heat capacity and coefficient of thermal expansion in calculated results. We also discuss the availability of the present neural-network potential to analyze the high-temperature properties of CaF.