Neural network potential for molecular dynamics calculation of UO2

Abstract

This study employed a machine learning approach to develop a neural network potential for molecular dynamics simulations of uranium dioxide (UO 2 ). The results of first-principles calculations were used as training data. The calculation results of the physical properties of UO 2 showed that this potential is widely applicable to the evaluation of physical properties. It is particularly effective for calculating thermophysical properties near the melting point, where experiments are difficult due to extremely high temperatures. The calculations of diffusion constant suggest that the melting of the oxygen sublattice in UO2 occurs at temperatures beyond 2600 K.

Topics & Concepts