Deep Learning of Accurate Interatomic Potentials for Uranium, Zirconium and Uranium-Zirconium Alloy

Abstract

Uranium-zirconium alloy is an important nuclear fuel in Integral Fast Reactor, which is of great significance to study its basic physical properties at high temperature by using advanced calculation methods. This work used deep potential molecular dynamics method, which combines the high accuracy of the first principles with the high efficiency of the classical molecular dynamics, to perform an evaluation of the static and thermophysical properties of body-centered cubic phase zirconium, uranium, and uranium-zirconium alloy. Firstly, the deep potential(DP) models of body-centered cubic zirconium(Zr-BCC), body-centered cubic uranium(U-BCC), and body-centered cubic uranium-zirconium alloy(U-Zr（BCC）) were trained by using deep neural network machine learning. Secondly, the DP models were used to predict equilibrium state equation, lattice constant, elastic properties, and phonon spectrum of the three systems, and the predicted results can reach the accuracy of the first principles. Then, the variation of heat capacity and density at constant pressure of Zr-BCC, U-BCC, and U-Zr（BCC） with temperature were predicted by using DP models, and the results are in good agreement with the experimental values. The research results show that the machine learning method provides an important path for successfully exploring more complex nuclear fuel properties.

Keywords

• uranium-zirconium alloy
• machine learning
• first principles
• molecular dynamics
• thermodynamic properties