Recent Development of Atomic Interaction Potential and Its Application in U and Uranium Alloys

Wensheng Lai*，Yi Wang，Zheten Yu，Zhengjun Zhang

Tsinghua University, Beijing 100084，China

ABSTRACT：Ab initio molecular dynamics (MD) simulation is the most accurate method to carry out MD simulation at the atomic scale. However, due to its large amount of calculation and time-consuming, the calculation scale is small (generally no more than hundreds of atoms). In order to carry out large-scale simulation, classical molecular dynamics simulation based on atomic interaction potential, which can be used not only in MD simulation, but also in molecular statics, Monte Carlo and transition state (NEB) calculations, has been proposed. This presentation introduces the recent development of atomic interaction potential from based on analytical function, force field matching to machine learning method. By analyzing various kinds of atomic interaction potentials, it can be found that the traditional atomic interaction potential and machine learning potential can be regarded as the mathematical expansion approximation of embedded atom method at different precision levels. The main problem of developing atomic interaction potential suitable for high-precision and large-scale atomic simulation is the introduction of atomic bond order correlation term and a systematic complete and efficient basis function set for describing electron density in this expansion approximation. The presentation will also give the introduction of the atomic bond-order correlation term into the atomic interaction potential and the application of systematic complete and efficient electron density description basis set to study the thermal and mechanical properties of uranium and uranium niobium alloys.

Keywords: embedded atom method；atomic interaction potential；machine learning potential；molecular dynamics simulation；

* Corresponding author: wslai@tsinghua.edu.cn