(660b) High-Dimensional Neural Network Potentials for Molecular Dynamics Simulations of Crystal Thermodynamics and Phase Transition

A high-dimensional neural network potential was developed for molecular dynamics simulations of condensed phase Ni and its phase transition. The reference data set was generated by exploring a wide range of temperature-pressure space around atmospheric pressure using ab initio MD simulations. A unified neural network potential was obtained, producing reliable performance on both solid and liquid phases of nickel crystals. Excellent agreements were achieved from molecular dynamics simulations on thermal expansion of the face-centered nickel bulk. The same potential yields accurate molten structures characterized by radial distribution functions and diffusivities for the liquid state. Simulations of the phase transition between liquid and solid phases were performed using the two-phase interface method. The predicted melting point temperature is within a few kelvins of the literature value. In this talk, the general methodology will be discussed for the development of neural network potentials that can be applied to describe crystals with much more complex phase behaviors.