The development of neural network potentials for cubic sodium chloride crystals with vacancies

Abstract

We implement a method for computing the interatomic potentials by training and ftting neural networks with the data obtain from molecular dynamics simulations. We construct a longrange neural network potential and apply it to NaCl system including ideal and defected systems. We keep the short-range part with Behler type and the long-range part is calculated using the Ewald summation. The reason for choosing the Ewald sum is because it provides high accuracy and feasible computational speed when estimating the long-range potential, due to the rapid convergence between long-range contribution in reciprocal space and the shortrange contribution in real space. In this work, we not only compare the result to molecular dynamics simulations but also adopt them into the training set of neural network.