[9a-W321-1~11] 23.1 Joint Session N "Informatics"

[9a-W321-2] Construction of Neural Network Potentials to Analyze Atom Dynamics under Electric Fields and Phonon Properties

Masayoshi Ogura¹, Takanori Moriya¹, Wenwen Li², Koji Shimizu¹, Emi Minamitani¹, 〇Satoshi Watanabe^1,3 (1.UTokyo, 2.AIST, 3.NIMS)

Keywords:neural network, phonon dispersion, Born effective charge

Interatomic potentials constructed with neural network (NN) and first-principles calculation data can achieve computational efficiency and reliability simultaneously and are powerful in computational analyses of problems related to atom dynamics. However, NN potentials so far developed are insufficient to analyze atom migration under electric fields and phonon properties. In this presentation, we report our studies which enables us to analyze these two issues. Specifically, we present our results on the development of NN potential that can predict the phonon dispersion of crystalline GaN accurately, and NN that can predict Born effective charges of atoms in amorphous Li₃PO₄.