Presentation Information
[8p-P11-5]Prediction of Formation Energy of Oxygen Defects in SrTiO3 Using Machine-Learning Molecular Dynamics Simulations
〇Kazutaka Nishiguchi1,2, Ryota Yamamoto2, Meguru Yamazaki1, Naoki Matsumura1, Yuta Yoshimoto1, Seiichiro Ten-no2, Yasufumi Sakai1 (1.Fujitsu Ltd., 2.Kobe Univ.)
Keywords:
neural network potential,formation energy,oxygen defect
The formation energy of oxygen defects in SrTiO3 was predicted by NNP-MD simulations using GeNNIP4MD, an NNP construction framework using active learning. We constructed NNP models using different data sets (Dataset1,2 3) for accuracy verification. The formation energy of oxygen defects predicted by NNP-MD reproduced the results of DFT calculations with high accuracy when the data set of oxygen defects was considered. In particular, NNP-MD reproduces well the extrapolated value of DFT even in the supercell size (5×5×5), which is difficult to calculate DFT.