In recent years, there has been significant progress in predicting atomic interactions using machine learning. However, machine learning models are difficult to extrapolate, and there have been issues with predicting areas near defects, where structures are highly disordered. This research has built a machine learning potential that can reproduce atomic interactions near dislocation cores with high precision by learning the dislocation defect structure of barium titanate (BaTiO3), a lead-free ferroelectric material, through active learning. The molecular dynamics simulations on BaTiO3 material using this potential have showed for the first time that the asymmetry of the dislocation core structure causes asymmetry in the formation of reversed polarization domains from the dislocation core.

Website containing the published article：physica status solidi rapid research letter
Related website：Hot Topic: Artificial Intelligence and Machine Learning

Ogata Lab.