Application of machine learning-based selective sampling to determine BaZrO3 grain boundary structures

Abstract

A selective sampling procedure is applied to reduce the number of density functional theory calculations needed to find energetically favorable grain boundary structures. The procedure is based on a machine learning algorithm involving a Gaussian process, and uses statistical modelling to map the energies of the all grain boundaries. Using the procedure, energetically favorable grain boundaries in BaZrO3 are identified with up to 85% lower computational cost than the brute force alternative of calculating all possible structures. Furthermore, our results suggest that using a grid size of 0.3 Å in each dimension is sufficient when creating grain boundary structures using such sampling procedures.