Evolutionary computing and machine learning for discovering of low-energy defect configurations

Abstract Density functional theory (DFT) has become a standard tool for the study of point defects in materials. However, finding most stable defective structures remains very challenging task as it involves solution multimodal optimization problem with high-dimensional objective function. Hitherto, approaches commonly used to tackle this have been mostly empirical, heuristic, and/or based on domain knowledge. In contribution, we describe an approach exploring potential energy surface (PES) covariance matrix adaptation evolution strategy (CMA-ES) and supervised unsupervised machine learning models. The resulting algorithm depends only limited set physically interpretable hyperparameters offers systematic way low-energy configurations isolated solids. We demonstrate its applicability different systems show ability find known discover additional ones well.