Crystal Structure Prediction of Cs–Te with Supervised Machine Learning

Abstract

Crystal structure prediction methods aim to determine the ground-state crystal structure for a given material. The vast combinatorial space associated with this problem makes conventional methods computationally prohibitive for routine use. To overcome these limitations, a novel approach combining high-throughput density functional theory calculations with machine learning is proposed. It predicts stable crystal structures within binary and ternary systems by systematically evaluating various structural descriptors and machine learning algorithms. The superiority of models based on atomic coordination environments is shown, with transfer-learned graph neural networks emerging as a particularly promising technique. By validating the proposed method on Cs–Te crystals, its ability to generate stable crystal structures is proved, suggesting its potential for advancing established computational schemes.