[19a-B01-1~10] 23.1 Joint Session N "Informatics"

[19a-B01-3] Fast Prediction of Crystal Structure Stability by Machine Learning

〇(M1)Yuki Inada¹, Yukari Katsura^1,2,3, Masaya Kumagai^3,4, Kaoru Kimura¹ (1.Univ. of Tokyo, 2.NIMS, 3.RIKEN, 4.Sakura Internet Inc.)

Keywords:machine learning, crystal structure stability

We developed a high-speed crystal stability predictor by machine learning. Original element descriptors were automatically generated from the statistic occurrences of coordination polyhedra from over 30,000 crystal structures. We employed a set of two neural networks for the predictions of formation energies. Our machine learning model succeeded to predict energy above the convex hull only from the atomic coordinates, with mean absolute error 0.08 eV/atom.