Machine-Learning-Assisted Acceleration on High-Symmetry Materials Search: Space Group Predictions from Band Structures

Efficiency of search of wanted materials with desired properties is limited by the huge search space. By deep learning methods, we demonstrate that space group information can be acquired from band structure inputs to reduce the search space. Despite atomic orbital or accidental degeneracies mixed with lattice degeneracies, band degeneracies as input can yield 96.0% prediction accuracy for cubic systems that leads to a 25.1-fold acceleration of searching speed overall. Additionally, for all space groups, the prediction accuracy is 82.0% with overall 36.9-fold acceleration in the search speed. In addition, valence band degeneracies as inputs can yield satisfactory results and may assist in structural analysis from ARPES results.