Exchange-correlation functionals for band gaps of solids: benchmark, reparametrization and machine learning

• DOI: 10.1038/s41524-020-00360-0
• Publication Date: 2020-07-10T10:04:06Z
• Authors: Pedro Borlido; Jonathan Schmidt; Ahmad W. Huran; Fabien Tran; Miguel A. L. Marques; Silvana Botti
• Keywords: Benchmark (surveying)

Abstract We conducted a large-scale density-functional theory study on the influence of exchange-correlation functional in calculation electronic band gaps solids. First, we use large materials data set that have recently proposed to benchmark 21 different functionals, with particular focus approximations meta-generalized-gradient family. Combining these results for 12 functionals our previous work, can analyze detail characteristics each approximation and identify its strong and/or weak points. Beside confirming mBJ, HLE16 HSE06 are most accurate gap calculations, reveal several other interesting chief among which local Slater potential approximation, GGA AK13LDA, meta-GGAs HLE17 TASK. also compare computational efficiency approximations. Relying data, investigate improvement promising subset by varying their internal parameters. The identified optimal parameters yield family fitted gaps. Finally, demonstrate how train machine learning models prediction, using as input structural composition well approximate obtained from theory.