Band gaps corrected by machine learning

Exchange-correlation functionals for band gaps of solids: Benchmark, reparametrization and machine learning
P. Borlido, J. Schmidt, A.W. Huran, F. Tran, M.A.L. Marques, and S. Botti
NPJ Comput. Mater. 6, 96 (2020)

This machine learning model uses Model Agnostic Supervised Local Explanations (MAPLE) based on gradient boosting trees to correct bandgaps calculated with the modified Becke-Johnson (mBJ) functional to be closer to experimental results. The model will output a prediction for the corrected bandgap and information about the most important training examples, that led the model to the correction. This information can be used to make an informed decision on whether to trust the model. The general expectation is that the materials with high weight will have a similar mBJ-bandgap or a similar chemical composition. For use of different models in the paper or for direct access to the model for high-throughput searches please contact us.