Integrating Density Functional Theory with Machine Learning for Enhanced Band Gap Prediction in Metal Oxides

Abstract

In this study, we used a combination of density functional theory with Hubbard U correction (DFT+U) and machine learning (ML) to accurately predict the band gaps and lattice parameters of metal oxides: TiO2 (rutile and anatase), cubic ZnO, cubic ZnO2, cubic CeO2, and cubic ZrO2. Our results show that including Up values for oxygen 2p orbitals alongside Ud/f for metal 3d or 4f orbitals significantly improves the band gap and lattice parameters predictions. Through extensive DFT+U calculations, we identified optimal (Up, Ud/f) integer pairs that closely matched the experimental values for band gaps and lattice parameters: (8 eV, 8 eV) for rutile TiO2, (3 eV, 6 eV) for anatase TiO2, (6 eV, 12 eV) for c-ZnO, (10 eV, 10 eV) for c-ZnO2, (9 eV, 5 eV) for c-ZrO2 and (7 eV, 12 eV) for c-CeO2. Our ML analysis demonstrated that simple supervised ML models can reliably achieve accuracy comparable to DFT+U calculations. These models have the potential to extend beyond the metal oxides used in training and to explore the effects and dependencies of U values on the bulk properties of materials. Our study not only identifies the best U pairs for predicting experimentally measured band gaps and lattice parameters but also highlights the effectiveness of straightforward regression ML models in predicting the band gaps and lattice parameters of metal oxides.