High-throughput hybrid-functional DFT calculations of bandgaps and formation energies and multifidelity learning with uncertainty quantification

The authors computed bandgaps and formation energy values of more than 1100 crystalline materials‬ using Density Functional Theory (DFT) with HSE‬ and PBE approximations of the pseudopotentials. They analyzed accuracies of HSE and PBE approximations among different classes of materials. They also built a multi-fidelity machine learning model to predict the bandgap at HSE accuracy when a material's PBE bandgap‬ is known. The new high-throughput DFT (HSE, PBE) data of more than 1100 materials and the predicted‬ HSE bandgap data of more than 21,000 materials are available publicly via a dedicated web app.