Insights into oxygen vacancies from high-throughput first-principles calculations

Oxygen vacancies play significant roles in various properties of oxide materials. Therefore, insights into the oxygen vacancies can facilitate the discovery of better oxide materials. To achieve this, we developed codes for high-throughput point-defect calculations and applied them to characterize oxygen vacancies in 937 oxides. From the resulting large dataset, we analyzed the vacancy structures and formation energies and constructed machine-learning regression models to predict vacancy formation energies. We have found that the vacancy formation energies are predicted using the random forest regression models with accuracies of 0.27--0.44 eV depending on the charge states. Analyses of the importance of the descriptors show that the formation energies of the neutral vacancies are mainly determined by the orbital characteristics of the conduction-band minima, the oxide stability, and the band gaps, whereas those of the doubly charged defects are determined by factors related to electrostatic energy. These codes and datasets are publicly available, and a graphical user interface is available to analyze the calculation results.