Abstract
Defect engineering of semiconductor photocatalysts is critical in reducing the reaction barriers. The generation of surface oxygen vacancies allows substantial tuning of the electronic structure of anatase titanium dioxide (TiO(2)), but disclosing the vacancy formation at the atomic level remains complex or time-consuming. Herein, we combine density functional theory calculations with machine learning to identify the main factors affecting the formation of oxygen defects and accelerate the prediction of vacancy formation. The results show that the first two-layer oxygen atoms on the typical surfaces of TiO(2,) including (100), (110), and (211) facets, are more likely to be activated when the gas is more reduced, the pressure is higher, and the reduction temperature is increased. Through machine learning, we can conveniently predict the formation of oxygen defects with high accuracy. Furthermore, we present an equation with acceptable accuracy for quantitatively describing the formation of oxygen vacancies in different chemical environments. Our work provides a fast and efficient strategy for characterizing the surface structure with atomic defects.