Exploring the Latent Chemical Space of Oxygen Vacancy Formation Energy by a Machine Learning Ensemble

Oxygen vacancy (V_O) is a fundamental and intrinsic defect in metal oxides since its formation is subject to various growth and annealing processes and it significantly affects the material properties. Therefore, the formation energy of an oxygen vacancy is of great interest in the fabrication and investigation of metal oxides. Traditional methods for obtaining the formation energy of an oxygen vacancy such as theoretical calculation and experiment require expensive costs, making it difficult to explore a large number of metal oxides. Here, machine learning (ML) models are built for rapid prediction of the formation energy of an oxygen vacancy. We show that an ensemble of multiple ML models allows the prediction of the formation energy of an oxygen vacancy in a wide regime of an unexplored latent chemical space from binary to quinary oxides.