High-throughput screening of SrxA1−xFeyB1−yO3 perovskites for low-temperature chemical looping air separation using graph neural networks
Low-temperature chemical looping air separation (CLAS) is a promising technology for producing oxygen-enriched gas streams, utilizing the redox properties of solid oxygen carriers to selectively capture and release oxygen from air. Oxygen vacancy formation energy (Eovf) is a key descriptor for evaluating the ease of oxygen release. In this study, the applicable range of Eovf for CLAS oxygen carriers was determined to be <2.3 eV via thermodynamic calculations. A graph neural network (GNN) model, specifically the ALIGNN architecture, was trained to predict Eovf with a mean absolute error (MAE) of 0.26 eV on the test set. Using this model, a high-throughput screening of 3,649 compositions of SrxA1−xFeyB1−yO3 perovskites was conducted to identify promising CLAS oxygen carriers. The predictions revealed that doping with Ba and Ca at the A-site and Co at the B-site effectively reduces Eovf. The screening criterion of Eovf < 2.3 eV successfully rediscovered several previously reported low-temperature CLAS oxygen carriers, validating the approach. This work demonstrates that GNN-based Eovf prediction can significantly accelerate the discovery of CLAS materials, with broader implications for other chemical looping applications such as full oxidation and syngas production.