Screening Oxygen-Adsorbed Single-Atom Alloys for Methane Initial Activation Using Machine Learning and Density Functional Theory Methods

C─H bond activation of methane is explored on single-atom alloys using density functional theory (DFT) and machine learning (ML). The DFT results indicate that the linear relationship between the activation energy and the reaction energy of methane oxygen-assisted dissociation is rough with RMSE = 0.25 eV. To predict the activation energy accurately and rapidly, gradient boosting decision tree (GBDT) algorithm is employed to build the ML model with six energetic, electronic, and structural descriptors. And the reaction energy DE is the most vital factor for prediction according to the importance analysis. This model is further applied to predict the activation energy of methane oxygen-assisted dissociation on 7625 local configurations from 2511 oxygen-adsorbed alloy surfaces. Through further screening based on structural rationality and reaction energy criteria, 89 configurations are remained. By searching the transition states, 14 configurations are identified with the activation energy lower than 0.5 eV. Moreover, comparing barriers of direct dehydrogenation and oxidative dehydrogenation pathways, chemisorbed oxygen atoms have promotion effects of methane activation over most of the 14 configurations. Finally, by AIMD simulations, 7 oxygen-adsorbed single-atom alloy surfaces including 12 local configurations are left as potential catalysts for methane initial oxidation due to structural stability.