Revealing the oxygen Reduction/Evolution reaction activity origin of Carbon-Nitride-Related Single-Atom catalysts: Quantum chemistry in artificial intelligence

The oxygen reduction reaction (ORR) and oxygen evolution reaction (OER) are two critical reactions for renewable energy applications, such as water electrolysers and fuel cells. During the last decade, single-atom catalysts (SACs) deposited on carbon nitrides have been a rising star as superior electrocatalysts for ORR and OER. However, either experiments or theoretical simulations cannot screen all the possible SACs at a high speed and low cost. Herein, with the aid of density functional theory (DFT), machine learning (ML) and cross validation scheme, the best performing ML models (root mean square error = 0.24 V/0.23 V for ORR/OER) are established and implemented to describe the underlying pattern of easily obtainable physical and chemical properties and the ORR/OER overpotentials of carbon-nitride-related SACs. The best SACs recommended by the ML models are further verified by DFT calculations to confirm the reliability and accuracy of models. Three promising oxygen electrocatalysts with higher activity than noble metals are identified including RhPc, Co-N-C, and Rh-C4N3. The electron number of d orbital of the metal active site is determined as the most effective descriptor by further model analysis. Finally, the universal mathematical expressions which can accurately predict the catalytic activity of carbon-nitride-related SACs without DFT calculations and ML process are obtained. The revolutionary DFT-ML hybrid scheme opens a new avenue of rational and low-cost design principles of desirable catalysts and even the exploration of recondite activity origin in an interdisciplinary view.