A general machine-learning framework for high-throughput screening for stable and efficient RuO2-based acidic oxygen evolution reaction catalysts

Abstract Doping guest elements is an effective way to increase activity and stability of RuO2 catalysts in acidic oxygen evolution reaction (OER). However, due to the vastness of doping space, it is challenging for either high-cost experiments or density functional theory (DFT) calculations to screen out the doped structures with the optimized catalytic performance. Herein, we reported a machine-learning (ML) framework that aims to realize high-throughput screening for both stability and activity of doped-RuO2 acidic OER catalysts from mono-doping to triple-doping at a low level of computational cost. Compared to the d-band theory and some other previous models, our ML model was constructed based on more general input features and realized higher prediction accuracy with mean absolute errors (MAEs) of 0.074, 0.142 and 0.082 eV for *OH, *O and *OOH adsorption, respectively. Through the ML models, three doping structures, Ru41Zn7O96, Ru41Zn4Fe3O96, and Ru39Zn4Cu4Co1O96 were found to possess the extraordinarily high stability and comparable or higher activity than the previously reported OER catalysts. This work provided an efficient study paradigm in fields of material screening and a useful guide for experimental synthesis.