Machine learning-assisted improving of C3N2 configurations: transition-metal doped for enhanced bifunctional oxygen electrocatalysis

• The bifunctional catalytic activity of SACs and DACs is explored by DFT and ML. • Pd-PC 3 N 2 exhibits the best bifunctional catalytic activity (BI = 0.66 V). • The adsorption capacity of DACs is stronger than that of SACs. • RFR machine learning model effectively predicts catalytic performance. To tackle the kinetic constraint of cathode oxygen reactions in energy conversion devices, this study designs high-performance bifunctional ORR/OER electrocatalysts via doping 3d and 4d transition metals into two C 3 N 2 configurations (PC 3 N 2 , IC 3 N 2 ). Single-atom doped IC 3 N 2 displays enhanced stability. For ORR, the ORR overpotential ( η ) values of Cu-PC 3 N 2 , Pd-PC 3 N 2 , and Pd-IC 3 N 2 are 0.41, 0.45, and 0.31 V, respectively, indicating that they possess good ORR catalytic activity. For OER, the η OER values of Pd-PC 3 N 2 and Pd-IC 3 N 2 are 0.21 and 0.53 V respectively, indicating that they possess good OER catalytic potential. For bifunctional catalytic activity, Pd-PC 3 N 2 exhibits the best bifunctional catalytic activity, with a bifunctional index value of only 0.66 V. Moreover, the Random Forest Regression model effectively predicts catalytic performance, achieving an R 2 of up to 0.906 and an RMSE of 0.213. This work provides guidance for designing efficient electrocatalysts and highlights machine learning’s utility in catalysis research.