Machine learning-driven shortening the screening process towards high-performance nitrogen reduction reaction electrocatalysts with four-step screening strategy

The urgent need to prepare clean energy by environmentally friendly and efficient methods, which has led to widespread attention on electrocatalytic nitrogen reduction reaction (NRR) for ammonia production. At present, single atom catalytic nitrogen reduction has become the earliest promising method for industrial production due to its high atomic utilization rate, high selectivity, high controllability, and high stability. However, how to quickly screen catalysts with high catalytic efficiency and selectivity in single-atom catalysts (SACs) remains a challenge. Herein, the 29 SACs are constructed from C6N2 nanosheets doped with transition metals (TM@C6N2), which are analyzed for stability, adsorption performance, NRR catalytic activity, electronic properties, and competitiveness using first-principles calculations. The results show that Mo@C6N2 and Re@C6N2 exhibit the most outstanding catalytic performances, with limiting potentials (UL) of −0.29 and −0.31 V, respectively, in the solvent model. Machine learning is used to derive descriptors from the intrinsic features to predict the free energy changes for the potential-determining step. The importance of features is calculated, with the first ionisation energy (IE1) being the most significant influencing factor. Based on the guidance of machine learning and considering that IE1 is related to the ability of metal atoms to donate electrons, a four-step screening strategy using the Integrated Crystal Orbital Hamilton Populations (ICOHP) to screen catalysts instead of the traditional five-step screening not only improves the screening efficiency but also obtains completely consistent screening results. This work presents a new approach to predicting the catalytic performance of SACs and provides new insights into the influence of intrinsic properties on catalytic activity.