Synergizing Machine Learning with High‐Throughput DFT to Design Efficient Single‐Atom Catalysts for Hydrogen Evolution Reaction

The development of efficient single-atom catalysts (SACs) for electrocatalytic hydrogen evolution (HER) has garnered significant attention within the scientific community. However, the extensive scope of material experimentation, coupled with high research and development costs and prolonged research cycles, severely hampers the efficient advancement of related materials. In this study, the HER activity of 90 types of SACs is systematically investigated, which consist of single transition metal (TM) and/or nonmetal (NM) atoms bonded in graphyne (TM-NM-GY), by synergizing machine learning algorithms with high-throughput DFT computations. The findings reveal that the HER catalytic activity of Fe-GY, Fe-B-GY, Ni-B-GY, Pd-B-GY, Sc-N-GY, Co-N-GY, Y-N-GY, and Pd-N-GY surpasses that of commercial Pt/C catalysts. Moreover, non-metallic B or N atom doping can effectively modulate the HER performance of SACs. Furthermore, it is confirmed that HER activity correlates with characteristic factors such as the bond length of the coordinating atoms, d-band center, metal binding height, charge transfer, and ICOHP. Finally, machine learning stacking models have proven efficient in predicting and designing superior HER SACs. It is anticipated that these insights will accelerate the prediction and design of corresponding SACs.