Machine Learning-Enhanced\nScreening of Single-Atom\nAlloy Clusters for Nitrogen Reduction Reaction

The electrochemical nitrogen reduction reaction (eNRR)\nunder ambient\nconditions is a promising method to generate ammonia (NH<sub>3</sub>), a crucial precursor for fertilizers and chemicals, without carbon\nemissions. Single-atom alloy catalysts (SAACs) have reinvigorated\ncatalytic processes due to their high activity, selectivity, and efficient\nuse of active atoms. Here, we employed density functional theory (DFT)\ncalculations integrated with machine learning (ML) to investigate\ndodecahedral nanocluster-based SAACs for analyzing structure–activity\nrelationships in eNRR. Over 300 nanocluster-based SAACs were screened\nwith all the transition metals as dopants to develop an ML model predicting\nstability and catalytic performance. Facet sites were identified as\noptimal doping positions, particularly with late group transition\nmetals showing superior stability and activity. Utilizing DFT+ML,\nwe identified 8 highly suitable SAACs for eNRR. Interestingly, the\nnumber of valence d-electrons in dopants proved crucial in screening\nfor eNRR activity. These catalysts exhibited low activity in hydrogen\nevolution reaction, further enhancing their suitability for eNRR.\nThis successful ML-driven approach accelerates catalyst design and\ndiscovery, holding significant practical implications.