Machine Learning Driven High-Throughput Screening of Asymmetric Dinuclear Cobalt for Nitrate-to-Ammonia Reduction with Near-100% Selectivity

The electrocatalytic nitrate reduction reaction (NO₃RR) offers a promising way to reduce pollutants and synthesis ammonia. However, it is fraught with problems such as unbalanced adsorption–desorption of intermediates, insufficient H* supply, and competing hydrogen evolution. Dual-atom catalysts (DACs) have emerged as promising candidates, but their rational design faces obstacles in balancing element selection and performance. To address this, we employed high-throughput calculations and machine learning to screen a series of DACs for NO₃RR. (Formula presented.), d_m1-dm2 and (Formula presented.) were identified as key features. Based on interpretable SHAP analysis, we predicted and developed a sulphur-doped, asymmetric, dinuclear cobalt catalyst (Co₂/NSC). Experiments show that this catalyst achieves a Faradaic efficiency of 99.95% for NH₃ production at −0.2 V vs. RHE, with a maximum yield of 101.19 mg h⁻¹ mgcat⁻¹ at −0.4 V vs. RHE. The underlying mechanism and DFT calculations indicate that dinuclear Co sites enhance *NO₃⁻ adsorption and reduce the energy barrier of the rate-determining step *NO→*NHO; sulphur doping facilitates active *H supply and stabilizes the *NHO intermediate via an S─H/O─H hydrogen bonding network, suppressing side reactions. In addition, the integrated membrane electrode assembly system and the technical-economic analysis confirm the practical applicability and economic viability of this process.