DFT and Machine Learning Unravel the Mechanism of BC3-Supported Single Atom Catalysts for Efficient N2 Reduction and Hydrogen Evolution

Abstract The electrocatalytic nitrogen reduction reaction (NRR) is a valuable green route for ammonia synthesis, yet it is still constrained by the lack of efficient catalysts and the impact of the competing hydrogen evolution reaction (HER). Density functional theory (DFT) is applied to explore the NRR and HER behaviors of metal single atoms embedded in pyridinic-N4-modified divacancy-defected BC3 monolayer structures (M-SACs@BC3) and reveal the underlying modulation mechanisms. DFT results suggest that Mo-SACs@BC3 delivers favorable NRR activity with a low limiting potential (UL) of –0.46 V while inhibiting the HER. In comparison, Tc-SACs@BC3 shows considerable HER activity, with its hydrogen adsorption free energy (ΔGH*) approaching –0.10 eV. Electronic structure analysis indicates that the high activity originates from strong orbital coupling between metal d-orbitals and reaction intermediates, as well as efficient charge transfer between the substrate and the metal sites. Random forest (RF) machine learning (ML) models are constructed to predict adsorption free energies for key NRR intermediates and ΔGH* for HER. The model exhibits good prediction capability on an external DFT- and literature-based dataset for rapid data-driven catalyst screening. This work offers theoretical insights for the rational design of BC3-based single-atom catalysts for NRR and HER.