Self-promoted ammonia selectivity for the electro-reduction of nitrogen on gt-C3N4 supported single metal catalysts: the machine learning model and physical insights

Abstract

Selectivity toward ammonia is an important indicator of a good electrocatalyst for the electrochemical nitrogen reduction reaction (eNRR). The multi-adsorption of N₂ on TM/gt-C₃N₄ greatly decreases the possibility of H binding, thus, self-promoting the selectivity toward NRR. Furthermore, the amount of nitrogen that can be trapped on the active sites of the studied catalysts is determined by the numbers of unoccupied d-orbitals of the supported single metal atom. The NRR selectivity on TM/gt-C₃N₄ (TM = V, Cr, Mn, Mo, Tc, W, and Re) is predicted to be 100% while three N₂ were adsorbed on TM (3N₂@TM/gt-C₃N₄). Furthermore, 3N₂@TM/gt-C₃N₄ is the dominant configuration under a high pressure region at room temperature. Multiple dinitrogen molecules can be stably adsorbed on the active site, which is a good indicator of thermal stability by AIMD simulation in the canonical ensemble. Machine-learning analysis indicates that the high selectivity toward ammonia is determined by the numbers of effectively bound N₂ molecules, and the low limiting-potential may correlate with the charging states of the supported metal atom, adsorption energy, and N–N bond length of the adsorbed N₂ molecule. W/N₃–G (W atom supported on three-pyrimidine-nitrogen-doped graphene) is predicted as a potential single atom catalyst with a low limiting-potential of −0.44 V and high selectivity based on the machine learning model, which is verified by further DFT calculations. This suggests a good generalization capability of the machine learning model.