Computational screening of M2N4-C-type dual-atom-catalysts for electrochemical ammonia synthesis by the first-principles DFT and machine learning

Abstract

Electrochemical ammonia synthesis is expected to complement the conventional Haber–Bosch method due to its low carbon emissions and stable operation under ambient conditions. However, due to the complexity of reaction pathways in the nitrogen reduction reaction (NRR), rapid identification and prediction of NRR electrocatalysts is computationally expensive and challenging. In this study, taking the graphene-based M₂N₄-C dual-atom-catalyst (DAC) family as an example, we investigated the NRR activity and mechanisms of 45 candidates with M from 3d transition metals. Six candidates were predicted to be promising NRR catalysts from DFT calculations. A universal descriptor Φ is trained from 4860 DFT-obtained data points to predict the NRR activity and path preference. The ML-trained descriptor Φ shows 84% probability in the correct qualitative prediction of NRR activity. Most importantly, the robustness and transferability of descriptor Φ are further confirmed in other M₂N₄-C DACs with M in 4d transition metals. This study shows a practical strategy for the fast computational screening of NRR catalysts based on DFT and an ML-trained universal descriptor, which could significantly benefit the development of electrochemical ammonia synthesis in industry.