Rational design of dual-atom δ5-borophene catalysts for nitrogen reduction reaction via density functional theory and machine learning

Electrocatalytic nitrogen reduction reaction (NRR) for synthesizing ammonia (NH3) is a promising strategy for sustainable NH3 production. Identifying appropriate NRR electrocatalysts is crucial for enhancing the efficiency and selectivity. Here, we employed density functional theory (DFT) and machine learning (ML) to investigate the performance of dual-atom catalysts (DACs) supported on δ5-borophene (δ5-BP) for NRR. Through four-step screening process and full reaction pathway calculations, we evaluated 23 homonuclear catalysts (MM@δ5-BP) and 21 stable heteronuclear candidates (MM∗@δ5-BP). The TiTi@δ5-BP and ZrZr@δ5-BP exhibited superior NRR catalytic activity with a limiting potential (UL) of −0.60 V among MM@δ5-BP, while NbHf@δ5-BP showed the lowest UL (−0.48 V) in MM∗@δ5-BP. We also investigated the influence of applied potential on the NRR through grand canonical DFT calculations. By constructing feature dataset and applying XGBoost Regressor (XGBR) and Gradient Boosting Regressor (GBR) algorithms with SHAP analysis, we achieved good agreement between ML-predicted UL and DFT-calculated UL. Our finding highlights the bonding interaction between two nitrogen atoms during N2 adsorption as the most critical feature. This work integrates DFT and ML approach to gain deep insights into complex dual-site activation and NRR mechanisms and to pave the way for accelerating the rational design of efficient DACs.