Accelerating design of phthalocyanine-based catalysts for electrocatalytic reduction of nitric oxide: A DFT and machine learning study

Nitrogen oxide reduction reaction (NORR) is a highly efficient method for reducing NO to synthesize ammonia. Herein, we designed 18 transition-metal (TM)-doped phthalocyanine materials and combined density functional theory (DFT) with machine learning (ML) to systematically study their catalytic performance for NORR through comprehensive evaluations of their NO adsorption energy, electronic structure, and Gibbs free energy. Our results indicate that NO can adsorb on V1@Pc with a moderate adsorption energy (−2.29 eV) and the lowest limiting potential (0.38 V) among the 18 TM1@Pc materials. Additionally, V1@Pc can effectively mitigate the competitive hydrogen evolution reaction (HER), showing that it has a remarkable performance of efficiency and selectivity for NORR. Finally, machine learning techniques were employed to construct predictive models by utilizing key feature descriptors derived from DFT calculations and intrinsic atomic properties. Among the five regression models, the Random Forest Regression (RFR) exhibits the highest accuracy (R² = 0.99) and the lowest RMSE (0.15 eV) in predicting Gibbs free energy. Our study may provide valuable theoretical insights for the rational design of high-performance electrocatalysts for NORR.