Tuning the MN4 plane in π-conjugated metal-phthalocyanine networks toward efficient CO2 electroreduction: a DFT-ML study

Abstract

Efficient CO₂ electrocatalysts are essential for achieving carbon neutrality, and there is an urgent need to move beyond the current range of metal phthalocyanines (MPcs) by developing next-generation, highly active, and cost-effective electrocatalysts. Herein, a combined density functional theory (DFT) and machine learning (ML) strategy was employed to systematically screen a library of M–N₂X₂ phthalocyanines (M = Sc–Zn; X = B/C/N/O/F) as single-atom catalysts for CO₂RR to CO and HCOOH. DFT calculations reveal that B, C, and O heteroatoms effectively modulate the electronic structure of MN₂X₂-Pc frameworks, yielding highly active and selective CO₂RR catalysts with low limiting potentials (U_L<0.76 V), outperforming noble-metal benchmarks (e.g., Au). Notably, CuN₂C₂-Pc and FeN₂O₂-Pc achieve record-low U_L values of 0.29 V (CO) and 0.28 V (HCOOH), respectively. By employing a gradient boosting regression (GBR) model, a highly accurate predictive model for U_L was established, achieving R² = 0.96 for U_{L_CO} with only three features and R² = 0.90 for U_{L_HCOOH} with seven features, underscoring the efficiency of feature selection and utilization. This integrated framework enables rapid catalyst screening and provides mechanistic insights, thereby accelerating the rational design of CO₂RR electrocatalysts.