High-Throughput Screening of MN4 Coordinated Macrocyclic Frameworks for CO Reduction by Combining First-Principles Calculations and Machine Learning

Abstract

The electrochemical CO reduction reaction (CORR) by single-atom catalysts (SACs) provides a promising route for converting CO into value-added C₁ products, but the rational discovery of efficient SACs remains challenging because of the large configurational space and complex reaction network. Herein, we chose two-dimensional periodic macrocyclic frameworks with MN₄ coordination centers as the SACs and established a multilevel screening strategy by combining first-principles calculations and machine learning (ML). After evaluating thermodynamic stability, dissolution resistance, *CO adsorption, and hydrogen evolution reaction suppression, 90 candidates were screened out from 156 SACs for further analysis. The hydrogenation of *CO to *CHO was identified as a key step for deep CO reduction, and ΔG_(*CHO-*CO) was selected as the main descriptor. An ExtraTrees regression model trained on the screened SACs achieved high predictive accuracy, with R₂ values of 0.947 and 0.925 for the training and test sets, respectively. By scanning the full reaction pathway of CORR (CH₄ as the final product) on the SACs with relatively low ΔG_(*CHO-*CO) value, the FeC₈N₈ was found to exhibit the highest catalytic activity among the examined candidates, with a low limiting potential (U_L) of −0.30 V. To accelerate catalyst evaluation, ΔG_(*CHO-*CO), which describes the first hydrogenation of adsorbed *CO, was selected as the key descriptor and predicted by ML algorithm. Full-pathway calculations were subsequently carried out for catalysts with relatively low ΔG_(*CHO-*CO) values to validate the descriptor-guided screening results.