What question did this study set out to answer?

The research aims to identify effective dual metal-site catalysts for CO2 reduction using machine learning and electronic descriptors.

March 21, 2026

Integrating Machine Learning and DFT Descriptors for Screening Dual Metal‐Site Catalysts for CO 2 Reduction to C 2 Products

Key Points

The research aims to identify effective dual metal-site catalysts for CO2 reduction using machine learning and electronic descriptors.
Screened 156 dual metal-site catalysts on nitrogen-doped carbon for stability and hydrogen adsorption.
Calculated CO dimerization energies to assess activity for C2 product formation.
Performed linear correlation analysis of CO dimerization energy against 71 electronic parameters.
Trained multiple machine learning models, focusing on the random forest regressor for predictions.
Identified 33 DMSCs with favorable CO dimerization energies below 0.75 eV.
O-dz2↓ orbital occupancy showed the highest correlation with CO dimerization energy (R2 = 0.69).
Random forest regressor achieved high predictive performance (R2 = 0.98 for training, 0.96 for testing).
O-dz2↓ occupancy also correlated strongly with overpotential for C2 production (R2 = 0.87).

Abstract

The electrochemical reduction of CO2 (CO2RR) offers a sustainable route to generate multi-carbon products (C2), but achieving high activity and selectivity remains a challenge. Dual metal-site catalysts (DMSCs), composed of two adjacent metal sites, provide unique active centers that enable simultaneous CO adsorption, a key step for C─C coupling. Here, we systematically investigated 156 DMSCs supported on nitrogen-doped carbon to identify promising candidates for CO2RR. Stability screening revealed that 8 DMSCs are unstable, while hydrogen adsorption calculations excluded 6 additional systems due to strong *H binding. Among the remaining catalysts, 33 DMSCs exhibit CO dimerization energies (∆G*CO dimer-2*CO) below 0.75 eV, indicating favorable activity toward C2 products. To explain these trends, we performed a linear correlation analysis of CO dimerization energy against 71 electronic parameters, which revealed that the occupancy of the dz2↓ orbital (denoted O-dz2↓) exhibits the highest correlation (R2 = 0.69). This suggested that combinations of electronic parameters could further improve the correlation and accurately predict CO2RR toward C2 products. To achieve this, multiple machine learning models were trained, with the random forest regressor (RFR) achieving superior performance (R2 = 0.98 for training and 0.96 for testing), demonstrating its suitability for predicting CO dimerization energy. Furthermore, the overpotential for C2 production of the 33 DMSCs was correlated with electronic descriptors, revealing that O-dz2↓ exhibited the highest correlation (R2 = 0.87), attributable to the substantial population of dz2↓ states near the Fermi level (EF), thereby underscoring its significance as a key descriptor. Overall, we emphasize the importance of using a multi-descriptor predictive model to accurately estimate CO dimerization energies, and we identify key electronic parameters of DMSCs that can predict the overpotential for C2 products. These insights offer a valuable framework for the rapid screening of low-cost materials with high selectivity toward C2 products.

Bookmark

Cite This Study

Sk et al. (Thu,) studied this question.

synapsesocial.com/papers/69be35836e48c4981c673d60 https://doi.org/https://doi.org/10.1002/smll.202513368

Bookmark