Machine-Learning-Accelerated Catalytic Activity Predictions of Transition Metal Phthalocyanine Dual-Metal-Site Catalysts for CO2 Reduction.

ABSTRACT

The highly active and selective carbon dioxide reduction reaction (CO2RR) can generate valuable products such as fuels and chemicals and reduce the emission of greenhouse gases. Single-atom catalysts (SACs) and dual-metal-sites catalysts (DMSCs) with high activity and selectivity are superior electrocatalysts for the CO2RR as they have higher active site utilization and lower cost than traditional noble metals. Herein, we explore a rational and creative density-functional-theory-based, machine-learning-accelerated (DFT-ML) method to investigate the CO2RR catalytic activity of hundreds of transition metal phthalocyanine (Pc) DMSCs. The gradient boosting regression (GBR) algorithm is verified to be the most desirable ML model and is used to construct catalytic activity prediction, with a root-mean-square error of only 0.08 eV. The results of ML prediction demonstrate Ag-MoPc as a promising CO2RR electrocatalyst with the limiting potential of only -0.33 V. The DFT-ML hybrid scheme accelerates the efficiency 6.87 times, while the prediction error is only 0.02 V, and it sheds light on the path to accelerate the rational design of efficient catalysts for energy conversion and conservation.