Constructing dual electron transfer channels to accelerate CO2 photoreduction guided by machine learning and first-principles calculation

Abstract

Designing dual electron transfer channels to achieve efficient carrier separation and understanding the corresponding mechanisms for CO2 photoreduction is of great significance. However, it is still challenging to find desirable model to achieve optimal photocatalytic performance. Herein, first-principles calculations and machine learning were combined to predict an optimized microstructure with dual electron transfer channels. The results indicate that the construction of BiOBr-Bi-g-C3N4 heterojunction has optimal free energy (|ΔG|) for H2O dissociation and CO2 reduction. Besides, the double electron transfer channels and excellent Bi active site can localize the photoexcited carriers at the interlayers rather than randomly distributing. These localized carriers generate intriguing superposition states at a particular timescale that optimize the multi-electronic reaction kinetics pathway of CO2 reduction, resulting in a 4.7 and 3.1 fold increase compared to pristine Bi-BiOBr and Bi-g-C3N4 with single electron transfer pathway. Machine learning was further used to optimize the experimental parameters, and the photocatalytic mechanism was verified by combining first-principles calculation with comprehensive experimental characterizations. This work provides experimental and theoretical references for the accurate prediction, rational design and ingenious fabrication of high-performance photocatalytic materials.