Publication date: 11th October 2022
Closing the carbon cycle by converting CO2 to long-chain hydrocarbons using renewable electricity is a beneficial process from both an environmental and economic perspective. Cu based catalysts exhibited unique performance for C-C bond formation in electrochemical carbon dioxide reduction reaction (eCO2RR), although with low product selectivity for C2+ products. Recently, oxide-derived Cu (OD-Cu) catalysts have shown the potential to produce multicarbon species with low overpotentials.[1] These OD-Cu catalysts are usually formed when Cu2O loses oxygen under reduction potentials. It is believed that its unique reconstructed surface is responsible for its superior performance. However, the structure of OD-Cu remains controversial from both experimental and computational point of view. In the experiments during characterization, the OD-Cu materials were easily re-oxidized, which led to difficulties in determining the ratio of oxygen to copper under reaction conditions. On the other hand, the computational studies with different oxygen ratio gave different surface structure and different proportion of active sites.[2-3] Therefore, understanding the structure of OD-Cu and the distribution of active sites is fundamental to find promising catalysts for generating long-chain hydrocarbons selectively.
In this study, large-scale Molecular Dynamics simulations based on Machine Learning Potentials were performed to understand the structure of the OD-Cu materials. We first constructed a Machine Learning Potential for OD-Cu materials through an active learning approach. The potential was constructed stepwise, from pure Cu, Cu2O to CuxO. After screening of 8.16×1010 structures, a database with 59491 structures was built. The Root-mean-square error (RMSE) of Energy is 4.6 meV/atom, and the RMSE of Force is 63.4 meV/Å. The potential showed good performance when applied to copper and copper oxide to various degrees of OD-Cu. The structures of the OD-Cu materials were obtained through Molecular Dynamics simulations on large spatial and temporal scales, and the most common ensembles were identified via coordination environment and graph theory analysis. The results reveal the structure of OD-Cu and provide a solid foundation for further research on the formation mechanism of long-chain hydrocarbons.
This work was funded by the Spanish Ministry of Science and Innovation (Ref. No. RTI2018-101394-BI00), and the CIN/AEI/10.13039/501100011033 (CEX2019-000925-S). The authors also thank the Barcelona Supercomputing Center (BSC-RES) for providing computational resources. Zan would like to acknowledge Marie Skłodowska-Curie Postdoctoral Fellowships (ref. MSCA-PF-2021 101064867-DESCRIPTOR).