Unveiling Methane Selectivity on Cu-Based Single-Atom Alloys for CO2 Electroreduction via Synergistic DFT and Machine Learning.

Electrochemical CO2 reduction offers a sustainable route for converting CO2 into value-added fuels and chemicals, yet achieving efficient methane production remains challenging because of the complex reaction network and insufficient catalyst selectivity. Herein, density functional theory combined with interpretable machine learning were employed to investigate the structure-activity relationship of Cu-based single-atom alloys constructed by incorporating 13 transition-metal atoms into Cu(100)/(111) surfaces. The calculations reveal that isolated metal atoms effectively regulate the electronic structures of both the dopant and neighboring Cu atoms, thereby altering the energetics of HCOO* formation. More importantly, the synergistic interaction between the isolated metal atom and adjacent Cu sites stabilizes the CH2OO*, redirecting the reaction pathway toward highly selective CH4, in sharp contrast to pristine Cu surfaces. Among all investigated catalysts, Ge1/Cu(100) is most active, with a limiting free energy as low as 0.24 eV while effectively suppressing the hydrogen evolution reaction. Machine learning analysis further identifies electron affinity and catalyst-support charge transfer as the two dominant descriptors governing catalytic activity, providing quantitative insight into the origin of the structure-activity relationship. This combined DFT-ML framework establishes descriptor-guided design principles for methane-selective CO2 reduction and offers a general strategy for the rational development of high-performance single-atom alloy electrocatalysts.