Machine Learning-Guided Two-Dimensional Single-Atom Catalysts for Electrochemical CO2 Utilisation

Abstract

Electrochemical CO2 reduction (CO2RR) offers a promising route to convert greenhouse gas emissions into value-added fuels and chemicals yet achieving high product selectivity while suppressing the competing hydrogen evolution reaction (HER) remains a central challenge. Two-dimensional (2D) material-supported single-atom catalysts (SACs) address this challenge through atomically defined coordination environment that enable precise tuning of the critical *COOH and *CO adsorption energies, the two descriptors that simultaneously govern CO2 activation and HER suppression. This review examines the convergence of 2D-SAC design and machine learning (ML) for CO2RR catalyst discovery. We first outline the mechanistic foundations of CO₂RR together with comparative insights from HER and OER, then discuss the structural and electronic principles governing single-atom stabilisation on nitrogen-rich carbon, transition metal dichalcogenide, and MXene supports. Recent catalytic advances are assessed with CO2RR as the primary focus, including proximity-engineered systems that promote C-C coupling and multicarbon product formation. Finally, emerging ML frameworks are critically reviewed, covering electronic, geometric, and spectroscopic descriptor design, data source strategies, and high-throughput screening workflows. Operando ML integration has revealed the dynamic evolution of active sites under reaction conditions, fundamentally reframing how training datasets must be constructed. Key priorities identified for the field are large-scale CO2RR-specific SAC database construction, multi-target descriptor development for full product selectivity, and closure of the synthesis-prediction feedback loop.