Computational advances in electrochemical CO2 reduction: design of copper-based and atomically dispersed catalysts, interfacial dynamics, and machine learning

Electrochemical CO2 reduction (eCO2R) offers a route to convert CO2 into fuels and chemicals using renewable electricity, but requires catalysts operating at low overpotential with high product selectivity. Computational studies have delivered essential mechanistic insight, yet a persistent gap remains between the idealised models used in most calculations and the structurally disordered, dynamically evolving catalysts encountered under operating conditions. This review surveys recent computational advances across three themes. The first addresses the methodological landscape, from fixed-electron density functional theory (DFT) and the computational hydrogen electrode (CHE) model through grand-canonical and constant-potential DFT, grand canonical Monte Carlo (GCMC), ab initio molecular dynamics (AIMD), machine-learning force fields (MLFFs), and data-driven machine-learning (ML) models. This establishes the capabilities and limitations of each. The second examines three catalyst classes: copper-based materials (high-index surfaces, amorphous systems, bimetallic alloys); single-atom catalysts (SACs), where coordination-environment engineering is the primary strategy for overcoming scaling-relation constraints; and dual-atom catalysts (DACs), where cooperative binding between adjacent metal centres may lower barriers along C2 pathways. Copper-based and atomically dispersed catalysts are the primary focus, representing the systems for which the computational literature is most extensive and the idealised-model/operando gap most acute. The third addresses operando phenomena such as potential-induced surface reconstruction, spectator co-adsorption, electrolyte cation effects, and pH-dependent selectivity that static slab models cannot capture but that critically determine performance. Current limitations are assessed and near-future directions proposed, including constant-potential MLFFs, nonlinear implicit solvation, multiscale atomistic-to-device workflows, and computation–operando spectroscopy integration.