Accurate adsorption energies of carbon-species on copper catalysts

Accurate prediction of adsorption energies for C-species on copper catalysts is critical for understanding catalytic behavior and catalyst design. However, conventional density functional theory often falls short, as exemplified by the "CO adsorption puzzle." Here, we demonstrate that the hybrid PBE-D3/M06 method overcomes these limitations, achieving near-chemical accuracy for C-species on Cu(100), Cu(110), and Cu(111) surfaces. This approach can resolve the CO puzzle by correctly predicting adsorption sites and energies in excellent agreement with experimental data. The method yields remarkably low mean absolute errors (MAEs) of 0.06 eV for reaction intermediates (vs. RPA calculations) and 0.04 eV for molecules (vs. experimental values). To accelerate these high-fidelity calculations, we further developed a machine learning model that rapidly predicts accurate adsorption energies (MAE: 0.08 eV) from standard PBE values. When applied to electrochemical CO₂-to-CO conversion, our approach predicts onset and equilibrium potentials to within 0.04 V of experimental measurements across all low-index copper facets. This combined computational strategy provides an efficient and reliable framework for the rational design of copper-based catalysts.