Construction of Sabatier Volcanoes for CO₂ Hydrogenation to C1-2 Oxygenates Using Data-Efficient Machine Learning

The development of technologies for CO₂ hydrogenation to valuable chemicals is prioritized globally due to their potential for large-scale CO₂ abatement. However, rational design of catalysts for this reaction is hindered by an incomplete understanding of the complex reaction network of CO₂ hydrogenation. This study addresses this gap by establishing a data- and computation-efficient framework for CO₂ hydrogenation to C1-2 oxygenates on ≈ 1.2 nm large Au, fcc-Co, Cu, Ni, Pd, Pt, and Rh nanoparticles. The obtained reaction networks could not be described using Brønsted-Evans-Polanyi relationships with sufficient accuracy, motivating the use of a nonlinear NN model for inference of activation energies. A pathfinder-like algorithm and an energetic span model are used to analyse these networks and to construct Sabatier volcano plots identifying optimal binding properties for a promising catalyst. The analysis indicates that efficient C2 oxygenate formation simultaneously requires CH_x formation, favorable C-C coupling and efficient protonation, as well as suppression of methanation and poisoning by CO, alcoholates, and carboxylates. These features are mutually exclusive on the studied monometallic nanoparticles, suggesting that multicomponent and multimetallic catalysts are more realistic design targets. This framework supports computational discovery of CO₂ hydrogenation catalysts challenged by data limitations and mechanistic complexity.