Data-driven massive reaction networks reveal new pathways underlying catalytic CO₂ hydrogenation
Automated discovery of catalytic pathways via DFT, ML-predicted barriers, and human-intelligence-inspired reaction enumeration
Heterogeneous catalytic pathways for clean energy conversion involve thousands of elementary steps, but most quantum-mechanical models involve only a few dozen reactions. This work combines extensive density functional theory (DFT) calculations, machine learning (ML) for activation barrier prediction, and human-intelligence-inspired reaction enumeration and elementary reaction identification to enable automated kinetic modeling of CO₂ hydrogenation on copper - a key process for producing fuels and chemicals.
Starting from a dataset of 152 elementary CO₂ reduction reactions, the approach expands to 9,389 elementary reactions, substantially reducing human bias in reaction pathway selection. The expanded network reveals 40-fold higher CO₂ conversion rates, following experimental trends of methanol and CO production. DFT calculations were validated against experimentally measured CO₂ conversion, confirming that even large networks with 100+ reactions are insufficient to capture the full reaction space.
A key ML-enabled discovery - validated post-facto - establishes the crucial roles of intermolecular hydrogen transfer and hydrogenation by molecular hydrogen, mechanisms that would be missed by smaller-scale analyses. The proposed strategy for comprehensively modeling complex catalytic mechanisms will significantly advance catalysis research and carbon conversion processes.