Nature of Reverse Water–Gas Shift Reactions at Metal–Oxide Interfaces Uncovered via Interpretable Machine Learning

Oxide-supported metal clusters are central to the reverse water–gas shift (RWGS) reaction, which converts CO2 to CO; however, the optimal interfacial properties governing activity remain unresolved. Although the oxygen vacancy formation energy (EV) is known to influence CO2 activation, its quantitative role and ideal value for catalysis have not been defined owing to the complexity of metal–oxide combinations and reaction pathways. Here, we integrate first-principles microkinetic modeling with interpretable machine learning across nine transition metal clusters on eight oxide supports to identify two key descriptors─EOV of the support and the atomic radius (r) of the metal cluster─that together control the RWGS reactivity. We reveal a volcano-type relationship between the turnover frequency (TOF) and EV, with optimal activity emerging at moderate vacancy formation energies (∼3.4 eV). A high EV suppresses vacancy formation, whereas a low EV limits CO2 activation. Additionally, larger metal radii systematically lower the barrier for lattice oxygen reduction, stabilizing the transition state and promoting vacancy regeneration. The reaction mechanism shifts from carboxylate-mediated to direct CO2 dissociation as EV increases. Our framework captures experimental trends across reported catalysts and provides a physically grounded, predictive strategy for designing efficient RWGS catalysts by engineering metal–oxide interfaces.