Deciphering the Interfacial Catalysis of Metal–Oxide Nanocatalysts in CO2 Hydrogenation through a Machine Learning Approach

Understanding the structure–property relationship is of pivotal importance for the rational design of efficient solid catalysts yet persists as a significant challenge due to the inherent complexity of solid materials. Here, we present an efficient strategy to decipher the interfacial catalysis in metal–oxide nanocatalysts during CO2 hydrogenation through synergistically integrated experimental and theoretical investigations with machine learning (ML) algorithms. Using model Pd/CeO2 catalysts as a proof-of-concept system, the matched experimental and ML-predicted results demonstrate that given sufficient oxygen vacancy concentrations on the support matrix the catalytic performance is primarily governed by the nature of supported metals. Specifically, atomically dispersed Pd species exhibit exceptional intrinsic activity for CO production, which is attributed to its enhanced H spillover and hydrogenation capacities but weakened CO binding affinity. Further ML analysis indicates the sum surface d charge of supported metal as the principal factor governing catalytic performance among four types of typical intrinsic features influencing catalytic hydrogenation processes, which could be directly evaluated by a geometric descriptor of average coordination number of the metal on the support, associated with the metal particle size. This work provides a generalizable theoretical framework for understanding metal–oxide interfaces in CO2 hydrogenation and opens up a novel approach for catalytically fundamental studies to unravel the complex nature of heterogeneous catalysis.