Machine learning assisted interfacial catalysis of metal-oxide nanocatalysts in CO2 hydrogenation

Understanding the structure-property relationship is of pivotal importance for the rational design of efficient solid catalysts, while this process is seriously blocked by their structural complexity. Here, we have developed a machine learning (ML) assisted strategy to deeply decouple the interfacial catalysis of metal-oxide nanocatalysts in CO ₂ hydrogenation by combining experimental and theoretical results with ML algorithm. Using representative Pd/CeO ₂ catalysts selected on basis of an experimental data-driven ML model, the matched experimental and ML-predicted results demonstrate that, besides surface oxygen vacancies of the support that are responsible for the adsorption and activation of CO ₂ , the catalytic activity and CO selectivity are determined by the H spillover and hydrogenation properties and the binding strength of CO intermediate, respectively, both tightly relevant to the nature of supported metals. Further feature-importance analysis indicates that the d states’ center of supported metals, associated with the intrinsic nature of metal atoms and the metal-support interactions, acts as the most important factor governing the catalytic performance. This work greatly deepens the fundamental understanding of metal-oxide NCs in CO ₂ hydrogenation and innovatively pens up a new paradigm for catalytically fundamental studies to unravel the complex nature of heterogeneous catalysis.