Collectivity effect in cluster catalysis under operational conditions

Subnanometer metal clusters, tiny assemblies of just a few atoms, hold great promise as next-generation catalysts. Yet, their potential has long been hindered by structural diversity and dynamic behavior under reaction conditions. In a breakthrough study, researchers combined artificial intelligence with multiscale modeling and statistical analysis to systematically map the catalytic sites of these clusters in realistic environments. The team discovered that catalytic activity arises not from a single active site, but from a collective ensemble of diverse sites spanning different sizes, compositions, isomers, and surface locations. Despite their distinct atomic arrangements and reaction pathways, these sites share high intrinsic activity and significant population, making them collectively responsible for overall performance. Machine learning revealed that this collective behavior is fundamentally governed by the balance between local atomic coordination and adsorption energy. Using CO oxidation on Cu/CeO₂ as a model system, the researchers validated their predictions through close agreement between computed mechanisms and experimental observations. This work offers fresh insights into the nature of active sites in heterogeneous catalysis and underscores the power of data-driven approaches to harness collective effects in cluster catalysts, paving the way for the rational design of highly efficient catalytic materials.