Almost two centuries after the word “catalysis” was first introduced by Berzelius in 1835, the field has been developed to the point where heterogeneous catalysis is at the heart of our today’s chemical industry. Nevertheless, one of the grand challenges in this area is being able to tune and design efficient catalysts for processes of interest. In order to do so, a molecular-level understanding of heterogeneous catalysts is of the utmost importance and indeed is a primary focus of modern catalysis research. Conventionally, the single most thermodynamically stable structure of the catalyst obtained under the reaction conditions had been considered as the reactive structure. However, catalysts in the subnano regime, in which there are only up to around 30 atoms per cluster, undergo structural dynamics under reaction conditions, which is triggered by high temperatures and pressures, and changing adsorbates. Using density functional theory and global optimization for structure prediction, in combination with statistical mechanics, we have recently shown that this dynamic fluxionality causes supported clusters to populate numerous distinct structural states under catalytic conditions. Furthermore, considering the single most stable structure gives unrealistic picture and inconsistent results with experiments. Therefore, the catalyst structure should be viewed as an evolving statistical ensemble of many structures. This new idea reforms the accepted models and calls for a new theory and modeling approaches leading to revised design strategies.

Our ensemble-average model along with careful sampling of relevant structures suggest that many earlier studies might have overlooked the actual active sites. Ensemble phenomena lead to surprising exceptions from established rules of catalysis such as scaling relations. Catalyst deactivation (sintering, poisoning) is also an ensemble property, and its extent of mitigation can be predicted through the new paradigm. For example, in collaboration with Scott Anderson (U. Utah), we showed that nano-alloying with Sn suppresses both sintering and coking of Pt clusters deposited on SiO2, and on Al2O3, in conditions of thermal dehydrogenation. Theoretically, we showed that this is an ensemble effect, whereby adding Sn quenches electronic spin in all thermally accessible Ptn isomers clusters, closing most of the reaction paths toward deeper dehydrogenation. The ensemble approach leads to a different view on the reaction kinetics and thermodynamics. Chemically distinct states of the catalyst get populated as T increases, and if these states have barriers significantly different from that of the global minimum the Arrhenius plot should be nonlinear. Therefore, we proposed a modification to the Arrhenius equation using an ensemble-average representation. Spectral signatures are also no longer those of a single structure. In this regard, we showed that for highly fluxional supported nanoclusters, the customary extraction of the oxidation state of the metal from X-ray Absorption Near Edge Structure (XANES) data by fitting to the bulk standards has to be revised. Fitting the experimental spectrum to the calculated spectra of computed ensembles of supported clusters can in contrast provide good agreement and insight on the spectrum-composition- structure relation.

These findings were enabled by advances in theory, such as global optimization and subsequent utilization of multiple local minima and pathways sampling as well as catalyst characterization under working condition. More importantly, our proposed model has been tested and confirmed by several experiments, as shown in joint publications with the experimental groups.