Abstract: Density functional theory (DFT) is widely used to predict the structures and properties of nanoparticles (up to about 3 nm wide), but its direct applications to nanocatalysts of experimentally relevant sizes can be prohibitively expensive. It has been demonstrated that this problem can be addressed through the use of cluster expansion models trained by DFT.
In this talk, four examples of using cluster expansions to better understand structure-property relationships in catalysts, from 0-D nanoparticles to 1-D nanorods and 2-D surfaces, are presented.
In the first example, the predicted Pt-Cu nanoparticle structures are compared with experimental characterization. It is demonstrated that the best agreement is achieved by constructing a novel cluster expansion for alloy nanoparticles of varying shape and size that explicitly includes adsorbates, enabling the prediction of nanoparticle structures in an oxidizing environment.
In the second example, a transition-state cluster expansion that explicitly includes a sublattice of sites is constructed to predict the activation energies and study atomic diffusion in Pt-Ni nanoparticles. This model is systematically improvable through the generation of training data to a point at which the prediction error is about half of that of commonly used simpler models, with a comparable overall execution speed in kinetic Monte Carlo simulations.
In the third example, a study of the CO2 reduction reaction on gold nanorods is presented. Nanostructures with the 4H phase show enhanced activity and selectivity relative to fcc nanorods experimentally. Cluster expansions are used to predict the equilibrium structures of fcc and 4H nanorods and further DFT calculations and kinetic models are used to identify the catalytically active sites. The enhanced activity of the 4H nanostructures is ascribed to their unique surface structures with under-coordinated sites, which may provide new design strategies for experimental research.
Lastly, a study of the hydrogen evolution reaction on transition metal phosphides and platinum surfaces is presented. Cluster expansions are used to predict structures and energetics of adsorbed hydrogen as a function of temperature and applied potential, allowing for the determination of the potential-dependent activity of different sites while fully accounting for the coverage effect. The challenge of using a simple descriptor for catalytic activity is discussed.