(488a) Grand Canonical Evolutionary Algorithm-Based Approach for Investigating Catalyst Surface Morphology

The surface structure of heterogeneous catalysts is well-known to dynamically respond to changes in the external reaction environment. As an example, the famous strong metal-support interaction involves full or partial covering of metal nanoparticles dispersed on oxide supports with reduced oxide films. Such structural perturbations introduce profound changes in their catalytic properties. The development of efficient strategies to generate catalyst models that incorporate this atomic-scale restructuring would, in turn, lead to improved understanding of the underlying structure-catalyst function relationships. To address this requirement, we present a grand-canonical evolutionary algorithm coupled with a geometric lattice-matching routine. We generate viable catalyst-surface morphologies through a series of geometric mutation and crossover operations, with the final fitness defined in terms of grand canonical surface-energy. Thus, we determine interfacial models with optimal composition and surface structure for specific external environments. We next present two examples of the application of this algorithm. First, we analyze Si(100), wherein the unreconstructed (100) facet is known to be metastable. The pairs of adjacent surface Si atoms buckle to form a dimer reconstruction. The evolutionary algorithm successfully captures this dimer configuration as a low energy motif without any explicit user input. Second, we investigate the oxidation of Pd(111) surfaces, which are known to develop thin-film surface-oxides. These oxides are the intermediate phases between simple oxygen adlayers and the bulk oxide that might be present as the reaction environment is made increasingly oxygen rich. We show that the oxidation of Pd surfaces proceeds through a Pd₆O₄ honeycomb-like structure before being oxidized to a PdO monolayer. Any additional increase in the oxygen pressure results in oxidation of the surface layers to PdO₂. We close by discussing strategies to extend the algorithm to larger, more compositionally complex, multi-element surface systems.