Learning the Structure of Ultradispersed Copper Clusters from X-ray Absorption Fine Structure data: Artificial Neural Network Approach

Unique structural motifs in metallic nanoparticles (NPs) result in properties that differ dramatically from those of bulk materials and that can be exploited in many applications, e.g., in the field of catalysis. The rational design of such materials, however, is hindered by the limitations of characterization methods that would allow detection of such motifs (undercoordinated surface sites, relaxation of interatomic distances) in ultrasmall (< 3 nm) NPs. Especially challenging is the problem of determination of NPs structure in operando conditions.

Here we explore a new route to solve this issue, by analyzing X-ray absorption near-edge structure (XANES) with an artificial neural network (NN), trained on a large set of ab-initio calculated XANES spectra. Trained NN can then map the XANES features to the descriptors of structure - particle-averaged coordination numbers and average interatomic distances -, and provide unique information about in-situ transformations of NPs structure. Here we employ this approach to follow in-situ the agglomeration of size-selected subnanometer copper clusters supported on zirconia (independently evidenced by grazing-incidence small-angle X-ray scattering (GISAXS)).