Determination of Local Structure Descriptors from X-ray Absorption Near Edge Structure in Bimetallic Nano-Alloys Using an Ab-Inito-Trained Neural Network

X-ray absorption spectroscopy (XAS) is a common method for the probing of nanoparticles (NPs). In conventional approaches, three-dimensional geometry is extracted from extended X-ray absorption fine structure (EXAFS) while the other portion of the XAS spectrum, the X-ray absorption near edge structure (XANES), is often neglected. XANES is sensitive to structure and is less perturbed by disorder, compared to EXAFS, but there's a lack of methodology preventing the quantitative determination of structure from XANES spectra. We have recently developed a method to fill this gap: an artificial neural network (NN) trained on ab-inito site-specific XANES calculations [1]. The NN was successful in the quantitative analysis of supported Pt NPs, motivating its application to bimetallic systems. Here we present a viable neural-network-based method for the extraction of local structure descriptors from experimental XANES measured in AuPd nano-alloys. By training the NN with a sufficiently diverse theoretical set to simulate the complexity of a bimetallic, one with a range of shapes, sizes, compositions, and lattice constants, we obtain results comparable to conventional EXAFS analysis.

[1] J. Timoshenko et al., J. Phys. Chem. Lett. 8 (20), 5091-5098. 2017