Radial Distribution Function from X-ray Absorption near Edge Structure with an Artificial Neural Network

Abstract

Extended X-ray absorption fine structure (EXAFS) is one of the common tools used to determine the local atomic coordination and bond length around an objective element via data conversion from the EXAFS profile to a radial distribution function (RDF). Thus, EXAFS has been widely used for the investigation of catalytic reactions, battery degradation, and other material developments. However, the data conversion from EXAFS to RDF involves three difficulties: weak signals, necessity of a wide energy range (∼1000 eV), and fitting parameters in reference samples. Recently, the direct estimation technique of RDF from EXAFS using machine learning was successfully developed, which overcomes the third problem. Here, we used an artificial neural network (ANN) to directly predict RDF from the near-edge region of the spectrum (X-ray absorption near-edge structure: XANES), which enables overcoming all of the difficulties. The ANN can correctly generate RDF only from XANES. XANES includes information about the bond length and coordination numbers in the range of ∼5 Å. We applied our prediction model to an experimental spectrum and confirmed its accuracy. The method proposed here is greatly beneficial for measuring the local bond length and coordination of materials whose extended energy region is difficult to be measured.