Compact Representation of the Local Atomic Structure of Matter for Machine Learning in XANES-Spectroscopy Data Processing

Abstract

The paper introduces a method for representing data on the local atomic structure as histograms of pair radial distribution functions categorized by atom types. This method is used to construct a structure descriptor essential for determining the material structure using machine learning and artificial intelligence methods. A distinctive feature of the approach is the simultaneous use of two sets of pair radial distribution functions: for pairs of all atom types and for pairs involving a selected absorbing atom. The developed approach is tested for determining the nearest environment of silver atoms in color centers in sodium-silicate glasses based on the spectra of X-ray absorption near the absorption edge of Ag. The informativeness of the proposed structure descriptor is demonstrated by its ability to recreate a three-dimensional model of the silver color center’s structure from the corresponding pair distance histograms. Using multiple machine-learning methods, we demonstrate that the proposed descriptor enables the high-quality reproduction of X-ray absorption near edge structure (XANES) spectra for color centers in glass within the framework of the finite-difference method, which results in a four-order-of-magnitude cut in the calculation time for the XANES spectra. The constructed machine-learning model establishes a fundamental connection between the atomic structure of color centers in glasses and the silver XANES spectrum, which is essential for determining the structure of glasses.