Search for Analytical Relations between X-Ray Absorption Spectra Descriptors and the Local Atomic Structure Using Machine Learning

In this paper, we develop a new technique for quantitative analysis of the near region of X-ray absorption spectra that is based on the extraction of spectrum descriptors and machine learning. The use of descriptors (edge position, intensity and curvature of minima and maxima, and tangent of the slope of the absorption edge) allows solution of the problem of systematic differences between theoretical calculations and experimental data, reducing the dimension of the problem and thereby improving the accuracy of machine-learning algorithms. We obtain analytical relations between the spectrum descriptors and the parameters of the local atomic structure of a substance, which extend the range of applicability of the empirical Natoli rule and analysis of the chemical shift of spectra to arbitrary classes of chemical compounds.