Automated estimation of materials parameter from X-ray absorption and electron energy-loss spectra with similarity measures

Abstract

Materials informatics has significantly accelerated the discovery and analysis of materials in the past decade. One of the key contributors to accelerated materials discovery is the use of on-the-fly data analysis with high-throughput experiments, which has given rise to the need for accelerated and accurate automated estimation of the properties of materials. In this regard, spectroscopic data are widely used for materials discovery because these data include essential information about materials. An important requirement for the realisation of the automated estimation of materials parameters is the selection of a similarity measure, or kernel function. The required measure should be robust in terms of peak shifting, peak broadening, and noise. However, the determination of appropriate similarity measures for spectra and the automated estimation of materials parameters from these spectra currently remain unresolved. We examined major similarity measures to evaluate the similarity of both X-ray absorption and electron energy-loss spectra. The similarity measures show good correspondence with the materials parameter, that is, the crystal-field parameter, in all measures. The Pearson's correlation coefficient was the highest for the robustness against noise and peak broadening. We obtained the regression model for the crystal-field parameter 10 Dq from the similarity of the spectra. The regression model enabled the materials parameter, that is, 10 Dq, to be automatically estimated from the spectra. With regard to research progress in similarity measures, this methodology would make it possible to extract the materials parameter from a large-scale dataset of experimental data.

Symmetry prediction and knowledge discovery from X-ray diffraction patterns using an interpretable machine learning approach

“…For example, experimental parameters such as scanning ranges, data point intervals, and measurement time can be objectively tailored for a specific purpose if an ML model indicates guidelines for these parameters to assure minimal but sufficient data quality for the purpose. An efficient experimental design is particularly crucial for experiments using synchrotron X-rays and neutron beams, where the efficient use of the measurement time is essential because of limited beamtime available 52 – 54 .…”

In Situ/Operando Electrocatalyst Characterization by X-ray Absorption Spectroscopy

“… 26 This situation has begun to change in the last years due to the advancements in XANES data modeling within different approximations 34 such as FMS, 33 , 42 time-dependent density-functional theory, 43 , 44 Bethe-Salpeter method, 45 , 46 and others. 47 − 50 Codes like FitIt 51 and MXAN 52 now provide the possibility to efficiently fit the parameters of simple structures and match theoretically simulated spectra to experimental data, while the recently developed machine learning approaches enable automatic spectra matching 53 , 54 and regression of structure parameters 12 , 34 , 38 based on the recognition of patterns in large data sets.…”

A Critical Review of Machine Learning of Energy Materials

“…Several works have tried to predict the local environments by combining ML models with high-throughput computational or experimental XAS data. [172,[290][291][292][293][294] Timoshenko et al [290] have used neural networks to predict the Pt nanoparticle structure from the L-edge X-ray absorption near-edge spectra. The authors first constructed an experimentally verified computational database of L-edge XANES.…”

“…Experimentally, atomic local environments can be inferred from X‐ray adsorption spectroscopy (XAS). Several works have tried to predict the local environments by combining ML models with high‐throughput computational or experimental XAS data . Timoshenko et al have used neural networks to predict the Pt nanoparticle structure from the L‐edge X‐ray absorption near‐edge spectra.…”