Robust Machine Learning Inference from X-Ray Absorption Near Edge Spectra through Featurization

Le résumé fourni par la source

X-ray absorption spectroscopy (XAS) is a powerful tool for probing local structures, oxidation states, and electronic properties of functional materials. Based on energy proximity to the absorption edge, spectra are categorized into XANES (near-edge) and EXAFS (extended) regions. However, interpreting XANES typically requires reference spectra, which are often limited in quality and availability. We first addressed this gap by generating a large-scale computational database of L-edge XANES spectra, using the FEFF9 code. Over 130,000 spectra for transition metal compounds were produced and made publicly accessible via the Materials Project, laying a foundation for machine learning (ML) applications in XAS. Then we explored how different spectral representations affect ML performance. We featurized the spectra and benchmarked the ML algorithms on oxidation state classification and bond length prediction tasks. The cumulative distribution function feature offered the best accuracy and robustness while remained explainable from physics. Experimental validation further confirmed the model's predictive ability on unseen data. Together, these projects establish a robust data resource and demonstrate how ML can accelerate and enhance XAS data interpretation.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.