Unlocking the Potential of EELS and XAFS with Machine Learning

Teruyasu Mizoguchi¹

¹Institute of Industrial Science, The Unviersity of Tokyo

Electron Energy Loss Spectroscopy (EELS) and X-ray Absorption Spectroscopy (XAFS) observed using STEM and synchrotron facilities are now indispensable tools for characterizing functional materials due to their superior spatial and time resolutions, high sensitivity, and abundant information. In particular, the near-edge structure in the EELS/XAFS spectrum (ELNES/XANES) reflects the partial density of states of the conduction band at the excited state. Recently, the application of machine learning to EELS/XAFS has been reported. Our work aims to transcend the traditional physics of spectrum generation through machine learning. First, we constructed a spectral database comprising more than 100,000 Carbon-K edges of organic molecules [1]. Using this database, we achieved the prediction of extensive properties, such as molecular weight and internal energy, which are typically considered unrelated to ELNES features, via machine learning [2] Additionally, we have attempted to extract valence band information, similar to that obtained from XPS, from the ELNES/XANES features [3,4]. Furthermore, the same information as EXAFS was also obtained from ELNES/XANES features [5]. In my presentation, I will discuss the applications of machine learning in unlocking the potential of these core-loss spectroscopies.

[1] Shibata, K., Kikumasa, K., Kiyohara, S., & Mizoguchi, T. (2022). Scientific Data, 9, 214.

[2] Kikumasa, K., Kiyohara, S., Shibata, K., & Mizoguchi, T. (2022). Advanced Intelligent Systems, 4, 2100103.

[3] Chen, P.Y., et al. (2023). Journal of Physical Chemistry Letters, 14, 4858.

[4] Takahara, I., Shibata, K., & Mizoguchi, T. submitted.

[5] Kiyohara, S., and Mizoguchi, T., (2020), J. Phys. Soc. Jpn (Letter), 89, 103001.