An exploration of machine learning for soft X-ray spectroscopic techniques applied to transition metals compounds

Johann Lueder^1,2*

¹Department of Materials and Optoelectronic Science, National Sun Yat-sen University, Kaohsiung, Taiwan
²Center for Theoretical and Computational Physics, National Sun Yat-sen University, Kaohsiung, Taiwan

* Presenter:Johann Lueder, email:johann.lueder@mail.nsysu.edu.tw

Machine learning has the potential to benefit research and accelerate our journey to new discoveries. Here, we apply deep learning with artificial neural networks and explore their capability to accelerate research using soft X-ray spectroscopic techniques. Soft X-ray absorption spectroscopy (XAS) is a versatile tool to characterize and understand materials’ properties. Applied to the L-edge of transition metal ions contained in samples, obtained insights into their electronic states and processes linked to electrons in d-orbitals drive the development of technologies in catalysis, energy storage, electronics and many more. However, the interpretation of L-edge spectra can be rather challenging due to the vector coupling of semicore p- and valence d-states, spin-orbit coupling and charge transfer effects. Our development of deep learning applied to XAS and its optimization offers the possibility for rapid analysis and prediction of L-edge spectra as well as the identification of oxidation states. We furthermore explore deep learning for related spectroscopic techniques such as X-ray photoelectron spectroscopy and distinguish the limitations, benefits and opportunities of deep learning for those techniques.

Keywords: machine learning, X-ray spectroscopy, transition metals