Izumi Takahara1,Poyen Chen1,Kiyou Shibata1,Teruyasu Mizoguchi1
The University of Tokyo1
Izumi Takahara1,Poyen Chen1,Kiyou Shibata1,Teruyasu Mizoguchi1
The University of Tokyo1
Energy-loss near-edge structure (ELNES) and X-ray absorption near-edge structure (XANES) are powerful tools for characterizing the local atomic and electronic structures of materials. ELNES/XANES is produced by the electronic transition from a core orbital to conduction bands, and thus it basically provides the information reflecting the partial density of states (PDOS) of the conduction band at the excited state of a given material. Although ELNES/XANES gives us valuable information such as chemical bonding and the charge state of a specific atom, the knowledge that can be extracted from the measurements is limited to the information reflecting the conduction band at the excited state. If the information reflecting the entire PDOS at the ground state can also be extracted from ELNES/XANES, materials properties based on PDOS can be evaluated directly from a single spectral measurement and materials developments would be accelerated.<br/>In this presentation, we introduce a novel machine learning (ML) based method for determining the PDOS of valence and conduction band at the ground state from the ELNES/XANES spectra simultaneously. We have carried out density functional theory (DFT) calculations on various structures of Si and prepared a database of Si-K ELNES/XANES and the corresponding PDOS at the ground state. On this database, we have trained a ML model to predict the entire PDOS spectra at the ground state from the ELNES/XANES spectra directly. In the result, our model successfully reproduced the characteristics of the PDOS spectra near the Fermi level. Therefore, we envision the utilization of the ML could be powerful for widening the application of the ELNES/XANES measurements including simultaneous mapping of structure and property in a transmission electron microscope.