Advancing the analysis of X-ray spectroscopy via machine learning

Abstract

X-ray absorption spectroscopy at the L2/3-edge can be used to obtain de- tailed information about the local electronic and geometric structure of transition metal complexes. By virtue of the dipole selection rules, the transition metal L2/3-edge usually exhibits two distinct spectral regions: i) the "white line", which is dominated by bound electronic transitions from metal 2p orbitals into unoccupied orbitals with d character; the intensity and shape of this band consequently reflects the d density of states and the nature of the surrounding ligands, and ii) the post-edge, where oscillations encode the local geometric structure around the X-ray absorption site. Interpreting the high information content held within X-ray spectra is a significant challenge, particularly at the L2/3-edge due to the high computational expense associated with performing the relevant calculations. Consequently, this thesis extends the recently-developed XANESNET deep neural network (DNN) be- yond the K-edge to predict X-ray spectra at the L2/3-edge of transition metals. It is demonstrated that XANESNET is able to predict L2/3-edge X-ray absorption spectra, including both the geometric and electronic parts with nothing but the geometric structure around the X-ray absorption site. As achieving simultaneously accurate and sufficiently comprehensive training data is challenging due to the aforementioned computational complexity, op- timal methods for constructing a training dataset were investigated and it was shown that with a much smaller training dataset, accurate predictions can still be achieved. Additionally, the DNN was applied to an ultrafast XAS case study where it was able to accurately predict transient X-ray spectra using ∆-machine learning, in which DNN learns the difference between two levels of theoretical spectra and corrects the resulting predictions to the higher level.