A machine learning approach to predict L-edge x-ray absorption spectra of light transition metal ion compounds

Abstract

Simulation of spectra of x-ray absorption spectroscopy (XAS) at the L-edge is a well-established and reliable computational tool that, in combination with experimental measurements, reveals details about the local electronic structure and the chemical environment of transition metal ions (TMI). For light TMIs, model 2p XAS model Hamiltonian approaches (MHA) are fast and accurate in computing spectra. Here, an alternative method using artificial neural networks (ANN) is employed. The new approach predicts x-ray absorption spectra at the L-edge of light transition metals ions for elements from Ti to Ni, including different oxidation states, from sets of physical input parameters. Computational performance and accuracy were optimized through a systematic exploration of ANN design and compared to MHA simulations. While the MHA remains more accurate due to the statistical nature of the developed machine learning method, the ANN-based prediction of 2p XAS spectra achieved good accuracy of more than 99%, in addition to a noticeable advantage in computational speed. This makes the method intrinsically more suitable for computational high-throughput setups as well as in support of data-intense experimental workflows. However, the extrapolation weakness of artificial neural networks remained for most cases and displayed a strong dependence on the TMI. For one TMI, an extrapolation accuracy of 78% was maintained.

References in corpus (7)

• Phase transitions of hybrid perovskites simulated by machine-learning force fields trained on-the-fly with Bayesian inference
• Controlling orbital moment and spin orientation in CoO layers by strain
• Bands, resonances, edge singularities and excitons in core level spectroscopy investigated within the dynamical mean field theory
• Classification of Local Chemical Environments from X-ray Absorption Spectra using Supervised Machine Learning
• Spin and orbital degrees of freedom in transition metal oxides and oxide thin films studied by soft x-ray absorption spectroscopy
• Data-driven kinetic energy density fitting for orbital-free DFT: linear vs Gaussian process regression
• A Deep Learning Model for Atomic Structures Prediction Using X-ray Absorption Spectroscopic Data