A Deep Neural Network for the Rapid Prediction of X-ray Absorption Spectra

Abstract

X-ray spectroscopy delivers strong impact across the physical and biological sciences by providing end-users with highly-detailed information about the electronic and geometric structure of matter. To decode this information in challenging cases, e.g. in operando catalysts, batteries, and temporally-evolving systems, advanced theoretical calculations are necessary. The complexity and resource requirements often render these out of reach for end-users, and therefore data are often not interpreted exhaustively, leaving a wealth of valuable information unexploited. In this paper, we introduce supervised machine learning of X-ray absorption spectra, by developing a deep neural network (DNN) that is able to estimate Fe K -edge X-ray absorption near-edge structure spectra in less than a second with no input beyond geometric information about the local environment of the absorption site. We predict peak positions with sub-eV accuracy and peak intensities with errors over an order of magnitude smaller than the spectral variations that the model is engineered to capture. The performance of the DNN is promising, as illustrated by its application to the structural refinement 1 of iron(II)tris(bipyridine) and nitrosylmyoglobin, but also highlights areas for which future developments should focus.

Enhancing the analysis of disorder in X-ray absorption spectra: application of deep neural networks to T-jump-X-ray probe experiments

“…To address this, contemporary works have explored supervised machine learning/deep learning algorithms with a view towards mapping the relationship between XANES spectra and the electronic and geometric structures of the systems that they characterise. 8,[33][34][35][36][37][38][39][40][41][42][43][44][45][46][47] For an ab initio MD-based approach like that described in the present Article, our own deep neural network (DNN; introduced in ref. 46) could be used to accelerate the prediction of the X-ray spectra for each of the ab initio MD snapshots (the bottleneck of the strategy), opening up a fast and cost-effective route to the quantitative interpretation of T-jump pump/X-ray probe experiments.…”

“…The architecture of our DNN has been described in detail in ref. 46; we recommend this reference to the unfamiliar reader. The code is publicly available 48 and a schematic is given in Fig.…”

“…All hyperparameters were determined via Bayesian optimisation using the GPyOpt 56,57 module and, if unspecified, are the same as those detailed in ref. 46.…”

“…Throughout this work, this convolution has been carried out using an energy-dependent arctangent function via an empirical model close to the Seah-Dench formalism, 61 as detailed in ref. 46. The arctangent convolution is only applied as a post-processing step on absorption cross-sections estimated by our DNN; we stress that our reference dataset comprises only unconvoluted crosssections, and our DNN learns from these unconvoluted crosssections, as in ref.…”

“…The arctangent convolution is only applied as a post-processing step on absorption cross-sections estimated by our DNN; we stress that our reference dataset comprises only unconvoluted crosssections, and our DNN learns from these unconvoluted crosssections, as in ref. 46.…”