Faraday joint interest group conference 2023

Convolutional neural networks for extracting structural characteristics from X-ray absorption spectra Tudur David and Thomas Penfold Newcastle University, UK

X-ray absorption spectroscopy (XAS) is a core technique, widely used across the physical and biological sciences due to its ability to deliver electronic and geometric information about a sample in a huge range of different environments. However, understanding XAS requires high levels of theory, which can be challenging due to resource requirements and complexity of theory [1]. To enhance the accessibility of XAS, a deep neural network has been developed, based on the multilayer perceptron (MLP) model, which can provide instantaneous, qualitative predictions of XAS spectra using nothing more than the local geometry of an arbitrary absorption site [2,3]. In the present work, we present an extension of XANESNET to perform the reverse analysis, i.e. spectrum to structure. Using a convolution neural network (CNN) architecture, our new network can translate XAS directly into a radial distribution function describing the arrangement of atoms around the absorbing atom. Furthermore, we present an autoencoder and generative adversarial models which are able to perform consistent forward and reverse mapping of x-ray spectra within a single model. References 1. Joly, Y.; Grenier, S. Theory of X-Ray Absorption Near-Edge Structure. In X-Ray Absorption and X-Ray Emission Spectroscopy: Theory and Applications; Van Bokhoven, J. A., Lamberti, C., Eds.; Wiley, 2016; Vol. 1, Chapter 4, pp 73−97. 2. Rankine, C.D., Madkhali, M.M. and Penfold, T.J., 2020. A deep neural network for the rapid prediction of X-ray absorption spectra. The Journal of Physical Chemistry A , 124 (21), pp.4263-4270. 3. Rankine, C.D. and Penfold, T.J., 2022. Accurate, affordable, and generalizable machine learning simulations of transition metal x-ray absorption spectra using the XANESNET deep neural network. The Journal of Chemical Physics , 156 (16), p.164102.

P17

© The Author(s), 2023

Made with FlippingBook Learn more on our blog