Transforming the analysis of X-ray spectroscopy with deep learning: time resolved dynamical predictions Clelia Middleton 1 , Prof. Tom Penfold 1 , Dr. Conor Rankine 1,2 1 Department of Chemistry, School of Natural and Environmental Sciences, Newcastle University, UK, 2 Department of Chemistry, University of York, UK We present the application of the previously developed deep neural network XANESNET [1] to a system of time- resolved dynamical trajectories of the simple ring system dithiane. [2] With the application of an iterative learning method, we find that the neural network is able to successfully predict spectral dynamics for any of the trajectories up to 900fs with around 10% accuracy overall after only 100fs of data learned. We also introduce an ensembling method as a metric of dependability by which an end user may judge that the machine learning model is producing optimal predictions. The standard deviation of the ensemble converges after 100fs, consistent with the timestep at which good predictions are produced by the model. References 1. C. D. Rankine and T. J. Penfold, Accurate, affordable, and generalizable machine learning simulations of transition metal x-ray absorption spectra using the XANESNET deep neural network, J. Chem. Phys., 2022, 156 , 164102. 2. C. D. Rankine, J. P. F. Nunes, M. S. Robinson, P. D. Lane and D. A. Wann, A theoretical investigation of internal conversion in 1,2-dithiane using non-adiabatic multiconfigurational molecular dynamics, Phys. Chem. Chem. Phys., 2016, 18 , 27170– 27174.
P41
© The Author(s), 2023
Made with FlippingBook Learn more on our blog