Applying Super-Resolution Machine Learning Techniques to Simulated Opacity Measurements for Near-Continuous-in-Energy Modeling of Radiative Phenomena

Numerical simulations are limited in spectral resolution by the extreme cost of performing multigroup radiation transport calculations. A direct calculation of the radiation transport in an Opacity-on-NIF hohlraum with near-continuous energy resolution is intractable. For this reason, postprocessing has been the technique of choice for computer modeling of these experiments. Current state-of-the-art postprocessors make limiting assumptions regarding the radiation field to make computing spectra at a high spectral resolution feasible. We propose an alternative technique, whereby a deep convolutional neural network machine learning model is trained on a large dataset of synthetic transmission and opacity spectra computed using TOPS¹ at a 34-group and 1800-group discretization in energy. We demonstrate that such a model can upscale these spectra with a high degree of accuracy and apply the model to measurements taken directly from a CASSIO² calculation of an Opacity-on-NIF hohlraum fielded on the NIF experimental platform to compare to the near-continuous measurement.

1. Joseph Abdallah, Jr. and Robert E. H. Clark. TOPS: A Multigroup Opacity Code, Los Alamos Report LA-10454 (1985). https://aphysics2.lanl.gov/opacity/lanl.

2. B.M. Haines, D.E. Keller, J.A. Marozas, et al. Comput. Fluids, 201, 104478 (2020).