Inferring Mix and Hot-Spot Conditions from Spectroscopic Measurements with Machine Learning
Bruce Hammel | 22-FS-039
Experiments on the National Ignition Facility (NIF) have provided clear experimental evidence of ablator mix into the hot-spot (H-S), leading to degraded performance; however, inferring the amount of mix from the experimental observations is highly challenging, because different combinations of experimental conditions can lead to similar experimental "observables." To address this question, we have developed and successfully tested a method that uses machine-learning-assisted statistical analysis (Bayesian Markov-Chain Monte Carlo), to infer the amount of mix and its impact on the H-S, with uncertainties. The inference is made from experimental observables, including x-ray and neutron measurements. To assess this approach, we have tested it on synthetic data from simulations, where the true H-S conditions (e.g., temperature, density, and the amount of mix) are known. When applied to experimental data, this approach will lead to an improved understanding of H-S conditions and the impact of mix resulting from the growth of capsule perturbations due to hydrodynamic instabilities.
This research is impactful to both nuclear-weapons science and high-energy-density science (HEDS). It utilizes advanced modeling and simulation methods, which integrate multiphysics models, cognitive simulation, multiscale modeling, and advanced computing architectures, to address a highly-complex multidimensional problem associated with the pursuit of ignition. More generally, it provides a method to quantify uncertainties in the interpretation of HEDS experiments, including those addressing dense-plasma effects on atomic processes in hot, dense plasmas for basic-science and mission-science applications.