Machine Learned Surrogate for High-Z xRAGE Simulations

High Z dopants can be added to the fuel in inertial confinement fusion capsules for usage

by diagnostics such as spectroscopy. Between 2006 and 2009, a campaign at the Omega

Laser Facility investigated the effects of noble gas dopants in DD fuel which found that

increased dopant concentration led to a degradation of the neutron yield. The hydrocode

xRAGE is used to better understand these experiments, but larger parameter studies can

be computationally expensive. We developed an active learning framework to train a

neural network surrogate to predict simulation outputs such as the yield, bang time, and

burn width. This allows for a reduction in computation time which can be used for more

comprehensive studies and target design. Furthermore, we use techniques such as

transfer learning to integrate existing experimental data, which can help to learn aspects

the experiment that are not captured in simulation.