Predicting N221204 observables from an ensemble of N210808 simulations*

In October 2022 the Cognitive Simulation group at LLNL had generated tens-of-thousands of 2d radiation hydrodynamic simulations of the 1.35 MJ neutron yield shot (N210808) for a Bayesian-SuperPostShot (BSPS) analysis. The BSPS analysis informs the team what the likely range of input parameters (laser fluctuations, preheat, m-band fraction etc.) are and how these are correlated amongst themselves. Using experimental data from N210808 and the accompanying repeat shots led to the ability to create an input-variability model based on the 1.9 MJ laser energy design. The variability model was modified slightly to consider the design changes for the December 2022 2.05 MJ laser energy design. An efficient multivariate integration method known as stochastic collocation was used to sample the variability model which led to a small set of new hydrodynamic simulations. The new simulations were then used in conjunction with the machine-learning technique “transfer learning” to create a modified machine-learned surrogate appropriate for the 2.05 MJ laser-energy design (N221204). The modified surrogate ultimately formed the basis for the CogSim prediction that N221204 had approximately a 50% chance of igniting.