Optimizing New Inertial Confinement Fusion Designs Under Uncertainty

Before new Inertial Confinement Fusion (ICF) designs move to the experimental phase, they are first investigated using numerical simulations. The simulations are calibrated against the past experiments and used to predict the outcomes of future shots. Typically, such predictions are not able to take into account the variability of the experimental outcomes originating from random variations in laser delivery and target quality, and for the sensitivity of a given design to these variations. To capture this shot-to-shot variability, we leverage a series of repeat shots following the N210808 experiment, build a predictive model that is jointly informed by the simulations and experiments, and translate the variations into the modified designs using machine learning. The model includes the uncertainty for the key ICF diagnostic measurements in addition to uncertainty about input conditions (such as capsule quality and laser delivery fluctuations). In this presentation, we discuss the application of this method to investigate the designs with a modified ablator thickness, shock timing, rise, and varying laser energy.