Automated search for inertial confinement fusion designs using Bayesian optimization

Inertial confinement fusion (ICF) experiments rely on complex multi-physics simulations such as the LLNL-developed HYDRA to guide design work. Designers often hand-tune simulations to align with experimental measurements, or to optimize for characteristics like high yield or implosion symmetry. However, these simulations can be expensive and have several dozen tunable parameters. This makes searching the parameter space for an optimal design difficult. Recently developed automated tools utilize Bayesian optimization to search these high-dimensional spaces for designs.

In this project, we use the Bayesian optimization tools to tune ICF simulations to experimental measurements of a well-characterized shot, N210808, the first MJ yield shot at the National Ignition Facility. This optimization algorithm runs 2D integrated simulations in HYDRA to converge on a design that matches the target experiment. The algorithm quickly and autonomously adjusts simulation parameters to match measurements such as hotspot shape, bang time, and “keyhole” shock-timing. Building on this work, we explore the tool’s ability to search for designs with certain target characteristics. This design search capability can be used to produce designs that match specified scalar values or time series profiles.