Deep learning model successfully predicted ignition in inertial confinement fusion experiment

Brian Spears and colleagues built a generative machine learning model that was used to successfully predict the outcome of a recent fusion ignition experiment at the U.S. National Ignition Facility (NIF). Their model predicted, with a probability greater than 70%, that ignition was the most likely outcome of the experiment. The findings could guide researchers working on future inertial confinement fusion experiments, which use energetic lasers to compress and heat a capsule of hydrogen isotopes to create nuclear reactions that produce fusion energy. Ignition refers to the outcome where the fusion energy produced exceeds the laser energy used to conduct the experiment – a feat that NIF researchers accomplished in a small experiment in 2022. This success was predicted by the generative learning model developed by Spears et al., which was built using experimental data, radiation hydrodynamics simulations, and Bayesian statistics. Having a successful prediction model will provide swift guidance to fusion energy researchers as they modify experimental designs and determine whether future upgrades in laser energy and other variables might improve fusion output and efficiency, the authors note.