Employing generative AI to optimize fusion experiments

Inertial confinement fusion (ICF) is a method to induce nuclear fusion by focusing powerful lasers on a hydrogen target. In 2022, an ICF experiment at the National Ignition Facility successfully achieved ignition, but the process is still far from producing a net gain of energy.

One way to improve the efficiency of ICF reactions is to optimize the input parameters, a complex task requiring extensive experimental data and computational simulations. Even with those tools, the sheer size of the parameter space makes exploring it challenging.

To address this challenge, Ben Tayeb et al. employed generative AI to optimize laser illumination parameters, simplifying experimental design and improving efficiency.

“The physics involved in an ICF experiment is complex,” said author Morad Ben Tayeb. “Thanks to the use of generative AI, we were able to explore solutions that break away from the usual ones.”

A typical laser ignition scheme consists of a gradual ramp up of laser intensity, followed by a plateau, and concluding with a shock to trigger ignition. The generative AI employed by the researchers suggested a second plateau before the shock that doubled the energy yield while halving the required input energy. Subsequently, the AI suggested removing the shock altogether, simplifying the process without sacrificing energy gains.

The authors hope to improve their results by expanding the types of inputs given to the AI.

“Although these results are preliminary, they pave the way for further research in optimization,” said Ben Tayeb. “The next steps will involve integrating the composition and dimensions of the targets, as well as laser parameters, into the optimization process.”

Source: “ICF target optimization using generative AI,” by Morad Ben Tayeb, Vladimir T. Tikhonchuk, and Jean-Luc Feugeas, Physics of Plasmas (2024). The article can be accessed at https://doi.org/10.1063/5.0228824 .