Predicting fusion ignition at the National Ignition Facility with physics-informed deep learning

Area of Science:

• Nuclear Fusion
• Plasma Physics
• Machine Learning

Background:

• Inertial confinement fusion (ICF) experiments aim to achieve self-sustaining fusion reactions.
• Predicting the outcome of complex ICF experiments remains a significant challenge.

Purpose of the Study:

• To report the achievement of ignition in an ICF experiment at the National Ignition Facility.
• To demonstrate the predictive power of a novel machine learning model for fusion experiments.

Main Methods:

• Utilized radiation hydrodynamics simulations, deep learning algorithms, and Bayesian statistics.
• Integrated experimental data with advanced computational models.
• Developed a generative machine learning model for outcome prediction.

Main Results:

• The ICF experiment successfully generated fusion energy output exceeding the input laser energy, achieving ignition.
• The predictive machine learning model assigned a greater than 70% probability of ignition prior to the experiment.
• The model accurately predicted ignition as the most likely outcome.

Conclusions:

• This experiment marks a significant milestone in fusion energy research, demonstrating net energy gain.
• The integration of machine learning with physics simulations offers a powerful tool for advancing fusion science.
• Predictive modeling can enhance experimental design and increase the likelihood of achieving key fusion milestones.