Direct-drive fusion experiments gain insights from machine learning-driven 3D reconstructions

Direct-drive inertial confinement fusion reactions involve depositing large amounts of energy on a spherical target using precisely calibrated lasers to induce an implosion of fuel that will trigger a nuclear fusion reaction. Key to this process is perfect spherical symmetry in both the laser energy and the target geometry, as any anisotropy results in reduced performance and lower yields.

Churnetski et al. employed a physics-informed machine learning approach to create 3D reconstructions of direct drive plasmas, relying on data collected from a suite of diagnostic instruments at the University of Rochester’s OMEGA Laser System.

The researchers’ 3D reconstruction technique applies a machine learning model trained on radiation-hydrodynamic simulations. To ensure the plasma’s hot spot conditions were reconstructed, the simulations spanned a wide parameter space of many amplitudes and directions of low-mode plasma asymmetries.

The team used these reconstructions to study implosion asymmetries and quantify their effects. They observed that low-mode implosion asymmetries cause a decrease in the neutron yield, compression, and laser energy coupling.

This reconstruction technique also helps the researchers tune their parameters after every test for improved results overall.

“The technique is performed systematically after each OMEGA deuterium-tritium cryogenic implosion and can be used to optimize performance during the experimental campaign by adjusting the initial target and laser conditions based on the 3D reconstruction results,” said author Kristen Churnetski. “The reconstructed plasma profiles provide insight into the implosion performance and hot spot conditions such as compression, pressure, and temperature in 3D.”

In the future, the authors plan to improve this approach by integrating simulation results from multiple time steps and reconstructing the total primary neutron yield to compare with the measured yield.

Source: “Three-dimensional reconstruction of inertial confinement fusion hot-spot plasma from x-ray and nuclear diagnostics on OMEGA,” by Kristen Churnetski, Ka Ming Woo, Wolfgang Theobald, Riccardo Betti, Luke Ceurvorst, Chad James Forrest, Varchas Gopalaswamy, Peter Ver Bryck Heuer, Steven Ivancic, James P. Knauer, Aarne Lees, Michael Michalko, Michael Jonathan Rosenberg, Rahul Shah, Christian Stoeckl, Cliff Thomas, and Sean Patrick Regan, Physics of Plasmas (2025). The article can be accessed at https://doi.org/10.1063/5.0268312 .

This paper is part of the Papers from the 66th Annual Meeting of the APS Division of Plasma Physics Collection, learn more here .