Reconstruction of residual kinetic energy in inertial-confinement fusion implosions at the National Ignition Facility using multiple heterogeneous data sources and neural networks
Phys. Rev. E 113, 065213 – Published 29 June, 2026
DOI: https://doi.org/10.1103/wvj3-xd6k
Abstract
Three-dimensional (3D) asymmetries and their associated residual kinetic energy (RKE) are a major performance-degradation mechanism in inertial-confinement fusion (ICF) implosions at the National Ignition Facility (NIF). These asymmetries can be diagnosed with the three neutron imaging systems fielded with orthogonal views to the implosion, as well as a suite of real-time neutron activation diagnostics (RTNADs) and neutron spectrometers. Conventional tomographic reconstructions are typically used to reconstruct the 3D morphology of the hot spot and surrounding dense fuel in an implosion using neutron images, but the reconstruction problem is ill-posed with only three imaging lines of sight. In this work a 3D reconstruction technique has been developed and used to overcome these limitations by combining information from the neutron images, RTNAD data, and neutron spectra to infer RKE. Markov chain Monte Carlo is used along with a group of neural networks to fit a physics model to the observed data to robustly reconstruct the 3D morphology of the hot spot and dense fuel in a NIF implosion. Bayesian statistics provides quantification of the uncertainty in the inference. Initial inferences find that the normalized RKE on several NIF experiments is between 2% and 5% of the total implosion energy, and is inversely related to the yield of an implosion, as expected. This technique is being developed to help guide designs toward minimizing asymmetries in ICF implosions and achieving higher performance.
