Observation-guided diagnostics of one-dimensional inertial confinement fusion implosion dynamics with diffusion modeling

Abstract

Inertial confinement fusion (ICF) experiments require precise control over high-dimensional capsule-design and pulse-drive parameter spaces to achieve thermonuclear ignition. Sparse experimental diagnostics and partial observations limit the resolution of complex implosion dynamics and make high-precision inverse inference difficult. In this work, we adopt a conditional diffusion model framework as an advanced physics-diagnostic tool to model one-dimensional laser/radiation-driven ICF implosions and to quantify forward diagnostic recovery and inverse parameter design under partial observations. An Euler forward-gradient-corrected sampling method is introduced, and actual observational guidance is incorporated to ensure that the predicted results remain consistent with both sparse diagnostic information and the radiation-hydrodynamic state manifold represented by the simulation data. Using this framework, we perform global sensitivity analysis with accumulated local effects and vector projection-based sensitivity indices to identify the dominant drivers of systematic variations across 35 physical output variables, including neutron yield and hot-spot pressure. In addition, ensemble sampling is used to reveal how stochastic uncertainty propagates into the design space and to identify specific parameters that are prone to structural instability. These results provide an efficient diagnostic route for optimizing target-capsule performance and pulse-shape architecture.