Title: Causal Multi-fidelity Surrogate Forward and Inverse Models for ICF Implosions

Abstract:

Advancements in inertial confinement fusion (ICF) depend on resolving inverse problems that connect experimental data to simulation inputs, thereby facilitating design optimization. Yet, these high-dimensional, dynamic optimization problems constrained by partial differential equations are often too complex or even impossible to solve directly. Recent studies suggest that focusing on specific robust features can make solving such inverse problems feasible. In this work, we focus on the deuterium-tritium (DT) interface within ICF capsules, developing a causal, dynamic, multifidelity reduced-order surrogate model. This model maps the time-varying radiation temperature drive to the radius and velocity dynamics of the interface. Specifically, the surrogate addresses an ordinary differential equation (ODE) representation of DT interface behavior. It is built by training a controller on a foundational analytical model, utilizing both low- and high-fidelity simulation data that account for radiation energy group structures.

After validating the surrogate model’s high accuracy, we employ machine learning (ML) techniques with data generated by the surrogate to tackle inverse problems. These ML models optimize the radiation temperature drive to match observed interface dynamics. Furthermore, when dealing with sparse temporal snapshots, the ML framework identifies the most critical time points for sampling the dynamics. Ultimately, this study illustrates how integrating operator learning, causal architectures, and physical inductive biases can significantly speed up discovery, design, and diagnostic processes in high-energy-density systems.

Source: arXiv Generated at: 2026-06-04 00:00:00 UTC