Causal multi-fidelity surrogate forward and inverse models for ICF implosions.

Area of Science:

• Physics
• Computational Science
• Fusion Energy

Background:

• Inertial confinement fusion (ICF) research faces challenges in solving complex inverse problems for design optimization.
• High-dimensional dynamic PDE-constrained optimization problems are often intractable with traditional methods.

Purpose of the Study:

• To develop a novel approach for solving inverse problems in ICF by creating a dynamic, multifidelity reduced-order surrogate model.
• To optimize the radiation temperature drive for reproducing observed deuterium-tritium (DT) interface dynamics.

Main Methods:

• Constructed a causal, dynamic, multifidelity reduced-order surrogate model for the ICF capsule's DT interface dynamics.
• Employed machine learning models trained on surrogate-generated data to solve inverse problems.
• Utilized operator learning, causal architectures, and physical inductive bias for accelerated discovery.

Main Results:

• Demonstrated excellent accuracy of the surrogate interface model in predicting DT interface radius and velocity.
• Successfully used ML models to optimize radiation temperature drive, reproducing observed interface dynamics.
• Identified the most informative times for sampling dynamics from sparse temporal data.

Conclusions:

• The integrated approach of operator learning, causal architectures, and physical inductive bias accelerates discovery, design, and diagnostics in high-energy-density systems.
• This method offers a powerful tool for tackling inverse problems in fusion energy research.