GenICF: Benchmarking Generative Methods for Inverse Modeling in Inertial Confinement Fusion

Ricardo Luna Gutierrez, Vineet Gundecha, Rahman Ejaz, Varchas Gopalaswamy, Riccardo Betti, Sahand Ghorbanpour, Aarne Lees, Soumyendu Sarkar

Keywords: ICF, Fusion, Inverse Modeling, Generative models, Benchmarking, Laser Pulse

TL;DR: GenICF presents a systematic benchmark of generative AI methods for designing laser pulse shapes in inertial confinement fusion, comparing data-driven approaches with physics constraints against costly traditional simulation-based design methods.

Abstract: The realization of practical inertial fusion energy critically depends on the ability to design laser pulse shapes (LPs) that robustly drive implosions while satisfying stringent physical constraints. Conventional LP design relies on large-scale radiation-hydrodynamic simulations coupled with manual iterative refinement, resulting in high computational cost and limited scalability. Recent advances in generative modeling provide an alternative pathway for data-driven inverse design. In this work, we present the first systematic comparison of generative paradigms for LP design. To enforce physical plausibility, we introduce domain-specific loss formulations. Our results constitute the first principled comparison of generative methods for LP design in inertial confinement fusion, providing guidance for the development of scalable, physics-constrained design frameworks for fusion energy applications.

Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)

Submission Number: 2174