White Paper: “Multi-Fidelity Methods for Fusion Energy”
Fusion energy represents one of the most complex scientific and engineering challenges, characterized by multi-scale, multi-physics problems spanning wide ranges of physical and temporal scales. First-principles, high-fidelity simulations require the world’s largest supercomputers running for days or weeks to simulate a single plasma realization, severely restricting essential many-query tasks such as design optimization, uncertainty quantification (UQ), and real-time control.
Multi-Fidelity Methods (MFM) systematically leverage models of varying precision, accuracy, resolution, and physical completeness to accelerate convergence and reduce time-to-solution. These approaches have demonstrated computational cost reductions of orders of magnitude across diverse scientific domains. Fusion energy is well-positioned to benefit given its existing ecosystem of models at various fidelities and the urgent need for commercial deployment.
The program addressed critical research areas, beginning with the integration of traditional plasma physics hierarchies with modern data-driven approaches, including proper orthogonal decomposition, dynamic mode decomposition, neural networks, and foundation models. Participants explored verification, validation, and UQ methodologies for trustworthy modeling, differentiable programming, tensor networks, and formal correctness verification. The program examined multi-scale coupling, integrated systems engineering, and Bayesian optimization with high-fidelity metrics for device design. Discussions covered model-based control architectures, sparse-diagnostic state estimation, and controller orchestration for real-time plasma control. Finally, the program investigated AI agents for autonomous simulation, code development, and scientific discovery, demonstrating their potential to transform fusion research workflows.
This program identified opportunities for multi-fidelity methods to transform fusion modeling to achieve robust, computationally tractable frameworks for design, optimization, and control, thereby accelerating the path to commercial fusion energy.
