A multi-fidelity approach in boundary plasma modeling
Ben Zhu
Columbia University
Department of Applied Physics and Applied Mathematics
The boundary region of toroidally confined plasma - spanning from the inside of the last closed flux surface to the first wall - arguably represents one of the most formidable computational challenges in fusion research. Capturing the nonlinear interplay between turbulent dynamics, flows, background profiles, and neutral particle dynamics at transport timescales requires high-fidelity kinetic simulations that are often prohibitively expensive for routine use. Even standard fluid-based global simulations remain too computationally intensive for rapid iterative design or real-time control. Consequently, the field has traditionally relied on reduced models based on crude or oversimplified assumptions to achieve necessary speedups.
The multi-fidelity approach provides a synergistic solution to this computational bottleneck. By leveraging a hierarchy of self-consistent models across different fidelity levels, this approach utilizes either direct coupling or emerging techniques, such as transfer learning, to achieve high-accuracy predictions with significantly reduced computational overhead. Current applications in boundary modeling, including turbulence-transport coupling and quasi-real-time predictions of divertor plasma solutions, will be presented. Finally, outstanding challenges and opportunities, such as the integration of data-driven surrogates into sophisticated multi-fidelity hierarchies, will be discussed.