Features of gyrokinetic turbulence in tokamaks and possible implications for multi-fidelity AI models
Greg Hammett
Princeton Plasma Physics Lab
Computational Sciences
Greg Hammett, Princeton Plasma Physics Laboratory High-fidelity gyrokinetic simulations are fairly accurate for many core-turbulence regimes and are rapidly improving in the challenging edge/SOL region, but they are computationally expensive, particularly inside optimization loops for the design of fusion power plants. This motivates multi-fidelity strategies that combine high-fidelity simulation codes with AI techniques in various ways. I will survey some of the things being learned about plasma turbulence from high-fidelity simulations, including some results from the Gkeyll code (see https://gkeyll.readthedocs.io and the references below). We will discuss some of the challenges of tokamak turbulence for AI models and how multi-fidelity techniques might help. We will describe some of the important plasma physics problems remaining in fusion research where AI tools can help make progress. Example references: A. C. D. Hoffmann, T. N. Bernard, M. Francisquez, G. W. Hammett, A. Hakim, J. Boedo, R. Rizkallah, C. K. Tsui, the TCV team (2025). https://arxiv.org/abs/2510.11874 T. N. Bernard, F. D. Halpern, M. Francisquez, G.W. Hammett, A. Marinoni (2024). https://doi.org/10.1088/1361-6587/ad8186