5D Neural Surrogates for Nonlinear Gyrokinetic Simulations of Plasma Turbulence

• DOI: 10.48550/arxiv.2502.07469
• Publication Date: 2025-02-11
• Authors: Gianluca Galletti; Fabian Paischer; Paul Setinek; W. A. Hornsby; Lorenzo Zanisi; Nicola Carey; S. Pamela; J. Brandstetter
• Keywords: Gyrokinetics

Nuclear fusion plays a pivotal role in the quest for reliable and sustainable energy production. A major roadblock to achieving commercially viable power is understanding plasma turbulence, which can significantly degrade confinement. Modelling turbulence crucial design performing scenarios next-generation reactor-class devices current experimental machines. The nonlinear gyrokinetic equation underpinning modelling evolves 5D distribution function over time. Solving this numerically extremely expensive, requiring up weeks single run converge, making it unfeasible iterative optimisation control studies. In work, we propose method training neural surrogates simulations. Our extends hierarchical vision transformer five dimensions trained on adiabatic electron approximation. We demonstrate that our model accurately infer downstream physical quantities such as heat flux time trace electrostatic potentials single-step predictions two orders of magnitude faster than numerical codes. work paves way towards simulations accelerate deployment commercial production via nuclear fusion.