Description
Unlike current tokamaks, next-generation fusion devices like ITER will operate near marginal turbulence stability. ITER's Q=10 scenarios feature heating power densities of ~0.12 MW·m−3, which is an order of magnitude lower than ASDEX Upgrade or DIII-D (~2 MW·m−3). Under such conditions, slow, large-scale modes are likely to play as important a role as, if not more important than, in current-day machines. This raises questions about whether nonlinear saturation mechanisms remain the same in marginal regimes, and the impact this may have on transport and modelling.
To elucidate this physics, we systematically compare predictions from fully kinetic, flux-driven gyrokinetic simulations (GYSELA) with gradient-driven (GKW) and quasilinear (QuaLiKiz) models, all under identical plasma conditions with gyrokinetic ions and trapped kinetic electrons. As in the adiabatic electron case [1], all models agree in far-from-marginal regimes. However, near marginality, the gradient-driven and quasilinear approaches underpredict heat, momentum and particle transport by factors.
This discrepancy stems from a fundamental shift in turbulence saturation. Notably, we observe a 45% increase in the free energy stored in coherent zonal structures such as staircases, as well as a 60% rise in propagating turbulent patterns (avalanches and spreading). Both of these are enabled by profile relaxation, hence the intrinsic “multi-scale” aspect of this physics. The flux-gradient relationship becomes non-monotonic. Close to marginality, the relative energy content of organized flows over turbulence increases. These flows also become steadier and more long-lasting by at least 50%. Identical kinetic profiles can produce differences in predicted fluxes of over 40%. While Kubo numbers reach or exceed unity, the core assumptions of the quasilinear reduction remain robust. These findings highlight the need for new saturation rules. We are currently working on two promising new approaches to developing accurate, computationally efficient reduced transport models for near-marginal operation: one based on symbolic machine learning, and the other on a gas of interacting dipoles.
[1] C. Gillot, G. Dif-Pradalier, Y. Sarazin et al. Plasma Phys. Control. Fusion 65 (2023) 055012