AI-enhanced Large Eddy Simulation for plasma fluid turbulence

Turbulence in fluids, gases, and plasmas remains an open problem of both practical and fundamental importance. Its irreducible complexity usually cannot be tackled computationally in a brute-force style. Combining Large Eddy Simulation (LES) techniques with Machine Learning (ML) allows us to only retain the largest dynamics explicitly, while small-scale dynamics are described by an ML-based sub-grid-scale model. Applying this novel approach to self-driven plasma turbulence allows us to remove large parts of the inertial range, reducing the computational effort by about three orders of magnitude, all while retaining the full statistical description of the turbulent systems' physical properties. Furthermore, this approach exhibits remarkable robustness and precision even when extrapolating far beyond training parameter space. In this talk, I will discuss recent results and their implications for efficiently exploring high-dimensional parameter spaces in magnetic confined fusion.