Description
High-fidelity boundary turbulence models, which properly address the highly anisotropic, multi-scale, and multi-physics dynamics of this region, are essential for operating and guiding the design of future fusion devices. A commonly used framework to study turbulence in the boundary region is the two-fluid drift-reduced plasma model, which captures turbulent phenomena responsible for transport across the magnetic field, parallel streaming, and losses at the vessel walls.
The high computational cost of the two-fluid simulations severely limits their application to the simulation of large devices such as ITER. To overcome current limitations, time integration schemes should be tailored to these models, taking into account the strong coupling between fast and slow temporal scales that makes their stable resolution especially challenging.
In this work, we develop and implement the globally stiffly accurate IMEX Runge–Kutta scheme BPR(3,5,3) for the two-fluid plasma model in the GBS code. The fastest plasma dynamics, namely the shear Alfvén waves (SAWs), parallel viscosity, and parallel heat conduction, are treated implicitly, enabling larger time steps than the Courant–Friedrichs–Lewy (CFL) limit while avoiding the full cost of fully implicit integration.
To accelerate the implicit solution of the SAW system, we construct a 3D physics-based preconditioner tailored to this problem, inspired by techniques developed in the magnetohydrodynamic (MHD) context. Both the solver and the preconditioner are implemented using the PETSc library. The new time-stepping algorithm has been validated and assessed for scalability.
By combining this physics-based preconditioner with an efficient IMEX time integration strategy, we demonstrate both numerical stability and significant computational gains in simulations of two-fluid edge turbulence. This approach enables faster simulations while ensuring a proper treatment of parallel viscosity and parallel heat conduction, resulting in more accurate and reliable results.