Mesh Refinement for Anisotropic Diffusion in Magnetized Plasmas

Highly accurate simulation of plasma transport is needed for the successful design and operation of magnetically confined fusion reactors. Unfortunately, the extreme anisotropy present in magnetized plasmas results in thin boundary layers that are expensive to resolve. This work investigates how mesh refinement strategies might reduce that expense to allow for more efficient simulation. It is first verified that higher order discretization only realizes the proper rate of convergence once the mesh resolves the thin boundary layer, motivating the focusing of refinement on the boundary layer. Three mesh refinement strategies are investigated: one that focuses the refinement across the layer by using rectangular elements with a ratio equal to the boundary layer width, one that allows for exponential growth in mesh spacing away from the layer, and one adaptive strategy utilizing the established Zienkiewicz and Zhu error estimator. Across 4 two-dimensional test cases with high anisotropy, the adaptive mesh refinement strategy consistently achieves the same accuracy as uniform refinement using orders of magnitude less degrees of freedom. In the test case where the magnetic field is aligned with the mesh, the other refinement strategies also show substantial improvement in efficiency. This work also includes a discussion generalizing the results to larger magnetic anisotropy ratios and to three-dimensional problems. It is shown that isotropic mesh refinement requires degrees of freedom on the order of either the layer width (2D) or the square of the layer width (3D), whereas anisotropic refinement requires a number on the order of the log of layer width for all dimensions. It is also shown that the number of conjugate gradient iterations scales as a power of layer width when preconditioned with algebraic multigrid, whereas the number is independent of layer width when preconditioned with ILU.