We present an efficient MPI-parallel geometric multigrid library for quadtree (2D) or octree (3D) grids with adaptive refinement. Cartesian 2D/3D and cylindrical 2D geometries are supported, with second-order discretizations for the elliptic operators. Periodic, Dirichlet, and Neumann boundary conditions can be handled, as well as free-space boundary conditions for 3D Poisson problems, for which we use an FFT-based solver on the coarse grid. Scaling results up to 1792 cores are presented. The library can be used to extend adaptive mesh refinement frameworks with an elliptic solver, which we demonstrate by coupling it to MPI-AMRVAC. Several test cases are presented in which the multigrid routines are used to control the divergence of the magnetic field in magnetohydrodynamic simulations.

Additional Metadata
Keywords Adaptive mesh refinement, Divergence cleaning, Elliptic solver, Multigrid, Octree
Persistent URL dx.doi.org/10.1016/j.cpc.2019.106866
Journal Computer Physics Communications
Citation
Teunissen, H.J, & Keppens, R. (2019). A geometric multigrid library for quadtree/octree AMR grids coupled to MPI-AMRVAC. Computer Physics Communications. doi:10.1016/j.cpc.2019.106866