Multigrid Inputs

incflo uses AMReX’s multigrid solver, MLMG, for the projections and for the implicit diffusion solves. The AMReX documentation on Linear Solvers describes the solver and its parameters (e.g. the available bottom solvers).

Projections

The MAC and nodal projections are done by AMReX-Hydro’s MacProjector and NodalProjector classes, which read their own options (verbosity, maximum number of iterations, bottom solver, …) with the prefix “mac_proj” or “nodal_proj”. These are documented in AMReX-Hydro’s documentation on projections inputs. The cell-centered projection (incflo.use_cc_proj = 1) is also done with the MacProjector, so it uses the “mac_proj” options.

In addition, incflo reads the following, which must be preceded by “mac_proj” or “nodal_proj”:

	Description	Type	Default
mg_rtol	Relative tolerance	Real float	1.e-11 1.e-4
mg_atol	Absolute tolerance	Real float	1.e-14 1.e-7
mg_max_coarsening_level	Maximum number of coarser levels to allow. If set to 0, the bottom solver will be called at the current level	Int	100

Diffusion

The diffusion solves are set up by incflo itself. Their options must be preceded by “scalar_diffusion” (tracers, temperature and, with incflo.use_tensor_solve = false, the component-wise velocity solve) or “tensor_diffusion” (the velocity solve with incflo.use_tensor_solve = true):

	Description	Type	Default
verbose	Verbosity of incflo’s diffusion routines	Int	0
mg_verbose	Verbosity of the multigrid solver	Int	0
mg_bottom_verbose	Verbosity of the bottom solver (“scalar_diffusion” also accepts bottom_verbose)	Int	0
mg_max_iter	Maximum number of multigrid iterations	Int	100
mg_bottom_maxiter	Maximum number of iterations in the bottom solver	Int	100
mg_max_fmg_iter	Number of full multigrid cycles before switching to V-cycles	Int	0
mg_max_coarsening_level	Maximum number of coarser levels to allow. If set to 0, the bottom solver will be called at the current level	Int	100
mg_maxorder	Maximum order (2, 3 or 4) of the interpolation used at boundaries	Int	2
mg_rtol	Relative tolerance	Real float	1.e-11 1.e-4
mg_atol	Absolute tolerance	Real float	1.e-14 1.e-7
bottom_solver	Which bottom solver to use: smoother or hypre. Any other value gives MLMG’s default bottom solver for these operators, BiCGStab	String	bicgstab
num_pre_smooth	Number of smoother iterations when going down the V-cycle	Int	2
num_post_smooth	Number of smoother iterations when going up the V-cycle	Int	2
use_gauss_seidel	“tensor_diffusion” only: use Gauss-Seidel rather than Jacobi smoothing (not used when there is an EB)	Bool	true

The hypre bottom solver requires incflo to be built with HYPRE support (USE_HYPRE = TRUE or -DINCFLO_HYPRE=yes). hypre’s own options take the prefix “hypre”; see the External Solvers section of the AMReX documentation.