amrex::MLAlgMG Class Reference

Algebraic system of one MG level of AMR level 0 of an MLLinOp, solved with AlgMG. More...

#include <AMReX_MLAlgMG.H>

Classes
struct	Impl

Public Member Functions
	MLAlgMG (int mglev, BoxArray const &grids, DistributionMapping const &dmap, Geometry const &geom, iMultiFab const &owner_mask, iMultiFab const &dirichlet_mask, MLNodeLinOp const &linop)
	MLAlgMG (int mglev, BoxArray const &grids, DistributionMapping const &dmap, Geometry const &geom, FabFactory< FArrayBox > const &factory, iMultiFab const *overset_mask, Real ascalar, Real bscalar, MultiFab const *acoef, Array< MultiFab const *, 3 > const &bcoef, MultiFab const *eb_bcoef, LayoutData< GpuArray< int, 2 *3 > > const &bctype, LayoutData< GpuArray< Real, 2 *3 > > const &bcl, int maxorder)
	Cell-centered operator alpha*a - beta*div(b grad): rows are all cells except those of fully covered boxes.
	~MLAlgMG ()
	MLAlgMG (MLAlgMG const &)=delete
	MLAlgMG (MLAlgMG &&)=delete
MLAlgMG &	operator= (MLAlgMG const &)=delete
MLAlgMG &	operator= (MLAlgMG &&)=delete
void	solve (MultiFab &soln, MultiFab const &rhs, Real reltol, Real abstol, int maxiter)
	Solve A x = b with zero initial guess.
void	applyVcycle (MultiFab &soln, MultiFab const &rhs)
	One AlgMG V-cycle with zero initial guess: the preconditioner form.
AlgMG< Real > &	solver () noexcept
	The solver, for settings such as verbosity and Krylov acceleration.

Detailed Description

Algebraic system of one MG level of AMR level 0 of an MLLinOp, solved with AlgMG.

Built by MLLinOp::makeAlgMG under the ParallelContext communicator of that level. The matrix is assembled once; solve() can be called repeatedly.

Constructor & Destructor Documentation

MLAlgMG() [1/4]

amrex::MLAlgMG::MLAlgMG	(	int	mglev,
		BoxArray const &	grids,
		DistributionMapping const &	dmap,
		Geometry const &	geom,
		iMultiFab const &	owner_mask,
		iMultiFab const &	dirichlet_mask,
		MLNodeLinOp const &	linop
	)

Nodal operator: rows are the nodes owned by this rank that are not Dirichlet nodes.

MLAlgMG() [2/4]

amrex::MLAlgMG::MLAlgMG	(	int	mglev,
		BoxArray const &	grids,
		DistributionMapping const &	dmap,
		Geometry const &	geom,
		FabFactory< FArrayBox > const &	factory,
		iMultiFab const *	overset_mask,
		Real	ascalar,
		Real	bscalar,
		MultiFab const *	acoef,
		Array< MultiFab const *, 3 > const &	bcoef,
		MultiFab const *	eb_bcoef,
		LayoutData< GpuArray< int, 2 *3 > > const &	bctype,
		LayoutData< GpuArray< Real, 2 *3 > > const &	bcl,
		int	maxorder
	)

Cell-centered operator alpha*a - beta*div(b grad): rows are all cells except those of fully covered boxes.

The a coefficient may be nullptr (zero); the b coefficients are required. eb_bcoef is the EB Dirichlet b coefficient or nullptr. bctype and bcl give the domain BC type and location per box and face, and maxorder the order of the Dirichlet boundary stencil (2 or 3).

~MLAlgMG()

amrex::MLAlgMG::~MLAlgMG ( )

default

MLAlgMG() [3/4]

amrex::MLAlgMG::MLAlgMG ( MLAlgMG const &  )

delete

MLAlgMG() [4/4]

amrex::MLAlgMG::MLAlgMG ( MLAlgMG &&  )

delete

Member Function Documentation

applyVcycle()

void amrex::MLAlgMG::applyVcycle	(	MultiFab &	soln,
		MultiFab const &	rhs
	)

One AlgMG V-cycle with zero initial guess: the preconditioner form.

operator=() [1/2]

MLAlgMG & amrex::MLAlgMG::operator= ( MLAlgMG &&  )

delete

operator=() [2/2]

MLAlgMG & amrex::MLAlgMG::operator= ( MLAlgMG const &  )

delete

solve()

void amrex::MLAlgMG::solve	(	MultiFab &	soln,
		MultiFab const &	rhs,
		Real	reltol,
		Real	abstol,
		int	maxiter
	)

Solve A x = b with zero initial guess.

soln is overwritten. Converged when the residual 2-norm is below max(reltol times the rhs 2-norm, abstol).

solver()

AlgMG< Real > & amrex::MLAlgMG::solver ( )

noexcept

The solver, for settings such as verbosity and Krylov acceleration.

---

The documentation for this class was generated from the following files:

• /home/runner/work/amrex/amrex/Src/LinearSolvers/MLMG/AMReX_MLAlgMG.H
• /home/runner/work/amrex/amrex/Src/LinearSolvers/MLMG/AMReX_MLAlgMG.cpp