amrex::PCG< V, M > Class Template Reference

Preconditioned conjugate gradient solver. More...

#include <AMReX_PCG.H>

Public Types
using	RT = typename M::RT

Public Member Functions
	PCG ()=default
void	define (M &linop)
void	solve (V &a_sol, V const &a_rhs, RT a_tol_rel, RT a_tol_abs, int a_its=-1)
	Solve the linear system.
void	setInitialGuessNonzero (bool flag)
void	setVerbose (int v)
	Sets verbosity.
void	setMaxIters (int niters)
	Sets the max number of iterations.
int	getNumIters () const
	Gets the number of iterations.
int	getStatus () const
	Gets the solver status.
RT	getResidualNorm () const
	Gets the 2-norm of the residual.

Detailed Description

template<typename V, typename M>
class amrex::PCG< V, M >

Preconditioned conjugate gradient solver.

Requires a symmetric positive definite operator and preconditioner. The operator type M must provide the same interface as for amrex::GMRES: makeVecRHS, makeVecLHS, norm2, dotProduct, assign, increment, linComb, scale, setToZero, apply and precond.

Member Typedef Documentation

RT

template<typename V , typename M >

using amrex::PCG< V, M >::RT = typename M::RT

Constructor & Destructor Documentation

PCG()

template<typename V , typename M >

amrex::PCG< V, M >::PCG ( )

default

Member Function Documentation

define()

template<typename V , typename M >

void amrex::PCG< V, M >::define ( M &  linop )

inline

Defines with a reference to M. It's the user's responsibility to keep the M object alive for PCG to be functional. This function must be called before solve() can be called.

getNumIters()

template<typename V , typename M >

int amrex::PCG< V, M >::getNumIters ( ) const

inline

Gets the number of iterations.

getResidualNorm()

template<typename V , typename M >

RT amrex::PCG< V, M >::getResidualNorm ( ) const

inline

Gets the 2-norm of the residual.

getStatus()

template<typename V , typename M >

int amrex::PCG< V, M >::getStatus ( ) const

inline

Gets the solver status.

0 converged, 1 max iterations reached, 2 p.Ap is zero or has changed sign (A not definite), 3 r.z is zero or has changed sign (preconditioner not definite). The sign is taken from the first r.z, so negative definite systems are solved as well. A NaN or infinity counts as the matching breakdown.

setInitialGuessNonzero()

template<typename V , typename M >

void amrex::PCG< V, M >::setInitialGuessNonzero ( bool  flag )

inline

Use a_sol passed to solve() as the initial guess (default: false, i.e., start from zero).

setMaxIters()

template<typename V , typename M >

void amrex::PCG< V, M >::setMaxIters ( int  niters )

inline

Sets the max number of iterations.

setVerbose()

template<typename V , typename M >

void amrex::PCG< V, M >::setVerbose ( int  v )

inline

Sets verbosity.

solve()

template<typename V , typename M >

void amrex::PCG< V, M >::solve	(	V &	a_sol,
		V const &	a_rhs,
		RT	a_tol_rel,
		RT	a_tol_abs,
		int	a_its = -1
	)

Solve the linear system.

The initial guess is zero unless setInitialGuessNonzero(true) was called, in which case it is the value of a_sol on entry.

Parameters

a_sol	unknowns, i.e., x in A x = b.
a_rhs	RHS, i.e., b in A x = b.
a_tol_rel	relative tolerance, relative to the initial residual norm, which is ||b|| for a zero initial guess.
a_tol_abs	absolute tolerance.
a_its	optional argument specifying the maximum number of iterations.

---

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

• /home/runner/work/amrex/amrex/Src/LinearSolvers/AMReX_PCG.H