conjugate gradient method More...

#include <dune/istl/solvers.hh>

Public Types
typedef X	domain_type
	Type of the domain of the operator to be inverted.

typedef X	range_type
	Type of the range of the operator to be inverted.

typedef X::field_type	field_type
	The field type of the operator.

typedef FieldTraits< field_type >::real_type	real_type
	The real type of the field type (is the same if using real numbers, but differs for std::complex)

typedef SimdScalar< real_type >	scalar_real_type
	scalar type underlying the field_type

Public Member Functions
	CGSolver (LinearOperator< X, X > &op, Preconditioner< X, X > &prec, scalar_real_type reduction, int maxit, int verbose, bool condition_estimate)
	Constructor to initialize a CG solver. More...

	CGSolver (LinearOperator< X, X > &op, ScalarProduct< X > &sp, Preconditioner< X, X > &prec, scalar_real_type reduction, int maxit, int verbose, bool condition_estimate)
	Constructor to initialize a CG solver. More...

virtual void	apply (X &x, X &b, InverseOperatorResult &res)
	Apply inverse operator. More...

virtual void	apply (X &x, X &b, double reduction, InverseOperatorResult &res)
	Apply inverse operator with given reduction factor. More...

virtual SolverCategory::Category	category () const
	Category of the solver (see SolverCategory::Category)

Protected Member Functions
void	printHeader (std::ostream &s) const
	helper function for printing header of solver output

void	printOutput (std::ostream &s, const CountType &iter, const DataType &norm, const DataType &norm_old) const
	helper function for printing solver output

void	printOutput (std::ostream &s, const CountType &iter, const DataType &norm) const
	helper function for printing solver output

Detailed Description

template<class X>
class Dune::CGSolver< X >

conjugate gradient method

Constructor & Destructor Documentation

◆ CGSolver() [1/2]

template<class X >

Dune::CGSolver< X >::CGSolver	(	LinearOperator< X, X > &	op,
		Preconditioner< X, X > &	prec,
		scalar_real_type	reduction,
		int	maxit,
		int	verbose,
		bool	condition_estimate
	)

inline

Constructor to initialize a CG solver.

Parameters

condition_estimate Whether to calculate an estimate of the condition number. The estimate is given in the InverseOperatorResult returned by apply(). This is only supported for float and double field types.

◆ CGSolver() [2/2]

template<class X >

Dune::CGSolver< X >::CGSolver	(	LinearOperator< X, X > &	op,
		ScalarProduct< X > &	sp,
		Preconditioner< X, X > &	prec,
		scalar_real_type	reduction,
		int	maxit,
		int	verbose,
		bool	condition_estimate
	)

inline

Constructor to initialize a CG solver.

Parameters

condition_estimate Whether to calculate an estimate of the condition number. The estimate is given in the InverseOperatorResult returned by apply(). This is only supported for float and double field types.

Member Function Documentation

◆ apply() [1/2]

virtual void Dune::IterativeSolver< X, X >::apply	(	X &	x,
		X &	b,
		double	reduction,
		InverseOperatorResult &	res
	)

inlinevirtualinherited

Apply inverse operator with given reduction factor.

apply inverse operator, with given convergence criteria.

Warning: Right hand side b may be overwritten!

Parameters

x	The left hand side to store the result in.
b	The right hand side
reduction	The minimum defect reduction to achieve.
res	Object to store the statistics about applying the operator.

Exceptions

SolverAbort When the solver detects a problem and cannot continue

Implements Dune::InverseOperator< X, X >.

◆ apply() [2/2]

template<class X >

virtual void Dune::CGSolver< X >::apply	(	X &	x,
		X &	b,
		InverseOperatorResult &	res
	)

inlinevirtual

Apply inverse operator.

Apply inverse operator,.

Warning: Note: right hand side b may be overwritten!

Parameters

x	The left hand side to store the result in.
b	The right hand side
res	Object to store the statistics about applying the operator.

Exceptions

SolverAbort When the solver detects a problem and cannot continue

Note: Currently, the CGSolver aborts when a NaN or infinite defect is detected. However, -ffinite-math-only (implied by -ffast-math) can inhibit a result from becoming NaN that really should be NaN. E.g. numeric_limits<double>::quiet_NaN()*0.0==0.0 with gcc-5.3 -ffast-math.

Implements Dune::InverseOperator< X, X >.

References Dune::InverseOperatorResult::clear(), Dune::InverseOperatorResult::conv_rate, Dune::InverseOperatorResult::converged, DUNE_THROW, Dune::Timer::elapsed(), Dune::InverseOperatorResult::elapsed, Dune::Hybrid::ifElse(), Dune::InverseOperatorResult::iterations, Dune::InverseOperator< X, X >::printHeader(), Dune::InverseOperator< X, X >::printOutput(), Dune::InverseOperatorResult::reduction, and Dune::BCRSMatrix< B, A >::row_wise.