This class implements the GCRODR iteration, where a single-vector Krylov subspace is constructed. The QR decomposition of block, upper Hessenberg matrix is performed each iteration to update the least squares system and give the current linear system residuals. More...

#include <BelosGCRODRIter.hpp>

Inheritance diagram for Belos::GCRODRIter< ScalarType, MV, OP, DM >:

[legend]

Public Types
typedef MultiVecTraits< ScalarType, MV, DM >	MVT

typedef OperatorTraits< ScalarType, MV, OP >	OPT

typedef DenseMatTraits< ScalarType, DM >	DMT

typedef Teuchos::ScalarTraits< ScalarType >	SCT

typedef SCT::magnitudeType	MagnitudeType

Public Member Functions
Constructors/Destructor
	GCRODRIter (const Teuchos::RCP< LinearProblem< ScalarType, MV, OP, DM > > &problem, const Teuchos::RCP< OutputManager< ScalarType > > &printer, const Teuchos::RCP< StatusTest< ScalarType, MV, OP, DM > > &tester, const Teuchos::RCP< MatOrthoManager< ScalarType, MV, OP, DM > > &ortho, Teuchos::ParameterList &params)
	GCRODRIter constructor with linear problem, solver utilities, and parameter list of solver options.

virtual	~GCRODRIter ()
	Destructor.

Solver methods
void	iterate ()
	This method performs block Gmres iterations until the status test indicates the need to stop or an error occurs (in which case, an std::exception is thrown).

void	initialize (GCRODRIterState< ScalarType, MV, DM > &newstate)
	Initialize the solver to an iterate, providing a complete state.

void	initialize ()
	Initialize the solver with empty data. Calling this method will result in error, as GCRODRIter must be initialized with a valid state.

GCRODRIterState< ScalarType, MV, DM >	getState () const
	Get the current state of the linear solver.

Status methods
int	getNumIters () const
	Get the current iteration count.

void	resetNumIters (int iter=0)
	Reset the iteration count.

Teuchos::RCP< const MV >	getNativeResiduals (std::vector< MagnitudeType > *norms) const
	Get the norms of the residuals native to the solver.

Teuchos::RCP< MV >	getCurrentUpdate () const
	Get the current update to the linear system.

void	updateLSQR (int dim=-1)
	Method for updating QR factorization of upper Hessenberg matrix.

int	getCurSubspaceDim () const
	Get the dimension of the search subspace used to generate the current solution to the linear problem.

int	getMaxSubspaceDim () const
	Get the maximum dimension allocated for the search subspace.

Public Member Functions inherited from Belos::GCRODRIteration< ScalarType, MV, OP, DM >
virtual	~GCRODRIteration ()

Public Member Functions inherited from Belos::Iteration< ScalarType, MV, OP, DM >
	Iteration ()
	Default Constructor.

virtual	~Iteration ()
	Destructor.

virtual Teuchos::RCP< const MV >	getNativeResiduals (std::vector< typename Teuchos::ScalarTraits< ScalarType >::magnitudeType > *norms) const=0
	Get the residuals native to the solver.

Accessor methods
const LinearProblem< ScalarType, MV, OP, DM > &	getProblem () const
	Get a constant reference to the linear problem.

int	getNumBlocks () const
	Get the maximum number of blocks used by the iterative solver in solving this linear problem.

void	setNumBlocks (int numBlocks)
	Set the maximum number of blocks used by the iterative solver.

int	getBlockSize () const
	Get the blocksize to be used by the iterative solver in solving this linear problem.

void	setBlockSize (int blockSize)
	Set the blocksize.

void	setSize (int recycledBlocks, int numBlocks)
	Set the maximum number of blocks used by the iterative solver and the number of recycled vectors.

bool	isInitialized ()
	States whether the solver has been initialized or not.

Detailed Description

template<class ScalarType, class MV, class OP, class DM>
class Belos::GCRODRIter< ScalarType, MV, OP, DM >

This class implements the GCRODR iteration, where a single-vector Krylov subspace is constructed. The QR decomposition of block, upper Hessenberg matrix is performed each iteration to update the least squares system and give the current linear system residuals.

Author: Michael Parks and Heidi Thornquist

Definition at line 130 of file BelosGCRODRIter.hpp.

Member Typedef Documentation

◆ MVT

template<class ScalarType , class MV , class OP , class DM >

typedef MultiVecTraits<ScalarType,MV,DM> Belos::GCRODRIter< ScalarType, MV, OP, DM >::MVT

Definition at line 137 of file BelosGCRODRIter.hpp.

◆ OPT

template<class ScalarType , class MV , class OP , class DM >

typedef OperatorTraits<ScalarType,MV,OP> Belos::GCRODRIter< ScalarType, MV, OP, DM >::OPT

Definition at line 138 of file BelosGCRODRIter.hpp.

◆ DMT

template<class ScalarType , class MV , class OP , class DM >

typedef DenseMatTraits<ScalarType,DM> Belos::GCRODRIter< ScalarType, MV, OP, DM >::DMT

Definition at line 139 of file BelosGCRODRIter.hpp.

◆ SCT

template<class ScalarType , class MV , class OP , class DM >

typedef Teuchos::ScalarTraits<ScalarType> Belos::GCRODRIter< ScalarType, MV, OP, DM >::SCT

Definition at line 140 of file BelosGCRODRIter.hpp.

◆ MagnitudeType

template<class ScalarType , class MV , class OP , class DM >

typedef SCT::magnitudeType Belos::GCRODRIter< ScalarType, MV, OP, DM >::MagnitudeType

Definition at line 141 of file BelosGCRODRIter.hpp.

Constructor & Destructor Documentation

◆ GCRODRIter()

template<class ScalarType , class MV , class OP , class DM >

Belos::GCRODRIter< ScalarType, MV, OP, DM >::GCRODRIter	(	const Teuchos::RCP< LinearProblem< ScalarType, MV, OP, DM > > &	problem,
		const Teuchos::RCP< OutputManager< ScalarType > > &	printer,
		const Teuchos::RCP< StatusTest< ScalarType, MV, OP, DM > > &	tester,
		const Teuchos::RCP< MatOrthoManager< ScalarType, MV, OP, DM > > &	ortho,
		Teuchos::ParameterList &	params
	)

GCRODRIter constructor with linear problem, solver utilities, and parameter list of solver options.

This constructor takes pointers required by the linear solver, in addition to a parameter list of options for the linear solver. These options include the following:

"Num Blocks" - an int specifying the maximum number of blocks allocated for the solver basis. Default: 25
"Restart Timers" = a bool specifying whether the timers should be restarted each time iterate() is called. Default: false
"Keep Hessenberg" = a bool specifying whether the upper Hessenberg should be stored separately from the least squares system. Default: false

Definition at line 393 of file BelosGCRODRIter.hpp.

◆ ~GCRODRIter()

template<class ScalarType , class MV , class OP , class DM >

virtual Belos::GCRODRIter< ScalarType, MV, OP, DM >::~GCRODRIter ( )

inlinevirtual

Destructor.

Definition at line 161 of file BelosGCRODRIter.hpp.

Member Function Documentation

◆ iterate()

template<class ScalarType , class MV , class OP , class DM >

void Belos::GCRODRIter< ScalarType, MV, OP, DM >::iterate ( )

virtual

This method performs block Gmres iterations until the status test indicates the need to stop or an error occurs (in which case, an std::exception is thrown).

iterate() will first determine whether the solver is inintialized; if not, it will call initialize() using default arguments. After initialization, the solver performs block Gmres iterations until the status test evaluates as Passed, at which point the method returns to the caller.

The block Gmres iteration proceeds as follows:

The operator problem->applyOp() is applied to the newest blockSize vectors in the Krylov basis.
The resulting vectors are orthogonalized against the previous basis vectors, and made orthonormal.
The Hessenberg matrix is updated.
The least squares system is updated.

The status test is queried at the beginning of the iteration.

Possible exceptions thrown include the GCRODRIterOrthoFailure.

Implements Belos::Iteration< ScalarType, MV, OP, DM >.

Definition at line 546 of file BelosGCRODRIter.hpp.

◆ initialize() [1/2]

template<class ScalarType , class MV , class OP , class DM >

void Belos::GCRODRIter< ScalarType, MV, OP, DM >::initialize ( GCRODRIterState< ScalarType, MV, DM > & newstate )

virtual

Initialize the solver to an iterate, providing a complete state.

The GCRODRIter contains a certain amount of state, consisting of the current Krylov basis and the associated Hessenberg matrix.

initialize() gives the user the opportunity to manually set these, although this must be done with caution, abiding by the rules given below. All notions of orthogonality and orthonormality are derived from the inner product specified by the orthogonalization manager.

Postcondition: isInitialized() == true (see post-conditions of isInitialize())

The user has the option of specifying any component of the state using initialize(). However, these arguments are assumed to match the post-conditions specified under isInitialized(). Any necessary component of the state not given to initialize() will be generated.

Note: For any pointer in newstate which directly points to the multivectors in the solver, the data is not copied.

Implements Belos::GCRODRIteration< ScalarType, MV, OP, DM >.

Definition at line 512 of file BelosGCRODRIter.hpp.

◆ initialize() [2/2]

template<class ScalarType , class MV , class OP , class DM >

void Belos::GCRODRIter< ScalarType, MV, OP, DM >::initialize ( )

inlinevirtual

Initialize the solver with empty data. Calling this method will result in error, as GCRODRIter must be initialized with a valid state.

Implements Belos::Iteration< ScalarType, MV, OP, DM >.

Definition at line 217 of file BelosGCRODRIter.hpp.

◆ getState()

template<class ScalarType , class MV , class OP , class DM >

GCRODRIterState< ScalarType, MV, DM > Belos::GCRODRIter< ScalarType, MV, OP, DM >::getState ( ) const

inlinevirtual

Get the current state of the linear solver.

The data is only valid if isInitialized() == true.

Returns: A GCRODRIterState object containing const pointers to the current solver state.

Implements Belos::GCRODRIteration< ScalarType, MV, OP, DM >.

Definition at line 229 of file BelosGCRODRIter.hpp.

◆ getNumIters()

template<class ScalarType , class MV , class OP , class DM >

int Belos::GCRODRIter< ScalarType, MV, OP, DM >::getNumIters ( ) const

inlinevirtual

Get the current iteration count.

Implements Belos::Iteration< ScalarType, MV, OP, DM >.

Definition at line 250 of file BelosGCRODRIter.hpp.

◆ resetNumIters()

template<class ScalarType , class MV , class OP , class DM >

void Belos::GCRODRIter< ScalarType, MV, OP, DM >::resetNumIters ( int iter = 0 )

inlinevirtual

Reset the iteration count.

Implements Belos::Iteration< ScalarType, MV, OP, DM >.

Definition at line 253 of file BelosGCRODRIter.hpp.

◆ getNativeResiduals()

template<class ScalarType , class MV , class OP , class DM >

Teuchos::RCP< const MV > Belos::GCRODRIter< ScalarType, MV, OP, DM >::getNativeResiduals ( std::vector< MagnitudeType > * norms ) const

Get the norms of the residuals native to the solver.

Returns: A std::vector of length blockSize containing the native residuals.

Definition at line 492 of file BelosGCRODRIter.hpp.

◆ getCurrentUpdate()

template<class ScalarType , class MV , class OP , class DM >

Teuchos::RCP< MV > Belos::GCRODRIter< ScalarType, MV, OP, DM >::getCurrentUpdate ( ) const

virtual

Get the current update to the linear system.

Note: Some solvers, like GMRES, do not compute updates to the solution every iteration. This method forces its computation. Other solvers, like CG, update the solution each iteration, so this method will return a zero vector indicating that the linear problem contains the current solution.

Implements Belos::Iteration< ScalarType, MV, OP, DM >.

Definition at line 434 of file BelosGCRODRIter.hpp.

◆ updateLSQR()

template<class ScalarType , class MV , class OP , class DM >

void Belos::GCRODRIter< ScalarType, MV, OP, DM >::updateLSQR ( int dim = -1 )

virtual

Method for updating QR factorization of upper Hessenberg matrix.

Note: If dim >= getCurSubspaceDim() and dim < getMaxSubspaceDim(), then the dim-th equations of the least squares problem will be updated.

Implements Belos::GCRODRIteration< ScalarType, MV, OP, DM >.

Definition at line 692 of file BelosGCRODRIter.hpp.