Thyra::AztecOOLinearOpWithSolve Class Reference

Concrete LinearOpWithSolveBase subclass implemented using AztecOO. More...

#include <Thyra_AztecOOLinearOpWithSolve.hpp>

Inheritance diagram for Thyra::AztecOOLinearOpWithSolve:

[legend]

List of all members.

Constructors/initializers/accessors
	AztecOOLinearOpWithSolve (const int fwdDefaultMaxIterations=400, const double fwdDefaultTol=1e-6, const int adjDefaultMaxIterations=400, const double adjDefaultTol=1e-6, const bool outputEveryRhs=false)
	STANDARD_MEMBER_COMPOSITION_MEMBERS (int, fwdDefaultMaxIterations)
	The default maximum number of iterations for forward solves.
	STANDARD_MEMBER_COMPOSITION_MEMBERS (double, fwdDefaultTol)
	The default solution tolerance on the residual for forward solves.
	STANDARD_MEMBER_COMPOSITION_MEMBERS (int, adjDefaultMaxIterations)
	The default maximum number of iterations for adjoint solves.
	STANDARD_MEMBER_COMPOSITION_MEMBERS (double, adjDefaultTol)
	The default solution tolerance on the residual for adjoint solves.
	STANDARD_MEMBER_COMPOSITION_MEMBERS (bool, outputEveryRhs)
	Determine if output for every RHS will be printed or not.
void	initialize (const Teuchos::RCP< const LinearOpBase< double > > &fwdOp, const Teuchos::RCP< const LinearOpSourceBase< double > > &fwdOpSrc, const Teuchos::RCP< const PreconditionerBase< double > > &prec, const bool isExternalPrec, const Teuchos::RCP< const LinearOpSourceBase< double > > &approxFwdOpSrc, const Teuchos::RCP< AztecOO > &aztecFwdSolver, const bool allowInexactFwdSolve=false, const Teuchos::RCP< AztecOO > &aztecAdjSolver=Teuchos::null, const bool allowInexactAdjSolve=false, const double aztecSolverScalar=1.0)
	Sets up this object.
Teuchos::RCP< const LinearOpSourceBase< double > >	extract_fwdOpSrc ()
	Extract the forward LinearOpBase<double> object so that it can be modified.
Teuchos::RCP< const PreconditionerBase< double > >	extract_prec ()
	Extract the preconditioner.
bool	isExternalPrec () const
	Determine if the preconditioner was external or not.
Teuchos::RCP< const LinearOpSourceBase< double > >	extract_approxFwdOpSrc ()
	Extract the approximate forward LinearOpBase<double> object used to build the preconditioner.
void	uninitialize (Teuchos::RCP< const LinearOpBase< double > > *fwdOp=NULL, Teuchos::RCP< const LinearOpSourceBase< double > > *fwdOpSrc=NULL, Teuchos::RCP< const PreconditionerBase< double > > *prec=NULL, bool *isExternalPrec=NULL, Teuchos::RCP< const LinearOpSourceBase< double > > *approxFwdOpSrc=NULL, Teuchos::RCP< AztecOO > *aztecFwdSolver=NULL, bool *allowInexactFwdSolve=NULL, Teuchos::RCP< AztecOO > *aztecAdjSolver=NULL, bool *allowInexactAdjSolve=NULL, double *aztecSolverScalar=NULL)
	Uninitialize.
Overridden from LinearOpBase
Teuchos::RCP< const VectorSpaceBase< double > >	range () const
Teuchos::RCP< const VectorSpaceBase< double > >	domain () const
Teuchos::RCP< const LinearOpBase< double > >	clone () const
Overridden from Teuchos::Describable
std::string	description () const
void	describe (Teuchos::FancyOStream &out, const Teuchos::EVerbosityLevel verbLevel) const
Overridden from SingleScalarLinearOpBase
bool	opSupported (EOpTransp M_trans) const
Overridden from SingleRhsLinearOpBase
void	apply (const EOpTransp M_trans, const VectorBase< double > &x, VectorBase< double > *y, const double alpha, const double beta) const
Overridden from SingleScalarLinearOpWithSolveBase
bool	solveSupportsTrans (EOpTransp M_trans) const
bool	solveSupportsSolveMeasureType (EOpTransp M_trans, const SolveMeasureType &solveMeasureType) const
void	solve (const EOpTransp M_trans, const MultiVectorBase< double > &B, MultiVectorBase< double > *X, const int numBlocks, const BlockSolveCriteria< double > blockSolveCriteria[], SolveStatus< double > blockSolveStatus[]) const

---

Detailed Description

Concrete LinearOpWithSolveBase subclass implemented using AztecOO.

This subclass is designed to be very flexible and handle a number of different use cases. It supports forward and optionally adjoint (transpose) solves. I can support inexact solves based on a relative residual norm tolerance or just allow for a default (i.e. tight) linear solve tolerance.

This subclass is not designed to be used directly by users but instead by subclasses of LinearOpWithSolveFactoryBase. One standard implementation that is fairly flexible (and will be make more flexible in the future) is AztecOOLinearOpWithSolveFactory.

This subclass allows for user-defined preconditioners or for built-in aztec preconditioners.

ToDo: Finish documentation!

Definition at line 66 of file Thyra_AztecOOLinearOpWithSolve.hpp.

---

Constructor & Destructor Documentation

Thyra::AztecOOLinearOpWithSolve::AztecOOLinearOpWithSolve	(	const int	fwdDefaultMaxIterations = 400,
		const double	fwdDefaultTol = 1e-6,
		const int	adjDefaultMaxIterations = 400,
		const double	adjDefaultTol = 1e-6,
		const bool	outputEveryRhs = false
	)

Construct uninitialized but with default option values.

Note, these defaults where taken from NOX::EpetraNew::LinearSystemAztecOO::applyJacobianInverse(...) on 2005/08/15.

Definition at line 182 of file Thyra_AztecOOLinearOpWithSolve.cpp.

---

Member Function Documentation

Thyra::AztecOOLinearOpWithSolve::STANDARD_MEMBER_COMPOSITION_MEMBERS	(	int	,
		fwdDefaultMaxIterations
	)

The default maximum number of iterations for forward solves.

Thyra::AztecOOLinearOpWithSolve::STANDARD_MEMBER_COMPOSITION_MEMBERS	(	double	,
		fwdDefaultTol
	)

The default solution tolerance on the residual for forward solves.

Thyra::AztecOOLinearOpWithSolve::STANDARD_MEMBER_COMPOSITION_MEMBERS	(	int	,
		adjDefaultMaxIterations
	)

The default maximum number of iterations for adjoint solves.

Thyra::AztecOOLinearOpWithSolve::STANDARD_MEMBER_COMPOSITION_MEMBERS	(	double	,
		adjDefaultTol
	)

The default solution tolerance on the residual for adjoint solves.

Thyra::AztecOOLinearOpWithSolve::STANDARD_MEMBER_COMPOSITION_MEMBERS	(	bool	,
		outputEveryRhs
	)

Determine if output for every RHS will be printed or not.

void Thyra::AztecOOLinearOpWithSolve::initialize	(	const Teuchos::RCP< const LinearOpBase< double > > &	fwdOp,
		const Teuchos::RCP< const LinearOpSourceBase< double > > &	fwdOpSrc,
		const Teuchos::RCP< const PreconditionerBase< double > > &	prec,
		const bool	isExternalPrec,
		const Teuchos::RCP< const LinearOpSourceBase< double > > &	approxFwdOpSrc,
		const Teuchos::RCP< AztecOO > &	aztecFwdSolver,
		const bool	allowInexactFwdSolve = false,
		const Teuchos::RCP< AztecOO > &	aztecAdjSolver = Teuchos::null,
		const bool	allowInexactAdjSolve = false,
		const double	aztecSolverScalar = 1.0
	)

Sets up this object.

Parameters:

	fwdOp	[in] The forward operator object that defines this objects LinearOpBase interface. interface.
	fwdOpSrc	[in] The source for the forward operator object fwdOp. This also should be the exact same object that is passed in through a LinearOpWithSolveFactoryBase interface.
	prec	[in] The original abstract preconditioner object that was passed through the LinearOpWithSolveFactoryBase interface. This object is not used for anything and can be set as prec==Teuchosnull.
	isExternalPrec	[in] True if the precondition was created externally from the LinearOpWithSolveFactoryBase object, false otherwise.
	approxFwdOpSrc	[in] The source for the original abstract approximate forward operator object that was passed through the LinearOpWithSolveFactoryBase interface. This object is not used for anything and can be set as approxFwdOpSrc==Teuchosnull.
	aztecFwdSolver	[in] The AztecOO object used to perform forward solves. This object must be be ready to call aztecFwdSolver->SetRHS() and aztecFwdSolver->SetLHS() and then call aztecFwdSolver->Solve().
	allowInexactFwdSolve	[in] Determines if this->solveSupportsSolveTolType(NOTRANS,SOLVE_TOL_REL_RESIDUAL_NORM) returns true or not. With the current design, an inexact forward solve can not be supported if there is left scaling or a left preconditioner aggregated with *aztecFwdOp.
	aztecAdjSolver	[in] The AztecOO object used to perform adjoint solves. This object must be be ready to call aztecAdjSolver->SetRHS() and aztecAdjSolver->SetLHS() and then call aztecAdjSolver->Solve().
	allowInexactAdjSolve	[in] Determines if this->solveSupportsSolveTolType(TRANS,SOLVE_TOL_REL_RESIDUAL_NORM) returns true or not. With the current design, an inexact forward solve can not be supported if there is left scaling or a left preconditioner aggregated with *aztecFwdOp.
	linearSystemTransformer	[in] This is a transformation object that is called to pre-preprocess the linear problem before a forward and adjoint linear solver and post-process the linear problem after forward and adjoint linear solve. This abstract object is used to deal with scaling and aggregated preconditioners. It is what makes this implementation fairly flexible.

Preconditions:

• fwdOp.get()!=NULL
• fwdOpSrc.get()!=NULL
• fwdFwdSolver.get()!=NULL

Postconditions:

• this->range() == fwdOp->range()
• this->domain() == fwdOp->domain()
• this->opSupports(M_trans) == opSupports(*fwdOp,M_trans)
• this->solveSupportsTrans(M_trans) == (aztecAdjSolver.get()!=NULL)
• this->solveSupportsSolveTolType([NOTRANS,CONJ], SolveMeasureType(SOLVE_MEASURE_NORM_RESIDUAL,SOLVE_MESURE_NORM_RHS)) == allowInexactFwdSolve
• this->solveSupportsSolveTolType([TRANS,CONJTRANS], SolveMeasureType(SOLVE_MEASURE_NORM_RESIDUAL,SOLVE_MESURE_NORM_RHS)) == (aztecAdjSolver.get()!=NULL&&allowInexactAdjSolve)

ToDo: Finish documentation!

Definition at line 201 of file Thyra_AztecOOLinearOpWithSolve.cpp.

RCP< const LinearOpSourceBase< double > > Thyra::AztecOOLinearOpWithSolve::extract_fwdOpSrc

(

)

Extract the forward LinearOpBase<double> object so that it can be modified.

Definition at line 236 of file Thyra_AztecOOLinearOpWithSolve.cpp.

RCP< const PreconditionerBase< double > > Thyra::AztecOOLinearOpWithSolve::extract_prec

(

)

Extract the preconditioner.

Definition at line 246 of file Thyra_AztecOOLinearOpWithSolve.cpp.

bool Thyra::AztecOOLinearOpWithSolve::isExternalPrec

(

)

const

Determine if the preconditioner was external or not.

Definition at line 255 of file Thyra_AztecOOLinearOpWithSolve.cpp.

RCP< const LinearOpSourceBase< double > > Thyra::AztecOOLinearOpWithSolve::extract_approxFwdOpSrc

(

)

Extract the approximate forward LinearOpBase<double> object used to build the preconditioner.

Definition at line 262 of file Thyra_AztecOOLinearOpWithSolve.cpp.

void Thyra::AztecOOLinearOpWithSolve::uninitialize	(	Teuchos::RCP< const LinearOpBase< double > > *	fwdOp = NULL,
		Teuchos::RCP< const LinearOpSourceBase< double > > *	fwdOpSrc = NULL,
		Teuchos::RCP< const PreconditionerBase< double > > *	prec = NULL,
		bool *	isExternalPrec = NULL,
		Teuchos::RCP< const LinearOpSourceBase< double > > *	approxFwdOpSrc = NULL,
		Teuchos::RCP< AztecOO > *	aztecFwdSolver = NULL,
		bool *	allowInexactFwdSolve = NULL,
		Teuchos::RCP< AztecOO > *	aztecAdjSolver = NULL,
		bool *	allowInexactAdjSolve = NULL,
		double *	aztecSolverScalar = NULL
	)

Uninitialize.

Definition at line 271 of file Thyra_AztecOOLinearOpWithSolve.cpp.

RCP< const VectorSpaceBase< double > > Thyra::AztecOOLinearOpWithSolve::range

(

)

const [virtual]

Implements Thyra::LinearOpBase< double, DomainScalar >.

Definition at line 312 of file Thyra_AztecOOLinearOpWithSolve.cpp.

RCP< const VectorSpaceBase< double > > Thyra::AztecOOLinearOpWithSolve::domain

(

)

const [virtual]

Implements Thyra::LinearOpBase< double, DomainScalar >.

Definition at line 319 of file Thyra_AztecOOLinearOpWithSolve.cpp.

RCP< const LinearOpBase< double > > Thyra::AztecOOLinearOpWithSolve::clone

(

)

const [virtual]

Reimplemented from Thyra::LinearOpBase< double, DomainScalar >.

Definition at line 326 of file Thyra_AztecOOLinearOpWithSolve.cpp.

std::string Thyra::AztecOOLinearOpWithSolve::description

(

)

const [virtual]

Reimplemented from Teuchos::Describable.

Definition at line 335 of file Thyra_AztecOOLinearOpWithSolve.cpp.

void Thyra::AztecOOLinearOpWithSolve::describe	(	Teuchos::FancyOStream &	out,
		const Teuchos::EVerbosityLevel	verbLevel
	)			const [virtual]

Reimplemented from Teuchos::Describable.

Definition at line 348 of file Thyra_AztecOOLinearOpWithSolve.cpp.

bool Thyra::AztecOOLinearOpWithSolve::opSupported

(

EOpTransp

M_trans

)

const [protected, virtual]

Implements Thyra::SingleScalarLinearOpBase< double >.

Definition at line 429 of file Thyra_AztecOOLinearOpWithSolve.cpp.

void Thyra::AztecOOLinearOpWithSolve::apply	(	const EOpTransp	M_trans,
		const VectorBase< double > &	x,
		VectorBase< double > *	y,
		const double	alpha,
		const double	beta
	)			const [protected, virtual]

Implements Thyra::SingleRhsLinearOpBase< double >.

Definition at line 438 of file Thyra_AztecOOLinearOpWithSolve.cpp.

bool Thyra::AztecOOLinearOpWithSolve::solveSupportsTrans

(

EOpTransp

M_trans

)

const [protected, virtual]

Implements Thyra::SingleScalarLinearOpWithSolveBase< double >.

Definition at line 453 of file Thyra_AztecOOLinearOpWithSolve.cpp.

bool Thyra::AztecOOLinearOpWithSolve::solveSupportsSolveMeasureType	(	EOpTransp	M_trans,
		const SolveMeasureType &	solveMeasureType
	)			const [protected, virtual]

Implements Thyra::SingleScalarLinearOpWithSolveBase< double >.

Definition at line 460 of file Thyra_AztecOOLinearOpWithSolve.cpp.

void Thyra::AztecOOLinearOpWithSolve::solve	(	const EOpTransp	M_trans,
		const MultiVectorBase< double > &	B,
		MultiVectorBase< double > *	X,
		const int	numBlocks,
		const BlockSolveCriteria< double >	blockSolveCriteria[],
		SolveStatus< double >	blockSolveStatus[]
	)			const [protected, virtual]

Implements Thyra::SingleScalarLinearOpWithSolveBase< double >.

Definition at line 533 of file Thyra_AztecOOLinearOpWithSolve.cpp.

---

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

• Thyra_AztecOOLinearOpWithSolve.hpp
• Thyra_AztecOOLinearOpWithSolve.cpp