Ifpack_Chebyshev: class for preconditioning with Chebyshev polynomials in Ifpack. More...

#include <Ifpack_Chebyshev.h>

Inheritance diagram for Ifpack_Chebyshev:

Public Member Functions
virtual int	SetUseTranspose (bool UseTranspose_in)

	Ifpack_Chebyshev (const Epetra_Operator *Matrix)
	Ifpack_Chebyshev constructor with given Epetra_Operator/Epetra_RowMatrix. More...

	Ifpack_Chebyshev (const Epetra_RowMatrix *Matrix)
	Ifpack_Chebyshev constructor with given Epetra_Operator/Epetra_RowMatrix. More...

virtual	~Ifpack_Chebyshev ()
	Destructor. More...

virtual int	Apply (const Epetra_MultiVector &X, Epetra_MultiVector &Y) const
	Applies the matrix to an Epetra_MultiVector. More...

virtual int	ApplyInverse (const Epetra_MultiVector &X, Epetra_MultiVector &Y) const
	Applies the preconditioner to X, returns the result in Y. More...

virtual double	NormInf () const
	Returns the infinity norm of the global matrix (not implemented) More...

virtual const char *	Label () const

virtual bool	UseTranspose () const
	Returns the current UseTranspose setting. More...

virtual bool	HasNormInf () const
	Returns true if the this object can provide an approximate Inf-norm, false otherwise. More...

virtual const Epetra_Comm &	Comm () const
	Returns a pointer to the Epetra_Comm communicator associated with this operator. More...

virtual const Epetra_Map &	OperatorDomainMap () const
	Returns the Epetra_Map object associated with the domain of this operator. More...

virtual const Epetra_Map &	OperatorRangeMap () const
	Returns the Epetra_Map object associated with the range of this operator. More...

virtual int	Initialize ()
	Computes all it is necessary to initialize the preconditioner. More...

virtual bool	IsInitialized () const
	Returns true if the preconditioner has been successfully initialized, false otherwise. More...

virtual bool	IsComputed () const
	Returns `true` if the preconditioner has been successfully computed. More...

virtual int	Compute ()
	Computes the preconditioners. More...

virtual double	GetLambdaMax ()
	Returns an approximation to the largest eigenvalue. More...

virtual double	GetLambdaMin ()
	Contains an approximation to the smallest eigenvalue. More...

virtual const Epetra_RowMatrix &	Matrix () const
	Returns a pointer to the matrix to be preconditioned. More...

virtual double	Condest (const Ifpack_CondestType CT=Ifpack_Cheap, const int MaxIters=1550, const double Tol=1e-9, Epetra_RowMatrix *Matrix_in=0)
	Computes the condition number estimates and returns the value. More...

virtual double	Condest () const
	Returns the condition number estimate, or -1.0 if not computed. More...

virtual int	SetParameters (Teuchos::ParameterList &List)
	Sets all the parameters for the preconditioner. More...

virtual std::ostream &	Print (std::ostream &os) const
	Prints object to an output stream. More...

virtual int	NumInitialize () const
	Returns the number of calls to Initialize(). More...

virtual int	NumCompute () const
	Returns the number of calls to Compute(). More...

virtual int	NumApplyInverse () const
	Returns the number of calls to ApplyInverse(). More...

virtual double	InitializeTime () const
	Returns the time spent in Initialize(). More...

virtual double	ComputeTime () const
	Returns the time spent in Compute(). More...

virtual double	ApplyInverseTime () const
	Returns the time spent in ApplyInverse(). More...

virtual double	InitializeFlops () const
	Returns the number of flops in the initialization phase. More...

virtual double	ComputeFlops () const
	Returns the number of flops in the computation phase. More...

virtual double	ApplyInverseFlops () const
	Returns the number of flops for the application of the preconditioner. More...

static int	PowerMethod (const Epetra_Operator &Operator, const Epetra_Vector &InvPointDiagonal, const int MaximumIterations, double &LambdaMax, const unsigned int *RngSeed=0)
	Simple power method to compute lambda_max. More...

static int	CG (const Epetra_Operator &Operator, const Epetra_Vector &InvPointDiagonal, const int MaximumIterations, double &lambda_min, double &lambda_max, const unsigned int *RngSeed=0)
	Uses AztecOO's CG to estimate lambda_min and lambda_max. More...

bool	IsInitialized_
	If `true`, the preconditioner has been computed successfully. More...

bool	IsComputed_
	If `true`, the preconditioner has been computed successfully. More...

int	NumInitialize_
	Contains the number of successful calls to Initialize(). More...

int	NumCompute_
	Contains the number of successful call to Compute(). More...

int	NumApplyInverse_
	Contains the number of successful call to ApplyInverse(). More...

double	InitializeTime_
	Contains the time for all successful calls to Initialize(). More...

double	ComputeTime_
	Contains the time for all successful calls to Compute(). More...

double	ApplyInverseTime_
	Contains the time for all successful calls to ApplyInverse(). More...

double	ComputeFlops_
	Contains the number of flops for Compute(). More...

double	ApplyInverseFlops_
	Contain sthe number of flops for ApplyInverse(). More...

virtual void	SetLabel ()
	Sets the label. More...

	Ifpack_Chebyshev (const Ifpack_Chebyshev &)
	Copy constructor (PRIVATE, should not be used) More...

Ifpack_Chebyshev &	operator= (const Ifpack_Chebyshev &)
	operator = (PRIVATE, should not be used) More...

int	PolyDegree_
	Contains the degree of Chebyshev polynomial. More...

bool	UseTranspose_
	If true, use the tranpose of `Matrix_`. More...

double	Condest_
	Contains the estimated condition number. More...

double	EigRatio_
	Contains the ratio such that [LambdaMax_ / EigRatio_, LambdaMax_] is the interval of interest for the Chebyshev polynomial. More...

int	EigMaxIters_
	Max number of iterations to use in eigenvalue estimation (if automatic). More...

std::string	Label_
	Contains the label of this object. More...

double	LambdaMin_
	Contains an approximation to the smallest eigenvalue. More...

double	LambdaMax_
	Contains an approximation to the largest eigenvalue. More...

double	MinDiagonalValue_
	Contains the minimum value on the diagonal. More...

int	NumMyRows_
	Number of local rows. More...

int	NumMyNonzeros_
	Number of local nonzeros. More...

long long	NumGlobalRows_
	Number of global rows. More...

long long	NumGlobalNonzeros_
	Number of global nonzeros. More...

Teuchos::RefCountPtr< const Epetra_Operator >	Operator_
	Pointers to the matrix to be preconditioned as an Epetra_Operator. More...

Teuchos::RefCountPtr< const Epetra_RowMatrix >	Matrix_
	Pointers to the matrix to be preconditioned as an Epetra_RowMatrix. More...

Teuchos::RefCountPtr < Epetra_Vector >	InvDiagonal_
	Contains the inverse of diagonal elements of `Matrix`. More...

bool	UseBlockMode_
	Use Block Preconditioning. More...

bool	SolveNormalEquations_
	Run on the normal equations. More...

bool	IsRowMatrix_
	If `true`, the Operator_ is an Epetra_RowMatrix. More...

Teuchos::RefCountPtr< Epetra_Time >	Time_
	Time object to track timing. More...

bool	ZeroStartingSolution_
	If `true`, the starting solution is always the zero vector. More...

Detailed Description

Ifpack_Chebyshev: class for preconditioning with Chebyshev polynomials in Ifpack.

The Ifpack_Chebyshev class enables the construction of preconditioners based on Chebyshev polynomials for an Epetra_RowMatrix. Ifpack_Chebyshev is derived from the Ifpack_Preconditioner class, which is itself derived from Epetra_Operator. Therefore this object can be used as preconditioner everywhere an ApplyInverse() method is required in the preconditioning step.

The class is an adaptation of the routine ML_Cheby in Smoother/ml_smoother.h

(04/04/06) Flops are not counted in the routine ApplyInverse()

(04/04/06) The switch to use the transpose matrix is not used in ApplyInverse()

The list of parameters is

EigRatio_ = List.get("chebyshev: ratio eigenvalue", EigRatio_); this is the ratio to define the lower bound on the spectrum; lambda^* = LambdaMax_ / EigRatio_; a typical value used in ML is 30.0 (30.0 is the default value).
LambdaMin_ = List.get("chebyshev: min eigenvalue", LambdaMin_); this is the smallest eigenvalue; this parameter is optional and is only accessed to check whether the input matrix is equal to identity.
LambdaMax_ = List.get("chebyshev: max eigenvalue", LambdaMax_); this is the largest eigenvalue of the matrix.
PolyDegree_ = List.get("chebyshev: degree",PolyDegree_); this is the polynomial degree.
MinDiagonalValue_ = List.get("chebyshev: min diagonal value", MinDiagonalValue_); this defines the threshold for diagonal values under which they are not inverted
ZeroStartingSolution_ = List.get("chebyshev: zero starting solution", ZeroStartingSolution_); this flag allows to set a non-zero initial guess.

Author: Ulrich Hetmaniuk. SNL 1414.

Date: Last modified on 04-Apr-06.

Definition at line 110 of file Ifpack_Chebyshev.h.

Constructor & Destructor Documentation

Ifpack_Chebyshev::Ifpack_Chebyshev ( const Epetra_Operator * Matrix )

Ifpack_Chebyshev constructor with given Epetra_Operator/Epetra_RowMatrix.

Creates an instance of Ifpack_Chebyshev class.

Parameters

Matrix - (In) Pointer to the operator to precondition.

Definition at line 83 of file Ifpack_Chebyshev.cpp.

Ifpack_Chebyshev::Ifpack_Chebyshev ( const Epetra_RowMatrix * Matrix )

Ifpack_Chebyshev constructor with given Epetra_Operator/Epetra_RowMatrix.

Creates an instance of Ifpack_Chebyshev class.

Parameters

Matrix - (In) Pointer to the matrix to precondition.

Definition at line 125 of file Ifpack_Chebyshev.cpp.

virtual Ifpack_Chebyshev::~Ifpack_Chebyshev ( )

inlinevirtual

Destructor.

Definition at line 132 of file Ifpack_Chebyshev.h.

Ifpack_Chebyshev::Ifpack_Chebyshev ( const Ifpack_Chebyshev & )

inlineprivate

Copy constructor (PRIVATE, should not be used)

Definition at line 351 of file Ifpack_Chebyshev.h.

Member Function Documentation

virtual int Ifpack_Chebyshev::SetUseTranspose ( bool UseTranspose_in )

inlinevirtual

This flag can be used to apply the preconditioner to the transpose of the input operator.

Returns: Integer error code, set to 0 if successful. Set to -1 if this implementation does not support transpose.

Implements Epetra_Operator.

Definition at line 142 of file Ifpack_Chebyshev.h.

int Ifpack_Chebyshev::Apply	(	const Epetra_MultiVector &	X,
		Epetra_MultiVector &	Y
	)		const

virtual

Applies the matrix to an Epetra_MultiVector.

Parameters

X	- (In) A Epetra_MultiVector of dimension NumVectors to multiply with matrix.
Y	- (Out) A Epetra_MultiVector of dimension NumVectors containing the result.

Returns: Integer error code, set to 0 if successful.

Implements Epetra_Operator.

Definition at line 228 of file Ifpack_Chebyshev.cpp.

int Ifpack_Chebyshev::ApplyInverse	(	const Epetra_MultiVector &	X,
		Epetra_MultiVector &	Y
	)		const

virtual

Applies the preconditioner to X, returns the result in Y.

Parameters

X	- (In) A Epetra_MultiVector of dimension NumVectors to be preconditioned.
Y	- (InOut) A Epetra_MultiVector of dimension NumVectors containing result.

Returns: Integer error code, set to 0 if successful.

Warning: This routine is NOT AztecOO complaint.

Public Member Functions

Detailed Description

Constructor & Destructor Documentation

Member Function Documentation

Member Data Documentation