A concrete implementation of a Broyden-type operator for NOX. More...

#include <NOX_Epetra_BroydenOperator.H>

Inheritance diagram for NOX::Epetra::BroydenOperator:

Collaboration diagram for NOX::Epetra::BroydenOperator:

Classes
class	ReplacementInterface

Public Member Functions
	BroydenOperator (Teuchos::ParameterList &nlParams, const Teuchos::RCP< NOX::Utils > &utils, Epetra_Vector &solnVec, const Teuchos::RCP< Epetra_CrsMatrix > &broydMat0, bool verbose=false)
	Constructor taking an initial matrix to be updated.

	BroydenOperator (const BroydenOperator &)
	Copy Constructor.

virtual	~BroydenOperator ()
	Destructor.

virtual const char *	Label () const
	Returns a character std::string describing the name of the operator.

virtual int	SetUseTranspose (bool UseTranspose)
	If set true, the transpose of this operator will be applied.

virtual int	Apply (const Epetra_MultiVector &X, Epetra_MultiVector &Y) const
	Return the result on an Epetra_Operator applied to an Epetra_MultiVector X in Y.

virtual int	ApplyInverse (const Epetra_MultiVector &X, Epetra_MultiVector &Y) const
	Return the result on an Epetra_Operator inverse applied to an Epetra_MultiVector X in Y.

virtual bool	UseTranspose () const
	Returns the current use transpose setting.

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

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

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

virtual bool	Filled () const
	See Epetra_RowMatrix documentation.

virtual int	NumMyRowEntries (int MyRow, int &NumEntries) const
	See Epetra_RowMatrix documentation.

virtual int	MaxNumEntries () const
	See Epetra_RowMatrix documentation.

virtual int	ExtractMyRowCopy (int MyRow, int Length, int &NumEntries, double Values, int Indices) const
	See Epetra_RowMatrix documentation.

virtual int	ExtractDiagonalCopy (Epetra_Vector &Diagonal) const
	See Epetra_RowMatrix documentation.

virtual int	Multiply (bool TransA, const Epetra_MultiVector &X, Epetra_MultiVector &Y) const
	See Epetra_RowMatrix documentation.

virtual int	Solve (bool Upper, bool Trans, bool UnitDiagonal, const Epetra_MultiVector &X, Epetra_MultiVector &Y) const
	See Epetra_RowMatrix documentation.

virtual int	InvRowSums (Epetra_Vector &x) const
	See Epetra_RowMatrix documentation.

virtual int	LeftScale (const Epetra_Vector &x)
	See Epetra_RowMatrix documentation.

virtual int	InvColSums (Epetra_Vector &x) const
	See Epetra_RowMatrix documentation.

virtual int	RightScale (const Epetra_Vector &x)
	See Epetra_RowMatrix documentation.

virtual double	NormInf () const
	See Epetra_RowMatrix documentation.

virtual double	NormOne () const
	See Epetra_RowMatrix documentation.

virtual int	NumGlobalNonzeros () const
	See Epetra_RowMatrix documentation.

virtual long long	NumGlobalNonzeros64 () const

virtual int	NumGlobalRows () const
	See Epetra_RowMatrix documentation.

virtual long long	NumGlobalRows64 () const

virtual int	NumGlobalCols () const
	See Epetra_RowMatrix documentation.

virtual long long	NumGlobalCols64 () const

virtual int	NumGlobalDiagonals () const
	See Epetra_RowMatrix documentation.

virtual long long	NumGlobalDiagonals64 () const

virtual int	NumMyNonzeros () const
	See Epetra_RowMatrix documentation.

virtual int	NumMyRows () const
	See Epetra_RowMatrix documentation.

virtual int	NumMyCols () const
	See Epetra_RowMatrix documentation.

virtual int	NumMyDiagonals () const
	See Epetra_RowMatrix documentation.

virtual bool	LowerTriangular () const
	See Epetra_RowMatrix documentation.

virtual bool	UpperTriangular () const
	See Epetra_RowMatrix documentation.

virtual const Epetra_Comm &	Comm () const
	See Epetra_RowMatrix documentation.

virtual const Epetra_Map &	RowMatrixRowMap () const
	See Epetra_RowMatrix documentation.

virtual const Epetra_Map &	RowMatrixColMap () const
	See Epetra_RowMatrix documentation.

virtual const Epetra_Import *	RowMatrixImporter () const
	See Epetra_RowMatrix documentation.

virtual const Epetra_BlockMap &	Map () const
	See Epetra_SrcDistObj documentation.

void	setStepVector (Epetra_Vector &vec)
	Set the current step vector, $y_k = x_{k+1} - x_k$ .

void	setStepVector (NOX::Epetra::Vector &vec)
	Set the current step vector, $y_k = x_{k+1} - x_k$ .

void	setYieldVector (Epetra_Vector &vec)
	Set the current yield vector, $y_k = F_{k+1} - F_k$ .

void	setYieldVector (NOX::Epetra::Vector &vec)
	Set the current yield vector, $y_k = F_{k+1} - F_k$ .

bool	computeSparseBroydenUpdate ()
	Compute the sparse Broyden update.

void	removeEntriesFromBroydenUpdate (const Epetra_CrsGraph &graph)
	Remove entries from being involved in Broyden updates.

const Epetra_CrsMatrix &	getBroydenMatrix ()
	Return a const reference to the Broyden matrix. The matrix is not owned but is obtained from the client at construction.

void	resetBroydenMatrix (const Epetra_CrsMatrix &mat)
	Reset the values of our matrix.

void	addReplacementInterface (ReplacementInterface *i)
	Register replacement interface.

"Is" functions
Checks to see if various objects have been computed. Returns true if the corresponding "compute" function has been called since the last update to the solution vector (via instantiation or computeX).
virtual bool	isStep () const

virtual bool	isYield () const

virtual bool	isBroyden () const

Public Member Functions inherited from NOX::Observer
	Observer ()
	Constructor.

virtual	~Observer ()
	Destructor.

virtual void	runPreSolutionUpdate (const NOX::Abstract::Vector &, const NOX::Solver::Generic &)
	User defined method that will be executed prior to the update of the solution vector during a call to NOX::Solver::Generic::step(). This is intended to allow users to adjust the direction before the solution update, typically based on knowledge of the problem formulation. The direction is const as we can't guarantee that changes to the direction won't violate assumptions of the solution algorithm. Users can change the update/direciton after a const cast, but NOX may not function as expected. Use at your own risk! More...

virtual void	runPostSolutionUpdate (const NOX::Solver::Generic &)
	User defined method that will be executed after the update of the solution vector during a call to NOX::Solver::Generic::step(). This is intended to allow users to adjust the direction after the solution update, typically based on knowledge of the problem formulation (e.g. clipping negative mass fractions). The direction is const as we can't guarantee that changes to the direction won't violate assumptions of the solution algorithm. Users can change the update/direciton after a const cast, but NOX may not function as expected. Use at your own risk! More...

virtual void	runPreLineSearch (const NOX::Solver::Generic &)
	User defined method that will be executed before a call to NOX::LineSearch::Generic::compute(). Only to be used in NOX::Solver::LineSearchBased!

virtual void	runPostLineSearch (const NOX::Solver::Generic &)
	User defined method that will be executed after a call to NOX::LineSearch::Generic::compute(). Only to be used in NOX::Solver::LineSearchBased!

Public Member Functions inherited from NOX::Epetra::Interface::Jacobian
	Jacobian ()
	Constructor.

virtual	~Jacobian ()
	Destructor.

Public Member Functions inherited from NOX::Epetra::Interface::Preconditioner
	Preconditioner ()
	Constructor.

virtual	~Preconditioner ()
	Destructor.

Protected Member Functions
virtual bool	initialize (Teuchos::ParameterList &nlParams, const Epetra_Vector &x)
	Initialize operator and data.

virtual bool	computeJacobian (const Epetra_Vector &x, Epetra_Operator &Jac)
	Allow for fill of Jacobian matrix whose values will replace corresponding entries in the Broyden matrix.

virtual bool	computePreconditioner (const Epetra_Vector &, Epetra_Operator &Prec, Teuchos::ParameterList *params=0)
	Allow for fill of preconditioning matrix whose values will replace corresponding entries in the Broyden matrix.

virtual void	runPreSolve (const NOX::Solver::Generic &solver)
	Set a flag to skip the first call to computeSparseBroydenUpdate as no valid step vector or residual vector is available.

virtual void	runPreIterate (const NOX::Solver::Generic &solver)
	Needed to preserve any existing call through to a user-defined Pre/Preost Operator.

virtual void	runPostIterate (const NOX::Solver::Generic &solver)
	Update the Broyden matrix using changes in residuals the solution vector from the most recent nnlinear iteration.

virtual void	runPostSolve (const NOX::Solver::Generic &solver)
	Needed to preserve any existing call through to a user-defined Pre/Preost Operator.

void	replaceBroydenMatrixValues (const Epetra_CrsMatrix &mat)
	Replace values in Broyden matrix with either Jacobian or preconditioning matrix entries.

Protected Attributes
bool	verbose

Teuchos::RCP< NOX::Epetra::Vector >	stepVec

Teuchos::RCP< NOX::Epetra::Vector >	yieldVec

Teuchos::RCP< NOX::Epetra::Vector >	workVec

Teuchos::RCP< NOX::Epetra::Vector >	oldX

Teuchos::RCP< NOX::Epetra::Vector >	oldF

Teuchos::RCP< Epetra_CrsMatrix >	crsMatrix

Teuchos::RCP < NOX::Epetra::Interface::Jacobian >	jacIntPtr

Teuchos::RCP< Epetra_CrsMatrix >	jacMatrixPtr

Teuchos::RCP < NOX::Epetra::Interface::Preconditioner >	precIntPtr

Teuchos::RCP< Epetra_CrsMatrix >	precMatrixPtr

Teuchos::ParameterList &	nlParams
	Reference to top-level nonlinear solver parameters list.

const Teuchos::RCP< NOX::Utils > &	utils
	Reference to NOX::Utils object.

Teuchos::RCP< NOX::Observer >	observer
	Pointer to a user defined NOX::Observer object.

std::string	label
	label for the Epetra_RowMatrix

std::vector< bool >	entriesRemoved
	Flag to signal removal of some entries.

std::map< int, std::list< int > >	retainedEntries
	Container of entries to omit from Broyden updates.

std::vector < ReplacementInterface * >	replacementInterfaces
	Container of entries to omit from Broyden updates.

IsValid flags
True if objects are current with respect to the currect stepVec.
bool	isValidStep

bool	isValidYield

bool	isValidBroyden

Detailed Description

A concrete implementation of a Broyden-type operator for NOX.

This operator is intended to allow cheap updates to an existing Jacobian or preconditioning matrix that would otherwise be difficult or impossible to obtain by other means. It computes updates using secant approximations emobdied in truncated Broyden updates that preserve matrix sparsity.

This class derives from NOX::Abstract::PrePostOperator in order to perform a Broyden-type update on an existing matrix that it holds but does not own. This update is performed after each nonlinear iteration within method runPostIterate(...) according to the recursive formula:

$\tilde{B}_{k+1} = \tilde{B}_k + \frac{({y_k - \tilde{B}_k s_k})s_k^T}{s^T s}$

where

$y_k = F_{k+1} - F_k$

and

$s_k = x_{k+1} - x_k$

The tilde on the matrices $ B $ indicates that the updates are constrained so that the nonzero structure of the original matrix passed into the constructor is preserved. Inasmuch as unconstrained Broyden updates produce dense matrices, these constrained updates lead to a loss of Broyden-matrix properties, e.g.

$\tilde{B}_{k+1} s_k \ne \tilde{B}_k + s_k$

$\tilde{B}_{k+1} q \ne \tilde{B}_k q \quad \forall q : s_k^T q = 0$

One could recover these properties by passing into the constructor a dense Epetra_CrsMatrix, though the cost of typical use of this matrix, e.g. applying ILU to it, would be significant. Additionally, "better" values obtained from another Jacobian or preconditioning matrix can be used to replace corresponding values in the updated Broyden matrix by passing the Jacobian or preconditioning matrix and its associated interface to the constructor. The structure of the Jacobain or preconditioning matrix typically represents a subset of the Broyden matrix, e.g. a block diagonal matrix.

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

NOX_Epetra_BroydenOperator.H
NOX_Epetra_BroydenOperator.C

Classes

Public Member Functions

Protected Member Functions

Protected Attributes

Detailed Description