Black-box multilevel adaptive smoothed aggregation preconditioner. More...

#include <MLAPI_MultiLevelAdaptiveSA.h>

Inheritance diagram for MLAPI::MultiLevelAdaptiveSA:

Collaboration diagram for MLAPI::MultiLevelAdaptiveSA:

Public Member Functions

	MultiLevelAdaptiveSA (const Operator &FineMatrix, Teuchos::ParameterList &List, const int NumPDEEqns, const int MaxLevels=20)
	Constructs the hierarchy for given Operator and parameters.

virtual	~MultiLevelAdaptiveSA ()
	Destructor.


const Space	GetOperatorDomainSpace () const
	Returns a copy of the internally stored domain space.

const Space	GetOperatorRangeSpace () const
	Returns a copy of the internally stored range space.

const Space	GetDomainSpace () const
	Returns a copy of the internally stored domain space.

const Space	GetRangeSpace () const
	Returns a copy of the internally stored range space.

Operator &	R (const int i)
	Returns a reference to the restriction operator of level `i`.

const Operator &	R (const int i) const
	Returns a reference to the restriction operator of level `i`.

Operator &	A (const int i)
	Returns a reference to the operator of level `i`.

const Operator &	A (const int i) const
	Returns a reference to the operator of level `i`.

Operator &	P (const int i)
	Returns a reference to the prolongator operator of level `i`.

const Operator &	P (const int i) const
	Returns a reference to the prolongator operator of level `i`.

InverseOperator &	S (const int i)
	Returns a reference to the inverse operator of level `i`.

const InverseOperator &	S (const int i) const
	Returns a reference to the inverse operator of level `i`.

int	GetMaxLevels () const
	Returns the actual number of levels.

void	SetMaxLevels (const int MaxLevels)
	Returns the actual number of levels.

const MultiVector	GetNullSpace () const
	Gets a reference to the internally stored null space.

void	SetNullSpace (MultiVector &NullSpace)
	Sets the null space multi-vector to `NullSpace`.

bool	IsComputed () const
	Returns `true` if the hierarchy has been successfully computed.

void	SetList (Teuchos::ParameterList &List)
	Sets the internally stored list to `List`.

std::string	GetSmootherType ()
	Returns the smoother solver type.

std::string	GetCoarseType ()
	Returns the coarse solver type.

void	SetInputNumPDEEqns (const int n)
	Returns the number of PDE equations on the finest level.

int	GetInputNumPDEEqns ()
	Returns the number of PDE equations on the current level.

int	GetNumPDEEqns ()
	Sets the number of PDE equations on the current level.

void	SetNumPDEEqns (const int NumPDEEqns)

int	GetMaxCoarseSize ()
	Returns the maximum allowed coarse size.

double	GetMaxReduction ()
	Returns the maximum allowed reduction.

int	GetNumItersCoarse ()
	Returns the maximum number of applications on the coarser levels.

int	GetNumItersFine ()
	Returns the maximum number of applications on the finest level.

double	GetComplexity ()
	Returns the multigrid preconditioner operator complexity.


void	Compute ()
	Creates an hierarchy using the provided or default null space.

void	AdaptCompute (const bool UseDefaultOrSpecified, int AdditionalCandidates)
	Setup the adaptive multilevel hierarchy.

void	SetupInitialNullSpace ()
	Computes the first component of the null space.

bool	IncrementNullSpace ()
	Increments the null space dimension by one.


int	Apply (const MultiVector &b_f, MultiVector &x_f) const
	Applies the preconditioner to `b_f`, returns the result in `x_f`.

int	SolveMultiLevelSA (const MultiVector &b_f, MultiVector &x_f, int level) const
	Recursively called core of the multi level preconditioner.


std::ostream &	Print (std::ostream &os, const bool verbose=true) const
	Prints basic information about `this` preconditioner.

Public Member Functions inherited from MLAPI::BaseOperator
virtual	~BaseOperator ()
	Virtual destructor.

Public Member Functions inherited from MLAPI::BaseObject
	BaseObject ()
	Constructor with empty label.

	BaseObject (const std::string &Label)
	Constructor with given Label.

virtual	~BaseObject ()
	Destructor.

void	SetLabel (const std::string &Label)
	Sets the Label of this object to `Label`.

const std::string &	GetLabel () const
	Returns the Label of this object.

Public Member Functions inherited from MLAPI::CompObject
	CompObject ()
	Constructor, set counter to 0.0.

	~CompObject ()
	Destructor.

double	GetFlops () const
	Returns the internal counter of flops.

void	SetFlops (double Flops) const
	Sets internal counter to `Flops`.

void	UpdateFlops (double Flops) const
	Updates internal counter by summing `Flops`.

Public Member Functions inherited from MLAPI::TimeObject
	TimeObject ()
	Constructor, set counter to 0.0.

	~TimeObject ()
	Destructor.

void	ResetTimer () const
	Resets the internal timer.

void	UpdateTime () const
	Updates the internal timer with the time spent since the last call to ResetTimer().

void	UpdateTime (double t) const
	Updates the internal timer with input value `t`.

double	GetTime () const
	Returns the internally stored counter.

Additional Inherited Members
Protected Attributes inherited from MLAPI::TimeObject
Epetra_Time	Time_
	Object used to track time.

double	TotalTime_
	Internal counter.

Detailed Description

Black-box multilevel adaptive smoothed aggregation preconditioner.

This class implements an adaptive smoothed aggregation preconditioner. An example of usage is reported in file AdaptiveSA.cpp. We note that the usage of this class is slightly different from that of MultiLevelSA.

An instance of this class can be created as follows:

int NumPDEEqns = 1;
int MaxLevels = 10;
MultiLevelAdaptiveSA Prec(FineMatrix, List, NumPDEEqns, MaxLevels);

Important methods of this class:

The number of PDE equations on the finest level can be queried using GetInputNumPDEEqns().
GetNumPDEEqns() returns the number of PDE equations on the current level. This value can be set via SetNumPDEEqns().
GetNullSpace() returns a reference to the internally stored null space; the null space is set using SetNullSpace().
GetMaxLevels() returns the number of levels. If called before Compute(), GetMaxLevels() returns the maximum number of levels used in the constructor, otherwise returns the actual number of levels.
GetSmootherType() and GetCoarseType() return the smoother and coarse type.
The number of application of the cycle in IncrementNullSpace() is given by GetNumItersCoarse() and GetNumItersFine().
Methods A(level), P(level), R(level) and S(level) return a reference to the internally stored operators.
Method SetList() can be used at any time to reset the internally stored list.

The general usage is:

Specify the null space using SetNullSpace(NS), where NS is a MultiVector, then compute the hierarchy using Compute(), or
Compute the first component of the null space using SetupInitialNullSpace(). This will define a single-vector null space, and store it using SetNullSpace(NS).
When a non-empty null space is provided, the user can increment by one the dimension of the null space by calling IncrementNullSpace().
Method AdaptCompute() performs all these operations.

Author: Marzio Sala, Ray Tuminaro, Jonathan Hu, Michael Gee, Marian Brezina.

Date: Last updated on Mar-05.

Todo:

store the structure of the aggregates for all phases.

Current implementation supposes zero threshold.

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

MLAPI_MultiLevelAdaptiveSA.h

Public Member Functions

Additional Inherited Members

Detailed Description