Interface for preconditioner variants that can be applied to quasi-Newton acceleration schemes. More...

#include <Preconditioner.hpp>

Inheritance diagram for precice::acceleration::impl::Preconditioner:

Collaboration diagram for precice::acceleration::impl::Preconditioner:

Public Member Functions
	Preconditioner (int maxNonConstTimeWindows)
virtual	~Preconditioner ()=default
	Destructor, empty.
virtual void	initialize (std::vector< size_t > &svs)
	initialize the preconditioner
void	apply (Eigen::MatrixXd &M, bool transpose)
	Apply preconditioner to matrix.
void	revert (Eigen::MatrixXd &M, bool transpose)
	Apply inverse preconditioner to matrix.
void	apply (Eigen::MatrixXd &M)
	To transform physical values to balanced values. Matrix version.
void	apply (Eigen::VectorXd &v)
	To transform physical values to balanced values. Vector version.
void	revert (Eigen::MatrixXd &M)
	To transform balanced values back to physical values. Matrix version.
void	revert (Eigen::VectorXd &v)
	To transform balanced values back to physical values. Vector version.
void	update (bool timeWindowComplete, const Eigen::VectorXd &oldValues, const Eigen::VectorXd &res)
	Update the scaling after every FSI iteration and require a new QR decomposition (if necessary)
bool	requireNewQR ()
	returns true if a QR decomposition from scratch is necessary
void	newQRfulfilled ()
	to tell the preconditioner that QR-decomposition has been recomputed
std::vector< double > &	getWeights ()
bool	isConst ()

Protected Member Functions
virtual void	_update_ (bool timeWindowComplete, const Eigen::VectorXd &oldValues, const Eigen::VectorXd &res)=0
	Update the scaling after every FSI iteration and require a new QR decomposition (if necessary)

Protected Attributes
std::vector< double >	_weights
	Weights used to scale the matrix V and the residual.
std::vector< double >	_invWeights
	Inverse weights (for efficiency reasons)
std::vector< size_t >	_subVectorSizes
	Sizes of each sub-vector, i.e. each coupling data.
std::vector< size_t >	_subVectorOffsets
	Offsets of each sub-vector in concatenated data, i.e. each coupling data.
int	_maxNonConstTimeWindows
	maximum number of non-const time windows, i.e., after this number of time windows, the preconditioner is frozen with the current weights and becomes a constant preconditioner
int	_nbNonConstTimeWindows = 0
	Counts the number of completed time windows with a non-const weighting.
bool	_requireNewQR = false
	True if a QR decomposition from scratch is necessary.
bool	_frozen = false
	True if _nbNonConstTimeWindows >= _maxNonConstTimeWindows, i.e., preconditioner is not updated any more.

Private Attributes
logging::Logger	_log {"acceleration::Preconditioner"}

Detailed Description

Interface for preconditioner variants that can be applied to quasi-Newton acceleration schemes.

Preconditioning (formerly also known as scaling) improves the balance between several sub-vectors for parallel or multi coupling.

apply() applies the weighting, i.e. transforms from physical values to balanced values. revert() reverts the weighting, i.e. transforms from balanced values back to physical values. update() updates the preconditioner, after every FSI iteration (though some variants might only be updated after a complete time window)

Definition at line 23 of file Preconditioner.hpp.

Constructor & Destructor Documentation

◆ Preconditioner()

precice::acceleration::impl::Preconditioner::Preconditioner ( int maxNonConstTimeWindows )

inline

Definition at line 25 of file Preconditioner.hpp.

◆ ~Preconditioner()

virtual precice::acceleration::impl::Preconditioner::~Preconditioner ( )

virtualdefault

Destructor, empty.

Member Function Documentation

◆ _update_()

virtual void precice::acceleration::impl::Preconditioner::_update_	(	bool	timeWindowComplete,
		const Eigen::VectorXd &	oldValues,
		const Eigen::VectorXd &	res )

protectedpure virtual

Update the scaling after every FSI iteration and require a new QR decomposition (if necessary)

Parameters

[in] timeWindowComplete True if this FSI iteration also completed a time windows

Implemented in precice::acceleration::impl::ConstantPreconditioner, precice::acceleration::impl::ResidualPreconditioner, precice::acceleration::impl::ResidualSumPreconditioner, and precice::acceleration::impl::ValuePreconditioner.

◆ apply() [1/3]

void precice::acceleration::impl::Preconditioner::apply ( Eigen::MatrixXd & M )

inline

To transform physical values to balanced values. Matrix version.

Definition at line 112 of file Preconditioner.hpp.

Here is the call graph for this function:

◆ apply() [2/3]

void precice::acceleration::impl::Preconditioner::apply	(	Eigen::MatrixXd &	M,
		bool	transpose )

inline

Apply preconditioner to matrix.

Parameters

transpose false = from left, true = from right

Definition at line 58 of file Preconditioner.hpp.

Here is the call graph for this function:

◆ apply() [3/3]

void precice::acceleration::impl::Preconditioner::apply ( Eigen::VectorXd & v )

inline

To transform physical values to balanced values. Vector version.

Definition at line 127 of file Preconditioner.hpp.

Here is the call graph for this function:

◆ getWeights()

std::vector< double > & precice::acceleration::impl::Preconditioner::getWeights ( )

inline

Definition at line 204 of file Preconditioner.hpp.

◆ initialize()

virtual void precice::acceleration::impl::Preconditioner::initialize ( std::vector< size_t > & svs )

inlinevirtual

initialize the preconditioner

Parameters

size	of the pp system (e.g. rows of V)

Reimplemented in precice::acceleration::impl::ConstantPreconditioner, and precice::acceleration::impl::ResidualSumPreconditioner.

Definition at line 37 of file Preconditioner.hpp.

Here is the call graph for this function:

◆ isConst()

bool precice::acceleration::impl::Preconditioner::isConst ( )

inline

Definition at line 209 of file Preconditioner.hpp.

◆ newQRfulfilled()

void precice::acceleration::impl::Preconditioner::newQRfulfilled ( )

inline

to tell the preconditioner that QR-decomposition has been recomputed

Definition at line 199 of file Preconditioner.hpp.

◆ requireNewQR()

bool precice::acceleration::impl::Preconditioner::requireNewQR ( )

inline

returns true if a QR decomposition from scratch is necessary

Definition at line 192 of file Preconditioner.hpp.

◆ revert() [1/3]

void precice::acceleration::impl::Preconditioner::revert ( Eigen::MatrixXd & M )

inline

To transform balanced values back to physical values. Matrix version.

Definition at line 140 of file Preconditioner.hpp.

Here is the call graph for this function:

◆ revert() [2/3]

void precice::acceleration::impl::Preconditioner::revert	(	Eigen::MatrixXd &	M,
		bool	transpose )

inline

Apply inverse preconditioner to matrix.

Parameters

transpose false = from left, true = from right

Definition at line 86 of file Preconditioner.hpp.

Here is the call graph for this function:

◆ revert() [3/3]

void precice::acceleration::impl::Preconditioner::revert ( Eigen::VectorXd & v )

inline

To transform balanced values back to physical values. Vector version.

Definition at line 155 of file Preconditioner.hpp.

Here is the call graph for this function:

◆ update()

void precice::acceleration::impl::Preconditioner::update	(	bool	timeWindowComplete,
		const Eigen::VectorXd &	oldValues,
		const Eigen::VectorXd &	res )

inline

Update the scaling after every FSI iteration and require a new QR decomposition (if necessary)

Parameters

[in] timeWindowComplete True if this FSI iteration also completed a time window

Definition at line 172 of file Preconditioner.hpp.

Here is the call graph for this function:

Member Data Documentation

◆ _frozen

bool precice::acceleration::impl::Preconditioner::_frozen = false

protected

True if _nbNonConstTimeWindows >= _maxNonConstTimeWindows, i.e., preconditioner is not updated any more.

Definition at line 239 of file Preconditioner.hpp.

◆ _invWeights

std::vector<double> precice::acceleration::impl::Preconditioner::_invWeights

protected

Inverse weights (for efficiency reasons)

Definition at line 219 of file Preconditioner.hpp.

◆ _log

logging::Logger precice::acceleration::impl::Preconditioner::_log {"acceleration::Preconditioner"}

private

Definition at line 249 of file Preconditioner.hpp.

◆ _maxNonConstTimeWindows

int precice::acceleration::impl::Preconditioner::_maxNonConstTimeWindows

protected

maximum number of non-const time windows, i.e., after this number of time windows, the preconditioner is frozen with the current weights and becomes a constant preconditioner

Definition at line 230 of file Preconditioner.hpp.

◆ _nbNonConstTimeWindows

int precice::acceleration::impl::Preconditioner::_nbNonConstTimeWindows = 0

protected

Counts the number of completed time windows with a non-const weighting.

Definition at line 233 of file Preconditioner.hpp.

◆ _requireNewQR

bool precice::acceleration::impl::Preconditioner::_requireNewQR = false

protected

True if a QR decomposition from scratch is necessary.

Definition at line 236 of file Preconditioner.hpp.

◆ _subVectorOffsets

std::vector<size_t> precice::acceleration::impl::Preconditioner::_subVectorOffsets

protected

Offsets of each sub-vector in concatenated data, i.e. each coupling data.

Definition at line 225 of file Preconditioner.hpp.

◆ _subVectorSizes

std::vector<size_t> precice::acceleration::impl::Preconditioner::_subVectorSizes

protected

Sizes of each sub-vector, i.e. each coupling data.

Definition at line 222 of file Preconditioner.hpp.

◆ _weights

std::vector<double> precice::acceleration::impl::Preconditioner::_weights

protected

Weights used to scale the matrix V and the residual.

Definition at line 216 of file Preconditioner.hpp.

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

src/acceleration/impl/Preconditioner.hpp

Public Member Functions

Protected Member Functions

Protected Attributes

Private Attributes

Detailed Description

Constructor & Destructor Documentation

◆ Preconditioner()

◆ ~Preconditioner()

Member Function Documentation

◆ _update_()

◆ apply() [1/3]

◆ apply() [2/3]

◆ apply() [3/3]

◆ getWeights()

◆ initialize()

◆ isConst()

◆ newQRfulfilled()

◆ requireNewQR()

◆ revert() [1/3]

◆ revert() [2/3]

◆ revert() [3/3]

◆ update()

Member Data Documentation

◆ _frozen

◆ _invWeights

◆ _log

◆ _maxNonConstTimeWindows

◆ _nbNonConstTimeWindows

◆ _requireNewQR

◆ _subVectorOffsets

◆ _subVectorSizes

◆ _weights