MGLevelOp< T > Class Template Reference

#include <MultiGrid.H>

Inheritance diagram for MGLevelOp< T >:

Detailed Description

template<typename T>
class MGLevelOp< T >

This class handles the additional tasks of coordinating operations between this level and the level one coarser than this 'level'. This class is also an MGLevelOpObserver, so objects of this type can observe other MGLevelOp objects.


Public Member Functions
	MGLevelOp ()
	Constructor.
virtual	~MGLevelOp ()
	Destructor.
virtual void	createCoarser (T &a_coarse, const T &a_fine, bool ghosted)=0
virtual void	relax (T &a_correction, const T &a_residual, int a_iterations)=0
virtual void	relaxNF (T &a_phi, const T *a_phiCoarse, const T &a_rhs, int a_iterations)
	specialized no-fine relax function, useful for full-multigrid schemes, defaults to regular relax
virtual void	restrictResidual (T &a_resCoarse, T &a_phiFine, const T &a_rhsFine)=0
virtual void	restrictResidual (T &a_resCoarse, T &a_phiFine, const T *a_phiCoarse, const T &a_rhsFine, bool homogeneous)
	full-multigrid version of restrictResidual, useful for FAS-type schemes. Defaults to standard restriction
virtual void	restrictR (T &a_phiCoarse, const T &a_phiFine)
	simple restriction function
virtual void	prolongIncrement (T &a_phiThisLevel, const T &a_correctCoarse)=0
void	addObserver (MGLevelOpObserver< T > *a_observer)
virtual void	applyOpMg (T &a_lhs, T &a_phi, T *a_phiCoarse, bool a_homogeneous)
	Apply an operator.
virtual void	residualNF (T &a_lhs, T &a_phi, const T *a_phiCoarse, const T &a_rhs, bool a_homogeneous=false)
void	removeObserver (MGLevelOpObserver< T > *a_observer)
void	addCoarserObserver (MGLevelOp< T > *a_operator, int a_coarseningFactor)
void	notifyObserversOfChange ()
	This should be called whenever the operator's data is updated.
virtual void	finerOperatorChanged (const MGLevelOp< T > &a_operator, int a_coarseningFactor)
int	numObservers () const
	Returns the number of objects observing this operator.
Private Member Functions
	MGLevelOp (const MGLevelOp &)
MGLevelOp &	operator= (const MGLevelOp &)
Private Attributes
std::vector< MGLevelOpObserver < T > * >	m_observers
	List of observers for the operator.
std::vector< int >	m_coarseningFactors
	Multigrid coarsening factors.

Constructor & Destructor Documentation

template<typename T>

MGLevelOp< T >::MGLevelOp ( ) [inline]

Constructor.

template<typename T>

virtual MGLevelOp< T >::~MGLevelOp ( ) [inline, virtual]

Destructor.

template<typename T>

MGLevelOp< T >::MGLevelOp ( const MGLevelOp< T > & ) [private]

Member Function Documentation

template<typename T>

virtual void MGLevelOp< T >::createCoarser	(	T &	a_coarse,
		const T &	a_fine,
		bool	ghosted
	)			`[pure virtual]`

Create a coarsened (by two) version of the input data. This does not include averaging the data. So if a_fine is over a Box of (0, 0, 0) (63, 63, 63), a_fine should be over a Box (0, 0, 0) (31, 31, 31).

Implemented in AMRNodeOp, AMRPoissonOp, NewPoissonOp, NewPoissonOp4, ViscousTensorOp, NWOViscousTensorOp, PoissonOp4, ResistivityOp, ViscousTensorOp, EBAMRPoissonOp, EBConductivityOp, EBPoissonOp, EBViscousTensorOp, NWOEBConductivityOp, NWOEBViscousTensorOp, and slowEBCO.

template<typename T>

virtual void MGLevelOp< T >::relax	(	T &	a_correction,
		const T &	a_residual,
		int	a_iterations
	)			`[pure virtual]`

Use your relaxation operator to remove the high frequency wave numbers from the correction so that it may be averaged to a coarser refinement. A point relaxtion scheme, for example takes the form a_correction -= lambda*(L(a_correction) - a_residual).

Implemented in AMRNodeOp, AMRPoissonOp, NewPoissonOp, NewPoissonOp4, ViscousTensorOp, NWOViscousTensorOp, PoissonOp4, ResistivityOp, ViscousTensorOp, EBAMRPoissonOp, EBConductivityOp, EBPoissonOp, EBViscousTensorOp, NWOEBConductivityOp, NWOEBViscousTensorOp, and slowEBCO.

Referenced by MGLevelOp< LevelData< NodeFArrayBox > >::relaxNF().

template<typename T>

virtual void MGLevelOp< T >::relaxNF	(	T &	a_phi,
		const T *	a_phiCoarse,
		const T &	a_rhs,
		int	a_iterations
	)			`[inline, virtual]`

specialized no-fine relax function, useful for full-multigrid schemes, defaults to regular relax

Necessary to implement FAS correctly

Reimplemented in AMRPoissonOp.

template<typename T>

virtual void MGLevelOp< T >::restrictResidual	(	T &	a_resCoarse,
		T &	a_phiFine,
		const T &	a_rhsFine
	)			`[pure virtual]`

calculate restricted residual a_resCoarse[2h] = I[h->2h] (rhsFine[h] - L[h](phiFine[h])

Implemented in AMRNodeOp, AMRPoissonOp, NewPoissonOp, NewPoissonOp4, ViscousTensorOp, NWOViscousTensorOp, PoissonOp4, ResistivityOp, VCAMRPoissonOp2, ViscousTensorOp, EBAMRPoissonOp, EBConductivityOp, EBPoissonOp, EBViscousTensorOp, NWOEBConductivityOp, NWOEBViscousTensorOp, and slowEBCO.

Referenced by MGLevelOp< LevelData< NodeFArrayBox > >::restrictResidual().

template<typename T>

virtual void MGLevelOp< T >::restrictResidual	(	T &	a_resCoarse,
		T &	a_phiFine,
		const T *	a_phiCoarse,
		const T &	a_rhsFine,
		bool	homogeneous
	)			`[inline, virtual]`

full-multigrid version of restrictResidual, useful for FAS-type schemes. Defaults to standard restriction

template<typename T>

virtual void MGLevelOp< T >::restrictR	(	T &	a_phiCoarse,
		const T &	a_phiFine
	)			`[inline, virtual]`

simple restriction function

template<typename T>

virtual void MGLevelOp< T >::prolongIncrement	(	T &	a_phiThisLevel,
		const T &	a_correctCoarse
	)			`[pure virtual]`

correct the fine solution based on coarse correction a_phiThisLevel += I[2h->h](a_correctCoarse)

Implemented in AMRNodeOp, AMRPoissonOp, NewPoissonOp, NewPoissonOp4, ViscousTensorOp, NWOViscousTensorOp, PoissonOp4, ResistivityOp, ViscousTensorOp, EBAMRPoissonOp, EBConductivityOp, EBPoissonOp, EBViscousTensorOp, NWOEBConductivityOp, NWOEBViscousTensorOp, and slowEBCO.

template<typename T>

void MGLevelOp< T >::addObserver ( MGLevelOpObserver< T > * a_observer ) [inline]

This adds a new observer to this operator. Note that this operator does not own the resources for the observer, so you must be careful to ensure that the observer does not go out of scope while the operator lives. If the observer is already observing the operator, this operation has no effect.

Parameters:

a_observer

An observer to be notified when the operator changes.

Referenced by MGLevelOp< LevelData< NodeFArrayBox > >::addCoarserObserver().

template<typename T>

virtual void MGLevelOp< T >::applyOpMg	(	T &	a_lhs,
		T &	a_phi,
		T *	a_phiCoarse,
		bool	a_homogeneous
	)			`[inline, virtual]`

Apply an operator.

useful for full-multigrid/FAS types of algorithms

template<typename T>

virtual void MGLevelOp< T >::residualNF	(	T &	a_lhs,
		T &	a_phi,
		const T *	a_phiCoarse,
		const T &	a_rhs,
		bool	a_homogeneous = `false`
	)			`[inline, virtual]`

Reimplemented in AMRPoissonOp.

template<typename T>

void MGLevelOp< T >::removeObserver ( MGLevelOpObserver< T > * a_observer ) [inline]

Removes the given observer from the operator's list of observers. This method has no effect if the observer is not found in the list.

Parameters:

a_observer

An observer to be removed from the operator's observer list.

template<typename T>

void MGLevelOp< T >::addCoarserObserver	(	MGLevelOp< T > *	a_operator,
		int	a_coarseningFactor
	)			`[inline]`

Adds a MGLevelOp observer to this operator that is notified when the operator changes so that it can coarsen its corresponding data. This is used to implement dependencies between multigrid operators at different grid levels.

Parameters:

	a_operator	The multigrid operator that will observe this operator. No operator may observe itself. If a_operator is already observing this operator, this method has no effect.
	a_coarseningFactor	The refinement factor between the grid level for a_operator and that for this one.

template<typename T>

void MGLevelOp< T >::notifyObserversOfChange ( ) [inline]

This should be called whenever the operator's data is updated.

Notify our observers that the operator's data has changed, taking special care to pass multigrid operators their level.

template<typename T>

virtual void MGLevelOp< T >::finerOperatorChanged	(	const MGLevelOp< T > &	a_operator,
		int	a_coarseningFactor
	)			`[inline, virtual]`

Overload this method to implement the response of a multigrid operator to a change in a finer operator that it is observing. This allows the coarser observer to coarsen any data that has been changed in the finer operator.

Parameters:

	a_operator	The operator whose state has changed.
	a_coarseningFactor	The factor by which this operator's data is coarsened with respect to that of a_operator.

Reimplemented in VCAMRPoissonOp2, EBConductivityOp, and EBViscousTensorOp.

Referenced by MGLevelOp< LevelData< NodeFArrayBox > >::notifyObserversOfChange().

template<typename T>

int MGLevelOp< T >::numObservers ( ) const [inline]

Returns the number of objects observing this operator.

template<typename T>

MGLevelOp& MGLevelOp< T >::operator= ( const MGLevelOp< T > & ) [private]