#include <LinearSystem.hpp>

Collaboration diagram for LinearSystem:

Public Member Functions
	LinearSystem (PetscInt lhsVectorSize, unsigned rowPreallocation=UINT_MAX)

	LinearSystem (Vec templateVector, unsigned rowPreallocation, bool newAllocationError=true)

	LinearSystem (Vec residualVector, Mat jacobianMatrix)

	LinearSystem (PetscInt lhsVectorSize, Mat lhsMatrix, Vec rhsVector)

	~LinearSystem ()

void	SetMatrixElement (PetscInt row, PetscInt col, double value)

void	AddToMatrixElement (PetscInt row, PetscInt col, double value)

void	AssembleFinalLinearSystem ()

void	AssembleIntermediateLinearSystem ()

void	FinaliseLhsMatrix ()

void	FinalisePrecondMatrix ()

void	SwitchWriteModeLhsMatrix ()

void	FinaliseRhsVector ()

void	SetMatrixIsSymmetric (bool isSymmetric=true)

bool	IsMatrixSymmetric ()

void	SetMatrixIsConstant (bool matrixIsConstant)

void	SetRelativeTolerance (double relativeTolerance)

void	SetAbsoluteTolerance (double absoluteTolerance)

void	SetKspType (const char *kspType)

void	SetPcType (const char *pcType, boost::shared_ptr< std::vector< PetscInt > > pBathNodes=boost::shared_ptr< std::vector< PetscInt > >())

void	DisplayMatrix ()

void	DisplayRhs ()

void	SetMatrixRow (PetscInt row, double value)

Vec	GetMatrixRowDistributed (unsigned rowIndex)

void	ZeroMatrixRowsWithValueOnDiagonal (std::vector< unsigned > &rRows, double diagonalValue)

void	ZeroMatrixRowsAndColumnsWithValueOnDiagonal (std::vector< unsigned > &rRowColIndices, double diagonalValue)

void	ZeroMatrixColumn (PetscInt col)

void	ZeroLhsMatrix ()

void	ZeroRhsVector ()

void	ZeroLinearSystem ()

Vec	Solve (Vec lhsGuess=nullptr)

void	SetRhsVectorElement (PetscInt row, double value)

void	AddToRhsVectorElement (PetscInt row, double value)

unsigned	GetSize () const

void	SetNullBasis (Vec nullbasis[], unsigned numberOfBases)

void	RemoveNullSpace ()

Vec &	rGetRhsVector ()

Vec	GetRhsVector () const

Mat &	rGetLhsMatrix ()

Mat	GetLhsMatrix () const

Mat &	rGetPrecondMatrix ()

Vec &	rGetDirichletBoundaryConditionsVector ()

void	GetOwnershipRange (PetscInt &lo, PetscInt &hi)

double	GetMatrixElement (PetscInt row, PetscInt col)

double	GetRhsVectorElement (PetscInt row)

unsigned	GetNumIterations () const

template<size_t MATRIX_SIZE>
void	AddLhsMultipleValues (unsigned *matrixRowAndColIndices, c_matrix< double, MATRIX_SIZE, MATRIX_SIZE > &rSmallMatrix)

template<size_t VECTOR_SIZE>
void	AddRhsMultipleValues (unsigned *vectorIndices, c_vector< double, VECTOR_SIZE > &smallVector)

void	SetPrecondMatrixIsDifferentFromLhs (bool precondIsDifferent=true)

void	SetUseFixedNumberIterations (bool useFixedNumberIterations=true, unsigned evaluateNumItsEveryNSolves=UINT_MAX)

void	ResetKspSolver ()

Private Member Functions
template<class Archive >
void	serialize (Archive &archive, const unsigned int version)

Private Attributes
Mat	mLhsMatrix

Mat	mPrecondMatrix

Vec	mRhsVector

PetscInt	mSize

PetscInt	mOwnershipRangeLo

PetscInt	mOwnershipRangeHi

MatNullSpace	mMatNullSpace

bool	mDestroyMatAndVec

KSP	mKspSolver

bool	mKspIsSetup

double	mNonZerosUsed

bool	mMatrixIsConstant

double	mTolerance

bool	mUseAbsoluteTolerance

std::string	mKspType

std::string	mPcType

Vec	mDirichletBoundaryConditionsVector

PCBlockDiagonal *	mpBlockDiagonalPC

PCLDUFactorisation *	mpLDUFactorisationPC

PCTwoLevelsBlockDiagonal *	mpTwoLevelsBlockDiagonalPC

boost::shared_ptr< std::vector < PetscInt > >	mpBathNodes

bool	mPrecondMatrixIsNotLhs

unsigned	mRowPreallocation

bool	mUseFixedNumberIterations

unsigned	mEvaluateNumItsEveryNSolves

void *	mpConvergenceTestContext

unsigned	mNumSolves

PetscReal	mEigMin

PetscReal	mEigMax

bool	mForceSpectrumReevaluation

Friends
class	TestLinearSystem

class	TestPCBlockDiagonal

class	TestPCTwoLevelsBlockDiagonal

class	TestPCLDUFactorisation

class	TestChebyshevIteration

class	boost::serialization::access

Detailed Description

Linear System class. Stores and solves a linear equation of the form Ax=b, where A is a square matrix and x and b are column vectors. The class uses PETSc.

Definition at line 65 of file LinearSystem.hpp.

Constructor & Destructor Documentation

LinearSystem::LinearSystem	(	PetscInt	lhsVectorSize,
		unsigned	rowPreallocation = `UINT_MAX`
	)

Constructor.

Parameters

lhsVectorSize	the size of the LHS vector
rowPreallocation	the max number of nonzero entries expected on a row a value of 0 is allowed: no preallocation is then done and the user must preallocate the memory for the matrix themselves. the default value allows for small size systems to be set as dense matrices automatically.

Todo:: : if we create a linear system object outside a cardiac solver, these are gonna be the default solver and preconditioner. Not consitent with ChasteDefaults.xml though...

Definition at line 54 of file LinearSystem.cpp.

References PetscTools::CreateVec(), EXCEPTION, mKspType, mLhsMatrix, mNumSolves, mOwnershipRangeHi, mOwnershipRangeLo, mPcType, mRhsVector, mRowPreallocation, mSize, and PetscTools::SetupMat().

LinearSystem::LinearSystem	(	Vec	templateVector,
		unsigned	rowPreallocation,
		bool	newAllocationError = `true`
	)

Alternative constructor.

Create a linear system, where the size is based on the size of a given PETSc vec.

The LHS & RHS vectors will be created by duplicating this vector's settings. This should avoid problems with using VecScatter on bidomain simulation results.

Parameters

templateVector	a PETSc vec
rowPreallocation	the max number of nonzero entries expected on a row
newAllocationError	tells PETSc whether to set the MAT_NEW_NONZERO_ALLOCATION_ERR. currently only used in PETSc 3.3 and later in PETSc 3.2 and earlier MAT_NEW_NONZERO_ALLOCATION_ERR defaults to false in PETSc 3.3 MAT_NEW_NONZERO_ALLOCATION_ERR defaults to true

Todo:: : if we create a linear system object outside a cardiac solver, these are gonna be the default solver and preconditioner. Not consistent with ChasteDefaults.xml though...

Definition at line 146 of file LinearSystem.cpp.

References mKspType, mLhsMatrix, mNumSolves, mOwnershipRangeHi, mOwnershipRangeLo, mPcType, mRhsVector, mRowPreallocation, mSize, and PetscTools::SetupMat().

LinearSystem::LinearSystem	(	Vec	residualVector,
		Mat	jacobianMatrix
	)

Alternative constructor.

Create a linear system which wraps the provided PETSc objects so we can access them using our API. Either of the objects may be NULL, but at least one of them must not be.

Useful for storing residuals and jacobians when solving nonlinear PDEs.

Parameters

residualVector	the residual vector
jacobianMatrix	the Jacobian matrix

Todo:: : if we create a linear system object outside a cardiac solver, these are gonna be the default solver and preconditioner. Not consitent with ChasteDefaults.xml though...

Definition at line 187 of file LinearSystem.cpp.

References mKspType, mLhsMatrix, mNumSolves, mOwnershipRangeHi, mOwnershipRangeLo, mPcType, mRhsVector, mRowPreallocation, and mSize.

LinearSystem::LinearSystem	(	PetscInt	lhsVectorSize,
		Mat	lhsMatrix,
		Vec	rhsVector
	)

Alternative constructor for archiving.

Parameters

lhsVectorSize	the size of the LHS vector
lhsMatrix	the RHS matrix
rhsVector	the RHS vector

Definition at line 109 of file LinearSystem.cpp.

References mLhsMatrix, mNumSolves, mOwnershipRangeHi, mOwnershipRangeLo, and mRhsVector.

LinearSystem::~LinearSystem ( )

Destructor.

Todo:: Never tested in linalg component

Definition at line 251 of file LinearSystem.cpp.

References PetscTools::AmMaster(), PetscTools::Destroy(), mDestroyMatAndVec, mDirichletBoundaryConditionsVector, mKspIsSetup, mKspSolver, mLhsMatrix, mMatNullSpace, mNumSolves, mpBlockDiagonalPC, mpConvergenceTestContext, mpLDUFactorisationPC, mPrecondMatrix, mPrecondMatrixIsNotLhs, mpTwoLevelsBlockDiagonalPC, mRhsVector, and PETSC_DESTROY_PARAM.

Member Function Documentation

template<size_t MATRIX_SIZE>

void LinearSystem::AddLhsMultipleValues	(	unsigned *	matrixRowAndColIndices,
		c_matrix< double, MATRIX_SIZE, MATRIX_SIZE > &	rSmallMatrix
	)

inline

Add multiple values to the matrix of linear system.

Parameters

matrixRowAndColIndices	mapping from index of the ublas matrix (see param below) to index of the PETSc matrix of this linear system
rSmallMatrix	Ublas matrix containing the values to be added

N.B. Values which are not local (ie the row is not owned) will be skipped.

Definition at line 540 of file LinearSystem.hpp.

References PetscMatTools::AddMultipleValues(), and mLhsMatrix.

template<size_t VECTOR_SIZE>

void LinearSystem::AddRhsMultipleValues	(	unsigned *	vectorIndices,
		c_vector< double, VECTOR_SIZE > &	smallVector
	)

inline

Add multiple values to the RHS vector.

Parameters

vectorIndices	mapping from index of the ublas vector (see param below) to index of the vector of this linear system
smallVector	Ublas vector containing the values to be added

N.B. Values which are not local (ie the row is not owned) will be skipped.

Definition at line 555 of file LinearSystem.hpp.

References PetscVecTools::AddMultipleValues(), and mRhsVector.

void LinearSystem::AddToMatrixElement	(	PetscInt	row,
		PetscInt	col,
		double	value
	)

Add the specified value to an entry of the matrix.

Parameters

row	the row index
col	the column index
value	the value for this entry

Definition at line 315 of file LinearSystem.cpp.

References PetscMatTools::AddToElement(), and mLhsMatrix.

void LinearSystem::AddToRhsVectorElement	(	PetscInt	row,
		double	value
	)

Add a value to an element of the right-hand side vector.

Parameters

row	the row index
value	the value to set this entry

Definition at line 357 of file LinearSystem.cpp.

References PetscVecTools::AddToElement(), and mRhsVector.

void LinearSystem::AssembleFinalLinearSystem ( )

Call this before Solve().

This calls FinaliseLhsMatrix() and FinaliseRhsVector().

Definition at line 320 of file LinearSystem.cpp.

References FinaliseLhsMatrix(), and FinaliseRhsVector().

Referenced by BoundaryConditionsContainer< ELEMENT_DIM, SPACE_DIM, PROBLEM_DIM >::ApplyDirichletToLinearProblem().

void LinearSystem::AssembleIntermediateLinearSystem ( )

Should be called before AddToMatrixElement.

This calls SwitchWriteModeLhsMatrix() and FinaliseRhsVector().

Definition at line 326 of file LinearSystem.cpp.

References FinaliseRhsVector(), and SwitchWriteModeLhsMatrix().

void LinearSystem::DisplayMatrix ( )

Display the left-hand side matrix.

Definition at line 362 of file LinearSystem.cpp.

References PetscMatTools::Display(), and mLhsMatrix.

void LinearSystem::DisplayRhs ( )

Display the right-hand side vector.

Definition at line 367 of file LinearSystem.cpp.

References PetscVecTools::Display(), and mRhsVector.

void LinearSystem::FinaliseLhsMatrix ( )

Sets up the PETSc matrix left-hand-side mLhsMatrix

Definition at line 332 of file LinearSystem.cpp.

References PetscMatTools::Finalise(), and mLhsMatrix.

Referenced by AssembleFinalLinearSystem(), and AbstractAssemblerSolverHybrid< ELEMENT_DIM, SPACE_DIM, PROBLEM_DIM, INTERPOLATION_LEVEL >::SetupGivenLinearSystem().

void LinearSystem::FinalisePrecondMatrix ( )

Sets up the PETSc matrix used for preconditioning.

Definition at line 342 of file LinearSystem.cpp.

References PetscMatTools::Finalise(), and mPrecondMatrix.

void LinearSystem::FinaliseRhsVector ( )

Sets up the PETSc vector right-hand-side mRhsVector

Definition at line 347 of file LinearSystem.cpp.

References PetscVecTools::Finalise(), and mRhsVector.

Referenced by AssembleFinalLinearSystem(), AssembleIntermediateLinearSystem(), and AbstractAssemblerSolverHybrid< ELEMENT_DIM, SPACE_DIM, PROBLEM_DIM, INTERPOLATION_LEVEL >::SetupGivenLinearSystem().

Mat LinearSystem::GetLhsMatrix ( ) const

Returns: access to the LHS matrix for archiving

Definition at line 524 of file LinearSystem.cpp.

References mLhsMatrix.

double LinearSystem::GetMatrixElement	(	PetscInt	row,
		PetscInt	col
	)

Returns: an element of the matrix. May only be called for elements you own.

Parameters

row	the row index
col	the column index

Definition at line 489 of file LinearSystem.cpp.

References PetscMatTools::GetElement(), and mLhsMatrix.

Vec LinearSystem::GetMatrixRowDistributed ( unsigned rowIndex )

Returns the i-th row of the LHS matrix as a distributed PETSc Vec

Parameters

rowIndex the row index

Returns: rowIndex-th row of the matrix in distributed format

Definition at line 377 of file LinearSystem.cpp.

References PetscMatTools::GetMatrixRowDistributed(), and mLhsMatrix.

Referenced by BoundaryConditionsContainer< ELEMENT_DIM, SPACE_DIM, PROBLEM_DIM >::ApplyDirichletToLinearProblem().

unsigned LinearSystem::GetNumIterations ( ) const

Returns: the number of iterations taken by the last Solve()

Definition at line 499 of file LinearSystem.cpp.

References mKspIsSetup, and mKspSolver.

void LinearSystem::GetOwnershipRange	(	PetscInt &	lo,
		PetscInt &	hi
	)

Returns: this process's ownership range of the contents of the system.

Parameters

lo	lowest index owned by this process
hi	highest index owned by this process

Definition at line 483 of file LinearSystem.cpp.

References mOwnershipRangeHi, and mOwnershipRangeLo.

Referenced by BoundaryConditionsContainer< ELEMENT_DIM, SPACE_DIM, PROBLEM_DIM >::ApplyDirichletToLinearProblem().

Vec LinearSystem::GetRhsVector ( ) const

Returns: access to the RHS vector for archiving

Definition at line 514 of file LinearSystem.cpp.

References mRhsVector.

double LinearSystem::GetRhsVectorElement ( PetscInt row )

Returns: an element of the RHS vector. May only be called for elements you own.

Parameters

row	the row index

Definition at line 494 of file LinearSystem.cpp.

References PetscVecTools::GetElement(), and mRhsVector.

unsigned LinearSystem::GetSize ( ) const

Returns: mSize.

Definition at line 413 of file LinearSystem.cpp.

References mSize.

Referenced by BoundaryConditionsContainer< ELEMENT_DIM, SPACE_DIM, PROBLEM_DIM >::ApplyDirichletToLinearProblem().

bool LinearSystem::IsMatrixSymmetric ( )

Get whether PETSc considers the matrix in this linear system as symmetric or not.

Returns: whether the matrix is symmetric or not.

Definition at line 568 of file LinearSystem.cpp.

References mLhsMatrix.

Referenced by BoundaryConditionsContainer< ELEMENT_DIM, SPACE_DIM, PROBLEM_DIM >::ApplyDirichletToLinearProblem().

void LinearSystem::RemoveNullSpace ( )

Remove the null space from the linear system.

Use for example if Dirichlet BC are applied to a singular system and, therefore, there's no null space anymore.

Definition at line 457 of file LinearSystem.cpp.

References mKspIsSetup, mKspSolver, mLhsMatrix, mMatNullSpace, PETSC_DESTROY_PARAM, and ResetKspSolver().

void LinearSystem::ResetKspSolver ( )

Method to regenerate all KSP objects, including the solver and the preconditioner (e.g. after changing the PDE time step when using time adaptivity).

Todo:: #1695 Store this number in a member variable.

Definition at line 1249 of file LinearSystem.cpp.

References mForceSpectrumReevaluation, mKspIsSetup, mKspSolver, PETSC_DESTROY_PARAM, and PetscTools::SetOption().

Referenced by RemoveNullSpace().

Vec & LinearSystem::rGetDirichletBoundaryConditionsVector ( )

Returns: access to the dirichlet boundary conditions vector.

Should only be used by the BoundaryConditionsContainer.

Definition at line 539 of file LinearSystem.cpp.

References mDirichletBoundaryConditionsVector.

Referenced by BoundaryConditionsContainer< ELEMENT_DIM, SPACE_DIM, PROBLEM_DIM >::ApplyDirichletToLinearProblem().

Mat & LinearSystem::rGetLhsMatrix ( )

Returns: access to the LHS matrix directly. Shouldn't generally need to be called.

Definition at line 519 of file LinearSystem.cpp.

References mLhsMatrix.

Referenced by BoundaryConditionsContainer< ELEMENT_DIM, SPACE_DIM, PROBLEM_DIM >::ApplyPeriodicBcsToLinearProblem(), OdeLinearSystemSolver::rGetLhsMatrix(), AbstractAssemblerSolverHybrid< ELEMENT_DIM, SPACE_DIM, PROBLEM_DIM, INTERPOLATION_LEVEL >::SetupGivenLinearSystem(), SimpleNewtonNonlinearSolver::Solve(), and OdeLinearSystemSolver::SolveOneTimeStep().

Mat & LinearSystem::rGetPrecondMatrix ( )

Returns: access to the matrix used for preconditioning.

Definition at line 529 of file LinearSystem.cpp.

References EXCEPTION, mPrecondMatrix, and mPrecondMatrixIsNotLhs.

Vec & LinearSystem::rGetRhsVector ( )

Returns: access to the RHS vector directly. Shouldn't generally need to be called.

Definition at line 509 of file LinearSystem.cpp.

References mRhsVector.

Referenced by BoundaryConditionsContainer< ELEMENT_DIM, SPACE_DIM, PROBLEM_DIM >::ApplyDirichletToLinearProblem(), AbstractAssemblerSolverHybrid< ELEMENT_DIM, SPACE_DIM, PROBLEM_DIM, INTERPOLATION_LEVEL >::SetupGivenLinearSystem(), SimpleNewtonNonlinearSolver::Solve(), and OdeLinearSystemSolver::SolveOneTimeStep().

template<class Archive >

void LinearSystem::serialize	(	Archive &	archive,
		const unsigned int	version
	)

inlineprivate

Archive the member variables.

Parameters

archive
version

Definition at line 162 of file LinearSystem.hpp.

References mKspType, mMatrixIsConstant, mNonZerosUsed, mPcType, mTolerance, and mUseAbsoluteTolerance.

void LinearSystem::SetAbsoluteTolerance ( double absoluteTolerance )

Set the absolute tolerance.

Parameters

absoluteTolerance the absolute tolerance

Definition at line 594 of file LinearSystem.cpp.

References mKspIsSetup, mKspSolver, mTolerance, and mUseAbsoluteTolerance.

void LinearSystem::SetKspType ( const char * kspType )

Set the KSP solver type (see PETSc KSPSetType() for valid arguments).

Parameters

kspType the KSP solver type

Definition at line 604 of file LinearSystem.cpp.

References mKspIsSetup, mKspSolver, and mKspType.

void LinearSystem::SetMatrixElement	(	PetscInt	row,
		PetscInt	col,
		double	value
	)

Change one of the entires of the matrix to the specified value.

Parameters

row	the row index
col	the column index
value	the value for this entry

Definition at line 310 of file LinearSystem.cpp.

References mLhsMatrix, and PetscMatTools::SetElement().

void LinearSystem::SetMatrixIsConstant ( bool matrixIsConstant )

Set mMatrixIsConstant.

Parameters

matrixIsConstant whether the matrix is constant

Definition at line 579 of file LinearSystem.cpp.

References mMatrixIsConstant.

void LinearSystem::SetMatrixIsSymmetric ( bool isSymmetric = true )

Force PETSc to treat the matrix in this linear system as symmetric from now on.

Parameters

isSymmetric whether the matrix is symmetric or not (defaults to true)

Todo:: : shall we allow modifying the symmetry flag anytime?

Definition at line 544 of file LinearSystem.cpp.

References mLhsMatrix, and PetscMatTools::SetOption().

void LinearSystem::SetMatrixRow	(	PetscInt	row,
		double	value
	)

Set all entries in a given row of a matrix to a certain value. This must be called by the process who owns the row, (but other processors will treat it as a null-op)

Parameters

row	the row index
value	the value to set each entry in this row

Definition at line 372 of file LinearSystem.cpp.

References mLhsMatrix, and PetscMatTools::SetRow().

void LinearSystem::SetNullBasis	(	Vec	nullbasis[],
		unsigned	numberOfBases
	)

Set the null basis of the linear system. In debug mode we test for orthonormality and throw EXCEPTIONs:

each direction is of unit length (to within a tolerance)
the directions are mutually orthogonal (to within a tolerance) Note that in NDEBUG (optimised mode) these tests are skipped and the parameters are passed directly into the LinearSystem.

Parameters

nullbasis	an array PETSc vectors containing orthogonal directions in the nullspace
numberOfBases	the number of directions (size of nullbasis array)

Definition at line 418 of file LinearSystem.cpp.

References EXCEPTION, and mMatNullSpace.

void LinearSystem::SetPcType	(	const char *	pcType,
		boost::shared_ptr< std::vector< PetscInt > >	pBathNodes = `boost::shared_ptr<std::vector<PetscInt> >()`
	)

Set the preconditioner type (see PETSc PCSetType() for valid arguments).

Parameters

pcType	the preconditioner type
pBathNodes	the list of nodes defining the bath

Todo:: : #1082 is this the way of defining a null pointer as the default value of pBathNodes?

Todo:: : #1082 use a single pointer to abstract class

Todo:: : #1082 use a single pointer to abstract class

Todo:: : #1082 use a single pointer to abstract class

Definition at line 614 of file LinearSystem.cpp.

References mKspIsSetup, mKspSolver, mpBathNodes, mpBlockDiagonalPC, mPcType, mpLDUFactorisationPC, mpTwoLevelsBlockDiagonalPC, and TERMINATE.

void LinearSystem::SetPrecondMatrixIsDifferentFromLhs ( bool precondIsDifferent = true )

Set method for mPrecondMatrixIsNotLhs.

Parameters

precondIsDifferent whether the matrix used for preconditioning is the same as the LHS.

Definition at line 1217 of file LinearSystem.cpp.

References mOwnershipRangeHi, mOwnershipRangeLo, mPrecondMatrix, mPrecondMatrixIsNotLhs, mRowPreallocation, mSize, NEVER_REACHED, and PetscTools::SetupMat().

void LinearSystem::SetRelativeTolerance ( double relativeTolerance )

Set the relative tolerance.

Parameters

relativeTolerance the relative tolerance

Definition at line 584 of file LinearSystem.cpp.

References mKspIsSetup, mKspSolver, mTolerance, and mUseAbsoluteTolerance.

void LinearSystem::SetRhsVectorElement	(	PetscInt	row,
		double	value
	)

Set an element of the right-hand side vector to a given value.

Parameters

row	the row index
value	the value to set this entry

Definition at line 352 of file LinearSystem.cpp.

References mRhsVector, and PetscVecTools::SetElement().

Referenced by BoundaryConditionsContainer< ELEMENT_DIM, SPACE_DIM, PROBLEM_DIM >::ApplyDirichletToLinearProblem(), and BoundaryConditionsContainer< ELEMENT_DIM, SPACE_DIM, PROBLEM_DIM >::ApplyPeriodicBcsToLinearProblem().

void LinearSystem::SetUseFixedNumberIterations	(	bool	useFixedNumberIterations = `true`,
		unsigned	evaluateNumItsEveryNSolves = `UINT_MAX`
	)

Set method for mUseFixedNumberIterations

Parameters

useFixedNumberIterations	whether to use fixed number of iterations
evaluateNumItsEveryNSolves	tells LinearSystem to perform a solve with convergence-based stop criteria every n solves to decide how many iterations perform for the next n-1 solves. Default is perfoming a single evaluation at the beginning of the simulation.

Definition at line 1242 of file LinearSystem.cpp.

References mEvaluateNumItsEveryNSolves, and mUseFixedNumberIterations.

Vec LinearSystem::Solve ( Vec lhsGuess = nullptr )

Solve the linear system.

Returns: the solution vector

Parameters

lhsGuess an optional initial guess for the solution (defaults to NULL)

Todo:: This is a magic number. Do we want a warning here?

Todo:: Should it be compulsory for the caller to supply this and manage the memory?

Todo:: #1695 Store this number in a member variable.

Definition at line 674 of file LinearSystem.cpp.

References PetscTools::AmMaster(), GenericEventHandler< 16, HeartEventHandler >::BeginEvent(), PetscTools::Destroy(), GenericEventHandler< 16, HeartEventHandler >::EndEvent(), EXCEPTION, mEigMax, mEigMin, mEvaluateNumItsEveryNSolves, mForceSpectrumReevaluation, mKspIsSetup, mKspSolver, mKspType, mLhsMatrix, mMatNullSpace, mMatrixIsConstant, mNonZerosUsed, mNumSolves, mpBathNodes, mpBlockDiagonalPC, mpConvergenceTestContext, mPcType, mpLDUFactorisationPC, mPrecondMatrix, mPrecondMatrixIsNotLhs, mpTwoLevelsBlockDiagonalPC, mRhsVector, mSize, mTolerance, mUseAbsoluteTolerance, mUseFixedNumberIterations, Timer::Print(), Timer::Reset(), PetscTools::SetOption(), TERMINATE, and PetscVecTools::Zero().

Referenced by SimpleNewtonNonlinearSolver::Solve(), and OdeLinearSystemSolver::SolveOneTimeStep().

void LinearSystem::SwitchWriteModeLhsMatrix ( )

Sets up the PETSc matrix left-hand-side mLhsMatrix

Definition at line 337 of file LinearSystem.cpp.

References mLhsMatrix, and PetscMatTools::SwitchWriteMode().

Referenced by AssembleIntermediateLinearSystem(), and AbstractAssemblerSolverHybrid< ELEMENT_DIM, SPACE_DIM, PROBLEM_DIM, INTERPOLATION_LEVEL >::SetupGivenLinearSystem().

void LinearSystem::ZeroLhsMatrix ( )

Zero all entries of the left-hand side matrix.

Definition at line 402 of file LinearSystem.cpp.

References mLhsMatrix, and PetscMatTools::Zero().

Referenced by ZeroLinearSystem().

void LinearSystem::ZeroLinearSystem ( )

Zero all entries of the left-hand side matrix and right-hand side vector.

Definition at line 407 of file LinearSystem.cpp.

References ZeroLhsMatrix(), and ZeroRhsVector().

Referenced by SimpleNewtonNonlinearSolver::Solve().

void LinearSystem::ZeroMatrixColumn ( PetscInt col )

Zero a column of the left-hand side matrix.

Unfortunately there is no equivalent method in Petsc, so this has to be done carefully to ensure that the sparsity structure of the matrix is not broken. Only owned entries which are non-zero are zeroed.

Parameters

col	the column index

Definition at line 392 of file LinearSystem.cpp.

References mLhsMatrix, and PetscMatTools::ZeroColumn().

void LinearSystem::ZeroMatrixRowsAndColumnsWithValueOnDiagonal	(	std::vector< unsigned > &	rRowColIndices,
		double	diagonalValue
	)

Zero several rows and columns of the matrix, putting a given value on the diagonal.

Parameters

rRowColIndices	A list of indices. All the rows with these indices, and all the columns with these indices, will be zeroed.
diagonalValue	value to put in the diagonal entries (of the zeroed rows)

Definition at line 387 of file LinearSystem.cpp.

References mLhsMatrix, and PetscMatTools::ZeroRowsAndColumnsWithValueOnDiagonal().

Referenced by BoundaryConditionsContainer< ELEMENT_DIM, SPACE_DIM, PROBLEM_DIM >::ApplyDirichletToLinearProblem().

void LinearSystem::ZeroMatrixRowsWithValueOnDiagonal	(	std::vector< unsigned > &	rRows,
		double	diagonalValue
	)

Zero several rows of the matrix, putting a given value in the diagonal entries.

*Massively* less expensive than zeroing each matrix row individually

Parameters

rRows	std::vector of rows to be zeroed
diagonalValue	value to put in the diagonal entries (of the zeroed rows)

Definition at line 382 of file LinearSystem.cpp.

References mLhsMatrix, and PetscMatTools::ZeroRowsWithValueOnDiagonal().

Referenced by BoundaryConditionsContainer< ELEMENT_DIM, SPACE_DIM, PROBLEM_DIM >::ApplyDirichletToLinearProblem(), and BoundaryConditionsContainer< ELEMENT_DIM, SPACE_DIM, PROBLEM_DIM >::ApplyPeriodicBcsToLinearProblem().

void LinearSystem::ZeroRhsVector ( )

Zero all entries of the right-hand side vector.

Definition at line 397 of file LinearSystem.cpp.

References mRhsVector, and PetscVecTools::Zero().

Referenced by ZeroLinearSystem().

Friends And Related Function Documentation

friend class boost::serialization::access

friend

Needed for serialization.

Definition at line 154 of file LinearSystem.hpp.

Member Data Documentation

bool LinearSystem::mDestroyMatAndVec

private

Whether we need to destroy the PETSc matrix and vector in our destructor

Definition at line 90 of file LinearSystem.hpp.

Referenced by ~LinearSystem().

Vec LinearSystem::mDirichletBoundaryConditionsVector

private

Storage for efficient application of Dirichlet BCs, see AbstractBoundaryConditionsContainer

Definition at line 105 of file LinearSystem.hpp.

Referenced by rGetDirichletBoundaryConditionsVector(), and ~LinearSystem().

PetscReal LinearSystem::mEigMax

private

Preconditioned operator largest eigenvalue

Definition at line 144 of file LinearSystem.hpp.

Referenced by Solve().

PetscReal LinearSystem::mEigMin

private

Preconditioned operator smallest eigenvalue

Definition at line 141 of file LinearSystem.hpp.

Referenced by Solve().

unsigned LinearSystem::mEvaluateNumItsEveryNSolves

private

When using fixed number of iterations, a solve with residual-based stop criteria will be performed every mEvaluateNumItsEveryNSolves solves to decide how many iterations perform for the next mEvaluateNumItsEveryNSolves-1 solves

Definition at line 132 of file LinearSystem.hpp.

Referenced by SetUseFixedNumberIterations(), and Solve().

bool LinearSystem::mForceSpectrumReevaluation

private

Under certain circunstances you have to reevaluate the spectrum before the k*n-th, k=0,1,..., iteration

Definition at line 147 of file LinearSystem.hpp.

Referenced by ResetKspSolver(), and Solve().

bool LinearSystem::mKspIsSetup

private

Used by Solve method to track whether KSP has been used.

Definition at line 93 of file LinearSystem.hpp.

Referenced by GetNumIterations(), RemoveNullSpace(), ResetKspSolver(), SetAbsoluteTolerance(), SetKspType(), SetPcType(), SetRelativeTolerance(), Solve(), and ~LinearSystem().