TPSA matrix for linear elasticity. More...

#include <TpsaMatrix.hpp>

Public Types
using	IstlMatrix = TpsaMatrixView< Scalar >
	What the linear solver operates on. More...

using	MatrixBlock = Opm::MatrixBlock< Scalar, numTpsaEq, numTpsaEq >
	Dense local block the linearizer accumulates into. More...

using	BlockAddress = TpsaBlockRef< Scalar >
	What blockAddress() returns. More...

using	field_type = Scalar
	Field type of the matrix entries. More...

Public Member Functions
	TpsaMatrix (std::size_t rows, std::size_t columns)
	Construct a matrix of the given block dimensions. More...

template<class Simulator >
	TpsaMatrix (const Simulator &simulator)
	Construct a square matrix sized from a simulator's degrees of freedom. More...

	TpsaMatrix (const TpsaMatrix &)=delete

	TpsaMatrix (TpsaMatrix &&)=delete

TpsaMatrix &	operator= (const TpsaMatrix &)=delete

TpsaMatrix &	operator= (TpsaMatrix &&)=delete

template<class Set >
void	reserve (const std::vector< Set > &sparsityPattern)
	Allocate all sub-matrices from a common sparsity pattern. More...

BlockAddress	blockAddress (const std::size_t rowIdx, const std::size_t colIdx) const
	Handle on the block at (rowIdx, colIdx). More...

void	clear ()
	Set all matrix entries to zero. More...

void	clearRow (const std::size_t row, const Scalar diag=1.0)
	Set the given row to zero, except for the main diagonal. More...

void	makeOverlapRowsInvalid (const std::vector< int > &overlapRows)
	Zero out the overlap rows and put the identity on their diagonal. More...

void	scaleFields (const std::array< Scalar, numTpsaFields > &rowFac, const std::array< Scalar, numTpsaFields > &colFac)
	Scale the system by one factor per field: A <- D_row * A * D_col. More...

void	block (const std::size_t rowIdx, const std::size_t colIdx, MatrixBlock &value) const
	Fill `value` with the stored entries of the given block. More...

void	setBlock (const std::size_t rowIdx, const std::size_t colIdx, const MatrixBlock &value)
	Overwrite the given block with a dense 7x7 block. More...

void	addToBlock (const std::size_t rowIdx, const std::size_t colIdx, const MatrixBlock &value)
	Add a dense 7x7 block to the given block. More...

void	commit ()
	No local caching, so nothing to commit. More...

void	finalize ()
	The structure is already solver-ready after reserve(). More...

IstlMatrix &	istlMatrix ()
	The sub-matrix view the linear solver operates on. More...

const IstlMatrix &	istlMatrix () const
	The sub-matrix view the linear solver operates on. More...

std::size_t	rows () const
	Number of block rows. More...

std::size_t	cols () const
	Number of block columns. More...

std::size_t	N () const
	Number of block rows. More...

std::size_t	M () const
	Number of block columns. More...

std::size_t	nonzeroes () const
	Number of nonzero blocks in the shared sparsity pattern. More...

DispDispMatrix00T< Scalar > &	dd00 ()
	Access the u_x-u_x sub-matrix. More...

DispDispMatrix11T< Scalar > &	dd11 ()
	Access the u_y-u_y sub-matrix. More...

DispDispMatrix22T< Scalar > &	dd22 ()
	Access the u_z-u_z sub-matrix. More...

RotRotMatrixT< Scalar > &	rr ()
	Access the rotation-rotation sub-matrix. More...

SPresSPresMatrixT< Scalar > &	spsp ()
	Access the solid pressure-solid pressure sub-matrix. More...

const DispDispMatrix00T< Scalar > &	dd00 () const
	Access the u_x-u_x sub-matrix. More...

const DispDispMatrix11T< Scalar > &	dd11 () const
	Access the u_y-u_y sub-matrix. More...

const DispDispMatrix22T< Scalar > &	dd22 () const
	Access the u_z-u_z sub-matrix. More...

const RotRotMatrixT< Scalar > &	rr () const
	Access the rotation-rotation sub-matrix. More...

const SPresSPresMatrixT< Scalar > &	spsp () const
	Access the solid pressure-solid pressure sub-matrix. More...

Friends
class	TpsaBlockRef< Scalar >

Detailed Description

template<class Scalar>
class Opm::Linear::TpsaMatrix< Scalar >

TPSA matrix for linear elasticity.

This is the equivalent as IstlSparseMatrixAdapter class is for flow linearizations. The TPSA linearizer still provides dense 7x7 blocks, but internally in the class the entries go into 19 sub-matrices. Displacement-displacement sub-matrix have been divided up in 1x1 fields for the Hypre BoomerAMG preconditioner. Overview of sub-matrix fields:

field 0: u_x            (1 dof)
field 1: u_y            (1 dof)
field 2: u_z            (1 dof)
field 3: rot_x/y/z      (3 dofs)
field 4: p_solid        (1 dof)

Note that, displacement-displacement off-diagonals are zero in linear elasticity

Template Parameters

Scalar Field type of the matrix entries.

Member Typedef Documentation

◆ BlockAddress

template<class Scalar >

using Opm::Linear::TpsaMatrix< Scalar >::BlockAddress = TpsaBlockRef<Scalar>

What blockAddress() returns.

◆ field_type

template<class Scalar >

using Opm::Linear::TpsaMatrix< Scalar >::field_type = Scalar

Field type of the matrix entries.

◆ IstlMatrix

template<class Scalar >

using Opm::Linear::TpsaMatrix< Scalar >::IstlMatrix = TpsaMatrixView<Scalar>

What the linear solver operates on.

◆ MatrixBlock

template<class Scalar >

using Opm::Linear::TpsaMatrix< Scalar >::MatrixBlock = Opm::MatrixBlock<Scalar, numTpsaEq, numTpsaEq>

Dense local block the linearizer accumulates into.

Constructor & Destructor Documentation

◆ TpsaMatrix() [1/4]

template<class Scalar >

Opm::Linear::TpsaMatrix< Scalar >::TpsaMatrix	(	std::size_t	rows,
		std::size_t	columns
	)

inline

Construct a matrix of the given block dimensions.

No storage is allocated here; call reserve() with the sparsity pattern first.

Parameters

[in]	rows	Number of block rows, i.e. degrees of freedom.
[in]	columns	Number of block columns, i.e. degrees of freedom.

◆ TpsaMatrix() [2/4]

template<class Scalar >

template<class Simulator >

Opm::Linear::TpsaMatrix< Scalar >::TpsaMatrix ( const Simulator & simulator )

inlineexplicit

Construct a square matrix sized from a simulator's degrees of freedom.

Template Parameters

Simulator Simulator type exposing model().numTotalDof().

Parameters

[in] simulator Simulator the number of degrees of freedom is taken from. Only read during construction.

◆ TpsaMatrix() [3/4]

template<class Scalar >

Opm::Linear::TpsaMatrix< Scalar >::TpsaMatrix ( const TpsaMatrix< Scalar > & )

delete

◆ TpsaMatrix() [4/4]

template<class Scalar >

Opm::Linear::TpsaMatrix< Scalar >::TpsaMatrix ( TpsaMatrix< Scalar > && )

delete

Member Function Documentation

◆ addToBlock()

template<class Scalar >

void Opm::Linear::TpsaMatrix< Scalar >::addToBlock	(	const std::size_t	rowIdx,
		const std::size_t	colIdx,
		const MatrixBlock &	value
	)

inline

Add a dense 7x7 block to the given block.

Parameters

[in]	rowIdx	Block row index.
[in]	colIdx	Block column index.
[in]	value	Dense block added to the stored entries. Its displacement-displacement off-diagonals are ignored.

References Opm::Linear::TpsaMatrix< Scalar >::blockAddress().

◆ block()

template<class Scalar >

void Opm::Linear::TpsaMatrix< Scalar >::block	(	const std::size_t	rowIdx,
		const std::size_t	colIdx,
		MatrixBlock &	value
	)		const

inline

Fill value with the stored entries of the given block.

The displacement-displacement off-diagonals are not stored and come back as zero.

Parameters

[in]	rowIdx	Block row index.
[in]	colIdx	Block column index.
[out]	value	Dense block the entries are written to. Fully overwritten.

References Opm::Linear::TpsaMatrix< Scalar >::blockAddress(), and Opm::Linear::TpsaBlockRef< Scalar >::gather().

◆ blockAddress()

template<class Scalar >

BlockAddress Opm::Linear::TpsaMatrix< Scalar >::blockAddress	(	const std::size_t	rowIdx,
		const std::size_t	colIdx
	)		const

inline

Handle on the block at (rowIdx, colIdx).

Only called while the sparsity pattern is set up, so the linear scan over the row is not on any hot path.

Parameters

[in]	rowIdx	Block row index.
[in]	colIdx	Block column index.

Returns: Handle on the requested block.

Exceptions

std::logic_error If the block is not part of the sparsity pattern.

Referenced by Opm::Linear::TpsaMatrix< Scalar >::addToBlock(), Opm::Linear::TpsaMatrix< Scalar >::block(), and Opm::Linear::TpsaMatrix< Scalar >::setBlock().

◆ clear()

template<class Scalar >

void Opm::Linear::TpsaMatrix< Scalar >::clear ( )

Set all matrix entries to zero.

Does nothing when called before reserve().

◆ clearRow()

template<class Scalar >

void Opm::Linear::TpsaMatrix< Scalar >::clearRow	(	const std::size_t	row,
		const Scalar	diag = `1.0`
	)

Set the given row to zero, except for the main diagonal.

The main diagonal of the block on the diagonal is set to diag. Written as dense blocks so that the field split is applied by TpsaBlockRef, i.e. by the same index map assembly goes through, rather than by a second description of which slots are field-diagonal.

Parameters

[in]	row	Block row to clear.
[in]	diag	Value put on the main diagonal of the diagonal block.

◆ cols()

template<class Scalar >

std::size_t Opm::Linear::TpsaMatrix< Scalar >::cols ( ) const

inline

Number of block columns.

Returns: Number of block columns of the matrix.

◆ commit()

template<class Scalar >

void Opm::Linear::TpsaMatrix< Scalar >::commit ( )

inline

No local caching, so nothing to commit.

◆ dd00() [1/2]

template<class Scalar >

DispDispMatrix00T< Scalar > & Opm::Linear::TpsaMatrix< Scalar >::dd00 ( )

inline

Access the u_x-u_x sub-matrix.

Returns: Reference to the sub-matrix. Only meaningful after reserve().

◆ dd00() [2/2]

template<class Scalar >

const DispDispMatrix00T< Scalar > & Opm::Linear::TpsaMatrix< Scalar >::dd00 ( ) const

inline

Access the u_x-u_x sub-matrix.

Returns: Reference to the sub-matrix. Only meaningful after reserve().

◆ dd11() [1/2]

template<class Scalar >

DispDispMatrix11T< Scalar > & Opm::Linear::TpsaMatrix< Scalar >::dd11 ( )

inline

Access the u_y-u_y sub-matrix.

Returns: Reference to the sub-matrix. Only meaningful after reserve().

◆ dd11() [2/2]

template<class Scalar >

const DispDispMatrix11T< Scalar > & Opm::Linear::TpsaMatrix< Scalar >::dd11 ( ) const

inline

Access the u_y-u_y sub-matrix.

Returns: Reference to the sub-matrix. Only meaningful after reserve().

◆ dd22() [1/2]

template<class Scalar >

DispDispMatrix22T< Scalar > & Opm::Linear::TpsaMatrix< Scalar >::dd22 ( )

inline

Access the u_z-u_z sub-matrix.

Returns: Reference to the sub-matrix. Only meaningful after reserve().

◆ dd22() [2/2]

template<class Scalar >

const DispDispMatrix22T< Scalar > & Opm::Linear::TpsaMatrix< Scalar >::dd22 ( ) const

inline

Access the u_z-u_z sub-matrix.

Returns: Reference to the sub-matrix. Only meaningful after reserve().

◆ finalize()

template<class Scalar >

void Opm::Linear::TpsaMatrix< Scalar >::finalize ( )

inline

The structure is already solver-ready after reserve().

◆ istlMatrix() [1/2]

template<class Scalar >

IstlMatrix & Opm::Linear::TpsaMatrix< Scalar >::istlMatrix ( )

inline

The sub-matrix view the linear solver operates on.

Only valid after reserve(); the view points into this object.

Returns: Reference to the 5x5 view over the sub-matrices.

◆ istlMatrix() [2/2]

template<class Scalar >

const IstlMatrix & Opm::Linear::TpsaMatrix< Scalar >::istlMatrix ( ) const

inline

The sub-matrix view the linear solver operates on.

Only valid after reserve(); the view points into this object.

Returns: Reference to the 5x5 view over the sub-matrices.

◆ M()

template<class Scalar >

std::size_t Opm::Linear::TpsaMatrix< Scalar >::M ( ) const

inline

Number of block columns.

Returns: Number of block columns of the matrix.

◆ makeOverlapRowsInvalid()

template<class Scalar >

void Opm::Linear::TpsaMatrix< Scalar >::makeOverlapRowsInvalid ( const std::vector< int > & overlapRows )

Zero out the overlap rows and put the identity on their diagonal.

Parameters

[in] overlapRows Block rows that are not owned by this process.

◆ N()

template<class Scalar >

std::size_t Opm::Linear::TpsaMatrix< Scalar >::N ( ) const

inline

Number of block rows.

Returns: Number of block rows of the matrix.

◆ nonzeroes()

template<class Scalar >

std::size_t Opm::Linear::TpsaMatrix< Scalar >::nonzeroes ( ) const

inline

Number of nonzero blocks in the shared sparsity pattern.

Returns: Number of nonzero blocks, or zero before reserve() has been called.

◆ operator=() [1/2]

template<class Scalar >

TpsaMatrix & Opm::Linear::TpsaMatrix< Scalar >::operator= ( const TpsaMatrix< Scalar > & )

delete

◆ operator=() [2/2]

template<class Scalar >

TpsaMatrix & Opm::Linear::TpsaMatrix< Scalar >::operator= ( TpsaMatrix< Scalar > && )

delete

◆ reserve()

template<class Scalar >

template<class Set >

void Opm::Linear::TpsaMatrix< Scalar >::reserve ( const std::vector< Set > & sparsityPattern )

inline

Allocate all sub-matrices from a common sparsity pattern.

Also flattens the pattern, caches the value-array base pointer of every sub-matrix and sets up the solver view. Must be called before any block is accessed.

Template Parameters

Set	Ordered container of column indices, e.g. std::set<unsigned>.

Parameters

[in] sparsityPattern One entry per block row, holding that row's column indices in ascending order.

◆ rows()

template<class Scalar >

std::size_t Opm::Linear::TpsaMatrix< Scalar >::rows ( ) const

inline

Number of block rows.

Returns: Number of block rows of the matrix.

◆ rr() [1/2]

template<class Scalar >

RotRotMatrixT< Scalar > & Opm::Linear::TpsaMatrix< Scalar >::rr ( )

inline

Access the rotation-rotation sub-matrix.

Returns: Reference to the sub-matrix. Only meaningful after reserve().

◆ rr() [2/2]

template<class Scalar >

const RotRotMatrixT< Scalar > & Opm::Linear::TpsaMatrix< Scalar >::rr ( ) const

inline

Access the rotation-rotation sub-matrix.

Returns: Reference to the sub-matrix. Only meaningful after reserve().

◆ scaleFields()

template<class Scalar >

void Opm::Linear::TpsaMatrix< Scalar >::scaleFields	(	const std::array< Scalar, numTpsaFields > &	rowFac,
		const std::array< Scalar, numTpsaFields > &	colFac
	)

Scale the system by one factor per field: A <- D_row * A * D_col.

Every sub-matrix couples exactly one field row to one field column, so a field scaling is a single factor per sub-matrix and can be applied to the flat value arrays without walking the sparsity pattern.

Parameters

[in]	rowFac	Row (equation) factor of each field.
[in]	colFac	Column (unknown) factor of each field.

Note: Scales in place, so it must be called once per assembly. Does nothing when called before reserve().

◆ setBlock()

template<class Scalar >

void Opm::Linear::TpsaMatrix< Scalar >::setBlock	(	const std::size_t	rowIdx,
		const std::size_t	colIdx,
		const MatrixBlock &	value
	)

inline

Overwrite the given block with a dense 7x7 block.

Parameters

[in]	rowIdx	Block row index.
[in]	colIdx	Block column index.
[in]	value	Dense block the stored entries are copied from. Its displacement-displacement off-diagonals are ignored.

References Opm::Linear::TpsaMatrix< Scalar >::blockAddress().

◆ spsp() [1/2]

template<class Scalar >

SPresSPresMatrixT< Scalar > & Opm::Linear::TpsaMatrix< Scalar >::spsp ( )

inline

Access the solid pressure-solid pressure sub-matrix.

Returns: Reference to the sub-matrix. Only meaningful after reserve().

◆ spsp() [2/2]

template<class Scalar >

const SPresSPresMatrixT< Scalar > & Opm::Linear::TpsaMatrix< Scalar >::spsp ( ) const

inline

Access the solid pressure-solid pressure sub-matrix.

Returns: Reference to the sub-matrix. Only meaningful after reserve().

Friends And Related Function Documentation

◆ TpsaBlockRef< Scalar >

template<class Scalar >

friend class TpsaBlockRef< Scalar >

friend

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

TpsaMatrix.hpp

Public Types

Public Member Functions

Friends

Detailed Description

Member Typedef Documentation

◆ BlockAddress

◆ field_type

◆ IstlMatrix

◆ MatrixBlock

Constructor & Destructor Documentation

◆ TpsaMatrix() [1/4]

◆ TpsaMatrix() [2/4]

◆ TpsaMatrix() [3/4]

◆ TpsaMatrix() [4/4]

Member Function Documentation

◆ addToBlock()

◆ block()

◆ blockAddress()

◆ clear()

◆ clearRow()

◆ cols()

◆ commit()

◆ dd00() [1/2]

◆ dd00() [2/2]

◆ dd11() [1/2]

◆ dd11() [2/2]

◆ dd22() [1/2]

◆ dd22() [2/2]

◆ finalize()

◆ istlMatrix() [1/2]

◆ istlMatrix() [2/2]

◆ M()

◆ makeOverlapRowsInvalid()

◆ N()

◆ nonzeroes()

◆ operator=() [1/2]

◆ operator=() [2/2]

◆ reserve()

◆ rows()

◆ rr() [1/2]

◆ rr() [2/2]

◆ scaleFields()

◆ setBlock()

◆ spsp() [1/2]

◆ spsp() [2/2]

Friends And Related Function Documentation

◆ TpsaBlockRef< Scalar >