Block lower-triangular preconditioner for the field-split TPSA system. More...

#include <TpsaPreconditioner.hpp>

Inheritance diagram for Opm::TpsaPreconditioner< Scalar, DispOp, RotOp, SPresOp, Comm >:

Public Member Functions
	TpsaPreconditioner (const Linear::TpsaMatrixView< Scalar > &S, const PropertyTree &prm)
	Sequential constructor. More...

	TpsaPreconditioner (const Linear::TpsaMatrixView< Scalar > &S, const PropertyTree &prm, const Comm &comm)
	Parallel constructor. More...

void	pre (MultiVector &, MultiVector &) override
	Nothing to prepare, the sub-solvers set themselves up. More...

void	post (MultiVector &) override
	Nothing to clean up after the last apply(). More...

Dune::SolverCategory::Category	category () const override
	Solver category, overlapping in parallel and sequential otherwise. More...

void	update () override
	Recompute the preconditioners of all five diagonal block solvers. More...

bool	hasPerfectUpdate () const override
	The block solvers are rebuilt from the current matrix on update(). More...

void	apply (MultiVector &v, const MultiVector &d) override
	Apply the preconditioner, i.e. approximately solve S v = d. More...

Static Public Attributes
static constexpr bool	isParallel = !std::is_same_v<Comm, Dune::Amg::SequentialInformation>
	Whether this instance runs the parallel code path. More...

static constexpr auto	_0 = Dune::Indices::_0
	Field indices into the multi-vector: u_x, u_y, u_z, rotation, solid pressure. More...

static constexpr auto	_1 = Dune::Indices::_1

static constexpr auto	_2 = Dune::Indices::_2

static constexpr auto	_3 = Dune::Indices::_3

static constexpr auto	_4 = Dune::Indices::_4

static constexpr std::size_t	pressureIdx = 0
	The sub-solvers have no pressure equation to single out. More...

Detailed Description

template<class Scalar, class DispOp, class RotOp, class SPresOp, class Comm = Dune::Amg::SequentialInformation>
class Opm::TpsaPreconditioner< Scalar, DispOp, RotOp, SPresOp, Comm >

Block lower-triangular preconditioner for the field-split TPSA system.

Solves the three scalar displacement blocks, carries their contribution over to the rotation and solid pressure defects, and solves those in turn. Each diagonal block gets its own Dune::FlexibleSolver, configured through the disp_disp_solver, rot_rot_solver and spres_spres_solver sub-trees; the three displacement blocks share the disp_disp_solver configuration.

Template Parameters

Scalar	Field type of the matrix entries
DispOp	Linear operator type of a single scalar displacement block, SeqDispDispOperatorT or ParDispDispOperatorT
RotOp	Linear operator type of the rotation block, SeqRotRotOperatorT or ParRotRotOperatorT
SPresOp	Linear operator type of the solid pressure block, SeqSPresSPresOperatorT or ParSPresSPresOperatorT
Comm	Communication type of the sub-solvers. Defaults to Dune::Amg::SequentialInformation, which selects the sequential implementation

Constructor & Destructor Documentation

◆ TpsaPreconditioner() [1/2]

template<class Scalar , class DispOp , class RotOp , class SPresOp , class Comm >
requires (!isParallel)

Opm::TpsaPreconditioner< Scalar, DispOp, RotOp, SPresOp, Comm >::TpsaPreconditioner	(	const Linear::TpsaMatrixView< Scalar > &	S,
		const PropertyTree &	prm
	)

Sequential constructor.

Parameters

S	View of the TPSA system matrix.
prm	Configuration of the diagonal block solvers. Must hold the `disp_disp_solver`, `rot_rot_solver` and `spres_spres_solver` sub-trees

◆ TpsaPreconditioner() [2/2]

template<class Scalar , class DispOp , class RotOp , class SPresOp , class Comm >
requires (isParallel)

Opm::TpsaPreconditioner< Scalar, DispOp, RotOp, SPresOp, Comm >::TpsaPreconditioner	(	const Linear::TpsaMatrixView< Scalar > &	S,
		const PropertyTree &	prm,
		const Comm &	comm
	)

Parallel constructor.

Parameters

S	View of the TPSA system matrix.
prm	Configuration of the diagonal block solvers. Must hold the `disp_disp_solver`, `rot_rot_solver` and `spres_spres_solver` sub-trees
comm	Communication of a single field. All five fields live on the same dof partition, so the same object is used for each of them. Only referenced, so it must outlive the preconditioner

Member Function Documentation

◆ apply()

template<class Scalar , class DispOp , class RotOp , class SPresOp , class Comm >

void Opm::TpsaPreconditioner< Scalar, DispOp, RotOp, SPresOp, Comm >::apply	(	MultiVector &	v,
		const MultiVector &	d
	)

override

Apply the preconditioner, i.e. approximately solve S v = d.

Sweeps over the fields in the order u_x, u_y, u_z, rotation, solid pressure. The displacement updates are substituted into the rotation and solid pressure defects before those blocks are solved.

Parameters

v	Update, overwritten by the result
d	Defect to precondition

◆ category()

template<class Scalar , class DispOp , class RotOp , class SPresOp , class Comm >

Dune::SolverCategory::Category Opm::TpsaPreconditioner< Scalar, DispOp, RotOp, SPresOp, Comm >::category

override

Solver category, overlapping in parallel and sequential otherwise.

◆ hasPerfectUpdate()

template<class Scalar , class DispOp , class RotOp , class SPresOp , class Comm >

bool Opm::TpsaPreconditioner< Scalar, DispOp, RotOp, SPresOp, Comm >::hasPerfectUpdate

overridevirtual

The block solvers are rebuilt from the current matrix on update().

Implements Dune::PreconditionerWithUpdate< Linear::TpsaMultiVector< Scalar >, Linear::TpsaMultiVector< Scalar > >.

◆ post()

template<class Scalar , class DispOp , class RotOp , class SPresOp , class Comm >

void Opm::TpsaPreconditioner< Scalar, DispOp, RotOp, SPresOp, Comm >::post ( MultiVector & )

override

Nothing to clean up after the last apply().

◆ pre()

template<class Scalar , class DispOp , class RotOp , class SPresOp , class Comm >

void Opm::TpsaPreconditioner< Scalar, DispOp, RotOp, SPresOp, Comm >::pre	(	MultiVector &	,
		MultiVector &
	)

override

Nothing to prepare, the sub-solvers set themselves up.

◆ update()

template<class Scalar , class DispOp , class RotOp , class SPresOp , class Comm >

void Opm::TpsaPreconditioner< Scalar, DispOp, RotOp, SPresOp, Comm >::update

overridevirtual

Recompute the preconditioners of all five diagonal block solvers.

Implements Dune::PreconditionerWithUpdate< Linear::TpsaMultiVector< Scalar >, Linear::TpsaMultiVector< Scalar > >.