Class for solving linear systems using classical and Krylov-subspace iterative methods. More...

#include <CSysSolve.hpp>

Inheritance diagram for CSysSolve< ScalarType >:

Public Types
using	Scalar = ScalarType

using	VectorType = CSysVector< ScalarType >

using	MatrixType = CSysMatrix< ScalarType >

using	ProductType = CMatrixVectorProduct< ScalarType >

using	PrecondType = CPreconditioner< ScalarType >

Public Member Functions
	CSysSolve (LINEAR_SOLVER_MODE linear_solver_mode=LINEAR_SOLVER_MODE::STANDARD)
	default constructor of the class.

unsigned long	CG_LinSolver (const VectorType &b, VectorType &x, const ProductType &mat_vec, const PrecondType &precond, ScalarType tol, unsigned long m, ScalarType &residual, bool monitoring, const CConfig *config) const
	Conjugate Gradient method.

unsigned long	FGMRES_LinSolver (const VectorType &b, VectorType &x, const ProductType &mat_vec, const PrecondType &precond, ScalarType tol, unsigned long m, ScalarType &residual, bool monitoring, const CConfig *config) const
	Flexible Generalized Minimal Residual method.

unsigned long	RFGMRES_LinSolver (const VectorType &b, VectorType &x, const ProductType &mat_vec, const PrecondType &precond, ScalarType tol, unsigned long m, ScalarType &residual, bool monitoring, const CConfig *config) const
	Flexible Generalized Minimal Residual method with restarts (frequency comes from config).

unsigned long	FGCRODR_LinSolver (const VectorType &b, VectorType &x, const ProductType &mat_vec, const PrecondType &precond, ScalarType tol, unsigned long max_iter, ScalarType &residual, bool monitoring, const CConfig *config, FgcrodrMode mode=FgcrodrMode::NORMAL, unsigned long custom_m=0) const
	Flexible Generalized Conjugate Residual Method with Inner Orthogonalization and Deflated Restarting.

unsigned long	BCGSTAB_LinSolver (const VectorType &b, VectorType &x, const ProductType &mat_vec, const PrecondType &precond, ScalarType tol, unsigned long m, ScalarType &residual, bool monitoring, const CConfig *config) const
	Biconjugate Gradient Stabilized Method (BCGSTAB)

unsigned long	Smoother_LinSolver (const VectorType &b, VectorType &x, const ProductType &mat_vec, const PrecondType &precond, ScalarType tol, unsigned long m, ScalarType &residual, bool monitoring, const CConfig *config) const
	Generic smoother (modified Richardson iteration with preconditioner)

unsigned long	Solve (MatrixType &Jacobian, const CSysVector< su2double > &LinSysRes, CSysVector< su2double > &LinSysSol, CGeometry geometry, const CConfig config)
	Solve the linear system using a Krylov subspace method.

unsigned long	Solve_b (MatrixType &Jacobian, const VectorType &LinSysRes, VectorType &LinSysSol, CGeometry geometry, const CConfig config, bool directCall=true)
	Solve the adjoint linear system using a Krylov subspace method.

template<class OtherType , su2enable_if<!std::is_same_v< ScalarType, OtherType > > = 0>
unsigned long	Solve_b (MatrixType &Jacobian, const CSysVector< OtherType > &LinSysRes, CSysVector< OtherType > &LinSysSol, CGeometry geometry, const CConfig config, bool directCall=true)

unsigned long	GetIterations () const
	Get the number of iterations.

ScalarType	GetResidual () const
	Get the final residual.

void	SetToleranceType (LinearToleranceType type)
	Set the type of the tolerance for stoping the linear solvers (RELATIVE or ABSOLUTE).

void	SetxIsZero (bool isZero)
	Assume the initial solution is 0 to save one product, or don't.

void	SetRecomputeResidual (bool recompRes)
	Set whether to recompute residuals at the end (while monitoring only).

void	SetMonitoringFrequency (bool frequency)
	Set the screen output frequency during monitoring.

void	ResetDeflation () const
	Discard FGCRODR's deflation vectors for the next solve.

Detailed Description

template<class ScalarType>
class CSysSolve< ScalarType >

Class for solving linear systems using classical and Krylov-subspace iterative methods.

The individual solvers could be stand-alone subroutines, but by creating CSysSolve objects we can more easily assign different matrix-vector products and preconditioners to different problems that may arise in a hierarchical solver (i.e. multigrid).

The methods of this class are designed to be called by multiple OpenMP threads. Beware of writes to class member variables, for example "Residual" should only be modified by one thread.

Member Typedef Documentation

◆ MatrixType

template<class ScalarType >

using CSysSolve< ScalarType >::MatrixType = CSysMatrix<ScalarType>

◆ PrecondType

template<class ScalarType >

using CSysSolve< ScalarType >::PrecondType = CPreconditioner<ScalarType>

◆ ProductType

template<class ScalarType >

using CSysSolve< ScalarType >::ProductType = CMatrixVectorProduct<ScalarType>

◆ Scalar

template<class ScalarType >

using CSysSolve< ScalarType >::Scalar = ScalarType

◆ VectorType

template<class ScalarType >

using CSysSolve< ScalarType >::VectorType = CSysVector<ScalarType>

Constructor & Destructor Documentation

◆ CSysSolve()

template<class ScalarType >

CSysSolve< ScalarType >::CSysSolve ( LINEAR_SOLVER_MODE linear_solver_mode = LINEAR_SOLVER_MODE::STANDARD )

default constructor of the class.

Parameters

[in] linear_solver_mode - enum, to let CSysSolve know in what context it is

Member Function Documentation

◆ BCGSTAB_LinSolver()

template<class ScalarType >

unsigned long CSysSolve< ScalarType >::BCGSTAB_LinSolver	(	const VectorType &	b,
		VectorType &	x,
		const ProductType &	mat_vec,
		const PrecondType &	precond,
		ScalarType	tol,
		unsigned long	m,
		ScalarType &	residual,
		bool	monitoring,
		const CConfig *	config
	)		const

Biconjugate Gradient Stabilized Method (BCGSTAB)

Parameters

[in]	b	- the right hand size vector
[in,out]	x	- on entry the intial guess, on exit the solution
[in]	mat_vec	- object that defines matrix-vector product
[in]	precond	- object that defines preconditioner
[in]	tol	- tolerance with which to solve the system
[in]	m	- maximum size of the search subspace
[out]	residual	- final normalized residual
[in]	monitoring	- turn on priting residuals from solver to screen
[in]	config	- Definition of the particular problem.

◆ CG_LinSolver()

template<class ScalarType >

unsigned long CSysSolve< ScalarType >::CG_LinSolver	(	const VectorType &	b,
		VectorType &	x,
		const ProductType &	mat_vec,
		const PrecondType &	precond,
		ScalarType	tol,
		unsigned long	m,
		ScalarType &	residual,
		bool	monitoring,
		const CConfig *	config
	)		const

Conjugate Gradient method.

Parameters

[in]	b	- the right hand size vector
[in,out]	x	- on entry the intial guess, on exit the solution
[in]	mat_vec	- object that defines matrix-vector product
[in]	precond	- object that defines preconditioner
[in]	tol	- tolerance with which to solve the system
[in]	m	- maximum size of the search subspace
[out]	residual	- final normalized residual
[in]	monitoring	- turn on priting residuals from solver to screen
[in]	config	- Definition of the particular problem.

◆ FGCRODR_LinSolver()

template<class ScalarType >

unsigned long CSysSolve< ScalarType >::FGCRODR_LinSolver	(	const VectorType &	b,
		VectorType &	x,
		const ProductType &	mat_vec,
		const PrecondType &	precond,
		ScalarType	tol,
		unsigned long	max_iter,
		ScalarType &	residual,
		bool	monitoring,
		const CConfig *	config,
		FgcrodrMode	mode = `FgcrodrMode::NORMAL`,
		unsigned long	custom_m = `0`
	)		const

Flexible Generalized Conjugate Residual Method with Inner Orthogonalization and Deflated Restarting.

Parameters

[in]	b	- the right hand size vector
[in,out]	x	- on entry the intial guess, on exit the solution
[in]	mat_vec	- object that defines matrix-vector product
[in]	precond	- object that defines preconditioner
[in]	tol	- tolerance with which to solve the system
[in]	max_iter	- maximum number of iterations
[out]	residual	- final normalized residual
[in]	monitoring	- turn on priting residuals from solver to screen
[in]	config	- Definition of the particular problem.
[in]	mode	- See FgcrodrMode.
[in]	custom_m	- alternative maximum size of the search subspace, overrides the config value if != 0.

◆ FGMRES_LinSolver()

template<class ScalarType >

unsigned long CSysSolve< ScalarType >::FGMRES_LinSolver	(	const VectorType &	b,
		VectorType &	x,
		const ProductType &	mat_vec,
		const PrecondType &	precond,
		ScalarType	tol,
		unsigned long	m,
		ScalarType &	residual,
		bool	monitoring,
		const CConfig *	config
	)		const

Flexible Generalized Minimal Residual method.

Parameters

[in]	b	- the right hand size vector
[in,out]	x	- on entry the intial guess, on exit the solution
[in]	mat_vec	- object that defines matrix-vector product
[in]	precond	- object that defines preconditioner
[in]	tol	- tolerance with which to solve the system
[in]	m	- maximum size of the search subspace
[out]	residual	- final normalized residual
[in]	monitoring	- turn on priting residuals from solver to screen
[in]	config	- Definition of the particular problem.

◆ GetIterations()

template<class ScalarType >

unsigned long CSysSolve< ScalarType >::GetIterations ( ) const

inline

Get the number of iterations.

Returns: The number of iterations done by Solve or Solve_b

◆ GetResidual()

template<class ScalarType >

ScalarType CSysSolve< ScalarType >::GetResidual ( ) const

inline

Get the final residual.

Returns: The residual at the end of Solve or Solve_b

◆ ResetDeflation()

template<class ScalarType >

void CSysSolve< ScalarType >::ResetDeflation ( ) const

inline

Discard FGCRODR's deflation vectors for the next solve.

◆ RFGMRES_LinSolver()

template<class ScalarType >

unsigned long CSysSolve< ScalarType >::RFGMRES_LinSolver	(	const VectorType &	b,
		VectorType &	x,
		const ProductType &	mat_vec,
		const PrecondType &	precond,
		ScalarType	tol,
		unsigned long	m,
		ScalarType &	residual,
		bool	monitoring,
		const CConfig *	config
	)		const

Flexible Generalized Minimal Residual method with restarts (frequency comes from config).

◆ SetMonitoringFrequency()

template<class ScalarType >

void CSysSolve< ScalarType >::SetMonitoringFrequency ( bool frequency )

inline

Set the screen output frequency during monitoring.

◆ SetRecomputeResidual()

template<class ScalarType >

void CSysSolve< ScalarType >::SetRecomputeResidual ( bool recompRes )

inline

Set whether to recompute residuals at the end (while monitoring only).

◆ SetToleranceType()

template<class ScalarType >

void CSysSolve< ScalarType >::SetToleranceType ( LinearToleranceType type )

inline

Set the type of the tolerance for stoping the linear solvers (RELATIVE or ABSOLUTE).

◆ SetxIsZero()

template<class ScalarType >

void CSysSolve< ScalarType >::SetxIsZero ( bool isZero )

inline

Assume the initial solution is 0 to save one product, or don't.

◆ Smoother_LinSolver()

template<class ScalarType >

unsigned long CSysSolve< ScalarType >::Smoother_LinSolver	(	const VectorType &	b,
		VectorType &	x,
		const ProductType &	mat_vec,
		const PrecondType &	precond,
		ScalarType	tol,
		unsigned long	m,
		ScalarType &	residual,
		bool	monitoring,
		const CConfig *	config
	)		const

Generic smoother (modified Richardson iteration with preconditioner)

Parameters

[in]	b	- the right hand size vector
[in,out]	x	- on entry the intial guess, on exit the solution
[in]	mat_vec	- object that defines matrix-vector product
[in]	precond	- object that defines preconditioner
[in]	tol	- tolerance with which to solve the system
[in]	m	- maximum number of iterations
[out]	residual	- final normalized residual
[in]	monitoring	- turn on priting residuals from solver to screen
[in]	config	- Definition of the particular problem.

◆ Solve()

template<class ScalarType >

unsigned long CSysSolve< ScalarType >::Solve	(	MatrixType &	Jacobian,
		const CSysVector< su2double > &	LinSysRes,
		CSysVector< su2double > &	LinSysSol,
		CGeometry *	geometry,
		const CConfig *	config
	)

Solve the linear system using a Krylov subspace method.

Parameters

[in]	Jacobian	- Jacobian Matrix for the linear system
[in]	LinSysRes	- Linear system residual
[in,out]	LinSysSol	- Linear system solution
[in]	geometry	- Geometrical definition of the problem.
[in]	config	- Definition of the particular problem.

◆ Solve_b() [1/2]

template<class ScalarType >

template<class OtherType , su2enable_if<!std::is_same_v< ScalarType, OtherType > > = 0>

unsigned long CSysSolve< ScalarType >::Solve_b	(	MatrixType &	Jacobian,
		const CSysVector< OtherType > &	LinSysRes,
		CSysVector< OtherType > &	LinSysSol,
		CGeometry *	geometry,
		const CConfig *	config,
		bool	directCall = `true`
	)

inline

◆ Solve_b() [2/2]

template<class ScalarType >

unsigned long CSysSolve< ScalarType >::Solve_b	(	MatrixType &	Jacobian,
		const VectorType &	LinSysRes,
		VectorType &	LinSysSol,
		CGeometry *	geometry,
		const CConfig *	config,
		bool	directCall = `true`
	)

Solve the adjoint linear system using a Krylov subspace method.

Parameters

[in]	Jacobian	- Jacobian Matrix for the linear system
[in]	LinSysRes	- Linear system residual
[in,out]	LinSysSol	- Linear system solution
[in]	geometry	- Geometrical definition of the problem.
[in]	config	- Definition of the particular problem.
[in]	directCall	- If this method is called directly, or in AD context.

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

Common/include/linear_algebra/CSysSolve.hpp
Common/src/linear_algebra/CSysSolve.cpp

Public Types

Public Member Functions

Detailed Description

Member Typedef Documentation

◆ MatrixType

◆ PrecondType

◆ ProductType

◆ Scalar

◆ VectorType

Constructor & Destructor Documentation

◆ CSysSolve()

Member Function Documentation

◆ BCGSTAB_LinSolver()

◆ CG_LinSolver()

◆ FGCRODR_LinSolver()

◆ FGMRES_LinSolver()

◆ GetIterations()

◆ GetResidual()

◆ ResetDeflation()

◆ RFGMRES_LinSolver()

◆ SetMonitoringFrequency()

◆ SetRecomputeResidual()

◆ SetToleranceType()

◆ SetxIsZero()

◆ Smoother_LinSolver()

◆ Solve()

◆ Solve_b() [1/2]

◆ Solve_b() [2/2]