oft_native_solvers Module Reference

Detailed Description

Abstract solver interfaces and select native implementations.

Abstract interface definitions

• Abstract solver class
• Abstract eigensolver class
• Abstract orthogonalization class
• Field creation prototype
• Boundary condition prototype

Native solver implementations

• Newton iteration

Preconditioner implementations

• Diagonal scaling
• Symmetric Point-Jacobi (Richardson iteration with diagonal scaling)
• Block-Jacobi
• Multi-Grid

See also

oft_cg, oft_gmres, oft_petsc_solvers

Authors

Chris Hansen

Date

August 2011

Data Types
type	oft_bjprecond
	Block-Jacobi preconditioner. More...
type	oft_diag_cscale
	Diagonal preconditioner. More...
type	oft_diag_scale
	Diagonal preconditioner. More...
type	oft_identity_inv
	Identity matrix inversion. More...
type	oft_jblock_precond
	Symmetric Jacobi smoother. More...
type	oft_ml_precond
	Multi-level preconditioner level. More...
type	oft_ml_trans
	Multi-level transfer level. More...
type	oft_native_cg_eigsolver
	CG eigensolver class. More...
type	oft_native_cg_solver
	CG solver class. More...
type	oft_native_gmres_csolver
	GMRES solver class (complex) More...
type	oft_native_gmres_solver
	GMRES solver class. More...
type	oft_nksolver
	Native Newton solver. More...
interface	oft_update_jacobian
	Needs docs. More...

Functions/Subroutines
recursive subroutine	bjprecond_apply (self, u, g)
	Precondition a linear system using a Block-Jacobi method.
subroutine	bjprecond_delete (self)
	Destroy Block-Jacobi preconditioner and deallocate all internal storage.
subroutine	bjprecond_setup_xml (self, solver_node, level)
	Setup block-Jacobi smoother from XML definition.
recursive subroutine	bjprecond_update (self, new_pattern)
	Update solver after changing settings/operators.
subroutine	cdiag_scale_apply (self, u, g)
	Solve the linear system for a diagonal matrix (complex)
subroutine	cdiag_scale_delete (self)
	Destroy diagonal preconditioner and deallocate all internal storage.
subroutine	cg_delete (self)
	Destroy diagonal preconditioner and deallocate all internal storage.
subroutine	cg_eig_delete (self)
	Destroy diagonal preconditioner and deallocate all internal storage.
subroutine	cg_eigsolver_apply (self, u, alam)
	Solve a general eigenvalue system using the Conjugate-Gradient method.
subroutine	cg_setup_xml (self, solver_node, level)
	Setup CG solver from XML definition.
recursive subroutine	cg_solver_apply (self, u, g)
	Solve a linear system using the Conjugate-Gradient method.
subroutine	cgmres_delete (self)
	Destroy diagonal preconditioner and deallocate all internal storage.
subroutine	cgmres_setup_xml (self, solver_node, level)
	Setup GMRES solver from XML definition.
recursive subroutine	cgmres_solver_apply (self, u, g)
	Solve a linear system using the flexible GMRES method (complex)
subroutine	diag_scale_apply (self, u, g)
	Solve the linear system for a diagonal matrix.
logical function, public	diag_scale_cast (self, source)
	Cast a solver object to a oft_diag_scale.
subroutine	diag_scale_delete (self)
	Destroy diagonal preconditioner and deallocate all internal storage.
subroutine	gmres_delete (self)
	Destroy diagonal preconditioner and deallocate all internal storage.
subroutine	gmres_setup_xml (self, solver_node, level)
	Setup GMRES solver from XML definition.
recursive subroutine	gmres_solver_apply (self, u, g)
	Solve a linear system using the flexible GMRES method.
subroutine	identity_inv_apply (self, u, g)
	Solve the trivial linear system for the identity matrix.
subroutine	identity_inv_delete (self)
	Destroy diagonal preconditioner and deallocate all internal storage.
subroutine	jblock_precond_apply (self, u, g)
	Apply 1-step of a symmetric Jacobi smoother with native CRS matrices.
logical function, public	jblock_precond_cast (self, source)
	Cast a solver object to a jblock_precond_cast.
subroutine	jblock_precond_delete (self)
	Destroy symmetric Jacobi preconditioner and deallocate all internal storage.
subroutine	jblock_setup_xml (self, solver_node, level)
	Setup symmetric Jacobi smoother from XML definition.
recursive subroutine	ml_precond_apply (self, u, g)
	Apply 1-step of a multi-level preconditioner.
logical function, public	ml_precond_cast (self, source)
	Cast a solver object to a oft_ml_precond.
recursive subroutine	ml_precond_delete (self)
	Destroy Multi-Level preconditioner and deallocate all internal storage.
recursive subroutine	ml_precond_update (self, new_pattern)
	Update solver after changing settings/operators.
recursive subroutine	ml_precond_view (self)
	Print solver configuration.
recursive subroutine	ml_trans_apply (self, u, g)
	Transfer solution between distributed and shared levels as part of a ML preconditioner.
logical function, public	ml_trans_cast (self, source)
	Cast a solver object to a oft_ml_trans.
subroutine	ml_trans_delete (self)
	Destroy Multi-Level preconditioner and deallocate all internal storage.
logical function	native_cg_eigsolver_cast (self, source)
	Cast an eigensolver object to a oft_native_cg_eigsolver.
logical function, public	native_cg_solver_cast (self, source)
	Cast a solver object to a oft_native_cg_solver.
logical function, public	native_gmres_solver_cast (self, source)
	Cast a solver object to a oft_native_gmres_solver.
subroutine	nksolver_apply (self, u, g)
	Solve a nonlinear system Newton's method.
subroutine	nksolver_delete (self)
	Destroy Newton solver and deallocate all internal storage.

Function/Subroutine Documentation

bjprecond_apply()

recursive subroutine bjprecond_apply ( class(oft_bjprecond), intent(inout)  self, class(oft_vector), intent(inout)  u, class(oft_vector), intent(inout)  g  )

private

Precondition a linear system using a Block-Jacobi method.

Parameters

[in,out]	self	Solver object
[in,out]	u	Guess (input), Solution (output)
[in,out]	g	RHS (input), Residual (output)

bjprecond_delete()

subroutine bjprecond_delete ( class(oft_bjprecond), intent(inout)  self )

private

Destroy Block-Jacobi preconditioner and deallocate all internal storage.

Parameters

[in,out]

self

Solver object

bjprecond_setup_xml()

subroutine bjprecond_setup_xml ( class(oft_bjprecond), intent(inout)  self, type(xml_node), intent(in), pointer  solver_node, integer(i4), intent(in), optional  level  )

private

Setup block-Jacobi smoother from XML definition.

Parameters

[in,out]	self	Solver object
[in]	solver_node	XML element containing solver definition
[in]	level	Level in MG hierarchy (optional)

bjprecond_update()

recursive subroutine bjprecond_update ( class(oft_bjprecond), intent(inout)  self, logical, intent(in), optional  new_pattern  )

private

Update solver after changing settings/operators.

Parameters

[in,out]	self	Solver object
[in]	new_pattern	Update matrix non-zero pattern (optional)

cdiag_scale_apply()

subroutine cdiag_scale_apply ( class(oft_diag_cscale), intent(inout)  self, class(oft_cvector), intent(inout)  u, class(oft_cvector), intent(inout)  g  )

private

Solve the linear system for a diagonal matrix (complex)

Parameters

[in,out]	self	Solver object
[in,out]	u	Guess (input), Solution (output)
[in,out]	g	RHS (input), Residual (output)

cdiag_scale_delete()

subroutine cdiag_scale_delete ( class(oft_diag_cscale), intent(inout)  self )

private

Destroy diagonal preconditioner and deallocate all internal storage.

Parameters

[in,out]

self

Solver object

cg_delete()

subroutine cg_delete ( class(oft_native_cg_solver), intent(inout)  self )

private

Destroy diagonal preconditioner and deallocate all internal storage.

Parameters

[in,out]

self

Solver object

cg_eig_delete()

subroutine cg_eig_delete ( class(oft_native_cg_eigsolver), intent(inout)  self )

private

Destroy diagonal preconditioner and deallocate all internal storage.

Parameters

[in,out]

self

Solver object

cg_eigsolver_apply()

subroutine cg_eigsolver_apply ( class(oft_native_cg_eigsolver), intent(inout)  self, class(oft_vector), intent(inout)  u, real(r8), intent(inout)  alam  )

private

Solve a general eigenvalue system using the Conjugate-Gradient method.

This solver uses a Non-Linear Conjugate-Gradient method to minimize the Rayleigh Quotient ( \( R = \frac{x*A*x}{x*M*x} \)).

Parameters

[in,out]	self	Solver object
[in,out]	u	Guess (input), Solution (output)
[in,out]	alam	Eigenvalue

cg_setup_xml()

subroutine cg_setup_xml ( class(oft_native_cg_solver), intent(inout)  self, type(xml_node), intent(in), pointer  solver_node, integer(i4), intent(in), optional  level  )

private

Setup CG solver from XML definition.

Parameters

[in,out]	self	Solver object
[in]	solver_node	XML element containing solver definition
[in]	level	Level in MG hierarchy (optional)

cg_solver_apply()

recursive subroutine cg_solver_apply ( class(oft_native_cg_solver), intent(inout)  self, class(oft_vector), intent(inout)  u, class(oft_vector), intent(inout)  g  )

private

Solve a linear system using the Conjugate-Gradient method.

Parameters

[in,out]	self	Solver object
[in,out]	u	Guess (input), Solution (output)
[in,out]	g	RHS (input), Residual (output)

cgmres_delete()

subroutine cgmres_delete ( class(oft_native_gmres_csolver), intent(inout)  self )

private

Destroy diagonal preconditioner and deallocate all internal storage.

Parameters

[in,out]

self

Solver object

cgmres_setup_xml()

subroutine cgmres_setup_xml ( class(oft_native_gmres_csolver), intent(inout)  self, type(xml_node), intent(in), pointer  solver_node, integer(i4), intent(in), optional  level  )

private

Setup GMRES solver from XML definition.

Parameters

[in,out]	self	Solver object
[in]	solver_node	XML element containing solver definition
[in]	level	Level in MG hierarchy (optional)

cgmres_solver_apply()

recursive subroutine cgmres_solver_apply ( class(oft_native_gmres_csolver), intent(inout)  self, class(oft_cvector), intent(inout)  u, class(oft_cvector), intent(inout)  g  )

private

Solve a linear system using the flexible GMRES method (complex)

Parameters

[in,out]	self	Solver object
[in,out]	u	Guess (input), Solution (output)
[in,out]	g	RHS (input), Residual (output)

diag_scale_apply()

subroutine diag_scale_apply ( class(oft_diag_scale), intent(inout)  self, class(oft_vector), intent(inout)  u, class(oft_vector), intent(inout)  g  )

private

Solve the linear system for a diagonal matrix.

Parameters

[in,out]	self	Solver object
[in,out]	u	Guess (input), Solution (output)
[in,out]	g	RHS (input), Residual (output)

diag_scale_cast()

logical function, public diag_scale_cast	(	type(oft_diag_scale), intent(out), pointer	self,
		class(oft_solver), intent(in), target	source
	)

Cast a solver object to a oft_diag_scale.

The source matrix must be oft_diag_scale or a child class, otherwise pointer will be returned as null and success == .FALSE.

Parameters

[out]	self	Reference to source object with desired class
[in]	source	Source solver to cast

Returns

Cast success flag

diag_scale_delete()

subroutine diag_scale_delete ( class(oft_diag_scale), intent(inout)  self )

private

Destroy diagonal preconditioner and deallocate all internal storage.

Parameters

[in,out]

self

Solver object

gmres_delete()

subroutine gmres_delete ( class(oft_native_gmres_solver), intent(inout)  self )

private

Destroy diagonal preconditioner and deallocate all internal storage.

Parameters

[in,out]

self

Solver object

gmres_setup_xml()

subroutine gmres_setup_xml ( class(oft_native_gmres_solver), intent(inout)  self, type(xml_node), intent(in), pointer  solver_node, integer(i4), intent(in), optional  level  )

private

Setup GMRES solver from XML definition.

Parameters

[in,out]	self	Solver object
[in]	solver_node	XML element containing solver definition
[in]	level	Level in MG hierarchy (optional)

gmres_solver_apply()

recursive subroutine gmres_solver_apply ( class(oft_native_gmres_solver), intent(inout)  self, class(oft_vector), intent(inout)  u, class(oft_vector), intent(inout)  g  )

private

Solve a linear system using the flexible GMRES method.

Parameters

[in,out]	self	Solver object
[in,out]	u	Guess (input), Solution (output)
[in,out]	g	RHS (input), Residual (output)

identity_inv_apply()

subroutine identity_inv_apply ( class(oft_identity_inv), intent(inout)  self, class(oft_vector), intent(inout)  u, class(oft_vector), intent(inout)  g  )

private

Solve the trivial linear system for the identity matrix.

Note

Typically used for simple cases when a solver wrapper is required but the trivial case of \( u = g \) is the desired result

Parameters

[in,out]	self	Solver object
[in,out]	u	Guess (input), Solution (output)
[in,out]	g	RHS (input), Residual (output)

identity_inv_delete()

subroutine identity_inv_delete ( class(oft_identity_inv), intent(inout)  self )

private

Destroy diagonal preconditioner and deallocate all internal storage.

Parameters

[in,out]

self

Solver object

jblock_precond_apply()

subroutine jblock_precond_apply ( class(oft_jblock_precond), intent(inout)  self, class(oft_vector), intent(inout)  u, class(oft_vector), intent(inout)  g  )

private

Apply 1-step of a symmetric Jacobi smoother with native CRS matrices.

Parameters

[in,out]	self	Solver object
[in,out]	u	Guess (input), Solution (output)
[in,out]	g	RHS (input), Residual (output)

jblock_precond_cast()

logical function, public jblock_precond_cast	(	type(oft_jblock_precond), intent(out), pointer	self,
		class(oft_solver), intent(in), target	source
	)

Cast a solver object to a jblock_precond_cast.

The source matrix must be jblock_precond_cast or a child class, otherwise pointer will be returned as null and success == .FALSE.

Parameters

[out]	self	Reference to source object with desired class
[in]	source	Source solver to cast

Returns

Cast success flag

jblock_precond_delete()

subroutine jblock_precond_delete ( class(oft_jblock_precond), intent(inout)  self )

private

Destroy symmetric Jacobi preconditioner and deallocate all internal storage.

Parameters

[in,out]

self

Solver object

jblock_setup_xml()

subroutine jblock_setup_xml ( class(oft_jblock_precond), intent(inout)  self, type(xml_node), intent(in), pointer  solver_node, integer(i4), intent(in), optional  level  )

private

Setup symmetric Jacobi smoother from XML definition.

Parameters

[in,out]	self	Solver object
[in]	solver_node	XML element containing solver definition
[in]	level	Level in MG hierarchy (optional)

ml_precond_apply()

recursive subroutine ml_precond_apply ( class(oft_ml_precond), intent(inout)  self, class(oft_vector), intent(inout)  u, class(oft_vector), intent(inout)  g  )

private

Apply 1-step of a multi-level preconditioner.

Parameters

[in,out]	self	Solver object
[in,out]	u	Guess (input), Solution (output)
[in,out]	g	RHS (input), Residual (output)

ml_precond_cast()

logical function, public ml_precond_cast	(	class(oft_ml_precond), intent(out), pointer	self,
		class(oft_solver), intent(in), target	source
	)

Cast a solver object to a oft_ml_precond.

The source matrix must be oft_ml_precond or a child class, otherwise pointer will be returned as null and success == .FALSE.

Parameters

[out]	self	Reference to source object with desired class
[in]	source	Source solver to cast

Returns

Cast success flag

ml_precond_delete()

recursive subroutine ml_precond_delete ( class(oft_ml_precond), intent(inout)  self )

private

Destroy Multi-Level preconditioner and deallocate all internal storage.

Parameters

[in,out]

self

Solver object

ml_precond_update()

recursive subroutine ml_precond_update ( class(oft_ml_precond), intent(inout)  self, logical, intent(in), optional  new_pattern  )

private

Update solver after changing settings/operators.

Parameters

[in,out]

self

Solver object

ml_precond_view()

recursive subroutine ml_precond_view ( class(oft_ml_precond), intent(inout)  self )

private

Print solver configuration.

Parameters

[in,out]

self

Solver object

ml_trans_apply()

recursive subroutine ml_trans_apply ( class(oft_ml_trans), intent(inout)  self, class(oft_vector), intent(inout)  u, class(oft_vector), intent(inout)  g  )

private

Transfer solution between distributed and shared levels as part of a ML preconditioner.

Parameters

[in,out]	self	Solver object
[in,out]	u	Guess (input), Solution (output)
[in,out]	g	RHS (input), Residual (output)

ml_trans_cast()

logical function, public ml_trans_cast	(	class(oft_ml_trans), intent(out), pointer	self,
		class(oft_solver), intent(in), target	source
	)

Cast a solver object to a oft_ml_trans.

The source matrix must be oft_ml_trans or a child class, otherwise pointer will be returned as null and success == .FALSE.

Parameters

[out]	self	Reference to source object with desired class
[in]	source	Source solver to cast

Returns

Cast success flag

ml_trans_delete()

subroutine ml_trans_delete ( class(oft_ml_trans), intent(inout)  self )

private

Destroy Multi-Level preconditioner and deallocate all internal storage.

Parameters

[in,out]

self

Solver object

native_cg_eigsolver_cast()

logical function native_cg_eigsolver_cast ( type(oft_native_cg_eigsolver), intent(out), pointer  self, class(oft_eigsolver), intent(in), target  source  )

private

Cast an eigensolver object to a oft_native_cg_eigsolver.

The source matrix must be oft_native_cg_eigsolver or a child class, otherwise pointer will be returned as null and success == .FALSE.

Parameters

[out]	self	Reference to source object with desired class
[in]	source	Source solver to cast

Returns

Cast success flag

native_cg_solver_cast()

logical function, public native_cg_solver_cast	(	type(oft_native_cg_solver), intent(out), pointer	self,
		class(oft_solver), intent(in), target	source
	)

Cast a solver object to a oft_native_cg_solver.

The source matrix must be oft_native_cg_solver or a child class, otherwise pointer will be returned as null and success == .FALSE.

Parameters

[out]	self	Reference to source object with desired class
[in]	source	Source solver to cast

Returns

Cast success flag

native_gmres_solver_cast()

logical function, public native_gmres_solver_cast	(	type(oft_native_gmres_solver), intent(out), pointer	self,
		class(oft_solver), intent(in), target	source
	)

Cast a solver object to a oft_native_gmres_solver.

The source matrix must be oft_native_gmres_solver or a child class, otherwise pointer will be returned as null and success == .FALSE.

Parameters

[out]	self	Reference to source object with desired class
[in]	source	Source solver to cast

Returns

Cast success flag

nksolver_apply()

subroutine nksolver_apply ( class(oft_nksolver), intent(inout)  self, class(oft_vector), intent(inout)  u, class(oft_vector), intent(inout)  g  )

private

Solve a nonlinear system Newton's method.

Parameters

[in,out]	self	Solver object
[in,out]	u	Guess (input), Solution (output)
[in,out]	g	RHS (input), Residual (output)

nksolver_delete()

subroutine nksolver_delete ( class(oft_nksolver), intent(inout)  self )

private

Destroy Newton solver and deallocate all internal storage.

Parameters

[in,out]

self

Solver object