#include <deal.II/lac/slepc_solver.h>

Inheritance diagram for SLEPcWrappers::SolverBase:

Public Member Functions
	SolverBase (SolverControl &cn, const MPI_Comm &mpi_communicator)

virtual	~SolverBase ()

template<typename OutputVector >
void	solve (const PETScWrappers::MatrixBase &A, std::vector< PetscScalar > &eigenvalues, std::vector< OutputVector > &eigenvectors, const unsigned int n_eigenpairs=1)

template<typename OutputVector >
void	solve (const PETScWrappers::MatrixBase &A, const PETScWrappers::MatrixBase &B, std::vector< PetscScalar > &eigenvalues, std::vector< OutputVector > &eigenvectors, const unsigned int n_eigenpairs=1)

template<typename OutputVector >
void	solve (const PETScWrappers::MatrixBase &A, const PETScWrappers::MatrixBase &B, std::vector< double > &real_eigenvalues, std::vector< double > &imag_eigenvalues, std::vector< OutputVector > &real_eigenvectors, std::vector< OutputVector > &imag_eigenvectors, const unsigned int n_eigenpairs=1)

template<typename Vector >
void	set_initial_space (const std::vector< Vector > &initial_space)

void	set_transformation (SLEPcWrappers::TransformationBase &this_transformation)

void	set_target_eigenvalue (const PetscScalar &this_target)

void	set_which_eigenpairs (EPSWhich set_which)

void	set_problem_type (EPSProblemType set_problem)

void	get_solver_state (const SolverControl::State state)

SolverControl &	control () const

Static Public Member Functions
static ::ExceptionBase &	ExcSLEPcWrappersUsageError ()

static ::ExceptionBase &	ExcSLEPcError (int arg1)

static ::ExceptionBase &	ExcSLEPcEigenvectorConvergenceMismatchError (int arg1, int arg2)

Protected Member Functions
void	solve (const unsigned int n_eigenpairs, unsigned int *n_converged)

void	get_eigenpair (const unsigned int index, PetscScalar &eigenvalues, PETScWrappers::VectorBase &eigenvectors)

void	get_eigenpair (const unsigned int index, double &real_eigenvalues, double &imag_eigenvalues, PETScWrappers::VectorBase &real_eigenvectors, PETScWrappers::VectorBase &imag_eigenvectors)

void	set_matrices (const PETScWrappers::MatrixBase &A)

void	set_matrices (const PETScWrappers::MatrixBase &A, const PETScWrappers::MatrixBase &B)

Protected Attributes
SolverControl &	solver_control

const MPI_Comm	mpi_communicator

EPS	eps

Static Private Member Functions
static int	convergence_test (EPS eps, PetscScalar real_eigenvalue, PetscScalar imag_eigenvalue, PetscReal residual_norm, PetscReal estimated_error, void solver_control)

Private Attributes
EPSConvergedReason	reason

Detailed Description

Base class for solver classes using the SLEPc solvers. Since solvers in SLEPc are selected based on flags passed to a generic solver object, basically all the actual solver calls happen in this class, and derived classes simply set the right flags to select one solver or another, or to set certain parameters for individual solvers.

For examples of how this and its derived classes can be used, including how to provide preconditioners to the matrix of which eigenvalues are to be computed, see the documentation of the SolverBase class as well as the extensive discussions in the documentation of the SLEPcWrappers namespace.

Definition at line 146 of file slepc_solver.h.

Constructor & Destructor Documentation

◆ SolverBase()

SolverBase< VectorType >::SolverBase	(	SolverControl &	cn,
		const MPI_Comm &	mpi_communicator
	)

Constructor. Takes the MPI communicator over which parallel computations are to happen.

Definition at line 35 of file slepc_solver.cc.

◆ ~SolverBase()

SolverBase< VectorType >::~SolverBase ( )

virtual

Destructor.

Definition at line 62 of file slepc_solver.cc.

Member Function Documentation

◆ solve() [1/4]

template<typename OutputVector >

void SolverBase< OutputVector >::solve	(	const PETScWrappers::MatrixBase &	A,
		std::vector< PetscScalar > &	eigenvalues,
		std::vector< OutputVector > &	eigenvectors,
		const unsigned int	n_eigenpairs = `1`
	)

Composite method that solves the eigensystem \(Ax=\lambda x\). The eigenvector sent in has to have at least one element that we can use as a template when resizing, since we do not know the parameters of the specific vector class used (i.e. local_dofs for MPI vectors). However, while copying eigenvectors, at least twice the memory size of eigenvectors is being used (and can be more). To avoid doing this, the fairly standard calling sequence executed here is used: Set up matrices for solving; Actually solve the system; Gather the solution(s).

Note: Note that the number of converged eigenvectors can be larger than the number of eigenvectors requested; this is due to a round off error (success) of the eigenproblem solver context. If this is found to be the case we simply do not bother with more eigenpairs than requested, but handle that it may be more than specified by ignoring any extras. By default one eigenvector/eigenvalue pair is computed.

This is declared here to make it possible to take a std::vector of different PETScWrappers vector types

Definition at line 704 of file slepc_solver.h.

◆ solve() [2/4]

template<typename OutputVector >

void SolverBase< OutputVector >::solve	(	const PETScWrappers::MatrixBase &	A,
		const PETScWrappers::MatrixBase &	B,
		std::vector< PetscScalar > &	eigenvalues,
		std::vector< OutputVector > &	eigenvectors,
		const unsigned int	n_eigenpairs = `1`
	)

Same as above, but here a composite method for solving the system \(A x=\lambda B x\), for real matrices, vectors, and values \(A, B, x, \lambda\).

Definition at line 736 of file slepc_solver.h.

◆ solve() [3/4]

template<typename OutputVector >

void SolverBase< OutputVector >::solve	(	const PETScWrappers::MatrixBase &	A,
		const PETScWrappers::MatrixBase &	B,
		std::vector< double > &	real_eigenvalues,
		std::vector< double > &	imag_eigenvalues,
		std::vector< OutputVector > &	real_eigenvectors,
		std::vector< OutputVector > &	imag_eigenvectors,
		const unsigned int	n_eigenpairs = `1`
	)

Same as above, but here a composite method for solving the system \(A x=\lambda B x\) with real matrices \(A, B\) and imaginary eigenpairs \(x, \lambda\).

Definition at line 774 of file slepc_solver.h.

◆ set_initial_space()

template<typename Vector >

void SolverBase< Vector >::set_initial_space ( const std::vector< Vector > & initial_space )

Set the initial vector space for the solver.

By default, SLEPc initializes the starting vector or the initial subspace randomly.

Definition at line 830 of file slepc_solver.h.

◆ set_transformation()

void SolverBase< VectorType >::set_transformation ( SLEPcWrappers::TransformationBase & this_transformation )

Set the spectral transformation to be used.

Definition at line 98 of file slepc_solver.cc.

◆ set_target_eigenvalue()

void SolverBase< VectorType >::set_target_eigenvalue ( const PetscScalar & this_target )

Set target eigenvalues in the spectrum to be computed. By default, no target is set.

Definition at line 126 of file slepc_solver.cc.

◆ set_which_eigenpairs()

void SolverBase< VectorType >::set_which_eigenpairs ( EPSWhich set_which )

Indicate which part of the spectrum is to be computed. By default largest magnitude eigenvalues are computed.

Note: For other allowed values see the SLEPc documentation.

Definition at line 138 of file slepc_solver.cc.

◆ set_problem_type()

void SolverBase< VectorType >::set_problem_type ( EPSProblemType set_problem )

Specify the type of the eigenspectrum problem. This can be used to exploit known symmetries of the matrices that make up the standard/generalized eigenspectrum problem. By default a non-Hermitian problem is assumed.

Note: For other allowed values see the SLEPc documentation.

Definition at line 148 of file slepc_solver.cc.

◆ get_solver_state()

void SolverBase< VectorType >::get_solver_state ( const SolverControl::State state )

Take the information provided from SLEPc and checks it against deal.II's own SolverControl objects to see if convergence has been reached.

Definition at line 307 of file slepc_solver.cc.

◆ control()

SolverControl & SolverBase< VectorType >::control ( ) const

Access to the object that controls convergence.

Definition at line 335 of file slepc_solver.cc.

◆ solve() [4/4]

void SolverBase< VectorType >::solve	(	const unsigned int	n_eigenpairs,
		unsigned int *	n_converged
	)

protected

Solve the linear system for n_eigenpairs eigenstates. Parameter n_converged contains the actual number of eigenstates that have converged; this can be both fewer or more than n_eigenpairs, depending on the SLEPc eigensolver used.

Definition at line 157 of file slepc_solver.cc.

◆ get_eigenpair() [1/2]

void SolverBase< VectorType >::get_eigenpair	(	const unsigned int	index,
		PetscScalar &	eigenvalues,
		PETScWrappers::VectorBase &	eigenvectors
	)

protected

Access the real parts of solutions for a solved eigenvector problem, pair index solutions, \(\text{index}\,\in\,0\dots \mathrm{n\_converged}-1\).

Definition at line 260 of file slepc_solver.cc.

◆ get_eigenpair() [2/2]

void SolverBase< VectorType >::get_eigenpair	(	const unsigned int	index,
		double &	real_eigenvalues,
		double &	imag_eigenvalues,
		PETScWrappers::VectorBase &	real_eigenvectors,
		PETScWrappers::VectorBase &	imag_eigenvectors
	)

protected

Access the real and imaginary parts of solutions for a solved eigenvector problem, pair index solutions, \(\text{index}\,\in\,0\dots \mathrm{n\_converged}-1\).

Definition at line 273 of file slepc_solver.cc.

◆ set_matrices() [1/2]

void SolverBase< VectorType >::set_matrices ( const PETScWrappers::MatrixBase & A )

protected

Initialize solver for the linear system \(Ax=\lambda x\). (Note: this is required before calling solve ())

Definition at line 77 of file slepc_solver.cc.

◆ set_matrices() [2/2]

void SolverBase< VectorType >::set_matrices	(	const PETScWrappers::MatrixBase &	A,
		const PETScWrappers::MatrixBase &	B
	)

protected

Same as above, but here initialize solver for the linear system \(A x=\lambda B x\).

Definition at line 87 of file slepc_solver.cc.

◆ convergence_test()

int SolverBase< VectorType >::convergence_test	(	EPS	eps,
		PetscScalar	real_eigenvalue,
		PetscScalar	imag_eigenvalue,
		PetscReal	residual_norm,
		PetscReal *	estimated_error,
		void *	solver_control
	)

staticprivate

A function that can be used in SLEPc as a callback to check on convergence.

Note: This function is not used currently.

Definition at line 343 of file slepc_solver.cc.

Member Data Documentation

◆ solver_control

SolverControl& SLEPcWrappers::SolverBase::solver_control

protected

Reference to the object that controls convergence of the iterative solver.

Definition at line 297 of file slepc_solver.h.

◆ mpi_communicator

const MPI_Comm SLEPcWrappers::SolverBase::mpi_communicator

protected

Copy of the MPI communicator object to be used for the solver.

Definition at line 302 of file slepc_solver.h.

◆ eps

EPS SLEPcWrappers::SolverBase::eps

protected

Objects for Eigenvalue Problem Solver.

Definition at line 354 of file slepc_solver.h.

◆ reason

EPSConvergedReason SLEPcWrappers::SolverBase::reason

private

Convergence reason.

Definition at line 360 of file slepc_solver.h.

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

include/deal.II/lac/slepc_solver.h
source/lac/slepc_solver.cc

Public Member Functions

Static Public Member Functions

Protected Member Functions

Protected Attributes

Static Private Member Functions

Private Attributes

Detailed Description

Constructor & Destructor Documentation

◆ SolverBase()

◆ ~SolverBase()

Member Function Documentation

◆ solve() [1/4]

◆ solve() [2/4]

◆ solve() [3/4]

◆ set_initial_space()

◆ set_transformation()

◆ set_target_eigenvalue()

◆ set_which_eigenpairs()

◆ set_problem_type()

◆ get_solver_state()

◆ control()

◆ solve() [4/4]

◆ get_eigenpair() [1/2]

◆ get_eigenpair() [2/2]

◆ set_matrices() [1/2]

◆ set_matrices() [2/2]

◆ convergence_test()

Member Data Documentation

◆ solver_control

◆ mpi_communicator

◆ eps

◆ reason