Concrete class implementing Polak-Ribiere-Polyak Conjugate Gradient non-linear algebraic solver. More...

#include <cgPRPNonLinearSolver.h>

Inheritance diagram for dftfe::cgPRPNonLinearSolver:

Public Member Functions
	cgPRPNonLinearSolver (const unsigned int maxNumberIterations, const unsigned int debugLevel, const MPI_Comm &mpi_comm_parent, const double lineSearchTolerance=1.0e-6, const unsigned int lineSearchMaxIterations=10, const double lineSeachDampingParameter=1.0, const double maxIncrementSolLinf=1e+6, const bool isCurvatureOnlyLineSearchStoppingCondition=false)
	Constructor.

	~cgPRPNonLinearSolver ()
	Destructor.

nonLinearSolver::ReturnValueType	solve (nonlinearSolverProblem &problem, const std::string checkpointFileName="", const bool restart=false)
	Solve non-linear problem using Polak-Ribiere-Polyak nonlinar conjugate gradient method.

void	save (const std::string &checkpointFileName)
	Create checkpoint file for current state of the cg solver.

Public Member Functions inherited from dftfe::nonLinearSolver
virtual	~nonLinearSolver ()=0
	Destructor.

Private Member Functions
void	initializeDirection ()
	Initialize direction.

nonLinearSolver::ReturnValueType	lineSearch (nonlinearSolverProblem &problem, const double tolerance, const unsigned int maxNumberIterations, const unsigned int debugLevel, const std::string checkpointFileName="", const int startingIter=-1, const bool isCheckpointRestart=false)
	Perform line search.

std::pair< double, double >	computeDeltaD ()
	Compute delta_d and eta_p.

double	computeEta ()
	Compute eta.

void	computeDeltas ()
	Compute delta new and delta mid.

void	updateDirection ()
	Update direction.

double	computeResidualL2Norm () const
	Compute residual L2-norm.

unsigned int	computeTotalNumberUnknowns () const
	Compute the total number of unknowns in all processors.

bool	updateSolution (const double alpha, const std::vector< double > &direction, nonlinearSolverProblem &problem)
	Update solution x -> x + \alpha direction.

void	load (const std::string &checkpointFileName)
	Load cg solver state from checkpoint file.

Private Attributes
std::vector< double >	d_conjugateDirection
	storage for conjugate direction

std::vector< double >	d_gradient
	storage for the gradient of the nonlinear problem in the current cg step

std::vector< double >	d_steepestDirectionOld

double	d_deltaNew
	intermediate variable for beta computation

double	d_deltaMid
	intermediate variable for beta computation

double	d_deltaOld
	intermediate variable for beta computation

double	d_beta

double	d_gradMax

unsigned int	d_numberUnknowns
	storage for number of unknowns to be solved for in the nonlinear problem

unsigned int	d_iter
	storage for current nonlinear cg iteration count

std::vector< unsigned int >	d_unknownCountFlag

const double	d_lineSearchTolerance
	line search stopping tolerance

const unsigned int	d_lineSearchMaxIterations
	maximum number of line search iterations

double	d_lineSearchDampingParameter

bool	d_isCGRestartDueToLargeIncrement

double	d_maxSolutionIncrementLinf

double	d_alphaChk
	line search data

double	d_etaPChk
	line search data

double	d_etaChk
	line search data

double	d_eta
	line search data

double	d_etaAlphaZeroChk
	line search data

double	d_functionValueChk
	line search data

double	d_functionalValueAfterAlphUpdateChk
	line search data

int	d_lineSearchRestartIterChk
	line search iter

bool	d_useSingleAtomSolutionsInitialGuess

bool	d_isCurvatureOnlyLineSearchStoppingCondition

MPI_Comm	mpi_communicator

const unsigned int	n_mpi_processes

const unsigned int	this_mpi_process

dealii::ConditionalOStream	pcout

Additional Inherited Members
Public Types inherited from dftfe::nonLinearSolver
enum	ReturnValueType { SUCCESS = 0 , FAILURE , LINESEARCH_FAILED , MAX_ITER_REACHED , RESTART }

Protected Member Functions inherited from dftfe::nonLinearSolver
	nonLinearSolver (const unsigned int debugLevel, const unsigned int maxNumberIterations, const double tolerance=0)
	Constructor.

double	getTolerance () const
	Get tolerance.

unsigned int	getMaximumNumberIterations () const
	Get maximum number of iterations.

unsigned int	getDebugLevel () const
	Get debug level.

Protected Attributes inherited from dftfe::nonLinearSolver
const unsigned int	d_debugLevel
	controls the verbosity of the printing

const unsigned int	d_maxNumberIterations
	maximum number of nonlinear solve iterations

const double	d_tolerance
	nonlinear solve stopping tolerance

Detailed Description

Concrete class implementing Polak-Ribiere-Polyak Conjugate Gradient non-linear algebraic solver.

Author: Sambit Das

Constructor & Destructor Documentation

◆ cgPRPNonLinearSolver()

dftfe::cgPRPNonLinearSolver::cgPRPNonLinearSolver	(	const unsigned int	maxNumberIterations,
		const unsigned int	debugLevel,
		const MPI_Comm &	mpi_comm_parent,
		const double	lineSearchTolerance = 1.0e-6,
		const unsigned int	lineSearchMaxIterations = 10,
		const double	lineSeachDampingParameter = 1.0,
		const double	maxIncrementSolLinf = 1e+6,
		const bool	isCurvatureOnlyLineSearchStoppingCondition = false )

Constructor.

Parameters

maxNumberIterations	Maximum number of iterations.
debugLevel	Debug output level: 0 - no debug output 1 - limited debug output 2 - all debug output.
lineSearchTolerance	Tolereance required for line search convergence.
lineSearchMaxIterations	Maximum number of iterations for the line search.
lineSearchDampingParameter	scales the initial line search step

◆ ~cgPRPNonLinearSolver()

dftfe::cgPRPNonLinearSolver::~cgPRPNonLinearSolver ( )

Destructor.

Member Function Documentation

◆ computeDeltaD()

std::pair< double, double > dftfe::cgPRPNonLinearSolver::computeDeltaD ( )

private

Compute delta_d and eta_p.

Returns: Pair containing delta_d and eta_p.

◆ computeDeltas()

void dftfe::cgPRPNonLinearSolver::computeDeltas ( )

private

Compute delta new and delta mid.

◆ computeEta()

double dftfe::cgPRPNonLinearSolver::computeEta ( )

private

Compute eta.

Returns: Value of eta.

◆ computeResidualL2Norm()

double dftfe::cgPRPNonLinearSolver::computeResidualL2Norm ( ) const

private

Compute residual L2-norm.

Returns: Value of the residual L2-norm.

◆ computeTotalNumberUnknowns()

unsigned int dftfe::cgPRPNonLinearSolver::computeTotalNumberUnknowns ( ) const

private

Compute the total number of unknowns in all processors.

Returns: Number of unknowns in all processors.

◆ initializeDirection()

void dftfe::cgPRPNonLinearSolver::initializeDirection ( )

private

Initialize direction.

◆ lineSearch()

nonLinearSolver::ReturnValueType dftfe::cgPRPNonLinearSolver::lineSearch	(	nonlinearSolverProblem &	problem,
		const double	tolerance,
		const unsigned int	maxNumberIterations,
		const unsigned int	debugLevel,
		const std::string	checkpointFileName = "",
		const int	startingIter = -1,
		const bool	isCheckpointRestart = false )

private

Perform line search.

Parameters

problem	nonlinearSolverProblem object (functor) to compute energy and forces.
tolerance	Tolerance (relative) required for convergence.
maxNumberIterations	Maximum number of iterations.
debugLevel	Debug output level: 0 - no debug output 1 - limited debug output 2 - all debug output.

Returns: Return value indicating success or failure.

◆ load()

void dftfe::cgPRPNonLinearSolver::load ( const std::string & checkpointFileName )

private

Load cg solver state from checkpoint file.

◆ save()

void dftfe::cgPRPNonLinearSolver::save ( const std::string & checkpointFileName )

virtual

Create checkpoint file for current state of the cg solver.

Implements dftfe::nonLinearSolver.

◆ solve()

nonLinearSolver::ReturnValueType dftfe::cgPRPNonLinearSolver::solve	(	nonlinearSolverProblem &	problem,
		const std::string	checkpointFileName = "",
		const bool	restart = false )

virtual

Solve non-linear problem using Polak-Ribiere-Polyak nonlinar conjugate gradient method.

Parameters

problem[in]	nonlinearSolverProblem object.
checkpointFileName[in]	if string is non-empty, creates checkpoint file named checkpointFileName for every nonlinear iteration. If restart is set to true, checkpointFileName must match the name of the checkpoint file. Empty string will throw an error.
restart[in]	boolean specifying whether this is a restart solve using the checkpoint file specified by checkpointFileName.

Returns: Return value indicating success or failure.

Implements dftfe::nonLinearSolver.

◆ updateDirection()

void dftfe::cgPRPNonLinearSolver::updateDirection ( )

private

Update direction.

◆ updateSolution()

bool dftfe::cgPRPNonLinearSolver::updateSolution	(	const double	alpha,
		const std::vector< double > &	direction,
		nonlinearSolverProblem &	problem )

private

Update solution x -> x + \alpha direction.

Parameters

alpha	Scalar value.
direction	Direction vector.
problem	nonlinearSolverProblem object.

Returns: bool true if valid update and false if increment bound exceeded

Member Data Documentation

◆ d_alphaChk

double dftfe::cgPRPNonLinearSolver::d_alphaChk

private

line search data

◆ d_beta

double dftfe::cgPRPNonLinearSolver::d_beta

private

storage for beta- the parameter for updating the conjugate direction d_beta = (d_deltaNew - d_deltaMid)/d_deltaOld

◆ d_conjugateDirection

std::vector<double> dftfe::cgPRPNonLinearSolver::d_conjugateDirection

private

storage for conjugate direction

◆ d_deltaMid

double dftfe::cgPRPNonLinearSolver::d_deltaMid

private

intermediate variable for beta computation

◆ d_deltaNew

double dftfe::cgPRPNonLinearSolver::d_deltaNew

private

intermediate variable for beta computation

◆ d_deltaOld

double dftfe::cgPRPNonLinearSolver::d_deltaOld

private

intermediate variable for beta computation

◆ d_eta

double dftfe::cgPRPNonLinearSolver::d_eta

private

line search data

◆ d_etaAlphaZeroChk

double dftfe::cgPRPNonLinearSolver::d_etaAlphaZeroChk

private

line search data

◆ d_etaChk

double dftfe::cgPRPNonLinearSolver::d_etaChk

private

line search data

◆ d_etaPChk

double dftfe::cgPRPNonLinearSolver::d_etaPChk

private

line search data

◆ d_functionalValueAfterAlphUpdateChk

double dftfe::cgPRPNonLinearSolver::d_functionalValueAfterAlphUpdateChk

private

line search data

◆ d_functionValueChk

double dftfe::cgPRPNonLinearSolver::d_functionValueChk

private

line search data

◆ d_gradient

std::vector<double> dftfe::cgPRPNonLinearSolver::d_gradient

private

storage for the gradient of the nonlinear problem in the current cg step

◆ d_gradMax

double dftfe::cgPRPNonLinearSolver::d_gradMax

private

◆ d_isCGRestartDueToLargeIncrement

bool dftfe::cgPRPNonLinearSolver::d_isCGRestartDueToLargeIncrement

private

flag which restarts the CG if large increment to the solution vector happens during line search

◆ d_isCurvatureOnlyLineSearchStoppingCondition

bool dftfe::cgPRPNonLinearSolver::d_isCurvatureOnlyLineSearchStoppingCondition

private

◆ d_iter

unsigned int dftfe::cgPRPNonLinearSolver::d_iter

private

storage for current nonlinear cg iteration count

◆ d_lineSearchDampingParameter

double dftfe::cgPRPNonLinearSolver::d_lineSearchDampingParameter

private

damping parameter (0,1] to be multiplied with the steepest descent direction, which controls the initial guess to the line search iteration.

◆ d_lineSearchMaxIterations

const unsigned int dftfe::cgPRPNonLinearSolver::d_lineSearchMaxIterations

private

maximum number of line search iterations

◆ d_lineSearchRestartIterChk

int dftfe::cgPRPNonLinearSolver::d_lineSearchRestartIterChk

private

line search iter

◆ d_lineSearchTolerance

const double dftfe::cgPRPNonLinearSolver::d_lineSearchTolerance

private

line search stopping tolerance

◆ d_maxSolutionIncrementLinf

double dftfe::cgPRPNonLinearSolver::d_maxSolutionIncrementLinf

private

maximum allowed increment (measured as L_{inf}(delta x)) in solution vector beyond which CG is restarted

◆ d_numberUnknowns

unsigned int dftfe::cgPRPNonLinearSolver::d_numberUnknowns

private

storage for number of unknowns to be solved for in the nonlinear problem

◆ d_steepestDirectionOld

std::vector<double> dftfe::cgPRPNonLinearSolver::d_steepestDirectionOld

private

storage for the steepest descent direction of the nonlinear problem in the previous cg step

◆ d_unknownCountFlag

std::vector<unsigned int> dftfe::cgPRPNonLinearSolver::d_unknownCountFlag

private

Storage for vector of flags (0 or 1) with size equal to the size of the solution vector of the nonlinear problem. If the flag value is 1 for an index in the vector, the corresponding entry in the solution vector is allowed to be updated and vice-versa if flag value is 0 for an index.

◆ d_useSingleAtomSolutionsInitialGuess

bool dftfe::cgPRPNonLinearSolver::d_useSingleAtomSolutionsInitialGuess

private

◆ mpi_communicator

MPI_Comm dftfe::cgPRPNonLinearSolver::mpi_communicator

private

◆ n_mpi_processes

const unsigned int dftfe::cgPRPNonLinearSolver::n_mpi_processes

private

◆ pcout

dealii::ConditionalOStream dftfe::cgPRPNonLinearSolver::pcout

private

◆ this_mpi_process

const unsigned int dftfe::cgPRPNonLinearSolver::this_mpi_process

private

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

workspace/include/cgPRPNonLinearSolver.h

Public Member Functions

Private Member Functions

Private Attributes

Additional Inherited Members

Detailed Description

Constructor & Destructor Documentation

◆ cgPRPNonLinearSolver()

◆ ~cgPRPNonLinearSolver()

Member Function Documentation

◆ computeDeltaD()

◆ computeDeltas()

◆ computeEta()

◆ computeResidualL2Norm()

◆ computeTotalNumberUnknowns()

◆ initializeDirection()

◆ lineSearch()

◆ load()

◆ save()

◆ solve()

◆ updateDirection()

◆ updateSolution()

Member Data Documentation

◆ d_alphaChk

◆ d_beta

◆ d_conjugateDirection

◆ d_deltaMid

◆ d_deltaNew

◆ d_deltaOld

◆ d_eta

◆ d_etaAlphaZeroChk

◆ d_etaChk

◆ d_etaPChk

◆ d_functionalValueAfterAlphUpdateChk

◆ d_functionValueChk

◆ d_gradient

◆ d_gradMax

◆ d_isCGRestartDueToLargeIncrement

◆ d_isCurvatureOnlyLineSearchStoppingCondition

◆ d_iter

◆ d_lineSearchDampingParameter

◆ d_lineSearchMaxIterations

◆ d_lineSearchRestartIterChk

◆ d_lineSearchTolerance

◆ d_maxSolutionIncrementLinf

◆ d_numberUnknowns

◆ d_steepestDirectionOld

◆ d_unknownCountFlag

◆ d_useSingleAtomSolutionsInitialGuess

◆ mpi_communicator

◆ n_mpi_processes

◆ pcout

◆ this_mpi_process