Detailed Description

Convergence criterion applying absolute or relative tolerances individually to each component of the whole solution increment vector.

A residual check is not done. If both an absolute and a relative tolerances are specified, at least one of them has to be satisfied.

Definition at line 26 of file ConvergenceCriterionPerComponentDeltaX.h.

#include <ConvergenceCriterionPerComponentDeltaX.h>

Inheritance diagram for NumLib::ConvergenceCriterionPerComponentDeltaX:

[legend]

Collaboration diagram for NumLib::ConvergenceCriterionPerComponentDeltaX:

[legend]

Public Member Functions
	ConvergenceCriterionPerComponentDeltaX (std::vector< double > &&absolute_tolerances, std::vector< double > &&relative_tolerances, const MathLib::VecNormType norm_type)

bool	hasDeltaXCheck () const override

bool	hasResidualCheck () const override

void	checkDeltaX (const GlobalVector &minus_delta_x, GlobalVector const &x) override

void	checkResidual (const GlobalVector &) override
	Check if the residual satisfies the convergence criterion.

void	reset () override

void	setDOFTable (const LocalToGlobalIndexMap &dof_table, MeshLib::Mesh const &mesh) override
	Sets the d.o.f. table used to extract data for a specific component.

Public Member Functions inherited from NumLib::ConvergenceCriterionPerComponent
	ConvergenceCriterionPerComponent (const MathLib::VecNormType norm_type)

Public Member Functions inherited from NumLib::ConvergenceCriterion
	ConvergenceCriterion (const MathLib::VecNormType norm_type)

virtual void	preFirstIteration ()

virtual void	setNoFirstIteration ()

virtual bool	isSatisfied () const
	Tell if the convergence criterion is satisfied.

MathLib::VecNormType	getVectorNormType () const

virtual	~ConvergenceCriterion ()=default

Private Attributes
const std::vector< double >	_abstols

const std::vector< double >	_reltols

LocalToGlobalIndexMap const *	_dof_table = nullptr

MeshLib::Mesh const *	_mesh = nullptr

Additional Inherited Members
Protected Attributes inherited from NumLib::ConvergenceCriterion
bool	_satisfied = true

bool	_is_first_iteration = true

const MathLib::VecNormType	_norm_type

Constructor & Destructor Documentation

◆ ConvergenceCriterionPerComponentDeltaX()

NumLib::ConvergenceCriterionPerComponentDeltaX::ConvergenceCriterionPerComponentDeltaX	(	std::vector< double > &&	absolute_tolerances,
		std::vector< double > &&	relative_tolerances,
		const MathLib::VecNormType	norm_type )

Definition at line 22 of file ConvergenceCriterionPerComponentDeltaX.cpp.

    : ConvergenceCriterionPerComponent(norm_type),
      _abstols(std::move(absolute_tolerances)),
      _reltols(std::move(relative_tolerances))
{
    if (_abstols.size() != _reltols.size())
    {
        OGS_FATAL(
            "The number of absolute and relative tolerances given must be the "
            "same.");
    }
 
    if (_abstols.empty())
    {
        OGS_FATAL("The given tolerances vector is empty.");
    }
}

References _abstols, _reltols, and OGS_FATAL.

Member Function Documentation

◆ checkDeltaX()

void NumLib::ConvergenceCriterionPerComponentDeltaX::checkDeltaX	(	const GlobalVector &	minus_delta_x,
		GlobalVector const &	x )

overridevirtual

Check if the change of the solution between iterations satisfies the convergence criterion.

Parameters

minus_delta_x	the current solution update
x	the new solution from the current iteration

Remarks: The Newton-Raphson solver computes minus_delta_x. x is needed for relative tolerances.

Implements NumLib::ConvergenceCriterion.

Definition at line 43 of file ConvergenceCriterionPerComponentDeltaX.cpp.

{
    if ((!_dof_table) || (!_mesh))
    {
        OGS_FATAL("D.o.f. table or mesh have not been set.");
    }
 
    for (unsigned global_component = 0; global_component < _abstols.size();
         ++global_component)
    {
        // TODO short cut if tol <= 0.0
        auto error_dx = norm(minus_delta_x, global_component, _norm_type,
                             *_dof_table, *_mesh);
        auto norm_x =
            norm(x, global_component, _norm_type, *_dof_table, *_mesh);
 
        INFO(
            "Convergence criterion, component {:d}: |dx|={:.4e}, |x|={:.4e}, "
            "|dx|/|x|={:.4e}",
            global_component, error_dx, norm_x,
            (norm_x == 0. ? std::numeric_limits<double>::quiet_NaN()
                          : (error_dx / norm_x)));
 
        bool const satisfied_abs = error_dx < _abstols[global_component];
        bool const satisfied_rel = checkRelativeTolerance(
            _reltols[global_component], error_dx, norm_x);
 
        _satisfied = _satisfied && (satisfied_abs || satisfied_rel);
    }
}

References _abstols, _dof_table, _mesh, NumLib::ConvergenceCriterion::_norm_type, _reltols, NumLib::ConvergenceCriterion::_satisfied, NumLib::checkRelativeTolerance(), INFO(), NumLib::norm(), and OGS_FATAL.

◆ checkResidual()

void NumLib::ConvergenceCriterionPerComponentDeltaX::checkResidual ( const GlobalVector & residual )

inlineoverridevirtual

Check if the residual satisfies the convergence criterion.

Implements NumLib::ConvergenceCriterion.

Definition at line 40 of file ConvergenceCriterionPerComponentDeltaX.h.

40{}

◆ hasDeltaXCheck()

bool NumLib::ConvergenceCriterionPerComponentDeltaX::hasDeltaXCheck ( ) const

inlineoverridevirtual

Tells if the change of the solution between iterations is checked.

Remarks: This method allows to save some computations if no such check will be done.

Implements NumLib::ConvergenceCriterion.

Definition at line 35 of file ConvergenceCriterionPerComponentDeltaX.h.

35{ return true; }

◆ hasResidualCheck()

bool NumLib::ConvergenceCriterionPerComponentDeltaX::hasResidualCheck ( ) const

inlineoverridevirtual

Tells if the residual is checked.

Remarks: This method allows to save some computations if no such check will be done.

Implements NumLib::ConvergenceCriterion.

Definition at line 36 of file ConvergenceCriterionPerComponentDeltaX.h.

36{ return false; }

◆ reset()

void NumLib::ConvergenceCriterionPerComponentDeltaX::reset ( )

inlineoverridevirtual

Indicate that a new iteration now starts.

A concrete implementation of ConvergenceCriterion might want to check both delta x and the residual. I.e., in a new iteration both checks have to be done anew. This method will make the ConvergenceCriterion forget the result of all checks already done, s.t. all necessary checks will have to be repeated in order to satisfy the ConvergenceCriterion.

Reimplemented from NumLib::ConvergenceCriterion.

Definition at line 42 of file ConvergenceCriterionPerComponentDeltaX.h.

42{ this->_satisfied = true; }

References NumLib::ConvergenceCriterion::_satisfied.

◆ setDOFTable()

void NumLib::ConvergenceCriterionPerComponentDeltaX::setDOFTable	(	const LocalToGlobalIndexMap &	dof_table,
		MeshLib::Mesh const &	mesh )

overridevirtual

Sets the d.o.f. table used to extract data for a specific component.

Implements NumLib::ConvergenceCriterionPerComponent.

Definition at line 75 of file ConvergenceCriterionPerComponentDeltaX.cpp.

{
    _dof_table = &dof_table;
    _mesh = &mesh;
 
    if (_dof_table->getNumberOfGlobalComponents() !=
        static_cast<int>(_abstols.size()))
    {
        OGS_FATAL(
            "The number of components in the DOF table and the number of "
            "tolerances given do not match.");
    }
}