#include <newton_solver.h>

Inheritance diagram for Hermes::Hermes2D::NewtonSolver< Scalar >:

Public Member Functions
	NewtonSolver (DiscreteProblem< Scalar > *dp)

	NewtonSolver (const WeakForm< Scalar > wf, const Space< Scalar > space)

	NewtonSolver (const WeakForm< Scalar > wf, Hermes::vector< const Space< Scalar > > spaces)

void	init_linear_solver ()

virtual bool	isOkay () const
	State querying helpers.

std::string	getClassName () const
	Get class name, for the purpose of messaging.

void	solve (Scalar *coeff_vec=NULL)

void	solve (Solution< Scalar > *initial_guess)

void	solve (Hermes::vector< Solution< Scalar > * > initial_guess)

void	solve_keep_jacobian (Scalar *coeff_vec=NULL)

void	solve_keep_jacobian (Solution< Scalar > *initial_guess)

void	solve_keep_jacobian (Hermes::vector< Solution< Scalar > * > initial_guess)

void	set_max_allowed_residual_norm (double max_allowed_residual_norm_to_set)

void	set_min_allowed_damping_coeff (double min_allowed_damping_coeff_to_set)

virtual void	set_iterative_method (const char *iterative_method_name)
	Call NonlinearSolver::set_iterative_method() and set the method to the linear solver (if applicable).

virtual void	set_preconditioner (const char *preconditioner_name)
	Call NonlinearSolver::set_preconditioner() and set the method to the linear solver (if applicable).

void	set_residual_as_function ()

void	set_newton_tol (double newton_tol)

void	set_newton_max_iter (int newton_max_iter)

virtual void	set_time (double time)

virtual void	set_time_step (double time_step)

virtual void	set_spaces (Hermes::vector< const Space< Scalar > * > spaces)
	See the class Hermes::Hermes2D::Mixins::SettableSpaces.

virtual void	set_space (const Space< Scalar > *space)

virtual Hermes::vector< const Space< Scalar > * >	get_spaces () const
	Get all spaces as a Hermes::vector.

void	set_manual_damping_coeff (bool onOff, double coeff=1.0)

void	set_initial_auto_damping_coeff (double coeff)

void	set_auto_damping_ratio (double ratio)

void	set_sufficient_improvement_factor (double ratio)

void	set_necessary_successful_steps_to_increase (unsigned int steps)

void	set_weak_formulation (const WeakForm< Scalar > *wf)
	Set the weak forms.

Public Member Functions inherited from Hermes::Hermes2D::Mixins::SettableSpaces< Scalar >
virtual const Space< Scalar > *	get_space (int n) const

Public Member Functions inherited from Hermes::Hermes2D::Mixins::MatrixRhsOutput< Scalar >
	MatrixRhsOutput ()

void	output_matrix (int firstIterations=-1)

void	set_print_zero_matrix_entries (bool to_set)
	Sets this instance to output matrix entries even though they are zero or not.

void	set_matrix_filename (std::string name)

void	set_matrix_varname (std::string name)

void	set_matrix_E_matrix_dump_format (EMatrixDumpFormat format)

void	set_matrix_number_format (char *number_format)

void	output_rhs (int firstIterations=-1)

void	set_rhs_filename (std::string name)

void	set_rhs_varname (std::string name)

void	set_rhs_E_matrix_dump_format (EMatrixDumpFormat format)

void	set_rhs_number_format (char *number_format)

Public Member Functions inherited from Hermes::Hermes2D::Mixins::StateQueryable
void	check () const
	Method to handle the state.

Protected Member Functions
void	init_attributes ()
	Internal setting of default values (see individual set methods).

Protected Attributes
const bool	own_dp
	This instance owns its DP.

SparseMatrix< Scalar > *	kept_jacobian
	Used by method solve_keep_jacobian().

SparseMatrix< Scalar > *	jacobian
	Jacobian.

Vector< Scalar > *	residual
	Residual.

LinearMatrixSolver< Scalar > *	linear_solver
	Linear solver.

double	newton_tol

int	newton_max_iter

bool	residual_as_function

double	max_allowed_residual_norm

double	min_allowed_damping_coeff

double	currentDampingCofficient

bool	manual_damping
	Manual / auto.

double	auto_damping_ratio
	The ratio between two damping coeffs when changing.

double	initial_auto_damping_ratio
	The initial (and maximum) damping coefficient.

double	sufficient_improvement_factor
	Sufficient improvement for continuing.

unsigned int	necessary_successful_steps_to_increase
	necessary number of steps to increase back the damping coeff.

Protected Attributes inherited from Hermes::Hermes2D::Mixins::MatrixRhsOutput< Scalar >
bool	print_matrix_zero_values

bool	output_matrixOn

int	output_matrixIterations

std::string	matrixFilename

std::string	matrixVarname

EMatrixDumpFormat	matrixFormat

char *	matrix_number_format

bool	output_rhsOn

int	output_rhsIterations

std::string	RhsFilename

std::string	RhsVarname

EMatrixDumpFormat	RhsFormat

char *	rhs_number_format

Detailed Description

template<typename Scalar>
class Hermes::Hermes2D::NewtonSolver< Scalar >

Class for Newton's method.
Typical usage:
// Initialize Newton's solver.
// Here wf is Hermes2D::WeakForm<double>, space is Hermes2D::Space<double>
Hermes::Hermes2D::NewtonSolver<double> newton_solver(&wf, &space);
Set a whole bunch of parameters according to your liking.
See the class documentation for all possible parameters.
newton_solver.set_newton_tol(1e-6);
newton_solver.set_newton_max_iter(15);
newton_solver.set_max_allowed_residual_norm(1e6);
newton_solver.set_min_allowed_damping_coeff(1e-3);

// Solve the linear problem.
try
{
// Just call solve().
newton_solver.solve();

// Get the solution vector from the solver.
double* sln_vector = newton_solver.get_sln_vector();

// Translate the solution vector into the previously initialized Solution<double> using the static method vector_to_solution.
Hermes::Hermes2D::Solution<double>::vector_to_solution(sln_vector, &space, &sln);
}
// All kinds of Exceptions may happen (Linear algebraic solver, some bad parameters, some data not initialized...)
catch(Hermes::Exceptions::Exception& e)
{
e.print_msg();
return -1;
}
// For illustrative purposes, otherwise one can just catch std::exception directly, as Hermes::Exceptions::Exception derive from it.
catch(std::exception& e)
{
std::cout << e.what();
return -1;
}

Definition at line 71 of file newton_solver.h.

Member Function Documentation

template<typename Scalar >

void Hermes::Hermes2D::NewtonSolver< Scalar >::set_auto_damping_ratio ( double ratio )

Set the ratio to the automatic damping. When the damping coefficient is decided to be descreased or increased, this is the ratio how it will be changed (this is the bigger ( > 1.0 ) of the two possible values). I.e. when the damping coefficient is shortened 3 times if deemed too big, make the parameter not 0.333333, but 3.0. Default: 2.0

Parameters

[in] ratio The ratio (again, it must be > 1.0, and it represents the inverse of the shortening factor).

Definition at line 213 of file newton_solver.cpp.

template<typename Scalar >

void Hermes::Hermes2D::NewtonSolver< Scalar >::set_initial_auto_damping_coeff ( double coeff )

Make the automatic damping start with this coefficient. This will also be the top bound for the coefficient. Default: 1.0

Parameters

[in] coeff The initial damping coefficient. Must be > 0 and <= 1.0.

Definition at line 202 of file newton_solver.cpp.

template<typename Scalar >

void Hermes::Hermes2D::NewtonSolver< Scalar >::set_manual_damping_coeff	(	bool	onOff,
		double	coeff = `1.0`
	)

Turn on or off manual damping (default is the automatic) and optionally sets manual damping coefficient. Default: default is the automatic damping, default coefficient if manual damping used is set by this method.

Parameters

[in]	onOff	on(true)-manual damping, off(false)-automatic damping.
[in]	coeff	The (perpetual) damping coefficient in the case of manual damping. Ignored in the case of automatic damping.

Definition at line 188 of file newton_solver.cpp.

template<typename Scalar >

void Hermes::Hermes2D::NewtonSolver< Scalar >::set_max_allowed_residual_norm ( double max_allowed_residual_norm_to_set )

Sets the maximum allowed norm of the residual during the calculation. Default: 1E9

Definition at line 102 of file newton_solver.cpp.

template<typename Scalar >

void Hermes::Hermes2D::NewtonSolver< Scalar >::set_min_allowed_damping_coeff ( double min_allowed_damping_coeff_to_set )

Sets minimum damping coefficient. Default: 1E-4

Definition at line 110 of file newton_solver.cpp.

template<typename Scalar >

void Hermes::Hermes2D::NewtonSolver< Scalar >::set_necessary_successful_steps_to_increase ( unsigned int steps )

Set how many successful steps are necessary for the damping coefficient to be increased, by multiplication by the parameter set by set_auto_damping_ratio(). The coefficient is then increased after each 'successful' step, if the sequence of such is not interrupted by an 'unsuccessful' step. Default: 1

Parameters

[in] steps Number of steps.

Definition at line 233 of file newton_solver.cpp.

template<typename Scalar >

void Hermes::Hermes2D::NewtonSolver< Scalar >::set_newton_max_iter ( int newton_max_iter )

Set the maximum number of Newton's iterations. Default: 15

Definition at line 94 of file newton_solver.cpp.

template<typename Scalar >

void Hermes::Hermes2D::NewtonSolver< Scalar >::set_newton_tol ( double newton_tol )

Set the residual norm tolerance for ending the Newton's loop. Default: 1E-8.

Definition at line 82 of file newton_solver.cpp.

template<typename Scalar >

void Hermes::Hermes2D::NewtonSolver< Scalar >::set_residual_as_function ( )

Interpret the residual as a function. Translate the residual vector into a residual function (or multiple functions) in the corresponding finite element space(s) and measure their norm(s) there. This is more meaningful than just measuring the l2-norm of the residual vector, since in the FE space not all components in the residual vector have the same weight. On the other hand, this is slower as it requires global norm calculation, and thus numerical integration over the entire domain. Therefore this option is off by default.

Definition at line 118 of file newton_solver.cpp.

template<typename Scalar >

void Hermes::Hermes2D::NewtonSolver< Scalar >::set_sufficient_improvement_factor ( double ratio )

Set the ratio of the current residual norm and the previous residual norm necessary to deem a step 'successful'. It can be either > 1.0, meaning that even if the norm increased, the step will be 'successful', or < 1.0, meaning that even though the residual norm goes down, we will further decrease the damping coefficient. Default: 0.95 param[in] ratio The ratio, must be positive.

Definition at line 223 of file newton_solver.cpp.

template<typename Scalar >

void Hermes::Hermes2D::NewtonSolver< Scalar >::set_time ( double time )

virtual

Set time information for time-dependent problems. See the class Hermes::Mixins::TimeMeasurable.

Definition at line 124 of file newton_solver.cpp.

template<typename Scalar >

void Hermes::Hermes2D::NewtonSolver< Scalar >::solve ( Scalar * coeff_vec = NULL )

Solve.

Parameters

[in] coeff_vec Ceofficient vector to start from.

Definition at line 261 of file newton_solver.cpp.

template<typename Scalar >

void Hermes::Hermes2D::NewtonSolver< Scalar >::solve ( Solution< Scalar > * initial_guess )

Solve.

Parameters

[in] initial_guess Solution to start from (which is projected to obtain the initial coefficient vector.

Definition at line 243 of file newton_solver.cpp.

template<typename Scalar >

void Hermes::Hermes2D::NewtonSolver< Scalar >::solve ( Hermes::vector< Solution< Scalar > * > initial_guess )

Solve.

Parameters

[in] initial_guess Solutions to start from (which is projected to obtain the initial coefficient vector.

Definition at line 251 of file newton_solver.cpp.

template<typename Scalar >

void Hermes::Hermes2D::NewtonSolver< Scalar >::solve_keep_jacobian ( Scalar * coeff_vec = NULL )

Solve which keeps jacobian. A solve() method where the jacobian is reused.

Definition at line 524 of file newton_solver.cpp.

template<typename Scalar >

void Hermes::Hermes2D::NewtonSolver< Scalar >::solve_keep_jacobian ( Solution< Scalar > * initial_guess )

Solve which keeps jacobian.

Parameters

[in] initial_guess Solution to start from (which is projected to obtain the initial coefficient vector.

Definition at line 506 of file newton_solver.cpp.

template<typename Scalar >

void Hermes::Hermes2D::NewtonSolver< Scalar >::solve_keep_jacobian ( Hermes::vector< Solution< Scalar > * > initial_guess )

Solve which keeps jacobian.

Parameters

[in] initial_guess Solutions to start from (which is projected to obtain the initial coefficient vector.

Definition at line 514 of file newton_solver.cpp.

Member Data Documentation

template<typename Scalar >

double Hermes::Hermes2D::NewtonSolver< Scalar >::max_allowed_residual_norm

protected

Maximum allowed residual norm. If this number is exceeded, the methods solve() return 'false'. By default set to 1E6. Possible to change via method set_max_allowed_residual_norm().

Definition at line 214 of file newton_solver.h.

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

Public Member Functions

Protected Member Functions

Protected Attributes

Detailed Description

template<typename Scalar> class Hermes::Hermes2D::NewtonSolver< Scalar >

Member Function Documentation

Member Data Documentation

template<typename Scalar>
class Hermes::Hermes2D::NewtonSolver< Scalar >