SimpleNewtonNonlinearSolver Class Reference

#include <SimpleNewtonNonlinearSolver.hpp>

Collaboration diagram for SimpleNewtonNonlinearSolver:

Public Member Functions
	SimpleNewtonNonlinearSolver (double linearSolverRelativeTolerance=1e-6)
virtual	~SimpleNewtonNonlinearSolver ()
virtual Vec	Solve (PetscErrorCode(pComputeResidual)(SNES, Vec, Vec, void ), PetscErrorCode(pComputeJacobian)(SNES, Vec, Mat , Mat , MatStructure , void ), Vec initialGuess, unsigned fill, void pContext)
void	SetTolerance (double tolerance)
void	SetWriteStats (bool writeStats=true)
Private Attributes
double	mLinearSolverRelativeTolerance
double	mTolerance
bool	mWriteStats
std::vector< double >	mTestDampingValues

Detailed Description

A solver class that uses Newton's method with damping.

Definition at line 39 of file SimpleNewtonNonlinearSolver.hpp.

Constructor & Destructor Documentation

SimpleNewtonNonlinearSolver::SimpleNewtonNonlinearSolver ( double linearSolverRelativeTolerance = 1e-6 )

Constructor.

Parameters:

linearSolverRelativeTolerance

defaults to 1e-6

Definition at line 34 of file SimpleNewtonNonlinearSolver.cpp.

References mTestDampingValues.

SimpleNewtonNonlinearSolver::~SimpleNewtonNonlinearSolver ( ) [virtual]

Destructor.

Definition at line 48 of file SimpleNewtonNonlinearSolver.cpp.

Member Function Documentation

void SimpleNewtonNonlinearSolver::SetTolerance ( double tolerance )

Set a tolerance other than the default.

Parameters:

tolerance

Definition at line 159 of file SimpleNewtonNonlinearSolver.cpp.

References mTolerance.

void SimpleNewtonNonlinearSolver::SetWriteStats ( bool writeStats = true ) [inline]

Call to set the solver to write details as it solves.

Parameters:

writeStats

defaults to true

Definition at line 128 of file SimpleNewtonNonlinearSolver.hpp.

References mWriteStats.

Vec SimpleNewtonNonlinearSolver::Solve	(	PetscErrorCode()(SNES, Vec, Vec, void )	pComputeResidual,
		PetscErrorCode()(SNES, Vec, Mat , Mat , MatStructure , void *)	pComputeJacobian,
		Vec	initialGuess,
		unsigned	fill,
		void *	pContext
	)			`[virtual]`

Solve().

Solve a nonlinear system using Newton's method with damping. Newton's algorithm is

x_new = x_old - J^{-1} f

where J is the Jacobian matrix evaluated at x_old and f the residual evaluated at x_old. The Newton method with damping is

x_new = x_old - s J^{-1} f

where s is some damping factor. Here s is chosen by just looked at a fixed set of possible damping factors and choosing the one which gives the best x_new (the one for which the residual evaluated at x_new has the lowest norm).

The solver quits once the ||f||/numVariables

Parameters:

	pComputeResidual	points to the function which computes the residual, it must take arguments SNES (a PETSc nonlinear solver object), Vec (current guess - a vector of the correct size), Vec (a Vec of the correct size in which the residual is returned), void* (a pointer to anything you may need to refer to when calculating the residual)
	pComputeJacobian	points to the function which computes the Jacobian, it must take arguments SNES (a PETSc nonlinear solver * object), Mat* (a pointer to the Jacobian matrix) ,Mat* (a pointer to a preconditioner matrix), MatStructure* (points to the PETSc matrix type e.g. AIJ), void* (a pointer to anything you may need to refer to when calculating the residual).
	initialGuess	A PETSc Vec of the correct size, containing initial guesses for the nonlinear solver.
	pContext	[optional] A pointer to a class that may have to be used in the ComputeResidual and ComputeJacobian functions
	fill	the expected maximum number of nonzeros in a row of the Jacobian matrix