Detailed Description

Abstract globalization and step-acceptance strategy for Newton-Raphson iterations.

Concrete implementations (e.g. FixedDampingStrategy) decide how to scale the raw Newton direction \( -\Delta x \) before it is applied to the current solution. The NonlinearSolver<Newton> owns one strategy instance and calls applyStep() once per iteration.

Definition at line 57 of file NewtonStepStrategy.h.

#include <NewtonStepStrategy.h>

Inheritance diagram for NumLib::NewtonStepStrategy:

[legend]

Collaboration diagram for NumLib::NewtonStepStrategy:

[legend]

Public Member Functions
virtual void	initialize (GlobalVector const &)
virtual StepResult	applyStep (GlobalVector const &x, GlobalVector const &minus_delta_x, GlobalVector const &res, GlobalMatrix const &J, GlobalVector &x_new, NewtonStepContext &ctx, int iteration)=0
void	setDampingPolicy (DampingPolicy const *policy)
virtual	~NewtonStepStrategy ()=default

Protected Attributes
DampingPolicy const *	_damping_policy = nullptr

Constructor & Destructor Documentation

◆ ~NewtonStepStrategy()

virtual NumLib::NewtonStepStrategy::~NewtonStepStrategy ( )

virtualdefault

Member Function Documentation

◆ applyStep()

virtual StepResult NumLib::NewtonStepStrategy::applyStep	(	GlobalVector const &	x,
		GlobalVector const &	minus_delta_x,
		GlobalVector const &	res,
		GlobalMatrix const &	J,
		GlobalVector &	x_new,
		NewtonStepContext &	ctx,
		int	iteration )

pure virtual

Given the current solution x, the Newton direction minus_delta_x, the residual res, and the Jacobian J, compute and apply a step to x_new. The strategy is free to re-evaluate F(x) internally.

Parameters

[in]	x	current solution
[in]	minus_delta_x	full Newton step (J^{-1} r)
[in]	res	residual at x
[in]	J	Jacobian at x (may be used for curvature)
[out]	x_new	accepted new solution
[in]	ctx	equation system context for re-evaluation
[in]	iteration	current outer iteration number

Implemented in NumLib::DampingReductionStrategy, and NumLib::FixedDampingStrategy.

◆ initialize()

virtual void NumLib::NewtonStepStrategy::initialize ( GlobalVector const & )

inlinevirtual

Called once before the outer Newton loop starts. The default implementation is a no-op; override if the strategy needs to cache the initial solution (e.g. for trust-region radius initialisation).

Definition at line 64 of file NewtonStepStrategy.h.

64{}

◆ setDampingPolicy()

void NumLib::NewtonStepStrategy::setDampingPolicy ( DampingPolicy const * policy )

inline

Inject the damping policy from the convergence criterion. Called by NonlinearSolver::setEquationSystem() before the first solve. A null pointer means the criterion imposes no step-length constraint.

Definition at line 87 of file NewtonStepStrategy.h.

    {
        _damping_policy = policy;
    }

References _damping_policy.

Member Data Documentation

◆ _damping_policy

DampingPolicy const* NumLib::NewtonStepStrategy::_damping_policy = nullptr

protected

Definition at line 95 of file NewtonStepStrategy.h.

Referenced by NumLib::DampingReductionStrategy::applyStep(), NumLib::FixedDampingStrategy::applyStep(), and setDampingPolicy().

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

NewtonStepStrategy.h