Adaptive tolerance selection for linear solver. More...

Public Member Functions
	EisenstatWalkerStrategy (const double newtonTolerance, const double etaMax=0.99, const double gamma=0.1)

void	setTolerance (const double newtonTolerance)

void	setEtaMax (const double etaMax)

void	setGamma (const double gamma)

Detailed Description

Adaptive tolerance selection for linear solver.

Note: Prevents over-solving linear systems far away from solution. Source: "Globally Convergent Inexact Newton Methods", Stanley C. Eisenstat and Homer F. Walker, https://doi.org/10.1137/0804022.; etaMax and gamma are configurable via the constructor, via setEtaMax()/setGamma() setters, or via the parameter system: fem.solver.nonlinear.eisenstatwalker.etamax (default: 0.99) fem.solver.nonlinear.eisenstatwalker.gamma (default: 0.1) For degenerate problems consider etaMax~0.1-0.3, gamma~0.5-0.9.

Constructor & Destructor Documentation

Dune::Fem::EisenstatWalkerStrategy::EisenstatWalkerStrategy	(	const double	newtonTolerance,
		const double	etaMax = `0.99`,
		const double	gamma = `0.1`
	)

inline

constructor

Parameters

[in]	newtonTolerance	absolute tolerance of the Newton method
[in]	etaMax	maximum forcing term (default: 0.99)
[in]	gamma	safeguard parameter (default: 0.1)

void Dune::Fem::EisenstatWalkerStrategy::setEtaMax ( const double etaMax )

inline

Set the maximum allowed forcing term. Reduce from 0.99 towards e.g. 0.1 to make EW less aggressive.

void Dune::Fem::EisenstatWalkerStrategy::setGamma ( const double gamma )

inline

Set the safeguard / superlinear-convergence decay parameter. Increase from 0.1 towards e.g. 0.9 for more conservative behaviour.

void Dune::Fem::EisenstatWalkerStrategy::setTolerance ( const double newtonTolerance )

inline

Set the Newton (nonlinear) tolerance used in the lower bound of eta.

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