This class computes the solution to using the Nic-Day method. More...

#include <LOCA_SingularJacobianSolve_NicDay.H>

Inheritance diagram for LOCA::SingularJacobianSolve::NicDay:

Collaboration diagram for LOCA::SingularJacobianSolve::NicDay:

Public Member Functions
	NicDay (Teuchos::ParameterList &params)
	Constructor.
	NicDay (const NicDay &source)
	Copy constructor.
virtual	~NicDay ()
	Destructor.
virtual LOCA::SingularJacobianSolve::Generic *	clone () const
	Clone function.
virtual LOCA::SingularJacobianSolve::Generic &	operator= (const LOCA::SingularJacobianSolve::Generic &source)
	Assignment operator.
virtual NicDay &	operator= (const NicDay &source)
	Assignment operator.
virtual NOX::Abstract::Group::ReturnType	reset (Teuchos::ParameterList &params)
	Reset parameters.
virtual NOX::Abstract::Group::ReturnType	compute (Teuchos::ParameterList &params, LOCA::Continuation::AbstractGroup &grp, const NOX::Abstract::Vector &input, const NOX::Abstract::Vector &approxNullVec, const NOX::Abstract::Vector &jacApproxNullVec, NOX::Abstract::Vector &result)
	Computes the solution as described above.
virtual NOX::Abstract::Group::ReturnType	computeMulti (Teuchos::ParameterList &params, LOCA::Continuation::AbstractGroup &grp, const NOX::Abstract::Vector const inputs, const NOX::Abstract::Vector &approxNullVec, const NOX::Abstract::Vector &jacApproxNullVec, NOX::Abstract::Vector **results, int nVecs)
	Computes solution for multiple RHS.
Public Member Functions inherited from LOCA::SingularJacobianSolve::Generic
	Generic ()
	Constructor.
	Generic (const Generic &source)
	Copy constructor.
virtual	~Generic ()
	Destructor.

Detailed Description

This class computes the solution to using the Nic-Day method.

This singular solve method is a modification of the deflation idea implemented in LOCA::SingularJacobianSolve::Nic where deflation of the right hand side is used to improve the conditioning of the linear system. Typically a solution to when is nearly singular will have a large component in the direction of the null vector . The idea then is to deflate out of the right hand side . The complete algorithm used here is:

$\‍[ \begin{aligned} &\tilde{b} = b - \frac{b^T J v}{v^T J^T J v} Jv \\ &\text{Solve}\; J\tilde{x} = \tilde{b} \\ &x = \tilde{x} + \frac{b^T J v}{v^T J^T J v} v \end{aligned} \‍]$

The solve $J\tilde{x} = \tilde{b}$ uses the underlying group's applyJacobianInverse() method and therefore this is a generic technique for computing solutions to nearly singular system since it uses any supplied linear solver.

This algorithm is selected by setting the "Method" parameter of the "Singular Solve" sublist of the NOX linear solver parameter list to "Nic-Day". The idea for this algorithm is taken from: R. A. Nicolaides, "Deflation of Conjugate Gradients With Applications to Boundary Value Problems," SIAM J. Numer. Anal., 24(2), 1987.

Member Function Documentation

◆ clone()

LOCA::SingularJacobianSolve::Generic * LOCA::SingularJacobianSolve::NicDay::clone ( ) const

virtual

Clone function.

Implements LOCA::SingularJacobianSolve::Generic.

References NicDay().

◆ compute()

NOX::Abstract::Group::ReturnType LOCA::SingularJacobianSolve::NicDay::compute	(	Teuchos::ParameterList &	params,
		LOCA::Continuation::AbstractGroup &	grp,
		const NOX::Abstract::Vector &	input,
		const NOX::Abstract::Vector &	approxNullVec,
		const NOX::Abstract::Vector &	jacApproxNullVec,
		NOX::Abstract::Vector &	result )

virtual

Computes the solution as described above.

Implements LOCA::SingularJacobianSolve::Generic.

References LOCA::ErrorCheck::checkReturnType(), NOX::Abstract::Vector::clone(), NOX::DeepCopy, NOX::Abstract::Vector::innerProduct(), and NOX::Abstract::Vector::update().

◆ computeMulti()

NOX::Abstract::Group::ReturnType LOCA::SingularJacobianSolve::NicDay::computeMulti	(	Teuchos::ParameterList &	params,
		LOCA::Continuation::AbstractGroup &	grp,
		const NOX::Abstract::Vector const	inputs,
		const NOX::Abstract::Vector &	approxNullVec,
		const NOX::Abstract::Vector &	jacApproxNullVec,
		NOX::Abstract::Vector **	results,
		int	nVecs )