OPTPP::OptNIPSLike Class Reference

OptNIPSLike is a derived class of OptConstrNewtonLike. More...

Inheritance diagram for OPTPP::OptNIPSLike:

Collaboration diagram for OPTPP::OptNIPSLike:

Public Member Functions
virtual int	checkConvg ()
	Check to see if algorithm satisfies the convergence criterion.
virtual int	checkDeriv ()
	Compare the analytic gradient with the finite-difference approximation.
void	computeDirDeriv (NEWMAT::ColumnVector &sk, const NEWMAT::ColumnVector &xc, NEWMAT::ColumnVector &derivative)
	Takes three arguments and void return.
NEWMAT::ColumnVector	computeSearch2 (NEWMAT::Matrix &Jacobian, const NEWMAT::ColumnVector &rhs)
virtual int	computeStep (NEWMAT::ColumnVector step)
	Compute the steplength along the search direction.
double	dampenStep (NEWMAT::ColumnVector &step)
	Takes one arguments and returns a real value.
real	getMu () const
virtual void	initHessian ()
	Initialize Hessian of Lagrangian.
NEWMAT::ColumnVector	initMultipliers (const NEWMAT::ColumnVector &df, NEWMAT::Matrix &dcon)
	Takes two arguments and returns a NEWMAT::ColumnVector.
virtual void	initOpt ()
	Initialize algorithmic parameters.
virtual real	merit (int flag, const NEWMAT::ColumnVector &xc, const NEWMAT::ColumnVector &yc, NEWMAT::ColumnVector &zc, NEWMAT::ColumnVector &sc)
real	merit2 (int flag, const NEWMAT::ColumnVector &xc, const NEWMAT::ColumnVector &yc, NEWMAT::ColumnVector &zc, NEWMAT::ColumnVector &sc)
	Takes five arguments and returns a real value.
real	merit3 (int flag, const NEWMAT::ColumnVector &xc, NEWMAT::ColumnVector &zc, NEWMAT::ColumnVector &sc)
	Takes four arguments and returns a real value.
virtual void	optimize ()
	Call the interior-point method.
	OptNIPSLike (int n, TOLS t)
	OptNIPSLike (int n, UPDATEFCN u)
	OptNIPSLike (int n)
	OptNIPSLike ()
	Default Constructor.
virtual void	printStatus (char *s)
	Print status of opt. method.
virtual void	readOptInput ()
	Read user-specified input options.
void	recoverFeasibility (NEWMAT::ColumnVector xinit, CompoundConstraint *constraints, double ftol)
virtual void	reset ()
	Takes zero arguments with void return.
void	setCenteringParameter (real newSigma)
	Sets the value of the centering parameter.
virtual void	setMeritFcn (MeritFcn option)
	Specify the merit function to used in step acceptance test.
void	setMu (real newMu)
	Set the value of the perturbation parameter.
void	setStepLengthToBdry (real newTau)
	Sets the percentage of step taken towards the boundary.
virtual NEWMAT::Matrix	setupMatrix (const NEWMAT::ColumnVector &xc)
virtual NEWMAT::ColumnVector	setupRHS (const NEWMAT::ColumnVector &xplus, const NEWMAT::ColumnVector &yplus, const NEWMAT::ColumnVector &zplus, const NEWMAT::ColumnVector &splus, real mu)
virtual NEWMAT::ColumnVector	setupRHS (const NEWMAT::ColumnVector &xc, real mu)
virtual NEWMAT::SymmetricMatrix	updateH (NEWMAT::SymmetricMatrix &H, int k)=0
	Compute the Hessian of the Lagrangrian or its approximation at iteration k.
void	updateMu (int k)
	Takes one arguments and updates the perturbation parameter.
virtual	~OptNIPSLike ()
	Destructor.
Protected Member Functions
virtual NLP1 *	nlprob () const =0
	pointer to NLP1
Protected Attributes
real	beta_
	penalty parameter for merit function 3
real	dirder_
	directional derivative of a merit function
real	mu_
	pertubation parameter
real	penalty_
	penalty parameter for merit function 2
const real	rho_
	constant set to .5
real	sigmin_
	centering parameter
const real	sw_
	constant
real	taumin_
	percentage of steplength to boundary

Detailed Description

OptNIPSLike is a derived class of OptConstrNewtonLike.

OptNIPSLike provides common data and functionality for OptFDNIPS, OptQNIPS, and OptNIPS.

The OptNIPS algorithm is a C++ implementation of NIPSM, a nonlinear interior-point code developed under MATLAB by Amr El-Bakry at Rice University and NIPSF, a Fortran implementation of the same code written by Frederik Saaf. Additional features include the merit functions proposed by Miguel Argaez and Richard Tapia in "Global Convergence of a Primal-Dual Newton Interior-Point Method for Nonlinear Programming Using a Modified Augmented Lagrange Function" as well as Robert Vanderbei and David Shanno in "An Interior-Point Algorithm For Nonconvex Nonlinear Programming".

Author:: P.J. Williams, Sandia National Laboratories, pwillia@sandia.gov

Definition at line 37 of file OptNIPSLike.h.

Constructor & Destructor Documentation

OPTPP::OptNIPSLike::OptNIPSLike ( ) [inline]

Default Constructor.

See also:: OptNIPSLike(int n); OptNIPSLike(int n, UPDATEFCN u); OptNIPSLike(int n, TOLS t)

Definition at line 56 of file OptNIPSLike.h.

References OPTPP::OptimizeClass::method.

OPTPP::OptNIPSLike::OptNIPSLike ( int n ) [inline]

Parameters:

n

an integer argument.

Definition at line 63 of file OptNIPSLike.h.

References OPTPP::OptimizeClass::method.

OPTPP::OptNIPSLike::OptNIPSLike	(	int	n,
		UPDATEFCN	u
	)			`[inline]`

Parameters:

	n	an integer argument.
	u	a function pointer.

Definition at line 71 of file OptNIPSLike.h.

References OPTPP::OptimizeClass::method.

OPTPP::OptNIPSLike::OptNIPSLike	(	int	n,
		TOLS	t
	)			`[inline]`

Parameters:

	n	an integer argument.
	t	tolerance class reference.

Definition at line 79 of file OptNIPSLike.h.

References OPTPP::OptimizeClass::method.

virtual OPTPP::OptNIPSLike::~OptNIPSLike ( ) [inline, virtual]

Destructor.

Definition at line 87 of file OptNIPSLike.h.

Member Function Documentation

virtual int OPTPP::OptNIPSLike::checkConvg ( ) [virtual]

Check to see if algorithm satisfies the convergence criterion.

Reimplemented from OPTPP::OptConstrNewtonLike.

virtual int OPTPP::OptNIPSLike::checkDeriv ( ) [virtual]

Compare the analytic gradient with the finite-difference approximation.

Reimplemented from OPTPP::OptConstrNewtonLike.

Reimplemented in OPTPP::OptFDNIPS, and OPTPP::OptQNIPS.

void OPTPP::OptNIPSLike::computeDirDeriv	(	NEWMAT::ColumnVector &	sk,
		const NEWMAT::ColumnVector &	xc,
		NEWMAT::ColumnVector &	derivative
	)

Takes three arguments and void return.

Parameters:

	sk	a NEWMAT::ColumnVector with contains the search direction
	xc	a NEWMAT::ColumnVector of current point
	derivative	a NEWMAT::ColumnVector of derivative of cost function

virtual int OPTPP::OptNIPSLike::computeStep ( NEWMAT::ColumnVector sk ) [virtual]

Compute the steplength along the search direction.

If an acceptable step not found, returns an error code = -1.

Reimplemented from OPTPP::OptConstrNewtonLike.

double OPTPP::OptNIPSLike::dampenStep ( NEWMAT::ColumnVector & step )

Takes one arguments and returns a real value.

Parameters:

step

a NEWMAT::ColumnVector with contains the search direction

real OPTPP::OptNIPSLike::getMu ( ) const [inline]

Returns:: The value of mu_, the pertubation parameter

Definition at line 122 of file OptNIPSLike.h.

References mu_.

virtual void OPTPP::OptNIPSLike::initHessian ( ) [virtual]

Initialize Hessian of Lagrangian.

Reimplemented from OPTPP::OptConstrNewtonLike.

Reimplemented in OPTPP::OptDHNIPS, and OPTPP::OptNIPS.

NEWMAT::ColumnVector OPTPP::OptNIPSLike::initMultipliers	(	const NEWMAT::ColumnVector &	df,
		NEWMAT::Matrix &	dcon
	)

Takes two arguments and returns a NEWMAT::ColumnVector.

Parameters:

	df	a NEWMAT::ColumnVector - gradient of obj. function
	dcon	a Matrix - gradient of constraints

Returns:: The initial value of the Lagrange multiplier z

virtual void OPTPP::OptNIPSLike::initOpt ( ) [virtual]

Initialize algorithmic parameters.

Reimplemented from OPTPP::OptConstrNewtonLike.

real OPTPP::OptNIPSLike::merit2	(	int	flag,
		const NEWMAT::ColumnVector &	xc,
		const NEWMAT::ColumnVector &	yc,
		NEWMAT::ColumnVector &	zc,
		NEWMAT::ColumnVector &	sc
	)

Takes five arguments and returns a real value.

Parameters:

	flag	an integer argument
	xc	a NEWMAT::ColumnVector
	yc	a NEWMAT::ColumnVector of Lagrange multipliers
	zc	a NEWMAT::ColumnVector of Lagrange multipliers
	sc	a NEWMAT::ColumnVector of slack variables