SYNOPSIS

#include <qnewton.h>

Inherits sc::Optimize.

Public Member Functions

QNewtonOpt (const Ref< KeyVal > &)
The KeyVal constructor.
QNewtonOpt (StateIn &)

void save_data_state (StateOut &)
Save the base classes (with save_data_state) and the members in the same order that the StateIn CTOR initializes them.
void apply_transform (const Ref< NonlinearTransform > &)

void init ()
Initialize the optimizer.
int update ()
Take a step.

Protected Attributes

double maxabs_gradient

double accuracy_

RefSymmSCMatrix ihessian_

Ref< HessianUpdate > update_

Ref< LineOpt > lineopt_

int take_newton_step_

int print_hessian_

int print_x_

int print_gradient_

int linear_

int restrict_

int dynamic_grad_acc_

int force_search_

int restart_

Detailed Description

The QNewtonOpt implements a quasi-Newton optimization scheme.

Constructor & Destructor Documentation

sc::QNewtonOpt::QNewtonOpt (const Ref< KeyVal > &)

The KeyVal constructor. The KeyVal constructor reads the following keywords:

update

This gives a HessianUpdate object. The default is to not update the hessian.

hessian

By default, the guess hessian is obtained from the Function object. This keyword specifies an lower triangle array (the second index must be less than or equal to than the first) that replaces the guess hessian. If some of the elements are not given, elements from the guess hessian will be used.

lineopt

This gives a LineOpt object for doing line optimizations in the Newton direction. The default is to skip the line optimizations.

accuracy

The accuracy with which the first gradient will be computed. If this is too large, it may be necessary to evaluate the first gradient point twice. If it is too small, it may take longer to evaluate the first point. The default is 0.0001.

print_x

If true, print the coordinates each iteration. The default is false.

print_gradient

If true, print the gradient each iteration. The default is false.

print_hessian

If true, print the approximate hessian each iteration. The default is false.

restrict

Use step size restriction when not using a line search. The default is true.

Member Function Documentation

void sc::QNewtonOpt::save_data_state (StateOut &) [virtual]

Save the base classes (with save_data_state) and the members in the same order that the StateIn CTOR initializes them. This must be implemented by the derived class if the class has data.

Reimplemented from sc::Optimize.

int sc::QNewtonOpt::update () [virtual]

Take a step. Returns 1 if the optimization has converged, otherwise 0.

Implements sc::Optimize.

Author

Generated automatically by Doxygen for MPQC from the source code.