com.numericalmethod.suanshu.optimization.unconstrained.steepestdescent
Class SteepestDescent.LineSearch

java.lang.Object
  com.numericalmethod.suanshu.optimization.unconstrained.steepestdescent.SteepestDescent.LineSearch

Direct Known Subclasses:

QuasiNewton.QuasiNewtonImpl

Enclosing class:

SteepestDescent

public abstract class SteepestDescent.LineSearch
extends java.lang.Object

A steepest-descent method, in each iteration, searches along a direction to find the next best minimizer along a direction. We want to find out how much the minimizer moves in this direction.

By making LineSearch an inner class, its methods will have access to the members in SteepestDescent, e.g., f. This is to avoid storing the same datum in multiple places, hence no inconsistency problem. For the very same reason, this class cannot be made static in order to access the non-static members in SteepestDescent, e.g., f. TODO: Consequently, (a problem?) the member storing an instance of LineSearch, e.g., lsearch, cannot be passed in the constructors. A non-static inner class cannot be created before the enclosing class is created.

Field Summary

protected Vector	dk the line search direction at the k-th iteration
protected Vector	gk the gradient evaluated at xk
RntoMatrix	H the Hessian function, H(x)
protected Matrix	Hk the Hessian evaluated at xk

Constructor Summary

protected

SteepestDescent.LineSearch(RntoMatrix H) Construct a line search instance.

Method Summary

protected abstract double	minimize(Vector xk) Perform a line search to minimize a function, starting from xk.
protected void	reset() Reset the transient states of the instance.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

H

public final RntoMatrix H

the Hessian function, H(x)

gk

protected Vector gk

the gradient evaluated at xk

Hk

protected Matrix Hk

the Hessian evaluated at xk

dk

protected Vector dk

the line search direction at the k-th iteration

Constructor Detail

SteepestDescent.LineSearch

protected SteepestDescent.LineSearch(RntoMatrix H)

Construct a line search instance.

Parameters:

H - the function to compute the Hessian at x

Method Detail

reset

protected void reset()

Reset the transient states of the instance.

minimize

protected abstract double minimize(Vector xk)

Perform a line search to minimize a function, starting from xk.

Parameters:

xk - the current minimizer

Returns:

ak+1