powell.3.man

COOOL：CWP面向对象最优化库（CWP Object Oriented Optimization Library） COOOL是C++类的一个集合

POWELL(derived)   OPTIMIZATION ALGORITHM    POWELL(derived)

                                                     Jun  1 15:19
NAME
    Powell - Powell Optima

SYNOPSIS
    #include <Powell.hh>

    class PowellOptima : public LineSearchOptima

        \fIPublic members\fP
            PowellOptima(LineSearch*,  int, double, double, double);
            PowellOptima(LineSearch*,  int, double, double, double, int);
            ~PowellOptima();
            Model<double> optimizer(Model<double>&);
            Model<long> optimizer(Model<long>&);
            const char* className() const;

        \fIPrivate members\fP
            double 			delta;
            double 			change;

DESCRIPTION
    Powell()
    Powell's method can be considered a derivative-free version
    of the conjugate gradient algorithm (Powell, M., Computer
     Journal, 7, 155-162). Here the objective function
    is minimized from an initial model along a set of conjugate
    directiions generatied by the procedure without resorting
    to the gradient of the objective function.
    

DESCRIPTION
    Constructors: 
     PowellOptima(LineSearch* ls,  int iter, double tol, 
                 double changeOf, double delta);
      ls: Defines the line search used in the optimization
         (At the present version you should use the 
         BrentLineSearch procedure)
      iter: Maximum number of iterations
      tol: Tolerance used in the line search
      changeOf: This is the minimum relative change of the 
        objective function value
        to stop the optimization (stopping criterion)
      delta: Initial step used in line search
    Methods:
     Model<double> optimizer(Model<double>&model0);
      model0: Initial model for the Powell's conjugate 
             gradient procedure
     The optimum model is returned by the function.
    

CAVEATS
    Although an interesting algorithm, Powell's method has 
    demonstrated to have a high computational cost, due 
    the excessive number of line search required to compute 
    the conjugate directions (see Applied Nonlinear Programming, 
    by David Himmelblau for details. It seems that the standard
    conjugate gradient by Hesteness and Stiefel with 
    numerical derivatives is a more efficient procedure.

DEFINED MACROS
    POWELL_OPTIMA_HH

INCLUDED FILES
    "LSearchOptima.hh"

SOURCE FILES
    Powell.cc