Re: Multidimentional function optimization

Does anyone know of any implemetations of Powell's method?
There is one in Numerical Recipes, but it is a very poor one.

"Mark Vaughan" <Mark.Vaughan@xxxxxxxxxxxxxxx> wrote in message
news:43a9812b$1@xxxxxxxxxxxxxxxxxxxxxxxxx
>
>
> "Lior Fainshil" <nospam@xxxxxxxxxx> wrote:
>>In many cases where minimization is needed, the function is not elemental.
>>It can be a result of some complicated simulation. In such cases there is
>>no
>>good way of evaluating the derivatives and it is better to work with
>>algorithms that don't need derivatives.
>
>
>
> look for Nelder-Mead and/or Brents method
>
> there are several listings at the URL I mentioned previously
>
>
>
>
>
>
>>
>>"Ben Crain" <bcrain@xxxxxxxxxxx> wrote in message
>>news:43a9587b$1@xxxxxxxxxxxxxxxxxxxxxxxxx
>>> Lior Fainshil wrote:
>>>> Sorry if I wasn't clear. I mean finding the minimum/maximum of
>>>> multidimensional mathematical functions. I can just evaluate the
>>>> function. There is no information about derivatives. I am interested in
>>>> any optimization methods starting from purely mathematical which find
>>>> local maximum in a small number of iterations up to smarter ones which
>>>> attempt to find global maximum and need large numbers of iterations.
>>>> Any
>>>> help will be appreciated.
>>>>
>>>
>>> ok, though I'm afraid I can't be of any help. I've implemented several
>>> standard optimization routines in Delphi, but they all require
>>> derivatives. I intend to do some without derivatives (based, for
>>> starters, on Richard Brent's "Algorithms for Minimization Without
>>> Derivatives"), but haven't done them yet.
>>>
>>> Why can't you work with derivatives? I use a math expression parser
>>> that
>>> calculates derivatives symbolically, so I've got them along with the
>>> functions I'm evaluating. There are probably several that do that. The
>>> one I like is from Alexei Cioina
>>> (http://groups.yahoo.com/group/applied_mathematics/).
>>>
>>> good luck, bc
>>
>>
>


.

Relevant Pages

  • Re: Looking for suggestions on a root search strategy
    ... possible to avoid evaluating f or its derivatives at points ... Let's assume that both f'and f''have an upper bound. ... without an illegal function evaluation. ...
    (sci.math.num-analysis)
  • Re: Evaulate an arithmetic expression
    ... Your derivative code isn't evaluating ... In other words, differentiating ... derivative expression that we need concern ourselves with what X ... in your derivatives. ...
    (comp.lang.lisp)
  • Re: Linear Algebra
    ... I don't really understand why these functions would give poor results. ... converting to Fortran form. ... will do the symbolic derivatives, though not convert to Fortran. ...
    (comp.lang.fortran)
  • Re: Ford Focus
    ... >> What does everyone think of this car? ... Poor? ... > List of most derivatives of Focii. ...
    (uk.rec.cars.misc)