[Date Index]
[Thread Index]
[Author Index]
looking inside FindMininum
*To*: mathgroup at smc.vnet.net
*Subject*: [mg29304] looking inside FindMininum
*From*: "Johannes Ludsteck" <johannes.ludsteck at wiwi.uni-regensburg.de>
*Date*: Tue, 12 Jun 2001 04:18:22 -0400 (EDT)
*Organization*: Universitaet Regensburg
*Sender*: owner-wri-mathgroup at wolfram.com
Dear Mathematica Gurus,
I would like to minimize a function f[x,y]with
respect to the arguments (x,y). The special
problem ist that the simple y = 1-x.x > 0 where x
is (x1,x2...,xn) must hold.
I do this by minimizing
f[x,y] + barrier[1-x.x]
where barrier is a linear function of 1-x.x and
grows if y approaches 0 or becomes < 0:
barrier[x_,z_]:=If[#<z,10(1-#/z),0.0]&[1-x.x];
I. e. barrier > 0 if 1-x.x < z. Where z is a
small positive number. z is set to a small number
when the optimization start, but should approach
0 in the course of the Optimization routine in
order to avoid distortions of the search.
I would like to decrease z after every
optimization step. But in order to do this I have
to ask FindMininum for the number of steps.
Until now I do this by
z=0.1
FindMinimum[f[y,x]+barrier[1x.x,z*=decrement],...]
but this is very unsatisfactory, because z is
decremented every time when f[]+barrier[] is
evaluated.
I would like to adjust z in a more sophisticated
way. But this is possible only if there is any
way to get more information out of FindMininum,
for example number of iterations etc.
Is there any way to milk FindMiniumum?
By the way: derivatives of f[y,x] cannot be
computed. Thus I provide two starting values for
every x and FindMinimum use a derivative-free
method.
Thank you and best regards,
Johannes Ludsteck
<><><><><><><><><><><><><><><><><><>
Johannes Ludsteck
Institut fuer Volkswirtschaftslehre
Universitaet Regensburg
Universitaetsstrasse 31
93053 Regensburg
Tel +49/0941/943-2741
Prev by Date:
**Re: JLink on Linux with Sun JDK 1.3.1**
Next by Date:
**Re: Re: OOP in Mathematica**
Previous by thread:
**Re: MathML**
Next by thread:
**Re: looking inside FindMininum**
| |