| Re: Constrained Optimization routine [message #45328 is a reply to message #45252] |
Mon, 29 August 2005 10:26  |
James Kuyper
Messages: 425
Registered: March 2000
|
Senior Member |
|
|
Gianluca Li Causi wrote:
> Thanks James, I'd not tried it!
>
> I've done it now upon your suggestion: it really does what I've
> described, but I've realized that it does not work with my specific
> problem, probably because one of my boundary G functions is a discrete
> one!
> In fact i have to minimize F so that a continuous G1 constraint is
> satisfied AND a second discrete constraint, G2, (which is the number of
> local maxima of F, so an integer value) is fixed.
>
> Can you or anybody else suggest me a similar routine which also handle
> discrete functions?
Would it be feasible to iterate over the plausible values of the
discrete constraint, doing a seperate constrained minimization for
each, and then select the value that has the lowest minimum? It's a
brute force method, but it should work, if it doesn't overstrain your
system.
I've seen minimization routines that allowed you to specify a minimum
step size; by setting that step size to 1 for G2, you could get it to
handle this pretty automatically. Unfortunately, CONSTRAINED_MIN
doesn't seem to have this feature.
If the brute-force method is too expensive, you might try writing a
function that takes a single argument, which it rounds to the nearest
integer, and uses as the number of local maxima. It would call
CONSTRAINED_MIN to find the mimimizing combination of the other
variables for that number of maxima, returning the value of that
minimum. This is a one-dimensional function that can itself be minized
by a routine which does have a minimum step size that can be set to 1,
such as NEWTON(..., TOLX=1) or FX_ROOT(..., TOL=1).
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
