Re: [Bug-gsl] Possible bug (or room for improvement) in gsl_mmin_simp2

[Oliver Jennrich]

On Mon, Jul 27, 2009 at 3:22 PM, Andrew W.
Steiner<address@hidden> wrote:
> Hello all,
>
>     I've been finding that the new simplex algorithm seems to fail
> to converge much more frequently than the old one. It took me
> awhile to generate a sufficiently portable case which demonstrates
> this, but attached is a program which shows that the simplex
> minimizer succeeds, but the simplex2 minimizer stalls and gets
> stuck (the true minimum is at (1,0,0)).

Well, that makes two of us. I have made a similar observation -
simplex2 stalls at some point, never improving the size of the
simplex, whereas th eoriginal simplex converges effortlessly.

>
>    Unfortunately I haven't yet figured out why it's stalling, but
> in the meantime I thought I'd post this code and see if anyone
> else has been having similar difficulties. It's a bit of an unusual
> function, but it's continuous everywhere and all of the other
> minimizers I try happily succeed without complaint.

One of the differences I found while briefly looking at the code is
the different calculation of the simplex size. 'simplex' uses and |..|
-norm, wheras simplex2 uses |..|^2. The latter is much more
susceptible to very asymmetric simplices.

Additional, the point updater is different, tracking the changes in
size and in the center point as well.

The critical thing seems to be the calculation of the size.




-- 
Space -- the final frontier