[SciPy-user] zeros of a function

[william ratcliff]

You must have a newer edition of the Recipes than I ;>  It's not immediately
obvious to me why this shouldn't work.  While there are local minima,  I
think that the ability of the global optimizer to find the global minima
depends on how deep the holes are.  I've also used fsolve and found in some
cases that it is unable to find a solution depending on how I set the
starting point.  I've had some success with reformulation of the problem as
a global optimization problem, but as you say, there are multiple local
minima and it becomes computationally pricey to search for the global
minima--my plan had been to use some more robust global optimization
strategies (rather than simple simulated annealing)--but before going that
route, it might be worthwhile to investigate other methods.    Has anyone
tried Broyden's Method?  If so, any feeling for its robustness?

Cheers,
William

On 5/23/07, Anne Archibald <peridot.faceted at gmail.com> wrote:
>
> On 23/05/07, william ratcliff <william.ratcliff at gmail.com> wrote:
> > What about re-expressing the problem as a nonlinear optimization problem
> > where you're fitting a function function to a vector of zeros?  Then,
> use
> > your favorite global optimizer to try to find the minima of the "chi^2"
> of
> > this function.  The issue here is that for some functions, even a global
> > optimizer could run into problems...
>
> Global optimization is *hard*. If you make a scalar function by (say)
> computing the squared length of a vector you want to find a zero of,
> you're almost certain to wind up with zillions of spurious local
> minima. They talk about this in Numerical Recipes too.
>
> Anne
> _______________________________________________
> SciPy-user mailing list
> SciPy-user at scipy.org
> http://projects.scipy.org/mailman/listinfo/scipy-user
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.scipy.org/pipermail/scipy-user/attachments/20070523/b86b29ee/attachment.html>