This is the mail archive of the
gsl-discuss@sourceware.cygnus.com
mailing list for the GSL project.
Re: multimin question
Hi Fabrice,
Thanks for the detailed reply. You're absolutely right, of course, that
something being mathematically true is only a small part of the picture
when you're working with numerical codes. Even I can understand that
although the same (x,y) mathematically minimizes f=x^2+y^2 and
f=x^2+y^2+1, it's not _necessarily_ equally easy for a numerical code to
find the minimum in both cases.
I've been wandering about in the multimin code for a couple of days and
I think I see the point that's been confusing me (a comment or two in
the code would have helped shorten my wandering about :-) ). Even
though I provide a calculation of the gradient for the multivariate
function I want to minimize, that gradient isn't used in the line search
which minimizes the function along the current direction of steepest
descent. It _is_ used in deciding whether the minimum the line search
finds is good enough or whether you need to embark on another round of
choosing a direction and minimizing.
I'm very interested to see what kind of clever stopping criteria you
come up with!
Cheers,
Dave
--
David Morrison Brookhaven National Laboratory phone: 631-344-5840
Physics Department, Bldg 510 C fax: 631-344-3253
Upton, NY 11973-5000 email: dave@bnl.gov