multifit non-linear least square fitting code

["Klassen, Timothy" <Timothy dot Klassen at gs dot com>]

Hi folks,

i had seem some remarks that there is a bug in the NumRec
Levenberg-Marquardt code, which might make it slow or even make it fail to
converge in not-too-difficult cases. Unfortunately i could never figure out
if this only refers to the code from the 1st ed, ie if it was fixed in the
2nd or what  (if one of you knows, I'd appreciate if you let me know...).

In any case, i figured i'd try the gsl code in the multifit project.  I used
the MGH17 from the NIST site as an example; cf. attachment. I found that the
gsl code does not converge to the right answer (even if started very close
to the answer) whereas the NumRec code does (albeit slowly sometimes).  I
then noticed that there is a commented out "nelson" example in the test.c
file, which also does not seem to converge. 

I'd appreciate any comments on issues with the gsl and num recipes
non-linear fitting routines.

Thanks,
Tim


 <<test_code.uu>>

Attachment: test_code.uu
Description: Binary data