Re: [Help-gsl] Levenberg-Marquardt misbehaves...

[Joakim Hove]

Federico Zenith <address@hidden> writes:


> How to I tell LM to stay away from the negative domain?

You could use squared parameters, i.e. if want to determine 'a' in
exp(-a*x) for strictly positive a, you could instead use a = b*b and
determine b. This guarantees that a = b*b is >= 0. Of course the _df
evaluations become slightly more involved in this way.


Good luck

Joakim




-- 
Joakim Hove
hove AT ift uib no
Tlf: +47 (55 5)8 27 90 
Fax: +47 (55 5)8 94 40
http://www.ift.uib.no/~hove/