This is the mail archive of the
gsl-discuss@sources.redhat.com
mailing list for the GSL project.
Re: multimin convergence problem
>>>>> "Brian" == Brian Gough <bjg@network-theory.co.uk> writes:
Brian> I think there are some terms missing in your derivatives
Brian> (e.g d(delta)/dr is missing, d(phi)/dr should be (1 +
Brian> (C/beta) cos(theta)) * d(theta)/dr). Also the derivatives
Brian> don't have the same overall scale as the function itself
Brian> --- I think that is why the covariance matrix looks bad.
Thanks for looking at this; I finally had time to go back and have a
second look. I have the scale problem fixed as well as the
derivatives and am now getting sensible values for the covariance. I
still have problems with getting NaNs during the solution but I
suspect this is a round-off and termination condition issue. Or it
may be that near the solution, the function is not really unimodal as
I've managed to get several different solutions, all similar but well
outside one another's variance estimates.
roland
--
PGP Key ID: 66 BC 3B CD
Roland B. Roberts, PhD RL Enterprises
roland@rlenter.com 76-15 113th Street, Apt 3B
roland@astrofoto.org Forest Hills, NY 11375