This is the mail archive of the
gsl-discuss@sources.redhat.com
mailing list for the GSL project.
Re: Bug in Levenbeg-Marquardt Algorithm
- From: Brian Gough <bjg at network-theory dot co dot uk>
- To: hans dot plesser at itf dot nlh dot no
- Cc: gsl-discuss at sources dot redhat dot com
- Date: Mon, 25 Feb 2002 19:02:46 +0000 (GMT)
- Subject: Re: Bug in Levenbeg-Marquardt Algorithm
- References: <15482.22639.87268.403076@itfhep.nlh.no>
Hans Ekkehard Plesser writes:
> I now had a look at the original MINPACK code (lmder.f). They set
> par_lower to 0 for rank deficient r, so here is a patch that does the
> same in lmder.c .
Thanks! That looks sensible. Do you have a test case with a
rank-deficient matrix that could be added to the test suite?
Brian
>
>
>
> 270c270,277
> < par_lower = fp / (delta * phider);
> ---
> >
> > /* w == zero if r rank-deficient,
> > then set lower bound to zero form MINPACK, lmder.f
> > Hans E. Plesser 2002-02-25 (hans.plesser@itf.nlh.no) */
> > if ( wnorm > 0 )
> > par_lower = fp / (delta * phider);
> > else
> > par_lower = 0.0;