This is the mail archive of the
gsl-discuss@sources.redhat.com
mailing list for the GSL project.
multimin convergence problem
- From: Roland Roberts <roland at astrofoto dot org>
- To: gsl-discuss at sources dot redhat dot com
- Date: 31 Jan 2002 14:59:32 -0500
- Subject: multimin convergence problem
I have a model function which I am trying to an ideal curve. The
problem stems from building a "barndoor tracker", see
http://hometown.aol.com/davetrott/page17.htm for details on what this
is.
The ideal function is just
phi(T) = K T
where K = 86400 * 180 / M_PI to measure the angle phi in arcseconds.
The model function is
L = R T
theta = 2 arcsin (2 L / r)
phi = theta + arcsin(sin(theta) / beta)
where R is a fixed parameter (set at R=1/1200) and (r,beta) are the
variables on which the minimization operates. T is in the range
0--7200 seconds.
What I'm finding is that the ability to get a solution at all is very
sensitive to how I initialize the line search parameters. The setting
of TOL can vary a lot, but the value of STEP has to fall in a fairly
narrow range or GSL starts spitting out NaNs. What is frustrating is
that neither the function nor the derivative is returning NaN at any
point; I've verified this by printout out all the function
evaluations.
Is there some magic to picking a good value for STEP? Shouldn't GSL
return an error rather than setting all my parameters to NaN? (BTW,
this doesn't seem to matter which solver i pick).
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