This is the mail archive of the
gsl-discuss@sources.redhat.com
mailing list for the GSL project.
Re: Questions on bsimp
- From: Brian Gough <bjg at network-theory dot co dot uk>
- To: Di Xiao <dxiao at physics dot utexas dot edu>
- Cc: Slaven Peles <peles at cns dot physics dot gatech dot edu>, <gsl-discuss at sources dot redhat dot com>
- Date: Mon, 5 Aug 2002 23:21:58 +0100 (BST)
- Subject: Re: Questions on bsimp
- References: <E17aJO1-0002pB-00@kaos.physics.gatech.edu><Pine.LNX.4.33.0208021040270.18244-100000@linux11.ph.utexas.edu>
Di Xiao writes:
> However, I still have a question. We use adaptive stepzie control
> because we want to deduce the computational effort. The algorithm
> should be smart to know at which region there is the smooth
> uninteresting curve, while in some region many small steps are
> needed. My question is why bsimp chose some unreasonably big steps
> in my code. It can jump from one period to another period without
> showing the details of each one. If you run the code, you will
> find from the results givin by bsimp, it is hard to tell what
> function it likes. I tried different gsl_odeiv_control, but none of
> them gave me good results. I guess the reason is when there are
> oscillations, bsimp is not a good one. Can anybody explain it?
Bsimp can take very large steps and still be highly accurate. See the
reference to the paper in the manual for details.