This is the mail archive of the
gsl-discuss@sourceware.org
mailing list for the GSL project.
Re: [Bug-gsl] Feature request for ODE integrators
- From: Brian Gough <bjg at network-theory dot co dot uk>
- To: "Kostrun, Marijan" <Marijan dot Kostrun at sylvania dot com>
- Cc: gsl-discuss at sources dot redhat dot com
- Date: Sat, 29 Jul 2006 16:43:51 +0100
- Subject: Re: [Bug-gsl] Feature request for ODE integrators
- References: <1150124618.8123.15.camel@rdfa0117578.sylvania.com><17552.16090.489386.922737@hp2.network-theory.co.uk><1153488406.8749.40.camel@rdfa0117578.sylvania.com>
At Fri, 21 Jul 2006 09:26:46 -0400,
Kostrun, Marijan wrote:
> I inspected the codes and have implemented them for my project rlabplus
> at http://rlabplus.sourceforge.net, however, the solution is not quite
> satisfactory.
> Let me explain why. Imagine you have ask for integration on the interval
> [t1,t2] but the NaN appears somewhere inside the interval. When gsl
> integrator hits NaN it does not integrate any further, an error flag is
> rised (from 0 to 2) and t1 is reset to its original value. I would like
> t1 to contain the last successful time and the array pointer the
> solution to the ODE at that time. This feature is implemented in other
> integrators, e.g., code ODE.F by Shampine and Gordon from netlib.
> If this is too complicated to implement, I will have to write the code
> to use some sort of bisection on the interval [t1,t2] to find the time
> up to which the integration can be performed.
Hello,
I think the only information that gsl_odeiv_evolve_apply can return
that it currently doesn't is the step size h that triggered the NaN. I
will modify it to do that. All the other information should be
available as it returns to the user after any successful single step.
--
best regards,
Brian Gough