On 9/30/17 3:29 AM, Tuomo Keskitalo wrote:
Hi,
Ok, interesting. So likely it's not GSL optimization issue. What happens
with a different ODE-solver on OSX?
No trip.
Also, maybe check that the solver is not making a huge number of small
steps (stiff problem?).
Not sure how I do that.
Not being able to reproduce(?) on Linux suggests something OSX specific,
some linked math library maybe?
I'm not being clear. I've got a case (call it case #1) that will
reproduce on OSX. This particular case #1 will not abort on Linux.
However, there are other cases which have indeed aborted on Linux, I've
just not yet captured the exact random seed which will enabled me to
reproduce those cases at will on Linux, though I imagine it wouldn't
take more than a couple of days of running to locate one.
The reason that I think it's not reproducing on Linux is simply
numerical roundoff error. With case #1, if I change dt=t1-t in
gsl_odeiv2_driver_apply() by 0.002%, the abort will not trip. For case
#1, the trip happens after the 1082th call to gsl_odeiv2_driver_apply(),
so the initial conditions of the 1083rd step are _just_ different enough
to cause the trip.
M
