Re: A timestep of size....

From: Florian Bürzle <florian.buerzle_at_uni-konstanz.de>
Date: Wed, 14 Apr 2010 11:04:19 +0200

Hi!


Am 13.04.2010 um 23:22 schrieb Guillermo Arreaga Garcia:

>
> hi all,
>
> the $Gadget2$ code stops evolving the system because
> the following run-time error occurs: "A timestep of size zero was assigned
> on the integer timeline".
>
> At a first try to continue the run, I put a non zero value for the parameter
> MinSizeTimestep of the latex parameter file, let us say, something like
> 1.0e-8. I also activated
> the option -DNOSTOP_WHEN_BELOW_MINTIMESTEP in the makefile. Despite of this,
> the run ends as I said.
>
> does anybody know what is happening in the code ?


Did it crash while your simulation was already running (as I suppose since you wrote "stops evolving")? Then, most likely, your accelerations grew too large. You can check this by inspecting the debugging output which follows directly after the error message - if there are very large numbers in the accelerations, or even "inf", then this was the problem.

If it happened already at the start, I'd suggest to have a look at the get_timestep function in timestep.c. Here you can check the definitions of dt_accel (and for gas particles dt_courant) - some of the things there are set in the parameter file, so you can do a sanity check if none of the values were erroneously set to zero (as was already suggested for the softenings in another answer to your question).


>
> does anybody have an idea how I can continue the run ?
>

Without knowing what you tried to do, it is almost impossible give some useful suggestion. You could check, if the softenings are appropriate for the simulations you want to do (if you do collapse simulations with gas particles only and ADAPTIVE_GRAVSOFT_FORGAS turned on, this is usually not the problem).

If the accelerations really grew too large, the reason for your simulation to crash may even be intrinsic to the physics of your model. However, if you use additional physics that modifies accelerations, then there are of course further possible sources for problems.

Hope that helps!

Florian


-----------------------------------------------------------------------------------------------------------
Florian Bürzle
University of Konstanz
Department of Physics, Group Prof. Nielaba
Universitätsstr. 10
78464 Konstanz, Germany
tel: +49 7531 88 3790
web: http://cms.uni-konstanz.de/physik/nielaba/mitarbeiter/florian-buerzle/
-----------------------------------------------------------------------------------------------------------
Received on 2010-04-14 11:04:21

This archive was generated by hypermail 2.3.0 : 2022-09-01 14:03:42 CEST