[aspect-devel] cryptic error in particle generation

[Rene Gassmoeller]

Hi Bart,

Did this happen in a parallel computation? Does it also happen in a 
serial model run? I briefly saw an issue last week with the parallel 
generation with the probability density function generation if the 
probability function is exactly zero over the whole volume of one the 
parallel domains. I did not had the time to fix it right away, and 
thought it would be safe to postpone it to the hackathon, guess I was 
wrong :-( ...

Anyway, could you check if it also happens if you use the original 
function, but instead of setting it to 0, set it to a very small number 
(e.g. 1e-6)? If this works it should be a decent workaround. Another fix 
would be to put the whole block in 
source/particle/generator/probability_density_function.cc lines 79-119 
into an

if (local_weight_integral > 0.0)
{

...

}


That should also fix the issue.


Let me know if that helps,

Best,

Rene


On 06/11/2018 03:35 PM, Bart Niday wrote:
> Hi all,
>
> I’m trying to use the particle probability density function to 
> generate a random uniform distribution over part of the model domain, 
> but I’ve been running into this error:
>
> Number of active cells: 16 (on 3 levels)
> Number of degrees of freedom: 268 (162+25+81)
>
> Assertion failed: (min_arg < max_arg), function 
> uniform_real_distribution, file 
> /Applications/deal.II-8.5-brew.app/Contents/Resources/brew/include/boost/random/uniform_real_distribution.hpp, 
> line 159.
>
> The attached parameter file replicates the problem, at least for me. 
> If I change the probability density function from if(y>0.5,1,0) to 
> if(x>0.5,1,0), it works normally. Changing y>0.5 to a different 
> number, sometimes it crashes and sometimes it works. The problem also 
> happens on my desktop, but not on my university cluster (running a 
> slightly different version of Aspect, but the particle code is unchanged).
>
> Any clues as to what this might be? Does anyone else see the same 
> problem running the attached file?
>
> Thanks,
>
>
> - Bart Niday
> Ph.D. Student
> Department of Earth Sciences, University of Oregon
>
>
> _______________________________________________
> Aspect-devel mailing list
> Aspect-devel at geodynamics.org
> http://lists.geodynamics.org/cgi-bin/mailman/listinfo/aspect-devel

-- 
Rene Gassmoeller
https://gassmoeller.github.io/

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.geodynamics.org/pipermail/aspect-devel/attachments/20180611/83807c61/attachment-0001.html>