Hi,
It's seems to me that bigParticles model is not managing well the calculus of voidFraction field with periodic BC or parallel computing. Indeed, voidFraction is not the same at each side of the boundary or between 2 processors.
In the image, it is voidfractionfield after first step of calculation where the rectangular mesh is divided in 4 processor (as a cross) and periodic at each side. One can clearly see the processor separation (orange cross) even if the particles are randomly distributed on the rectangle...
How to fix that ?
j
| Attachment | Size |
|---|---|
 voidfrac.jpg | 35.77 KB |
Maryam | Mon, 04/29/2013 - 02:09
My observation
I just wanted to add my observation. I generated two packings by running liggghts on 1 and 4 processors. Runs with 4 processors miss atoms at the processor boundaries. Decreasing the neighbour bin size did not help.
on 1 processor:
on 4 processors:
cgoniva | Tue, 04/30/2013 - 17:40
Hi Maryam,
Hi Maryam,
this is related to the parallel packing algorithm in LIGGGHTS - you might let the particles settle a little before running the simulation, or fill with an insert stream.
KR
Chris
Maryam | Wed, 05/01/2013 - 03:16
Thanks Chris,
Thanks Chris,
Seems like this has nothing to do with what j has described above.
cgoniva | Wed, 05/01/2013 - 09:16
bigParticles model and periodic or parallel computing
Hi J,
the bigParticle model is recursively looking for cells covered by a particle, starting at it's centre. It cannot cross processor BCs with the current algorithm. You might have a look at the IB voidfraction model which uses satellite points to overcome this problem. Still an implementation for periodic BCs would be tough.
Cheers,
Chris