Empty regions with insert/pack using more than one processor core

Submitted by gukargl on Mon, 05/27/2013 - 16:12

I try to fill a region with insert/pack using four diferent particle radii from the same template. However, if I try to use more processor cores e.g. 24 instead of 10 then there are empty regions where no particles are inserted. This seems to be associated with using more than one particle, since if only one particle size is used then all regions are filled normally without regard of the number of processors.
I attach two screenshots showing the difference. The first using 24 cores and the second using only 10 cores but the same script.

Additionally I noticed that out of four particle sizes only the first three are implemented and the fourth is completely absent in the distribution.

What I want to do with this model is to create a material sample with given properties which I can use as standard soil sample for further simulations.

Cheers
Günter

ckloss's picture

ckloss | Wed, 05/29/2013 - 16:27

Hi Günter,

which version of LIGGGHTS are you using? There has been a fix in LIGGGHTS for this recently with version 2.3.3 for the case where an insertion region is highly divided across multiple processors. I ran your case with the most recent version (2.3.4), and it looks ok there.

However, I get some warning that not all insertions could be performed. This is due to the fact that fix insert/pack can only reach a packing fraction of ~0.3. What you could do to overcome this is to insert with fix insert/pack overlap_check no. then run a few steps with fix nve/limit so that the particles relax and then change to fix nve/sphere

Cheers
Christoph

gukargl | Wed, 06/05/2013 - 16:33

Hi Christoph,
Apparently this was run still on the 2.2.4 version. I will upgrade to a newer version and try again. Thanks for the other recommendations.
Cheers
Günter

evansmuts | Tue, 07/16/2013 - 11:37

Hi Christoph

You say that fix insert/pack can only reach a packing fraction of ~0.3.

Is this a "hard limit" or is it just the practical limit of what the algorithm can do? Why I ask, is that I am using insert/pack, but have been struggling to get packing fraction greater than ~0.35.

Can you please comment on this in the documentation? I think it would be important to know the practical limits of a command before using it.

Thanks
Evan

PaulWinkler's picture

PaulWinkler | Tue, 07/16/2013 - 19:15

Hi,

which algorithm is used for insert packing? Something like Random Sequential Addition (RSA)?
The way to allow overlapping, use nve/limit and just wait a while and than start simulation is something like Jodrey-Tory/Force-Biased algorithm?
How homogenous are the generated packings? Can anyone recommend an easy way to analyse Liggghts results with voronoi cells?

Regards
Paul

ckloss's picture

ckloss | Thu, 07/18/2013 - 18:17

Hi Paul,

should be possible to use this either out-of-the-box or with slight modifications

Cheers
Christoph

ckloss's picture

ckloss | Thu, 07/18/2013 - 18:20

>>which algorithm is used for insert packing? Something like Random Sequential Addition (RSA)?
I don't know what name the algorithm has: It's just trying to insert the spheres one by one and checks for overlap with the existing ones

Cheers
Christoph

ckloss's picture

ckloss | Thu, 07/18/2013 - 18:17

>>Is this a "hard limit" or is it just the practical limit of what the algorithm can do?
>>Why I ask, is that I am using insert/pack, but have been struggling to get packing fraction greater than ~0.35.
It's a practical limit. To overcome it, you can either let the particles grow after insertion (fix adapt) or turn off overlapcheck and relax with nve/limit

Cheers
Christoph

ckloss's picture

ckloss | Thu, 07/18/2013 - 18:17

>>Is this a "hard limit" or is it just the practical limit of what the algorithm can do?
>>Why I ask, is that I am using insert/pack, but have been struggling to get packing fraction greater than ~0.35.
It's a practical limit. To overcome it, you can either let the particles grow after insertion (fix adapt) or turn off overlapcheck and relax with nve/limit

Cheers
Christoph