Problem moving particle [Updated 10/10/13]

Submitted by pfalkingham on Fri, 09/13/2013 - 20:01
AttachmentSize
Plain text icon Initial generation3.91 KB
Plain text icon Main file.3.57 KB

I'm trying to set up a volume of particles, then move the particle with ID 1 to a position above the volume, change it's diameter and density, and let it drop. Problem is it disappears.

I make the volume separately (create.txt), and then in the main file that (drop.txt) that file is read in and I use the following:


...
run 10

set atom 1 x 0 y 0 z 0 vx 0 vy 0 vz 0
set atom 1 diameter 0.03
set atom 1 density 4865

run 60000

The first dump file (produced before run 10) contains all the particles, the next output after the set atom commands simply doesn't contain particle 1 anymore.

Liggghts 2.3.7, both make fedora and make openmpi produce this problem.

[edit]

Here are input files with larger particles that should run in a reasonable time.

It looks like it could be a communication issue - in this case, it works fine on my local laptop, and it works fine on the HPC when only requesting 1 node (12 cores), but the moved particle disappears when I request 2 nodes or more (24+cores).

Tested using 2.3.8, compiled as xt5, and fedora.

ckloss's picture

ckloss | Thu, 09/19/2013 - 20:59

Hi Pete,

thansk - I'll have a look!

Cheers
Christoph

pfalkingham | Fri, 09/20/2013 - 21:31

My testing hasn't exactly been methodical recently, but a few more details:

This is only happening with lots of small particles (I'm seeing it in my 500,000 particles with radius 0.0006m), which makes systematic testing a pain, as you can imagine.

However, now rather than the particle disapearing, I'm getting an application crash, and the only error messages come from the machine (no errors in log.liggghts), and seem to indicate it's a memory issue.

ckloss's picture

ckloss | Tue, 10/08/2013 - 14:49

Hi Pete,

your case seems to be too large to be debugged on a "I'll have a quick look" basis (which is the only support we provide for the public version).
Can you post a smaller case that reporduces the issue?

thanks!
Christoph

pfalkingham | Wed, 10/09/2013 - 15:40

Hmmm... so far, no. This was only happening in large simulations... I'll do some exploring and see if i can isolate the issue in a smaller test case.

pfalkingham | Thu, 10/10/2013 - 19:25

Initial post updated - new files attached.

Here are input files with larger particles that should run in a reasonable time.

It looks like it could be a communication issue - in this case, it works fine on my local laptop, and it works fine on the HPC when only requesting 1 node (12 cores), but the moved particle disappears when I request 2 nodes or more (24+cores).

Tested using 2.3.8, compiled as xt5, and fedora.