Multisphere in parallel

Submitted by Zhenli on Fri, 03/24/2017 - 10:46

Dear all,

I am just by testing the multisphere method in LIGGGHTS3.6.0. The case was very simple, the multisphere particle consists only two particles.
some thousands of multispheres were simulated. A moving stl wall was available in the simulation.
What I found was, the single simulation went no problem, but the parallel simulation has a "explosion" of particles after 3-5 timesteps first at the boundary of each sub-domain.

Is this problem known and are there some methods to overcome this problem?

Thanks a lot!

best regards

Zhen

enrique | Fri, 03/24/2017 - 16:31

Hi, I don't know the answer but i'm interested in working with multisphere particles. I'd appreciate so much if you could share your code for the input file so I can implement in my case.

Zhenli | Mon, 03/27/2017 - 10:20

Hi, Enrique,
No problem, but I am trying at first to make a simplied case and load it up, it will take some short time.

regards

Zhen

estefan31 | Fri, 03/24/2017 - 18:04

I was actually about to post about this very problem. In my case, I'm simulating a direct shear test with the upper loading platen as a servo wall. There are 11700 multisphere particles simulated on 24 processors in a 4x3x2 grid. The simulation gets through the compaction stage just fine and comes to equilibrium. Usually about halfway through the shearing stage, out of nowhere a bunch of particles just disappear. It seems more like an implosion than an explosion for me. It's not even a gradual loss of particles, the total number of particles just instantly reduces to a lower total amount.

estefan31 | Fri, 03/24/2017 - 18:26

I'll post my logfile in a bit after running a simulation that does not include any of the code implementations I made so I can make sure the bug isn't on my end. Enrique, you can use this to learn how the servo wall works. Zhenli, since I think we're having a very similar problem, would you be willing to speak via email or skype about this in case any of us find a solution? My email is estefan31@berkeley.edu.

enrique | Mon, 03/27/2017 - 20:17

I think I'll have that problem too.
just to test a example, I used the multisphere example insertion from the directory Tutorials_public and I added a moving mesh. The test was just made with few multisphere particles and everything seems to be working properly.
I need to insert hundreds of thousands of particles so I'd appreciate you posted the solution or a hint.