|
From: | Axel Arnold |
Subject: | Re: [ESPResSo-users] Periodic box, bonds and mpi |
Date: | Fri, 10 Oct 2014 10:16:24 +0200 |
User-agent: | Mozilla/5.0 (Macintosh; Intel Mac OS X 10.9; rv:24.0) Gecko/20100101 Thunderbird/24.6.0 |
On 10.10.14 01:50, Aleksei Kabedev
wrote:
What that in practice means is that those three particles moved apart more than a processor domain. There has to be in addition an error about broken pair bonds, otherwise, you simply forgot to bind those particles by pair bonds.
Capping only helps if the problem are strongly overlapping particles. And with LB, you cannot simply reduce the time step, because if you don't reduce the mesh at the same time, you crank up the sound speed and break LB.
Don't worry, that is taken care of. Basically, a border shell of particles is exchanged between processors, where the shell is wide enough for any two particles bonded by pair bonds. So, as long as you have pair bonds for the both arms of your angular bond, the particles need to be on the same CPU, if the pair bonds are intact.
Hmm, agrid 1.0 with tau 0.001 is a bit bold, apart from that the magnitude of the external force is important.
Now that doesn't work. Effectively, you have particles of diameter scale, and linearly extrapolate the LJ potential below 0.6. However, there it is already pretty steep, so that capping barely helps. You should start at 1.1*0.6 and slowly and gradually reduce to about 0.9*0.6. Best, Axel -- JP Dr. Axel Arnold ICP, Universität Stuttgart Allmandring 3 70569 Stuttgart, Germany Email: address@hidden Tel: +49 711 685 67609 |
[Prev in Thread] | Current Thread | [Next in Thread] |