Possible effects of random seed on computation time

Hi, all:

I am now studying a system of Si and to obtain a good statistic average, for the same system, I conducted many simulations start with different random seed, but everything else is the same, simple NVT relaxation. However it happened to me that, let’s say if I throw 100 simulations into the cluster, 90 of them will be finished very quickly and almost ends at the same time, however there are always several of them proceed very slowly and takes almost ten time longer to reach the same tilmestep.

My question is that has anyone else encountered this problem? Is this a problem of LAMMPS at all? I am thinking about it can be either the problem of the cluster or LAMMPS, can anyone help?



To see if it’s cluster-related, it can help to track which jobs get sent to which (combination of) processors/nodes and see if they consistently perform slower or not. If so, it can be due to slower processors (although a difference of a factor ten seems like a lot) or due to some problems related to that node.

Try re-running one of the slow simulations (same seed).
If it doesn’t run consistently slowly, then I would guess
it has nothing to do with LAMMPS.