[lammps-users] Optimal Coulomb Cut Off

LAMMPS LAMMPS Mailing List Mirror
1

LAMMPS Community,

I had the fortune of receiving a helpful hint from Paul that the long/short range cut off for the Coulomb potential can be adjusted to optimize computational time. This has worked greatly for my ~800 particle system and I have found that the optimal cut off is about the same whether I run on 2 or 8 processors.

I am testing out a larger system (N~7000) on my 8 core machine, but later intend to run it on a cluster. Can I expect the optimal cutoff for 8 cores to be the same for however many cores I intend to run on the cluster (so anywhere between 12-36 cores)?

I can certainly determine this by adjusting the cut off, running on the cluster a bit, and looking. However, I am wondering if the optimal Coulomb cutoff should be dependent on the amount of processors the MD code is run with.

Thanks,

Brian Giera
Doctor of Philosophy Candidate
Squires and Shell Group
Department of Chemical Engineering
University of California Santa Barbara

2

LAMMPS Community,

I had the fortune of receiving a helpful hint from Paul that the long/short
range cut off for the Coulomb potential can be adjusted to optimize
computational time. This has worked greatly for my ~800 particle system and
I have found that the optimal cut off is about the same whether I run on 2
or 8 processors.

I am testing out a larger system (N~7000) on my 8 core machine, but later
intend to run it on a cluster. Can I expect the optimal cutoff for 8 cores
to be the same for however many cores I intend to run on the cluster (so
anywhere between 12-36 cores)?

no. you cannot expect this, because the scaling with the number of processors
of the short range part of coulomb and the long-range part of coulomb (=k-space)
is different.

axel.