[lammps-users] Scaling problems in LAMMPS

LAMMPS LAMMPS Mailing List Mirror
1

Hello all,

I am confronted with a rather strange problem. With all the parameters same, when I used
kspace = 0.0001 -------- on 64 procesors ---- Ran fine
kspace = 0.0001 -------- on 512 processors ---- ERROR: Cannot compute PPPM order. Reduce PPPM stencil…
kspace = 0.00006 -------on 512 processors — Ran fine. { 0.00006 or 0.00008 depending on the size of system}

So, I decided to do,
kspace = 0.0001 -------- on 64 procesors ---- Ran fine for a few picoseconds and then
kspace = 0.0001 -------- on 512 processors — Ran fine…

Any suggestions to the problem ?

Thanks,

Pavan

2

I'm not clear on what problem you're describing. If you are
saying that a 2nd run with a new kspace accuracy succeeds
when it is the 2nd run, but not when it is the 1st run, it
could be b/c the system volume changed. The kspace
command should print out the FFT lattice size it is using
and if the # of neighbors needed on that lattice is bigger
than the processor sub-domain size, then LAMMPS will
throw the error you are seeing.

Steve