[lammps-users] Dependence the thermal conductivity results of number of processors

LAMMPS LAMMPS Mailing List Mirror
1

Hi all. I found out that the result of thermal conductivity computing depends on the number of processors. I’ve used the example script for solid Ar from page http://lammps.sandia.gov/doc/compute_heat_flux.html

Results of computing:

num_process therm_conductivity [W/mK]
1 0.2894747897
2 0.3226395484
3 0.3266229955
4 0.3114426029

LAMMPS 13 Sep 2010
OS Ubuntu 10.04

How can I avoid this problem?

2

Hi all. I found out that the result of thermal
conductivity computing depends on the number of processors. I've used the
example script for solid Ar from
page http://lammps.sandia.gov/doc/compute_heat_flux.html
Results of computing:
num_process therm_conductivity [W/mK]
1 0.2894747897
2 0.3226395484
3 0.3266229955
4 0.3114426029
LAMMPS 13 Sep 2010
OS Ubuntu 10.04
How can I avoid this problem?

are you sure there actually _is_ a problem?

have you checked how well each of the
values are converged?

different number of processors will
result in different but equivalent
trajectories due to limitations of floating
point math.

3

The velocity command in that script will generate
velocites that differ (for a particular atom) depending
on the number of procs. See the loop option on the
velocity command, if you want to avoid that. Thus
running on different P will give you different trajectories
and results, but statistically similar. As Axel said, for
a small system, with a short runs, the variations you
are seeing seems normal.

Steve