IR intensity difference with different number of cores

Dear Lammps users,
I am trying to simulate the IR spectrum of cristobalite based on NVE molecular dynamics trajectory using lammps, but i got a problem with the number of cores to use. I run the same input script and data file with the same timestep, same dump frequency, but with one process, two processors and three processors respectively. And use the same post-processing package for the two cases, the resulting IR spectrum have the same wavenumber, however differs in the IR intensity. I have tried to create the initial velocity for the atoms the same after reading the lammps documentation, but still can’t fix the problem. I really have no idea how that intensity difference comes from, is that because the 150 ps NVE simulation is not long enough? or someting else wrong with my input scritp. I have post my input script as well as the resulting IR spectum. Any piece of help will be greatly appreciated!!!

Best regards

Jiasen Guo

lammps input script for Sodalite with MZHB

units metal
dimension 3
atom_style full

read_data MZHB_CRI_original.lmp

#velocity all create 0 87287 loop geom

potential

bond_style harmonic
bond_coeff 1 11.650000 1.6200000

angle_style harmonic
angle_coeff 1 3.4000000 109.47000
angle_coeff 2 1.1100000 149.80000

pair_style lj/cut/coul/long 12.000
pair_coeff 1 1 0.0085997000 1.959982864 12.0000
pair_coeff 1 2 0.0052661900 1.758034219 12.0000
pair_coeff 2 2 0.0032249000 1.576892023 12.0000

special_bonds lj 1.0 1.0 1.0 coul 1.0 1.0 1.0

kspace_style ewald 1.0e-4

ir_NVE_3cores.jpg

Dear Lammps users,
I am trying to simulate the IR spectrum of cristobalite based on NVE
molecular dynamics trajectory using lammps, but i got a problem with the
number of cores to use. I run the same input script and data file with the
same timestep, same dump frequency, but with one process, two processors and
three processors respectively. And use the same post-processing package for
the two cases, the resulting IR spectrum have the same wavenumber, however
differs in the IR intensity. I have tried to create the initial velocity for
the atoms the same after reading the lammps documentation, but still can't
fix the problem. I really have no idea how that intensity difference comes
from, is that because the 150 ps NVE simulation is not long enough? or
someting else wrong with my input scritp. I have post my input script as
well as the resulting IR spectum. Any piece of help will be greatly
appreciated!!!

apart from the fact that the "intensities" you are computing are not
that relevant since they are actually spectral densities and
disregarding any transition moments, the differences you are seeing
are very small.

it has been discussed many times on this mailing list that due to the
use of floating point math and its non-associative nature (i.e. the
exact result of a sum depends on the order of summing the elements),
trajectories will always diverge (exponentially) for computations with
different numbers of processors (cf. chaos theory, butterfly effect).
so what you are seeing is simply a manifestation of the statistical
uncertainty of the "intensities" (or rather spectral densities) you
are computing. so nothing to worry about. you shouldn't put too much
meaning into the intensities in the first place.

axel.