LAMMPS WWW Site - LAMMPS Documentation - LAMMPS Mailing List Archives
Re: [lammps-users] Different outcomes with same *.data & *.in while changing the parallel core numbers
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [lammps-users] Different outcomes with same *.data & *.in while changing the parallel core numbers

From: 高瞻 <sandy950111@...24...>
Date: Sun, 11 Jun 2017 15:11:01 +0800

Thanks for your suggestions, my LAMMPS version is lammps-17Nov16, and my input file is as follows:

# variables input

variable         DT equal 2.0              
variable         A equal 220               
variable         r equal 10                
variable         L equal 75                
variable         THERMO equal 5000              
variable         Pstart equal -0.5*10*$A
variable         Pstop equal -0.5*10*$A

# init

units            real
dimension        3
boundary         p p p
atom_style       full
timestep         ${DT}
neighbor         2.0 bin
neigh_modify     delay 0 every 1 check yes

# create geometry

pair_style       lj/long/tip4p/long long long 2 1 1 1 0.1546 9 8.5
kspace_style     pppm/disp/tip4p 1e-5
kspace_modify    force/disp/real 0.0001 force/disp/kspace 0.002 mix/disp geom 
bond_style       harmonic
angle_style      harmonic
read_data        data.droplet_npt

# variables output

variable         PRESS equal press
variable         VOL equal vol

# fixes 

fix              0 all indent 10.0 sphere $L $L $L $r units box
fix              1 all shake 0.0001 100 0 b 1 a 1
fix              barostat all npt temp 298.0 298.0 $(100.0*dt) iso ${Pstart} ${Pstop} $(1000.0*dt) drag 1.0
fix              print_press all print 1 "${PRESS}" file press.dat  screen no title ""
fix              print_vol all print 1 "${VOL}" file vol.dat screen no title ""

# run 

thermo_style     custom step temp press vol
thermo           ${THERMO}     
run              250000

The simulation is designed to mimic the cavitation in water at a low pressure, and the "outcome changes a lot" means the vol-time curves in the simulations are different. The vol variable in the 100 core simulation is increasing much quicker than in the 20 core simulation, which makes me confused.
So, what I want to ask for advice is that is there any other factors except the "floating point error" that cause this situation? Because I think the the "outcome change" in the two simulation is really beyond my expectation.
Thank you very much!

2017-06-11 13:27 GMT+08:00 Axel Kohlmeyer <akohlmey@...24...>:

On Sun, Jun 11, 2017 at 12:13 AM, 高瞻 <sandy950111@...24...> wrote:
Hello everyone,
       I use the TIP4P-2005 water model to simulate the two-phase process of water, but when I change the parallel core number from 100 to 20, the outcome of the simulation changes a lot.

​what do you mean by "outcome changes a lot"? is it just different positions of individual atoms, or significant differences in (averaged) properties?

also, *always* mention the LAMMPS version you are using and best provide your input file, too, so that people can check, whether there are obvious mistakes somewhere.
       I can't understand why, should I change the simulation parameters such as the timestep while I set it as 2fs now or any other parameters to solve this problem?

​diverging trajectories with different number of cores, even different implementations of the same potential, are quite normal, as LAMMPS employs floating point math and floating point math does not associate (e.g. the exact outcome can be slightly different for all properties that​ are summed up in different order) which will result in an exponentially growing difference between such runs.

so at this moment, it is not even clear, whether you are reporting a real issue, or just physics and numerical science at work.



Thanks a lot!

Check out the vibrant tech community on one of the world's most
engaging tech sites,!
lammps-users mailing list

Dr. Axel Kohlmeyer  akohlmey@...24...
College of Science & Technology, Temple University, Philadelphia PA, USA
International Centre for Theoretical Physics, Trieste. Italy.