Using fix npt with with long range Coulomb Interaction(kspace)

LAMMPS LAMMPS Mailing List Mirror
1

Hi,

I have a doubt regarding the use of fix npt with long range Coulomb Interaction.
I am simulating a polymer-water system where I am using fix npt to change the pressure few times during the simulation.

Below is a portion of my input script which does this job:

NPT at 0.5 atm

log cycle_1_npt_0.5
fix 1 all npt temp 400.0 400.0 5.0 iso 0.5 0.5 1000.0
run 500000

compression from 0.5 to 1.0 atm

log cycle_1_compression_0.5_to_1
unfix 1
fix 2 all npt temp 400.0 400.0 5.0 iso 0.5 1.0 1000.0
run 500000

I have run this earlier with an older lammps version (18 Jan 2011) and things were pretty fine.

Now while I am running the same thing on another machine with a newer version of lammps (17 Oct 2012), I get problem with pressure, Ecoul and Elong at the end of one run and the beginning of the other, rest all thermodynamic output remaining the same as expected. I am using “kspace_style ewald 1.0e-4” for long range Coulombic interaction.

Below is the themo output for the end and beginning of fixes:

End of fix 1:
Step Temp Press Volume PotEng KinEng TotEng E_vdwl E_coul E_long E_pair E_bond E_angle E_dihed E_impro E_mol CPU

500000 395.82769 536.24633 32096.629 -5131.664 2981.5748 -2150.0892 -755.93059 -687.33008 -8134.3703 -9577.631 1077.101 1903.8957 1220.2378 244.73247 4445.967 2945.0038

Begining of fix 2:
Step Temp Press Volume PotEng KinEng TotEng E_vdwl E_coul E_long E_pair E_bond E_angle E_dihed E_impro E_mol CPU
500000 395.82769 521.75017 32096.629 -5134.0405 2981.5748 -2152.4657 -755.93059 2270.395 -11094.472 -9580.0076 1077.101 1903.8957 1220.2378 244.73247 4445.967 0

The Ecoul and Elong are changed by almost same amount, so that the net potential energy is nearby the last value. Sometimes, in a later stage of simulation, the change is in decimal places, as below:

End of one fix:

1000000 401.23223 827.41872 31193.493 -5240.6336 3022.2845 -2218.3491 -791.75238 2257.482 -11098.919 -9633.1893 1066.0763 1847.8487 1225.0291 253.60156 4392.5556 816.22427

Begining of the other fix:
Step Temp Press Volume PotEng KinEng TotEng E_vdwl E_coul E_long E_pair E_bond E_angle E_dihed E_impro E_mol CPU
1000000 401.23223 827.6165 31193.493 -5240.6314 3022.2845 -2218.3468 -791.75238 2272.4813 -11113.916 -9633.187 1066.0763 1847.8487 1225.0291 253.60156 4392.5556 0

Also, I can see change in Ewald parameters ( G vector, RMS force accuracy, and relative force accuracy).

I came across the below mail thread(3 years old) which highlighted a similar problem. It was suggested told to be “the way LAMMPS adjusts the grid to
achieve the specfied accuracy” but later was found to be “a bug with fix box/relax and long-range Coulombics”.

http://lammps.sandia.gov/threads/msg12047.html

I am not using fix box/relax and also I hope that bug must have been fixed in the later lammps versions.
Could you please suggest if this is an allowed kspace grid adjustment by lammps or still a bug. As I have already ran this in an earlier version without any problem, I doubt if the newer version has some issues.

Any suggestions will be really appreciated.
Thank you so much in advance for your time and support.

Thanks,
Madhu.

2

Hi,

I have a doubt regarding the use of fix npt with long range Coulomb
Interaction.
I am simulating a polymer-water system where I am using fix npt to change
the pressure few times during the simulation.

Below is a portion of my input script which does this job:

# NPT at 0.5 atm
log cycle_1_npt_0.5
fix 1 all npt temp 400.0 400.0 5.0 iso 0.5 0.5 1000.0
run 500000

# compression from 0.5 to 1.0 atm
log cycle_1_compression_0.5_to_1
unfix 1
fix 2 all npt temp 400.0 400.0 5.0 iso 0.5 1.0 1000.0
run 500000

I have run this earlier with an older lammps version (18 Jan 2011) and
things were pretty fine.

Now while I am running the same thing on another machine with a newer
version of lammps (17 Oct 2012), I get problem with pressure, Ecoul and
Elong at the end of one run and the beginning of the other, rest all
thermodynamic output remaining the same as expected. I am using
"kspace_style ewald 1.0e-4" for long range Coulombic interaction.

have a look at the comments for the 2 Mar 2012 patch here:
http://lammps.sandia.gov/bug2012.html

this change and subsequent tweaks and improvements may be at the root
of the issues you are seeing.

axel.

p.s.: also you should upgrade to the current version with lots of
additional improvements and bugfixes.

3

Hi Axel,

Thanks for the suggestion.
I will install the latest version and try to run the same.
I will update if the problem still persists.

In the mean time, could you please suggest if the trajectories which I now get (with a little change in energy) is still acceptable?

Thanks,
Madhu.

4

Hi Axel,

Thanks for the suggestion.
I will install the latest version and try to run the same.
I will update if the problem still persists.

In the mean time, could you please suggest if the trajectories which I now
get (with a little change in energy) is still acceptable?

what are you going write in your paper? "some dude on the mailing list
said this simulation was ok, even without checking my data, so it must
be good."? what do you think a referee would say about this?

whether the simulation is useful or not is up to you to find out.
essentially you ran with a lower than desired accuracy on the kspace
forces. you noticed that something is different. what does this tell
you?

axel.