questions on "fix langervin"

Dear LAMMPS users:

I got question on the “fix langervin ”. My intension is to apply langervin thermal scale while keeping the total force unchanged. As manual instructed, I use “zero yes” as below:

fix 1 sub_g langevin 300 300 100 123456 zero yes

The group applied with langervin has only one atom type, thus the total force is supposed to be unchanged after applying “fix langervin”. However, after running, I found the total force on the group significantly changed. I tried to adjust the damping parameter but unable to help. Any suggestions? Thanks!

Thanks & Best Regards!

Dai Ling

Scientist

Institue of High Performance Computing, A*STAR

Singapore

Tel: +65-64191428

However,
after running, I found the total force on the group significantly changed.

How did you determine this, on a timestep by timestep basis?

Steve

Also: Please post the line from your input script containing fix ... nve?

(You said you applied langevin dynamics on only one atom, and then
zero the total force. This is a very strange test. Is the atom
moving?)

Cheers
Andrew

Hi:

Thanks for your kind comments.

Actually the ones applied with langervin has one type of atom (containing thousands of atoms actually), not one atom only.

I run two simulations with nve, one with langervin & one without. I output the total force information (all the force, including those from langervin setting) every 100 steps. As instructed in the manual, I zero out the total force that applied from the langervin setting. Thus, the total force values from the two runs shall be equal. However, it is not the case, there is always 5-20% differences between them. This is question that I raise.

Thanks for attention!

Input scripts:

# thermal equil
fix t_scale_1 therm_sub_g langevin 300 300 301 123456 zero yes

#Run sliding
fix 1 all nve
run 10000

Again I ask, how are you computing the total force?

Steve

Hi:

Thanks for your kind comments.

Actually the ones applied with langervin has one type of atom (containing thousands of atoms actually), not one atom only.

I run two simulations with nve, one with langervin & one without. I output the total force information (all the force, including those from langervin setting) every 100 steps. As instructed in the manual, I zero out the total force that applied from the langervin setting. Thus, the total force values from the two runs shall be equal. However, it is not the case, there is always 5-20% differences between them. This is question that I raise.

those two different simulations will *always* diverge
(usually in an exponential fashion). that is a consequence
of MD being a chaotic system and using floating point math
which is not associative and thus will always result in small
differences in the per atom total force, which will nucleate
the exponential divergence quickly. thus you can only expect
the equivalence of the two different simulations for a small
number of steps.

axel.