Dear all,
I have two questions in relation to calculating thermal conductivity of a polymer. I perform the following steps.
- Minimize the structure
- Run NPT
- Run NVT
- Run NVE
- Apply RNEMD scheme
a) I find that when I perform steps 2,3,4 for 5000 timesteps each, the system runs fine, but if I increase that to 10000 steps each, then errors such as ‘Dihedral atoms 890 891 892 899 missing on proc 4 at step 11957’ show up. I tried changing the parameters of the applied fixes but the same issue comes up during simulation of one of the three steps 2,3,4. I am unsure if my starting structure is incorrect or whether I am making a mistake while implementing either of the fixes. Any suggestions would really help.
b) When RNEMD is applied after a successful NVE run (5000 timesteps), for the first few timesteps temperature oscillates about the mean value imposed through the NPT and NVT fixes. But during a long run (~ 10000 timesteps), it gradually keeps on increasing. Nothing specific could be observed from the visualization. Any suggestions on how to account for the temp-rise or overcome this issue will be very helpful.
Thank you
Ganesh
Dear all,
I have two questions in relation to calculating thermal conductivity of a
polymer. I perform the following steps.
1. Minimize the structure
you may want to insert a period with fix nve plus fix langevin in here
2. Run NPT
3. Run NVT
4. Run NVE
5. Apply RNEMD scheme
a) I find that when I perform steps 2,3,4 for 5000 timesteps each, the
system runs fine, but if I increase that to 10000 steps each, then errors
such as 'Dihedral atoms 890 891 892 899 missing on proc 4 at step 11957'
show up. I tried changing the parameters of the applied fixes but the same
issue comes up during simulation of one of the three steps 2,3,4. I am
unsure if my starting structure is incorrect or whether I am making a
mistake while implementing either of the fixes. Any suggestions would really
help.
the lammps manual has a list of potential reasons for missing atoms.
50k or 100k steps each are not a lot of equilibration. you should not
determine the length of equilibration by picking a number but by
monitoring the total energy. during the first part of equilibration, it might
be advisable to reduce the length of the time step.
b) When RNEMD is applied after a successful NVE run (5000 timesteps), for
the first few timesteps temperature oscillates about the mean value imposed
through the NPT and NVT fixes. But during a long run (~ 10000 timesteps), it
gradually keeps on increasing. Nothing specific could be observed from the
visualization. Any suggestions on how to account for the temp-rise or
overcome this issue will be very helpful.
a) 10k time steps is not really a long run.
b) if you kinetic energy is rising that typically would
have one of two reasons: 1) your system is not yet
equilibrated or 2) your time step is too large. or both.
cheers,
axel.
Axel,
I was using a 1.0 fs timestep initially, but on reducing that to 0.5, 0.1 and 0.05 I find that the system generates similar error at different stages of the simulation. Is reducing the timestep to values of 0.001 fs advisable for equilibrating the system. Also, apparently by reducing the number of processors, I find that the system runs smoothly without any errors. Should a system execution be dependent on the number processors?
Ganesh
Also, apparently by reducing the number of
processors, I find that the system runs smoothly without any errors. Should
a system execution be dependent on the number processors?
in a statistical sense, no.
Steve
I was able to figure out the errors associated with the atoms being lost. The equilibration scheme had to be modified by including fix nve+langevin as suggested in a previous post by Axel. However, the temperature seems to keep on increasing when the RNEMD is applied after the system is well equilibrated (total energy constant over a simulation of 100 ps using NVE). Using a smaller timestep increases the temperature value, which means kinetic energy is increasing. I am unable to figure out where from the additional energy is getting into the system, and if a fix-viscous type of option needs to be implemented. My earlier experience of using RNEMD with crystalline materials did not require any such implementation though. Any suggestions will be helpful.
Do you exchange energies of the particles with the same masses then
you apply RNEMD?
German Samolyuk
I agree with German. Did you make sure that you are swapping KE for same masses of atom (with the RNEMD version that is default in lammps code)?
I second that. In case you have to swap velocities of atoms of unequal
masses, please refer to Craig Tenney's email to the mailing list on
May 25 in which he detailed the necessary (simple) code changes to
modify the RNEMD method for use with atoms of unequal masses.
Zhun-Yong
Posted a patch for this today as 16Jul - that contains Craig's code
for this.
Steve