Abrupt rise in thermal conductivity

LAMMPS LAMMPS Mailing List Mirror
1

Dear Users,

I want to measure thermal conductivity of Ar. For the convergence test I did several simulations with
cell size = 4x4x4unit cell(fcc) ,Temperature=10K

Question: 1) when timestep=5 fs
correlation length=200
and sample interval=100(so the correlation time here is 5x200x100fs=100ps)
then the value of tensor components(k11,k22,k33) of thermal conductivity are near about 1.9W/mK which is not equal to the already published value(~2.5W/mK)

But when timestep=10 fs
correlation length=200
and sample interval=50(so the correlation time here is 10x200x50fs=100ps)
then the value of tensor components(k11,k22,k33) of thermal conductivity are near about 2W/mK up to 4,300,000 steps(total steps is 10,000,000) but after that they

increase exponentially !!!

Could you suggest me about the reason behind that strange increment and how to get rid of that problem???

Question: 2) In my script the “thermo” command prints tensor components(k11,k22,k33) of thermal conductivity in log file along with other some variables and also for convenience
I have saved those components in a separate .txt file. Surprisingly the values read by log file and .txt file are not same !!!

Please help me by giving some suggestion.
I am attaching my input script files and output log files corresponds to above two questions

Question: 3) What is exact method to check the convergence since for different combination of “timestep”,“correlation length” and “sample interval” I am getting different values of
thermal conductivity along with problem which I face in Question: 1 ??

Thanks & Regards
Amal

in.therm_Ar_GK_5fs (1.89 KB)

in.therm_Ar_GK_10fs (1.89 KB)

log.thermoAr_GK_5fs (76.6 KB)

log.thermoAr_GK_10fs (148 KB)

2

Have you tested whether you can run a stable md at a 10fs time step for your system at all?
Axel

3

thank you Dr. Axel. I am not that much learned person of Lammps. Would you please tell me what is meant by stable md?

4

This has nothing to do with lammps but md in general. For example it means that you have to have an accaptable energy conservation runningban equilibratef system in nve ensemble.

Axel.

5

Sorry to annoy, Dr. Axel. But how to test this problem. From your 1st reply it is seemed to me that for longer timestep this problem may take place…is it true?

6

This is a question about md basics. Please consult a suitable textbook and teachers and experienced colleagues. A mailing list is no replacement for that and not good for it in the first place. It goes without saying that trying to run an advanced procedure such as a thermal conductivity determination without such fundamental knowledge is highly risky if not foolish.

Axel

7

Dear Amal

I haven't really any experience of calculating the thermal conductivity of solids using Green-Kubo but I would advise you to be careful when using the NVT ensemble for such small systems (And especially in the case of solids as there are very few modes of vibration). I would imagine that using a thermostat might remove some of the phonon modes which would contribute significantly to the thermal conductivity.

In any case in general for Green-Kubo calculations of such a small system I see no advantage in using the NVT ensemble over the NVE. I did a quick test on the system you provided the input for and it appears both time-steps will provide ample energy convergence at that temperature when in the NVE ensemble.

Obviously for solids one has to be very careful in extrapolating the thermal conductivity for various system sizes (finite size analysis) in order to account for the increasing amounts of phonon modes that may exist for different system sizes.

In general it is my understanding that calculation of the thermal conductivity of solids (and in particular amorphous) can be quite non-trivial so I would do the required reading of literature before attempting it.

All the best, Jeff.