Dear LAMMPS Users,
I am trying to calculate the thermal conductivity of a (10,10) CNT that is 5 nm in length using the Green-Kubo formalism. I modified the liquid Argon LAMMPS thermal conductivity example for CNT. First, I equilibrated the CNT structure at 300K, and then I ran the simulation to get the heat flux, heat current auto correlation function (HCACF) and thermal conductivity. To plot the HCACF, I averaged columns 4, 5, and 6 in the ‘J0Jt.dat’ file for every 1000 timesteps (which is my dump interval).
Currently I am having two problems -
- The HCACF does not decay to zero. I have tried running the simulation for up to 1200 ps, but the HCACF still did not decay to zero.
- The thermal conductivity value I attain differs based on the correlation length parameter I use in the ‘fix ave/correlate’ command. Should the conductivity be independent of correlation length? If not, then how should I determine the appropriate correlation length?
I checked the LAMMPS email archive, but I was not able to find anything specific to my issues. Attached is the LAMMPS script, as well as other input data files. There is also a graph of the HCACF over time attached.
I would appreciate any insights as to why the HCACF does not decay to zero.
Thank you,
Chad Daksha
in.cnt-edited.txt (2.44 KB)
CH.airebo (770 KB)
coordi-cnt.txt (31.9 KB)
HCACF_graph.pptx (122 KB)
You should subtract the average of total energy from particle’s total energy for the HCACF to decay to zero.
Thank you for your response. If you don’t mind, I have a few questions:
Would it be possible for you to provide any reference(s) for this subtraction? The sources I’ve looked through don’t apply this subtraction, at least to the best of my knowledge. I have attached them for viewing purpose. But, please correct me if I am wrong.
Also, if I wanted to implement the subtraction, could you suggest ways it may be possible in my script? I do not think it is possible to implement the subtraction prior to the ‘compute flux’ command, since it requires arguments that are also computes. Therefore, I would be unable to create variables for the new potential and kinetic energies, post subtraction.
Thank you once again,
Chad Daksha
Luke-JHT-2007-Thermal Conductivity of SWNT.pdf (963 KB)
Yao-2013-Thermal conduction of carbon nanotubes using molecular dynamics.pdf (582 KB)
Che-2000-Thermal Conductivity of CNTs.pdf (226 KB)
in.cnt-edited.txt (2.44 KB)
Thank you for your response. If you don't mind, I have a few questions:
Would it be possible for you to provide any reference(s) for this
subtraction? The sources I've looked through don't apply this subtraction,
at least to the best of my knowledge. I have attached them for viewing
purpose. But, please correct me if I am wrong.
Subtracting average quantity while calculating time correlation functions
is a general requirement, not specific to HCACF, but may not make any
difference in many cases. Specific to HCACF, you may want to see the
definitions in Eq. 2 and 3 of Physical Review B
pdf/10.1103/PhysRevB.43.6573. However, when I tested on a different system
using my code, I found no difference with and without subtracting the
average energy. So lammps code must give you the same answer without the
need of subtracting the average energy.
My suggestion in the previous reply was in fact based on an apparently
wrong data you provided, in which the HCACF fluctuates around non-zero
value even at long time. I suspect you wrongly interpreted J0Jt.dat output.
With sufficiently long enough correlation length of few picoseconds
(attached input), I got the following HCACF with your input file.
Regarding your other questions, (1) your HCACF was not decaying to zero
because you possibly used a short correlation length of 0.0002*100*10 = 0.2
ps (despite generating a nanosecond long trj); (2) you can see that the
attached HCACF decays to zero in about 2 ps and as a general rule of thumb,
you must need at least 10 such correlations long trj to get a decent
estimate of thermal conductivity.
!image.png|636x481
Also, if I wanted to implement the subtraction, could you suggest ways it
in.cnt-edited.txt (2.44 KB)
