I am, trying to calculate thermal conductivity using Green-Kubo method using compute heat/flux command. There is an input script given at the end of compute heat/flux command description. Can anybody please clarify why we have used f_JJ[3], f_JJ[4],f_JJ[5] in vaiables k11, k22, and k33. The input script is pasted below.

Thank you.

Truly,
Rajesh.

Sample LAMMPS input script for thermal conductivity of solid Ar

units real
variable T equal 70
variable V equal vol
variable dt equal 4.0
variable p equal 200 # correlation length
variable s equal 10 # sample interval
variable d equal $p*$s # dump interval

Fix JJ is an instance of fix ave/correlate. The
fix ave/correlate doc page explains its output (see
the bottom of the page). Part of the text says:

The global array has # of rows = Nrepeat and # of columns = Npair+2. The first column has the time delta (in timesteps) between the pairs of input values used to calculate the correlation, as described above. The 2nd column has the number of samples contributing to the correlation average, as described above. The remaining Npair columns are for I,J pairs of the N input values, as determined by the type keyword, as described above.

One more point I noticed is that in doc page description in the kappa formula volume term is in numerator while in the script given below it is in denominator. I think doc page formula has some error as in literature also volume term will be found in denominator. Please correct me if I am wrong.

Thank you for clarification. I got it. V factor in not included in heat/flux compute and thats why it has been divided twice by vol in lammps script. One more thing I have a question. Why do calculated Jx, Jy and Jz and average them even if we need to calculate kappa for a single direction only??

If you are modeling an isotropic system, ideally, the kappa values along x,y,z should be equal. Therefore, to reduce the uncertainty of the results of MD simulations, we could average the three kappas. However, if your system is anisotropic, this averaging would be a wrong practice.

It is worth mentioning that comparing the kappa values could provide a good means for evaluating the precision of the results of our simulations.

Thank you for your suggestion. Actually, my kappa values are different in all the three directions with large difference. So, it will be better for me to use direction wise kappa values.

One more query is regarding convergence. Of which quantity, the convergence should be attained?? J or kappa??

your welcome. Actually, you will have a different value for J at any time step while the entire simulation gives only one kappa (in any direction). Therefore, it is the convergence of kappa which you should be concern about. For a better perception of this topic you could read the following article (especially look at Fig. 7):