Dear LAMMPS Users,
To calculate the thermal conductivity, LAMMPS provides 4 methods, including Langevin thermostat, heat exchange, MP, and the GreenKubo method.
For the Langevin thermostat, I am confused by the calculation of energy flux dQ. According to the help document “REDEME.txt”, I CAN NOT find the socalled total in/out energy for 2 regions normalized by # of atoms in the log file. Where is the 0.905+0.947? I can not find them in the log file. I have run this example by myself, and compared the results with the benchmarks, but still failed. Can anyone help me?
Any help would be highly appreciated.
Thanks
Timmy
Yakang,
Perhaps there are typos in readme file (e.g., the numbers you mentioned were perhaps from another simulation, instead of the simulation corresponding to the log file). The two numbers you mentioned should correspond to the numbers at line 218 and columns 56 in the log file (I am referring to this link https://github.com/lammps/lammps/tree/master/examples/KAPPA)
0.89036148 0.88285227
In general, the data analysis methods given in the readme file just give a quick and rough estimate and are only adequate for systems with relatively small thermal conductivity, where the time for achieving steadystate is relatively short. For systems with relatively large thermal conductivity and phonon mean free path, one needs to analyze the temperature profile and energytime relation very carefully, to make sure that steadystate has been actually achieved.
There is also a related question: https://sourceforge.net/p/lammps/mailman/message/35710354/
Best,
Bruce
Dear Prof. Dr. FAN,
Thanks very much for your kind reply.
Using “0.89036148 0.88285227” to calculate the energy flux dQ is reasonable and consistent with the log file. That makes sense!
Thanks again for your help.
All in Best,
Timmy
Dear Prof. Dr. FAN,
Thanks very much for your kind reply.
Using “0.89036148 0.88285227” to calculate the energy flux dQ is reasonable and consistent with the log file. That makes sense!
I have a following question: How to determine the convergence of simulation for this Langevin thermostat method? As the simulation time goes, the in/out energy increases all the time. When should we stop the simulation to calculate this energy flux to ensure that we can get accurate Kappa?
Thanks again for your help.
All in Best,
Timmy
Yakang,

This depends on the system you study, particularly the length of the system. A longer system requires a longer time to achieve steady state (it takes time to transfer heat). For example, it requires about 10 ns for a 10 micron long graphene sheet. So you need to do some computer experiments to gain experiences.

To check whether steady state has been achieved, you can divide you total production time (the last run) into a few blocks (say, 10) and calculate the temperature profile and heat flux for each block. Then you can check the time evolution of the temperature profile and heat flux.

It (usually) also requires to do a few independent simulations to get a reasonable error bar.
For more suggestions, you can discuss with me privately.
Best,
Bruce