I’m trying to find phonon mean free path or phonon lifetime to explain the change in thermal conductivity. However, I can’t find a way to do so. But I see some paper can calculate this with lammps. I wonder if anyone has tried this. Could you please give me some hints?
Thank you.

You can obtain phonon lifetimes by post-processing outputs from MD (eg lammps) computations only, principally by normal mode decomposition (NMD). This requires two different computations: a) a fix-phonon to obtain the eigenfrequencies and eigenvectors b) a long NVE run at thermal equilibrium, keeping trace of the atomic displacement from equilibrium and velocities.

Traditionally, step a) is achieved with lattice dynamics, however the Green’s function MD approach of fix-phonon is a very interesting alternative.

Then, you need to map the eigenvectors on the atomic disp. and velocities, which will give you the normal mode coordinates. After that, you can form a suitable expression of the mode-wise harmonic heat flux or of the total energy as a function of the normal mode coordinates, perform autocorrelation, and estimate the phonon lifetime by integrating the autocorrelation function (numerically or analytically with a fit).

That’s roughly the time domain normal mode analysis (TDNMA) method, look at papers from McGaughey, Larkin, Henry & Chen etc for more details. You also have another approach which is frequency-based (called SED, for spectral energy density). There are also very recent papers, on a non-equilibrium approach: http://arxiv.org/abs/1511.02607

if you want to use nonequilibrium methods for determining the mean free paths, I suggest you take a look at my codes at https://github.com/ksaaskil/shc-python-tools The Python scripts and LAMMPS codes are for calculating the spectral heat current distribution in a-Si, and it should be easy to modify the codes for different materials. The mean free paths are obtained by calculating spectral currents for various lengths and determining the frequency-dependent mean free paths from the length-dependence as explained in our papers listed in readme.md. I hope that helps.