Why is the thermal conductivity of SiC obtained using LAMMPS too low?

[lzq857]

Dear all，
I am new to ALAMODE. By referring to the examples on the official tutorial, I successfully calculated the thermal conductivity of silicon (Si). However, for silicon carbide (SiC), although the phonon dispersion I calculated is accurate, the thermal conductivity is too low—around 50 W/mK at 300 K. I have tried adjusting the cutoff energy, k-point density, and supercell size, but none of these changes could helpe. Therefore, I am seeking assistance. Here is part of my input

units metal
atom_style atomic
boundary p p p

read_data tmp.lammps

pair_style tersoff
pair_coeff * * SiC.tersoff C Si

dump 1 all custom 1 XFSET id xu yu zu fx fy fz
dump_modify 1 format float "%20.15f"
run 0

[ttadano]

I do not see any obvious errors in the input files, so the reason for the underestimation is not clear to me. Perhaps, the accuracy of the third-order force constant may not be good.

For example, can you check how the thermal conductivity changes with the cutoff radius by gradually increasing it from the nearest-neighbor distance to 8.3148? This may help you understand the reason for the issue.

[lzq857]

I have set the cutoff radii to (none 3.6), (none 5.8), (none 8.3148), and (none 10), and the resulting thermal conductivities at 300 K are 913.0996, 175.9575, 113.0928, and 105.0978, respectively. The results seem to converge around 100. Among these, the value of 5.8 is slightly larger than the second-nearest-neighbor distance. Theoretically, I should choose the converged result of 105. However, this value still appears to be too low. I am also puzzled as to why the results did not converge just a little bit beyond the second-nearest-neighbor distance. I am using the Tersoff 1989 potential, I still haven't been able to identify where the issue lies.

[ttadano]

I have computed the lattice thermal conductivity of 3C-SiC using the Tersoff potential, but I could not reproduce the issue.

The thermal conductivity at 300 K was ~740 W/mK when the third-order force constants were computed up to the second-nearest neighbor. Please see the attached file for more details. SiC.zip

This value did not change when I increased the cutoff radius to ~8.3 Bohr (which includes pairs and triples up to the 5th nearest neighbor). This is reasonable because the Tersoff potential is short-ranged (the interaction range is only up to the second-nearest).

[lzq857]

Thank you very much for your help! Your cubic.fcs file is much more detailed than mine, and I have obtained the correct results using your fcs file. Since the phonon dispersion I calculated is correct, it indicates that the harmonic.xml file is fine. Therefore, the issue seems to lie with cubic_DFSET. To obtain cubic_DFSET, the correct in.sw and SiC222.lammps files are needed, but I couldn't find any errors in these two files (the fact that I can calculate the correct phonon dispersion should prove that these two files are fine). It really frustrates and intrigues me, so may I request your complete input files? Including (sic.tersoff, in.tersoff, SiC222.lammps, SiC_alm.in), as I would like to conduct a detailed comparison of each file.
Thank again.

[ttadano]

Here are my input files for the LAMMPS and ALAMODE calculations. to_share.zip

I used the SiC.tersoff file from https://c4science.ch/source/lammps/browse/master/potentials/SiC.tersoff.

[lzq857]

Dear Dr. Tadano,
Thank you so much，by comparing your files, I found that the issue lies in the LAMMPS data file. The lattice constant of 17.2842 in this file was obtained through Tersoff optimization by GULP. When I modified the lattice constant in data_file to the pre-optimized value of 17.4, while keeping other files unchanged, the calculated thermal conductivity was correct. However, I don't understand why this situation occurred.
In addition to this question, as a beginner, I would like to ask how to obtain the optimized lattice structure suitable for thermal conductivity calculations in ALAMODE, as well as the corresponding data file. In other words, I want to learn the process you used to obtain the data file to avoid encountering the problem of incorrect optimized crystal structures again. This is because I plan to perform substitution on SiC and carry out calculations in the future.
SiC444_lammps.txt