Phonon in LAMMPS

Dear LAMMPS users,

I have a problem which I cannot understand and therefore I cannot solve it. It would be very kind of you if you will explain me what I did wrong or give me some hint.

I want to calculate phonon DOS for diamond using LAMMPS fix command phonon with AIREBO potential. As I understand this type of calculation needs molecular dynamics simulation at finite temperature. My question is what temperature should I use to get a result comparable with experiment? Because I tried different temperatures from 300K to 1500K and results were not in an agreement with experimental data. Also I tried to use different sizes of diamond cell at fixed temperature but nothing changes. I tried to use ReaxFF potential as well, but it gets me even worse results.
Please find attached two pictures with phonon spectra obtained from AIREBO (AIREBO.jpg) and ReaxFF (ReaxFF.jpg).

Here is my input file for this calculation:

boundary p p p
units metal
atom_style atomic

pair_style airebo 3.0

pair_coeff * * CH.airebo C
neighbor 0.5 nsq
neigh_modify every 1 delay 0 check yes
fix 1 all npt temp 500 500 1 iso 1. 1. 1. pchain 8 drag 1.0

fix 2 all phonon 10 500 5000 Phonon
timestep 2e-4

thermo_style custom step temp etotal press vol
thermo 100
dump 1 all xyz 500

dump_modify 1 append yes element C
run 650000

Thank you in advance!



Direct comparison to experiment is a bad idea because many potential problems could easily end up hidden. Moreover, I wonder what kind of resemblance to the experimental results the airebo potential can render anyways. Why not trying something simpler to at least rule out other issues? You could try reproducing some of the phonon curves of the original paper by Kong, or try to obtain dispersion curves for more simpler systems/potential for which the results can be generated using alternative methods. A good test case can be found in a paper by Dario Alfe where he introduces his “Phon” code that gets phonon dispersions using a different approach (google it). There is also a similar open source code (fropho) by some japanese gentelman that you could employ to cross-validate your MD data. Once you are convinced
your methodology is bulletproof you can move on to get the DOS (which you should also test as well in a comparison to another SIMULATion study).