[lammps-users] Muller Plathe method for SiC

Dear Lammps users,
I'm trying to apply Muller-Plathe method to calculation of thermal
conductivity in SiC system with Tersoff potential.
I'm using Lammps version 6Dec08 on a linux cluster with 24 processors.
1) Could you please check it I correctly determine Tersoff potential
in my input file. The value of thermal
conductivity is ~30 times smaller compared to experiment.
2) Second question. Dear Steve you mentioned codes for calculation of
heat flux. I wonder if it's available.

German Samoluyk

Bellow see input file for M-P method.

units metal
boundary p p p

atom_style atomic

lattice sc 4.329

region box block 0 1 0 1 0 1

create_box 2 box

create_atoms 1 single 0 0 0
create_atoms 1 single 0.5 0.5 0
create_atoms 1 single 0.5 0 0.5
create_atoms 1 single 0 0.5 0.5
create_atoms 2 single 0.25 0.25 0.25
create_atoms 2 single 0.75 0.75 0.25
create_atoms 2 single 0.75 0.25 0.75
create_atoms 2 single 0.25 0.75 0.75

mass 1 28.086
mass 2 12.011

replicate 320 4 4

velocity all create 500.0 87287

#read_restart rstrt_2

pair_style tersoff
pair_coeff * * SiC.tersoff Si C

newton on
neighbor 0.3 bin
neigh_modify delay 10

fix 1 all nve

group Silicon type 1
fix 2 Silicon thermal/conductivity 200 x 20

compute ke all ke/atom
variable temp atom c_ke[]/(1.5*8.67e-5)

fix 3 all ave/spatial 1 8000 10000 x lower 69.264 v_temp file tmp.profile

thermo_style custom step f_2 temp

thermo 10000

timestep 0.0002

dump 1 all atom 500 dump.meam

#restart 200 rstrt_1 rstrt_2

run 4000000

You're using the Tersoff potential correctly. Have no
idea why your kappa is 30x too small.

Re: heat flux - we're close to releasing what has been
contributed, probably later this week.


Dear German,
One reason that your values are so low could be due to Tersoff potential itself. It is a short range potential with no van der Waals interactions in it. So, you might wanna add LJ interactions on the top of it to get better estimate.


Hello German,

Your block is one lattice lenght small, in md simulations there is size effects due the system size. If you want to compare your thermal conductivity to macroscopic bulk system you need to do simulations with increasing system sizes and then extrapolate from those results the bulk thermal conductivity. LJ-potentials are not very usefull for bulk systems. I have gotten very good results for Si and Ge with tersoff potential using MP, i think its genious potential.

And i think also that MP needs way bigger system to work. You have so many CPUs, try at minimum something like 0 4 0 4 0 10.

Tuomo Hyvönen

Dear Vikas,

What is confusing me is that thermal conductivity calculated by G-K
method for tersoff potential
in beta-SiC by Li, Porter and Yip is in good agreement with experiment.

All the best,

Hello Tuomo,
After definition of lattice I use in the input
replicate 320 4 4
it means that I'm working with a system 320x4x4, not 1x1x1
and I use tersoff potential not LJ.

2009/7/1 Tuomo Hyvönen <[email protected]>:

Hi German & folks,

Im sorry, i missed the replicate command. Other proposal: thing with MP in lammps is that it assumes that exhanged velocity vectors are from same weight atoms. So im not sure how your SiC is arranged, but that might be it. Somebody introduced a correcting code for this before in this forum, but i dont know how did it works.

Tuomo Hyvönen

2009/7/9 German Samolyuk <[email protected]…12…24…>

Hi Tuomo,

Yes, you are absolutely right.
and in order to exchange energy between the same mass atoms I've
introduce group Si for Si atom only and exchange energy between atoms
from this group

group Silicon type 1
fix 2 Silicon thermal/conductivity 200 x 20


2009/7/10 Tuomo Hyvönen <[email protected]>:

What about introducing group for carbon atoms, simlarly with similar frequency


2009/7/10 German Samolyuk <[email protected]…24…>

I did not think about it.
I've checked if energy is conserved and temperature is constant with Si group.
Do you think it could change result?
I'll try it. Thanks!

2009/7/10 Vikas Varshney <[email protected]>: