Hi Steve, Alex and every body else,
Has any of you ever performed thermal conductivity (\kappa)
calculations for organic molecules using TraPPE-UA in Lammps?
I'm interested in computing \kappa for certain organic compounds and
have been doing some tests on n-Hexane using this FF.
For \kappa I am getting values that underestimate the experimental
\kappa as well as the one computed by other people with alternative UA
models. (J. Phys. Chem. B 2005, 109, 15060-15067)
Tried using both NEMD and RNEMD in Lammps. For the NEMD I extended the
"compute_heat_flux" by computing the flux on atoms that belong to a
region and a group. For RNEMD I used the bare bones Lammps
implementation. Both type of calculations are giving me more or less
the same value of ~0.06 W/mK (0.12 W/mK is what to expect). Relevant
lines and variables in my input files:
I recall that the TraPPE model discards the hydrogen atoms in an
alkane molecule and adjusts the bonded and non-bonded interactions
accordingly.
I don't have any experience calculating thermal conductivity, but I wonder:
Is it possible to get realistic thermal behavior from this kind of model?
(The number of degrees of freedom is much lower. I don't think the
parameters of the TraPPE were chosen with this in mind. I remember
Axel had a similar point recently.) My wild guess is that perhaps you
should not use a "united-atom" model (at least not one which was not
designed for this application).
Have you tried using the same NEMD or RNEMD settings with an all-atom model?
Are the results more accurate?
Cheers
Andrew
Hi Andrew,
I recall that the TraPPE model discards the hydrogen atoms in an
alkane molecule and adjusts the bonded and non-bonded interactions
accordingly.
This is true but so does the united atom (UA) model of the reference
J. Phys. Chem. B 2005, 109, 15060-15067.
I don't have any experience calculating thermal conductivity, but I wonder:
Is it possible to get realistic thermal behavior from this kind of model?
Yes it is possible. See reference above.
(The number of degrees of freedom is much lower. I don't think the
parameters of the TraPPE were chosen with this in mind. I remember
Axel had a similar point recently.) My wild guess is that perhaps you
should not use a "united-atom" model (at least not one which was not
designed for this application).
Well, I am studying other properties for which UA is the way to go. It
could well be that
TraPPE-UA cannot get \kappa right. Not very familiar with the TraPPE
history but I am almost
positive \kappa is not one of the input data used in the
parametrization of the model. However,
they probably didn't use viscosities either and the model gets it
pretty OK for n-Hexane.
Have you tried using the same NEMD or RNEMD settings with an all-atom model?
Are the results more accurate?
I have tried RNEMD and currently trying NEMD with OPLS-AA. The values
are higher for the RNMED runs (NEMD still running) and much closer to
experiment. I was careful enough to use smaller timesteps with RNEMD,
because the implementation in Lammps swaps only atomic velocities (in
molecular systems should swap center of mass velocities) and the
fluxes could show drift from the harmonic bonds going a bit crazy
while doing NVE integration.
Anyways, may well be that TraPPE-UA indeed is not suited. Yet, as I am
not really a soft matter guy, I wanted to drop the ball on the gurus
court just in case I missed something.
Carlos
This is true but so does the united atom (UA) model of the reference
J. Phys. Chem. B 2005, 109, 15060-15067.I don't have any experience calculating thermal conductivity, but I wonder:
Is it possible to get realistic thermal behavior from this kind of model?Yes it is possible. See reference above.
Oh well, shows you what I know.
(The number of degrees of freedom is much lower. I don't think the
parameters of the TraPPE were chosen with this in mind. I remember
Axel had a similar point recently.) My wild guess is that perhaps you
should not use a "united-atom" model (at least not one which was not
designed for this application).Well, I am studying other properties for which UA is the way to go. It
could well be that
TraPPE-UA cannot get \kappa right. Not very familiar with the TraPPE
history but I am almost
positive \kappa is not one of the input data used in the
parametrization of the model. However,
they probably didn't use viscosities either and the model gets it
pretty OK for n-Hexane.Have you tried using the same NEMD or RNEMD settings with an all-atom model?
Are the results more accurate?I have tried RNEMD and currently trying NEMD with OPLS-AA. The values
are higher for the RNMED runs (NEMD still running) and much closer to
experiment. I was careful enough to use smaller timesteps with RNEMD,
because the implementation in Lammps swaps only atomic velocities (in
molecular systems should swap center of mass velocities) and the
fluxes could show drift from the harmonic bonds going a bit crazy
while doing NVE integration.
Anyways, may well be that TraPPE-UA indeed is not suited. Yet, as I am
not really a soft matter guy, I wanted to drop the ball on the gurus
court just in case I missed something.
Carlos
Alas, I'm not much of a guru. (But I post anyway.) Hopefully someone
else will eventually chime in.
(Email the TraPPE authors?)
Cheers!
Andrew