Thermostatting a region

Hello all,

I’m new to MD and LAMMPS. I’m trying to calculate the thermal conductivity of tip3p water model, in between two fcc lattice walls (a channel, basically). I would like to know how to thermostat the walls properly. I’ve been trying out various thermostatting techniques available in lammps but I’m not able to get to the equlibration point.

This is what I do :

my boundary is p p f, (used f because of the kspace solver with pppm)
used the typical sigma and epsilon values for tip3p model(including bond and angle parameters),
A spring/self potential for the walls.
set a temperature of 303K using the velocity command for the walls and water region.
Then minimized the energy of the system.
------hope up to here, nothing is wrong-------

Now, I was confused in choosing a thermostat. On referring to certain literature I found that the NVT would be fine, but I have got non-periodic boundary condition, so I guess the canonical ensemble would not be followed. I tried using the NVT command just for the liquid region, but the temperature of the system does not increase beyond 234K. I tried using fix langevin + nve and still, there isn’t much of a difference. I also tried the fix temp/rescale + nve, the system superficially looks equilibriated but the results (density/temperature profiles) were incorrect.

In theory i guess using a fix langevin + nve should work fine, right? Is it advisable to use a fix temp/rescale for the walls with this ?

yours sincerely,
Pinto.

Hello all,

I'm new to MD and LAMMPS. I'm trying to calculate the thermal conductivity
of tip3p water model, in between two fcc lattice walls (a channel,
basically). I would like to know how to thermostat the walls properly. I've
been trying out various thermostatting techniques available in lammps but
I'm not able to get to the equlibration point.

This is what I do :

my boundary is p p f, (used f because of the kspace solver with pppm)
used the typical sigma and epsilon values for tip3p model(including bond and
angle parameters),
A spring/self potential for the walls.
set a temperature of 303K using the velocity command for the walls and water
region.
Then minimized the energy of the system.
------hope up to here, nothing is wrong-------

Now, I was confused in choosing a thermostat. On referring to certain
literature I found that the NVT would be fine, but I have got non-periodic
boundary condition, so I guess the canonical ensemble would not be followed.

yes, but that doesn't matter. the logic is different. fix nvt will
result in an NVT ensemble, if - and only if - you have a bulk system
in equilibrium. that doesn't change the validity of the algorithm to
couple a non-periodic system to an external heat bath. the important
part about the nose-hoover thermostat is that it is best only used,
when the system is already in or close to equilibrium and then only
weakly coupled (the relaxation constant determines what the respective
characteristic frequency is). btw: similarly, fix nve does not expect
an NVE ensemble, it just is a plain integrator that will result in an
NVE ensemble, if no other manipulations are present.

I tried using the NVT command just for the liquid region, but the
temperature of the system does not increase beyond 234K. I tried using fix

how do you measure the temperature? on the water region or the entire system?

langevin + nve and still, there isn't much of a difference. I also tried the

did you time integrate the wall with that?

fix temp/rescale + nve, the system superficially looks equilibriated but the
results (density/temperature profiles) were incorrect.

temp/rescale should be outlawed. good uses for it are scarce, but
there are too many cases of abuse.

In theory i guess using a fix langevin + nve should work fine, right? Is it
advisable to use a fix temp/rescale for the walls with this ?

don't use temp/rescale at all.

when setting up a simulation from a completely non-equilibrated
configuration, i would recommend a 4-step procedure:
- minimization to reduce potential energy from overlaps and bad
(local) geometries
- fix nve + langevin with a short time constant to reach
equipartitioning of kinetic energy. this will dampen intrinsic
fluctuation of the system, so it is not advisable for production
- fix nve + langevin with a long time constant
- fix nvt with a somewhat long-sh time constant.

if your system is far from equilibrium, you may replace fix nve by fix
nve/limit in the second step for additional avoidance of problems

if your system equilibrates quickly, you may skip the third step.

for the final production, it may be worth considering to have the
walls and the liquid independently thermostatted. even with two
different methods.

in general, look for quantities that you can relate your simulation to
and try to measure them in the simulation and tweak parameters. ...and
don't be impatient. good equilibration always takes longer than you
expect. at least twice as long (and then some).

axel.

Hello Dr. Axel,

Thanks for the suggestion! It indeed works.

With the above method, the system runs. However I had noticed something unusual, one of the wall layer breaks apart. I then re-positioned the wall layers and this still happens. The strange thing is both the upper and lower wall layers are kept apart from equal distances from the water molecules (i,e. the lower wall is 8A below, and the upper wall is 8A above). Is this because of bad dynamics ?

Thanks and Regards,

Pinto

Hello Dr. Axel,

Thanks for the suggestion! It indeed works.
With the above method, the system runs. However I had noticed something
unusual, one of the wall layer breaks apart. I then re-positioned the wall

what does that mean "break apart"?

layers and this still happens. The strange thing is both the upper and lower
wall layers are kept apart from equal distances from the water molecules
(i,e. the lower wall is 8A below, and the upper wall is 8A above). Is this
because of bad dynamics ?

how should i know. visualize it and examine what is happening. if it
is happening too fast, then write out coordinates more often.

axel.