Hi, There:
Recently I come up with a problem about NPT ensemble while doing MD simulations.
For example, if I have a box of atoms, with periodic boundary condition in x-, y- and z- directions. Then I divide the box of atoms into 2 groups, ie the atoms in the lower half part of the box belong to group 1 and atoms in the upper half part of the box belong to group 2. Theoretically can I use NPT ensembles (two thermo-bath at two different temperature) to control the temperature of those 2 groups at 2 different temperatures, ie, group 1 at 300K and group 2 at 600K? In LAMMPS software it seems I can do that way, but I am not sure whether that’s right. Thanks.
Best regards,
Xuhang
Hi, There:
Recently I come up with a problem about NPT ensemble while doing MD
simulations.
For example, if I have a box of atoms, with periodic boundary condition in
x-, y- and z- directions. Then I divide the box of atoms into 2 groups, ie
the atoms in the lower half part of the box belong to group 1 and atoms in
the upper half part of the box belong to group 2. Theoretically can I use
NPT ensembles (two thermo-bath at two different temperature) to control the
temperature of those 2 groups at 2 different temperatures, ie, group 1 at
300K and group 2 at 600K? In LAMMPS software it seems I can do that way, but
I am not sure whether that's right. Thanks.
with LAMMPS you have three options to handle this case:
a) run fix nvt on both segments with separate thermostats and then fix
press/berendsen to control the overall volume
b) run fix nph on the total system to do time integration with box
motion and then use fix temp/csvr or temp/berendsen of langevin to
control temperature
c) realize that there is little to gain from box fluctuations if the
system is large enough and just run with fixed volume.
please also note, that you likely want some kind of "healing" region
between those two extremes.
there are several long discussions about such non-equilibrium
simulations in the mailing list archives and some discussion in the
manual, too. perhaps, you should review them some more?
axel.
You should also understand better what kind of phenomena you are trying to simulate and what is the correct way to do it. Note that just because Lammps allows you to do something doesn’t mean it is correct (the thinking is done at the human level). One could mix-and mingle potentials in Lammps to do some type of hybrid simulation. Yet, just because the code will run without errors doesn’t mean your options are properly justified. In your example you have a 300K temp diff between your two regions with a sharp interface… awkward…
Carlos