Parrinello Rahman MD (1980)

LAMMPS LAMMPS Mailing List Mirror
1

I’m studying, and trying to reproduce as best I can the study in Parrinello, and Rahman’s, “Crystal Structure and Pair Potentials: A Molecular Dynamics Study”, PRL, (45) no. 14, 1980, with lammps.

Among the numerous choices for deforming the simulation box, including the howto section of the doc on triclinic boxes,

http://lammps.sandia.gov/doc/Section_howto.html#howto_12

I’ve decided on using fix npt as the simplest choice in this self-training exercise.

How can I correctly implement the quantity in the paper, “W*”, which the authors call a mass parameter in my npt lammps simulation? (Parrinello & Rahman used a value of W* = 20)

Am I also using the npt fix correctly to indepently vary the x, y, and z dimensions of the simulation box correctly?

I output the box sides as thermo during the runs and they all look approximately equal throughout the entire run.

Any other guidance/comments appreciated.

respectfully,

test.in (1.09 KB)

2

I'm studying, and trying to reproduce as best I can the study in Parrinello,
and Rahman's, "Crystal Structure and Pair Potentials: A Molecular Dynamics
Study", PRL, (45) no. 14, 1980, with lammps.

Among the numerous choices for deforming the simulation box, including the
howto section of the doc on triclinic boxes,

http://lammps.sandia.gov/doc/Section_howto.html#howto_12

I've decided on using fix npt as the simplest choice in this self-training
exercise.

none of them provides an extended lagrangian mechanism.

How can I correctly implement the quantity in the paper, "W*", which the
authors call a mass parameter in my npt lammps simulation? (Parrinello &
Rahman used a value of W* = 20)

you have to write a parrinello-rahman extended lagrangian integrafor
that treats the box dimensions additional (fictitious) degrees of
freedom.

... and if you get this far, you could then also add lagrange
multipliers for charges and thus enable modeling polarizable models
that use this mechanism (as does car-parrinello MD, but that would be
beyond(?) the scope of LAMMPS)

Am I also using the npt fix correctly to indepently vary the x, y, and z
dimensions of the simulation box correctly?

the npt integrator is a different integrator. a parrinello-rahman
variable cell integrator is much simpler and does not sample the npt
ensemble properly.

I output the box sides as thermo during the runs and they all look
approximately equal throughout the entire run.

that is not a very thorough measure for sampling the correct ensemble.
that works with a berendsen scheme, which is entirely empirical, just
as well. try some property that is based on fluctuations, for example.

axel.

3

How can I correctly implement the quantity in the paper, “W*”, which the
authors call a mass parameter in my npt lammps simulation? (Parrinello &
Rahman used a value of W* = 20)

you have to write a parrinello-rahman extended lagrangian integrafor
that treats the box dimensions additional (fictitious) degrees of
freedom.

The doc page for fix npt references the Parinello/Rahman paper, but
fix npt is using eqs of motion and integration methods that are now

more standard in barostatted MD, and as Axel said perform better
sampling in the NPT ensemble. The Pdamp parameter you define
with fix npt is effectively a mass parameter (actually inverse mass)

which is used to update the extra degree of freedom associated
with a box dimension.

Steve