[lammps-users] simulation of thousands of rigid objects

Hi LAMMPS users,

I am simulating systems consisting of 10,000 to 50,000 rigid particles. Each rigid particle contains 9 atoms. As I see in the LAMMPS documentation of fix_rigid, such systems do not yield good scale in parallel. Is there any way around this problem, such that I can run simulations as fast as possible?

Any ideas and suggestions are appreciated.

Thank you,
-Trung

Hi LAMMPS users,

I am simulating systems consisting of 10,000 to 50,000 rigid particles.
Each rigid particle contains 9 atoms. As I see in the LAMMPS
documentation of fix_rigid, such systems do not yield good scale in
parallel. Is there any way around this problem, such that I can run
simulations as fast as possible?

a) you could write your own integrator where you integrate not the
equations of motion for each atom, but for each particle (similar to
the asphere integrators).
b) you could write a pair style so that the interactions of your
particles can be used with the existing asphere integrators.

or in short you don't treat your particles as collection of atoms
with constraints, as real rigid objects and then map the forces on
each atom of each particle into a force on the center of mass and a
torque around the center.

cheers,
   axel.

Fix rigid does what Axel suggests, treating each rigid body
as a sum of forces and torques to integrate it. It just may
be inefficient for lots of small bodies. The alternative is
to do the same thing but with more sophisticated communcation
for bodies that straddle two (or more) processors. Such
an enhanced fix rigid hasn't been written.

But I would just try the existing fix rigid and see how it does.
I'd be interested to see your timings and how much speed-up
you get for your system in parallel.

Steve

Thank you Steve and Axel. I'll send you the timings shortly.

-Trung

Quoting Steve Plimpton <[email protected]>:

Trung,

Dont forget on Aon you have myrinet, but not all the nodes have it. If your not sure how to use it let me know, This only applies if your going to have more than one node in the calculation, and your using aon.

Brock Palen
www.umich.edu/~brockp
Center for Advanced Computing
[email protected]
(734)936-1985