query on use of Gay-Berne potential in combination with other potentials in a "single molecule"

LAMMPS LAMMPS Mailing List Mirror
1

I had a quick query regarding lammps, which I didn't manage to resolve
via the manual.

I understood from the manual that the lammps code works well with
(say) single Gay-Berne particles and also with mixtures of Gay-Berne
particles or Gay-Berne's and spheres.

But... I wondered if it was now possible
to link Gay-Berne's (or other anisotropic particles)
together to form a singe molecule
(to provide, for example, a coarse-grained model of a polymer)

This might, for example look like...
GB-LJ-LJ-LJ-GB

where the lines indicate spring bonds, with the additional possibility
of using angle potentials also. (Ideally the bonds are not through the
centre of the Gay-Berne particles)

best wishes,
Mark

2

In principle, this would be no different than hooking
LJ monomers together into a chain via a harmonic
or FENE potential. The pair_style GB knows
nothing about this, so it does not change pairwise
interactions, except that it will respect the special_bonds
settings, same as other pair potentials do for normal
polymer chains. If an ellipsoid is one
of the monomers, the bond spring is simple attached
to its center. It would be free to rotate as it is torqued.

If you want your short chain to be a rigid body, which
would not allow the ellipsoids to rotate relative to the
rest of the body,you could use fix rigid, which can
include ellipsoids, point particles, and finite-size
spheres if you like.

Steve

3

mark,

I had a quick query regarding lammps, which I didn't manage to resolve
via the manual.

I understood from the manual that the lammps code works well with
(say) single Gay-Berne particles and also with mixtures of Gay-Berne
particles or Gay-Berne's and spheres.

But... I wondered if it was now possible
to link Gay-Berne's (or other anisotropic particles)
together to form a singe molecule
(to provide, for example, a coarse-grained model of a polymer)

This might, for example look like...
GB-LJ-LJ-LJ-GB

if you look at the gay-berne potential energy function, you should
see that it is just a aniosotropic generalization of the 12-6 lennard-jones
potential, so you can express this with the gayberne pair style alone
(there may even be optimizations in the pair style for that scenario,
but i am not 100% certain. it has been a while since i last looked at the code).

where the lines indicate spring bonds, with the additional possibility
of using angle potentials also. (Ideally the bonds are not through the
centre of the Gay-Berne particles)

the problem with that would be computational efficiency. gay-berne
potentials are extremely computationally demanding - which, btw,
makes them such an ideal target for GPU acceleration - so a better
approach might be to just overlay multiple LJ particles (like an expansion).
check out the work of the group of sharon glotzer, they have devised
some nifty schemes to optimize those overlays to use the minimum
amount of interactions to construct models for shaped particles.

cheers,
     axel.

4

thanks Steve,
that's interesting,

I guess in this case I could do a jointed chain with
flexible joints by having a series of rigid bodies defined
as an ellipsoid sphere combination with the flexible links
via the spheres

I guess an alternative would be if
I wrote an additional bond potential
that links the ellipsoid in a different place
(e.g. the end would be obviously
defined by geometry), this would
give a force and torque on the ellipsoid and a force on
an atom

best wishes,
Mark

5

the problem with that would be computational efficiency. gay-berne
potentials are extremely computationally demanding - which, btw,
makes them such an ideal target for GPU acceleration - so a better
approach might be to just overlay multiple LJ particles (like an expansion).
check out the work of the group of sharon glotzer, they have devised
some nifty schemes to optimize those overlays to use the minimum
amount of interactions to construct models for shaped particles.

cheers,
      axel.

thanks Axel,

I've seen the very nice work of Sharon Glotzer,
  and maybe rigid spheres combinations
can indeed do most things
& might be worth me trying out

I guess I was looking at this particularly in
relation to linked GB disks, which
might require a lot of spheres to get a decent aspect ratio,

I've not yet tested what the break even point is,
in relation to Gay-Berne's and spheres (for like-like
i.e. similar packing fraction) I'm guessing it might be
around 6 spheres to a GB (but don't really know)

certainly the GPU acceleration looks an exciting option
for GB particles, & is one of the reasons for currently
investigating lammps

we've also been playing with other, possibly cheaper,
longer time-step potentials for anisotropic systems
e.g.
A new anisotropic soft-core model for the simulation
of liquid crystal mesophases. Lintuvuori, J. S.,
Wilson M. R., J. Chem. Phys., 2008, 128, 044906.

A soft-core Gay-Berne model for the simulation of
liquid crystals by Hamiltonian replica exchange.
  Berardi, R., Zannoni, C., Lintuvuori, J. S., Wilson, M. R.
J. Chem. Phys. 131, 174107 (2009); DOI:10.1063/1.3254019

so if tests on linked GBs is successful with lammps,
I guess I could have a go at trying some of these
other potentials

thanks once again for your help

best wishes,
Mark

6

J. Chem. Phys. 132, 035105 (2010); doi:10.1063/1.3269994

might be of interest to you.

Gay-Berne is ~15 times the cost per particle pair vs lj (see: http://lammps.sandia.gov/bench.html )

With GPU acceleration, the cost per particle pair is much more competitive with accelerated LJ because of the higher arithmetic intensity. (see: http://users.nccs.gov/~wb8/gpu/kid_single.htm ) If you have a machine with enough GPU nodes, can maybe change this to a factor of 2 difference - possibly even better if the bond/angle/etc calculation times exceed the GPU non-bonded times (run concurrently).

Depending on your simulation, aspect ratios, sizes, etc. distance of closest approach calculations can also be a concern when using ellipsoids vs spheres ( see the Perram et al. ref on the doc page).

- Mike

7

J. Chem. Phys. 132, 035105 (2010); doi:10.1063/1.3269994

might be of interest to you.

thanks Michael, it is indeed of interest; reading this article the users seem to have
defined an atom and a Gay-Berne as a rigid body and used bond or
bond angle potentials through the atom (at least if I read the paper correctly)

Gay-Berne is ~15 times the cost per particle pair vs lj (see: http://lammps.sandia.gov/bench.html )

I wasn't expecting such a cost differential, I wonder if its because the
code in lammps is for the full biaxial Gay-Berne & the pure unixial one
can be coded more cheaply? I guess I should try (I was currently struggling
a little on the lammps learning curve as the Gay-Berne benchmark
doesn't run with the newest code - but will hopefully get there)

With GPU acceleration, the cost per particle pair is much more competitive with accelerated LJ because of the higher arithmetic intensity. (see: http://users.nccs.gov/~wb8/gpu/kid_single.htm ) If you have a machine with enough GPU nodes, can maybe change this to a factor of 2 difference - possibly even better if the bond/angle/etc calculation times exceed the GPU non-bonded times (run concurrently).

that's immensely impressive!!!!!
fantastic

best wishes,
Mark