Dear all:
I know it was asked in an earlier post about TIP5P, but I was curious if anyone was currently working on trying to implement a five-point model in LAMMPS right now. One of my students is interested in working with TIP5P, so it would be good to know if anybody is currently developing such a model, or if there is sufficient interest to be worth doing it ourselves.
Thanks,
—AEI
Dear all:
I know it was asked in an earlier post about TIP5P, but I was curious if
anyone was currently working on trying to implement a five-point model in
LAMMPS right now. One of my students is interested in working with TIP5P, so
it would be good to know if anybody is currently developing such a model, or
if there is sufficient interest to be worth doing it ourselves.
you can easily set up TIP5P using fix rigid/small. either disable the
test for per type mass == 0 or use something small like 1.0e-10.
only downside is the rather short time step required to have stable
time integration with rigid bodies.
axel.
I guess the big question here is how small does that time step need to be? With TIPxP styles, we can use SHAKE to get away with a 2 fs time step. I’d rather not have to do 20 or 30 time steps to replace that; I doubt that would be very efficient in the long run.
—AEI
> Dear all:
>
> I know it was asked in an earlier post about TIP5P, but I was curious if
> anyone was currently working on trying to implement a five-point model
> in
> LAMMPS right now. One of my students is interested in working with
> TIP5P, so
> it would be good to know if anybody is currently developing such a
> model, or
> if there is sufficient interest to be worth doing it ourselves.you can easily set up TIP5P using fix rigid/small. either disable the
test for per type mass == 0 or use something small like 1.0e-10.
only downside is the rather short time step required to have stable
time integration with rigid bodies.I guess the big question here is how small does that time step need to be?
With TIPxP styles, we can use SHAKE to get away with a 2 fs time step. I'd
rather not have to do 20 or 30 time steps to replace that; I doubt that
would be very efficient in the long run.
how small would depend on your system and the kind of dynamics
happening. i would expect you have to use about 0.5fs. on a well
equilibrated bulk system you may get away with a larger time step.
that is still a significant cost compared with 2fs, but not as drastic
as you suspect it could be.
i view efficiency from the perspective of time to solution. these
days, it is often easier to get more CPU time, than to get your hands
on a suitably motivated student or postdoc that can program well
enough, not to count the time it will take to implement and test such
a new pair style compared to being able to start a production run
right away. while you can in principle follow the same approach as for
TIP4P, things get a little bit more complex for TIP5P, as you have to
not just displace one position, but have to do this twice, and if you
want to use long-range electrostatics, you need a matching kspace
style as well.
axel.
Hi Ahmed - I agree with the discussion points on this. If someone
created a pair_tip5p style that was specific to the TIP5P model
(allowed long timesteps, etc), it would be a nice addition to LAMMPS.
Steve