On Fri, Dec 16, 2016 at 3:35 AM, Jason Swails <jason.swails.gmail.com>
wrote:
> On Thu, Dec 15, 2016 at 3:07 AM, Giovanni Bussi <giovanni.bussi.gmail.com>
> wrote:
>
> > Hi Chris,
> >
> > a few clarifications from the plumed side, which might be useful for
> other
> > people reading this list:
> >
> > 1. Plumed is compatible with amber16. We only have patches up to amber14
> > because starting with amber15 the support for plumed is included directly
> > in amber.
> > 2. Plumed only works with sander, and not with pmemd.
> > 3. Support for Hamiltonian replica exchange shipped with plumed is only
> for
> > gromacs. Actually, the changes required for this to work are in gromacs,
> > and not in plumed. So, even with sander+plumed, you cannot use
> Hamiltonian
> > replica exchange with sander+plumed. Even worst: sander+plumed does not
> > support any form of replica exchange.
> >
>
> Are you sure about this? The way Hamiltonian REMD is implemented, the
> coordinates are communicated between replicas, not the colvars. This makes
> replica exchange attempts more expensive (since you have to compute the
> whole potential energy instead of just the colvar), but that also makes it
> much more flexible (i.e., H-REMD will typically work without having to
> write code for different kinds of Hamiltonian modifications).
>
> I *believe* H-REMD with sander + plumed will work, but that should be
> validated first.
>
I see. If PLUMED is called again during the exchange to compute the energy
with the trial coordinates from the neighboring replica, and if the energy
computed by PLUMED is added to the force field energy to compute the
acceptance, than it will work when PLUMED has static potentials only (as in
replica-exchange umbrella sampling). This should be validated as you said.
To make it work with time dependent potentials (e.g. to allow parallel
tempering metadynamics) one should make sure that metadynamics potential,
which is history dependent, is not updated during the trial. This should be
a trivial fix, so if someone is interested and willing to test it I can
help implementing it.
Giovanni
>
>
>
> > 4. Anyway, sander and pmemd alone (without plumed)* I think* support
> > Hamiltonian replica exchange (I know people using it).
> >
>
> Yes, sander, pmemd, and pmemd.cuda all support H-REMD (but only at
> constant volume).
>
> HTH,
> Jason
>
> --
> Jason M. Swails
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Dec 16 2016 - 00:30:02 PST
