[gmx-developers] [Fwd: your kind advice needed]
hessb at mpip-mainz.mpg.de
Thu Mar 2 09:15:01 CET 2006
Berk Hess wrote:
> -------- Original Message --------
> Subject: your kind advice needed
> Date: Wed, 1 Mar 2006 21:47:25 +0000
> From: Marie <ma7 at soton.ac.uk>
> To: hessb at mpip-mainz.mpg.de
> Dear Sir
> I am intending to implement Hybrid Monte Carlo in Gromacs which simply
> collects energies (kinetic and potential) and accepts/rejects new
> coordinates based on total (kinetic + potential) energy using Monte Carlo
> test. If the new positions are failed, the old positions are restored and
> the momenta is regenerated randomly using gaussian distribution.
> My question is how easy would it be for me to get into the Gromacs
> code. I
> am not worried about the coding but about the dead ends where its not
> possible to implement such scheme into one integrator (HMC) and call MD
> integrator within it to make MD moves and perform what I described
> above. I
> have had bad experience with NAMD as its doesnt allow abstraction and
> flexibilty the way its integrator works so before I start my work on
> Gromacs, I would be really grateful to you for your kind advice.
A long time ago I have implemented a hybrid brownian dynamics Monte Carlo
scheme (but this has never been distributed).
So it is possible and no too hard. But with regenerating velocities one
have to take care of several things, such as for instance constraints.
But what kind of ensemble would you produce by using a criterion
that uses the total energy, microcanonical?
Also for any MC method that uses the forces to produce moves,
the forces enter into the acceptance criterion.
Finally I don't see the use of an hybrid MD/MC method, as for most
systems this would not be more efficient than plain MD.
Even hybrid BD/MD on a LJ liquid does not give an improvement over plain BD.
More information about the gromacs.org_gmx-developers