[gmx-users] Implicit solvent

[Justin Lemkul]

On 2/14/13 10:21 AM, Sebastien Cote wrote:
>
> Dear Justin,
> I am not sure to follow you. You essentially say that it is better to avoid using implicit solvent i.e. the generalized Born-formalism implemented in GROMACS?
> For the case of optimizing and relaxing a system (expecting short MD), I agree that it might be preferable to use explicit solvent.
> But for some systems, such as the oligomerization of a protein from a random state, explicit solvent can make this problem intractable in MD (inefficient sampling of conformational space due to too many degrees of freedom from the solvent) or REMD (too many processors to use due to too many degrees of freedom).
> I saw some tutorials and workshops suggesting larger cutoffs when using implicit solvent as the non-bonded function do tend to zero at long range. Then, domain decomposition can be used over particle decomposition.
> Could you clarify your point please? Especially, your statement saying that : "you can only run on 1-2 processors in implicit solvent".

There is an existing limitation in Gromacs, related to constraints, that causes 
the implicit solvent code to fail when run on CPU if you try to use more than 2 
processors.  It is my understanding that someone (probably Berk) is working on a 
better fix than what is in place now.  DD does not work with nonperiodic unit 
cells and simple grid searches; you have to use particle decomposition.  Therein 
lies another limitation.  I have found that only infinite cutoffs (i.e. 
rlist=rcoulomb=rvdw=0) lead to stable simulations, while any finite cutoff (even 
several nm in length) leads to poor energy conservation and unstable trajectories.

You can use GPU acceleration for implicit solvent simulations via OpenMM, but 
support for this feature is minimal, at best, at the present time.

Please do not take my comments as a criticism of implicit solvent methods; they 
are perfectly useful for a number of cases.

-Justin

-- 
========================================

Justin A. Lemkul, Ph.D.
Research Scientist
Department of Biochemistry
Virginia Tech
Blacksburg, VA
jalemkul[at]vt.edu | (540) 231-9080
http://www.bevanlab.biochem.vt.edu/Pages/Personal/justin

========================================