[gmx-users] Question regarding freeze groups and their use

[Mark Abraham]

On 15/02/2012 5:09 AM, J. Nathan Scott wrote:
> Hi all,
>
> I know there has been a *lot* of discussion on the mailing list on
> using freezegrps and potential pitfalls, but after having read much of
> this discussion the past couple of days and the manual I still find
> myself with questions and problems I'm hoping someone can help with.
>
> What I want to do, specifically, is have my protein frozen (*not* just
> constrained with some large force, I need actual freezing of every
> protein atom) but have the solvent still react to the frozen protein.
> The purpose behind this is to compare the water behavior near a
> particular part of a protein when it is dynamic and when it is not and
> thereby learn something about the "communication" between protein and
> solvent in this case. However, it seems that I'm not finding a
> rational set of mdp options to allow this, at least not a set for
> which dynamics will run with any speed. I have tried many of the
> things people have discussed on the mailing list, including turning
> off constraints, turning off pressure coupling, reducing the Protein
> heat bath temperature to 0 K, and using energygrp_excl. Here are the
> parameters I just tried:
>
> ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
> integrator      = md
> nsteps          = 1000000
> dt              = 0.001
> nstxout         = 1
> nstvout         = 1
> nstxtcout       = 1
> nstenergy       = 1
> nstlog          = 1
> continuation    = yes
> constraints     = none

dt = 0.0005 is normally used when there are no constraints.

> ns_type         = grid
> nstlist         = 5
> rlist           = 1.0
> rcoulomb        = 1.0
> rvdw            = 1.0
> coulombtype     = PME
> pme_order       = 4
> fourierspacing  = 0.16
> tcoupl          = V-rescale
> tc-grps         = Protein Non-Protein
> tau_t           = 0.1   0.1
> ref_t           = 0   300
> pcoupl          = no
> pbc             = xyz
> DispCorr        = EnerPres
> gen_vel         = no
> freezegrps      = Protein
> freezedim       = Y Y Y
> ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
>
> This ran, but *incredibly* slowly (I'm using a single 32 processor
> node, and only got 255 steps in 10 wallclock minutes). From the
> discussions I've read on the mailing list it seems that PME is
> somewhat less than ideal for simulations involving frozen atoms, but I
> used PME for electrostatics in my non-frozen simulations and would
> like to keep that consistent if possible (I do get a grompp warning of
> course). I am also getting warnings in the log file like: "DD  step
> 254  vol min/aver 0.183! load imb.: force 2395.3%  pme mesh/force
> 1.243" which I believe must be due to the frozen protein causing
> difficulties with domain decomposition.
>
> Can anyone offer any advice regarding these issues? I know this has
> been discussed often, but nothing I'm finding in the archives is
> particularly relevant.

Allowing a run to complete and inspecting the timing breakdown at the 
end of the .log file might be instructive. Manual 7.3.24 suggests the 
use of energy group exclusions.

Mark