[gmx-users] Domain Decomposision and Frozen Groups

Fri Jan 23 18:21:32 CET 2009

Actually I thought that if I don't exclude interactions, it would make mdrun crash. I read in the mailing list that for using freeze option, I have to exclude interactions, use comm_mode=none and use NVT conditions (although I tried NPT too and it worked). Maybe I was missing something.

Thank you Berk. I will do your suggestions.

________________________________
From: Berk Hess <gmx3 at hotmail.com>
To: Discussion list for GROMACS users <gmx-users at gromacs.org>
Sent: Friday, January 23, 2009 4:34:05 AM
Subject: RE: [gmx-users] Domain Decomposision and Frozen Groups

 Hi,

No, if you need to freeze two proteins you do not have to exclude interactions.
Excluding interactions will make you simulation slightly faster,
but freeze groups freeze a group, even if there are forces working on it
(if there were no forces working, you would not need freeze groups in the first place).

Berk

________________________________
Date: Thu, 22 Jan 2009 15:51:49 -0800
From: resal81 at yahoo.com
Subject: Re: [gmx-users] Domain Decomposision and Frozen Groups
To: gmx-users at gromacs.org

Thank you Berk!

Actually I need to freeze two proteins so I have to exclude interactions. I already tried to use PME and energygrps_excl together, but grompp used to give me this warning:
      Can not exclude the lattice Coulomb energy between energy groups
which is related to PME (I think it is necessary for me to stick with PME because I mainly want study electrostatic interactions). When I searched the mailng list, some people had suggested that if I want to use PME with frozen groups, I have to use [exclusions] in the topology file. So I did. Any hint for a better workaround? Because I am using a thermodynamic cycle, I am thinking of ignoring this warning...

I am trying to understand the domain decomposition method. Indeed I already encountered same error message for small peptides and I figured it out that if I make the box bigger, decomposition would work well. I will try bigger boxes for my proteins, but I have no clear idea about how much bigger the box should be...

Thanks

________________________________
From: Berk Hess <gmx3 at hotmail.com>
To: Discussion list for GROMACS users <gmx-users at gromacs.org>
Sent: Thursday, January 22, 2009 4:59:09 PM
Subject: RE: [gmx-users] Domain Decomposision and Frozen Groups

 Hi,

The exclusions in the topology introduce long range interactions
that require large domain decomposition cells.
The md.log of 4.0.3 will tell you this.
Remove all those exclusions and add energygrp exclusions in the mdp file.
(Or do not use exclusions at all)

Berk

________________________________
Date: Thu, 22 Jan 2009 12:37:24 -0800
From: resal81 at yahoo.com
To: gmx-users at gromacs.org
Subject: [gmx-users] Domain Decomposision and Frozen Groups

Hi everyone;

I am trying to do some simulation using frozen groups on an 8 core computer. I have tow boxes: 

1) Box 1 has two proteins and both of them are kept frozen (I used [exclusions] directive in topology file to exclude the interactions in each frozen group, NVT). 

2) Box 2 has those two proteins but without being frozen (no use of [exclusions] and freeze options). 

Here is the mdp file:
=========================================================
integrator     =     sd
dt     =      0.002
nsteps     =     50000
comm_mode = None
ns_type                  = grid
pbc                      = xyz
nstlist    = 10
rlist     =     1
coulombtype     =     PME
rcoulomb     =     1
fourierspacing     =     0.12
pme_order     =     6
ewald_rtol     =     1e-6
vdwtype     =     Switch
rvdw_switch     =     0.8
rvdw     =     0.9
DispCorr     =     EnerPres
nstxout     =     500
nstvout     =     500
nstlog     =     500
nstenergy     =     500
nstxtcout                = 5000
xtc-precision            = 1000
tc_grps                  = system
tau_t                    = 1
ref_t                    = 310
energygrps    = chainC chainF SOL  ; these three lines were commented out for the box 2 simulation
freezegrps    = chainC chainF
freezedim    = Y Y Y Y Y Y
=========================================================

I used this command to run the simulations:
mpiexec -n 8 mdrunmpi -s eqvspc.tpr -c eqvspc.gro -o eqvspc.trr -x eqvspc.xtc -g eqvspc.log -e eqvspc.edr

It works for the box two (no freeze), but for box 1 gives me this error:
=========================================================
Back Off! I just backed up eqvspc.log to ./#eqvspc.log.11#
Reading file eqvspc.tpr, VERSION 4.0.3 (single precision)
-------------------------------------------------------
Program mdrunmpi, VERSION 4.0.3
Source code file: domdec.c, line: 5858

Fatal error:
There is no domain decomposition for 8 nodes that is compatible with the given box and a minimum cell size of 6.25692 nm
Change the number of nodes or mdrun option -rdd or -dds
Look in the log file for details on the domain decomposition
-------------------------------------------------------
Error on node 0, will try to stop all the nodes
Halting parallel program mdrunmpi on CPU 0 out of 8
==========================================================

The serial version of mdrun works with box 1.
I searched the user mailing list and found that some people had similar problem with restraints in the past. I tried to use some -rdd and -dds values, but without any success. Can you please point out where is the source of this error and how I can get domain decomposition and frozen (or restraint) groups to work.

Any help or hint is highly appreciated!

Reza Salari

________________________________
Express yourself instantly with MSN Messenger! MSN Messenger

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20090123/db5f10aa/attachment.html>