[gmx-users] -multidir option in REMD (GROMACS v2016.1) -resolved

ABEL Stephane Stephane.ABEL at cea.fr
Wed Aug 2 19:08:01 CEST 2017

THank you Mark and Johannes, 

I finally resolved my problem by using your solutions and can launch my REMD simulations 




Message: 1
Date: Wed, 02 Aug 2017 10:59:10 +0000
From: Mark Abraham <mark.j.abraham at gmail.com>
To: gmx-users at gromacs.org
Subject: Re: [gmx-users] -multidir option in REMD (GROMACS v2016.1)
        <CAMNuMAQeP0DEuO08vqJyVTnpbbd3RjAaWa2bD_czH377EEAc3w at mail.gmail.com>
Content-Type: text/plain; charset="UTF-8"


On Wed, Aug 2, 2017 at 11:22 AM ABEL Stephane <Stephane.ABEL at cea.fr> wrote:

> Dear all,
> I would like to do a REMD of a system with 75 replicas (with 75
> directories named md_0,..., md_75) and I have two questions.
> 1) To launch the md calculation we should use the followng command :
> mpirun -np 4 mdrun_mpi -v -multidir sim[0123] -replex XXX
> My question is since I have 75 replicas (directories). How I should set
> the argument of -multidir ? Like this : -multidir md_[01234...75] or only
> "md_".

The -multidir option needs a list of all your directories. The best way to
do that is with shell (e.g. wildcard) expansion. Just using "md_" doesn't
get expanded by anything. I recently observed someone suggest that
"md_{0..75}" might do what you want, but I've never tried it. Anyway, md_*
will expand them all, and IIRC mdrun will sort out the ordering from the
values of the REMD control parameter.

> I have used the following command : mdrun_mpi -v -multidir
> ./"$Stage1_FileName"/md_ -replex 500, but the program stops with the
> following error : Need at least two replicas for replica exchange (option
> -multi)

Since there's only one directory named, because the shell couldn't expand
anything for you, mdrun doesn't know how to proceed. Pro tip - get it
working for a small number of replicas (e.g. 2) before worrying about using
all 75.

> 2) Second question It is not clear to me how I should set the number of
> step in the mdp file of each replica use in the production runs.  For
> instance I want to do a simulation of 50 ns, should I set nsteps =
> 5000000/75

Each simulation is independent and every e.g. mdp parameter is particular
to it. So choose your number of steps in each mdp file to be the length of
a single replica. Obviously nsteps * dt = time in the usual way, but since
we don't know dt, you'll have to compute your nsteps ;-)


Could you help me ?

More information about the gromacs.org_gmx-users mailing list