[gmx-users] R: Re: R: Re: User tabulated potentials with cutoff-scheme = verlet, GROMACS 5.0.1 and 5.0.2

Thu Oct 2 15:52:14 CEST 2014

On Thu, Oct 2, 2014 at 3:24 PM, Francesco Mambretti <
francesco.mambretti at studenti.unimi.it> wrote:

> Ok thank you. Another question. 108 particles was just a test for
> beginning, but we would like to study much bigger systems (from 1000 to
> 10000 particles at least). Is it possible and reasonable with the group
> scheme? Does it support parallel computing?
>

Sure, for the last 15 years. :-) The group scheme has only MPI support (no
OpenMP, no CUDA), so in practice you're limited to something like at least
500 atoms per rank(=core). YMMV because we don't study the performance on
such systems.

Mark

> Francesco
>
> Il 02/10/14 13:43, Mark Abraham  <mark.j.abraham at gmail.com> ha scritto:
> >
> > On Thu, Oct 2, 2014 at 1:13 PM, Francesco Mambretti <
> > francesco.mambretti at studenti.unimi.it> wrote:
> >
> > > So then, you have to update the pages on your website (
> > >
> http://www.gromacs.org/Documentation/How-tos/Tabulated_Potentials?highlight=tabulated+potential
> )
> > > about tabulated potentials,
> >
> >
> > Indeed, I have fixed that. Unfortunately, much of the Gromacs
> documentation
> > precedes the existence of the Verlet scheme, and we have not made a
> > systematic attempt to update all of it.
> >
> >
> > > and also in the user manual (5.0.1 version).
> > >
> >
> > Noted
> >
> >
> > > Otherwise, people risks misleading about this topic!
> > > In fact, I spent my last month in learning GROMACS, encouraged by the
> > > documentation about tabulated potentials (it seems so easy to do
> that!),
> > > and nothing is said about the cutoff scheme you have to use! And also
> the
> > > attached pdf should be updated.
> > >
> >
> > That would be up to the user who contributed it, unfortunately.
> >
> > This is a very strong limit, LJ and Buckingham potentials are often not
> so
> > > accurate potentials, if you really want to simulate physical systems.
> > >
> > > So the alternatives I have are:
> > > 1) use group charge (e.g.: I have 108 atoms of mixed Ar and Kr, and
> each
> > > of them is a group charge) and group cutoff - but the run will require
> more
> > > time
> > >
> >
> > I'd expect that a tiny VDW-only system like that will show little to no
> > performance difference between the cutoff schemes (assuming the group
> > scheme setup is buffered), because both make about the same use of SIMD
> > intrinsics and the effect spent in optimizing for your kind of systems is
> > about the same in both. Once tables work in Verlet, I'd expect even less
> > difference between the two (but GPU support will be great for larger
> > systems).
> >
> > 2) downgrade my GROMACS version, so that I am not forced to use verlet
> > > cutoff if I want a configuration that
> >
> > scales fast
> > >
> >
> > You're not forced to use Verlet now ;-) The default scheme changed in 5,
> > but the group scheme is still there (and works better than ever). There's
> > zero value in using an earlier version.
> >
> > Mark
> >
> > 3) use LJ potential
> > >
> > > Am I right?
> > >
> > > Francesco
> > >
> > > Il 02/10/14, Mark Abraham <mark.j.abraham at gmail.com> ha scritto:
> > > > On Thu, Oct 2, 2014 at 11:58 AM, Francesco Mambretti <
> > > > francesco.mambretti at studenti.unimi.it> wrote:
> > > >
> > > > > Dear all, I am a student in Physics at Milan University. For my
> > > thesis, I
> > > > > should run a simulation of a binary liquid mixture of Argon and
> > > Krypton.
> > > > >
> > > > > Until now, I have been learning how to use GROMACS (version 5.0.1
> and,
> > > > > since
> > > > > yesterady, 5.0.2) essentials, and my simulations were all based on
> LJ
> > > > > potential, and all was fine.
> > > > >
> > > > > Now I want to use a tabulated potential, that is a more detailed
> > > potential,
> > > > > different in form from LJ, taken from literature.
> > > > > I followed the steps of the instructions on GROMACS website
> > > > > (
> > > > >
> > >
> http://www.gromacs.org/Documentation/How-tos/Tabulated_Potentials?highlight=tabulated+potential
> > > > > )
> > > > > and the attached pdf, but it doesn't work.
> > > > > I ought to use cutoff-scheme = verlet (because the "group" scheme
> does
> > > not
> > > > > make sense, because my system has no charge),
> > > >
> > > >
> > > > VDW-only systems work fine in the group scheme, e.g. with a single
> atom
> > > per
> > > > charge group - but you will need to organize your own details for
> > > adequate
> > > > buffering.
> > > >
> > > >
> > > > > but I receive these errors
> > > > > from grompp:
> > > > >
> > > > > ERROR 1 [file Aziz.mdp]:
> > > > > With Verlet lists only cut-off and PME LJ interactions are
> supported
> > > > >
> > > > >
> > > > > ERROR 2 [file Aziz.mdp]:
> > > > > With Verlet lists only cut-off, reaction-field, PME and Ewald
> > > > > electrostatics are supported
> > > > >
> > > > > Aziz.mdp is my mdp file, that contains the instructions
> coulombtype =
> > > User
> > > > > and vdwtype = User.
> > > > >
> > > > > What am I missing?
> > > > > Or it is true that GROMACS is not able to use user-defined
> potentials
> > > with
> > > > > verlet's cutoff? This would be a serious handicap for this code!
> > > > > Am I forced to use charge groups?
> > > > >
> > > >
> > > > For now, yes. There is work in progress to implement user tables for
> the
> > > > Verlet scheme (& GPUs). Some help with alpha-stage testing would be
> very
> > > > welcome! But for that, you might like to have a working group-scheme
> > > setup
> > > > against which to validate.
> > > >
> > > > Mark
> > > >
> > > >
> > > > > Thanks in advance,
> > > > >
> > > > > Francesco Mambretti - University of Milan - LCP (Parallel Computing
> > > Lab)
> > > > >
> > > > > --
> > > > > Gromacs Users mailing list
> > > > >
> > > > > * Please search the archive at
> > > > > http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before
> > > > > posting!
> > > > >
> > > > > * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
> > > > >
> > > > > * For (un)subscribe requests visit
> > > > > https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users
> or
> > > > > send a mail to gmx-users-request at gromacs.org.
> > > > >
> > > > --
> > > > Gromacs Users mailing list
> > > >
> > > > * Please search the archive at
> > > http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before
> > > posting!
> > > >
> > > > * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
> > > >
> > > > * For (un)subscribe requests visit
> > > > https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users
> or
> > > send a mail to gmx-users-request at gromacs.org.
> > > >
> > > >
> > > --
> > > Gromacs Users mailing list
> > >
> > > * Please search the archive at
> > > http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before
> > > posting!
> > >
> > > * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
> > >
> > > * For (un)subscribe requests visit
> > > https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
> > > send a mail to gmx-users-request at gromacs.org.
> > >
> > --
> > Gromacs Users mailing list
> >
> > * Please search the archive at
> http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before
> posting!
> >
> > * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
> >
> > * For (un)subscribe requests visit
> > https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
> send a mail to gmx-users-request at gromacs.org.
> >
> --
> Gromacs Users mailing list
>
> * Please search the archive at
> http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before
> posting!
>
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>
> * For (un)subscribe requests visit
> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
> send a mail to gmx-users-request at gromacs.org.
>