[gmx-users] all-bonds constraints

[Mark Abraham]

On Thu, Apr 3, 2014 at 12:20 PM, Juan Munoz-Garcia <
juan.munoz-garcia at bioch.ox.ac.uk> wrote:

> Dear GROMACS users,
>
> I'm quite new with GROMACS. I come from using AMBER for a long time and
> after doing quite a few GROMACS tutorials I've found in all of them that
> the variable "constraints" is set to "all-bonds" even during the production
> run. I'm quite surprised about this. My understanding as I learnt from
> AMBER is that due to the high frequency of motion of the bonds with H atoms
> (aprox 1 fs timescale) those are fixed during the integration time step as
> this is longer (usually 2 fs) than the motion frequency. However, all other
> bonds are not fixed because their motions have larger timescales (of course
> this is a simplification) than the integration time step.


Chicken or egg? One should generally choose a model physics that is useful
for the problem at hand, and then pick a maximally large timestep that can
work with it. Modelling a bond with a Hookes' Law spring is an
approximation, and so is modelling it as rigid. Neither is a particularly
good description of the underlying QM physics. But if one approximation
permits a longer timestep, an improved sampling rate might be a sufficient
reason for a choice between poor approximations. Consensus for most
problems of biochemical interest is that the effects of H-bond vibrations
aren't of interest for the problems at hand; if you want to model bond
formation, you can't accept a model that has no "electron vibrations," and
*that* drives your choice of time step.

I find that constraining all bonds is too "artificial" as the  bond length
> is supposed to be oscillating between the equilibrium values given by the
> force field. As this is a quite important difference compared to what I was
> used to in AMBER, I'd like you to please tell me why is it "constraints =
> all-bonds" generally used during the production run instead of "constraints
> = h-bonds" ?
>

As Justin said, there are force-field parameterization issues here. Force
field parameters are (unfortunately) co-optimized with and dependent on
implementation issues like size of time step and lengths of cut-offs. The
use of constraints=all-bonds couples constraints across a whole polymer; if
that's a protein simulated using a spatial domain decomposition, those
atoms are widely distributed on the machine and the communication required
for the iterative constraint solver can become rate limiting at extreme
scale. Reducing the constraint iteration to a single node is unattractive
if the rest of the machine lies idle - which is part of why we'd like to
improve the ability to deploy task parallelism in GROMACS at scale. The
main alternative is using shorter time steps (maybe no constraints at all!)
- if they only way you can get 10x greater sampling rate for a single
simulation is by using 100x the hardware at one tenth of the time step,
then that could be a serious option.

I think that feeling that "constrained H bonds" is natural and "constrained
heavy-atom bonds" is artificial is wrongly conflating experience with
correctness. We're supposed to be scientists here - we should only be
satisfied with a model if can be shown to work at least as well as the
alternative. So if timestep studies like that reported in
http://pubs.acs.org/doi/abs/10.1021/ct900549r are not convincing for your
problem of interest, then you need to decide where the limit is - by
observing it, or accepting others' observations, and not by drinking their
Kool-Aid!

Mark


> Thank you.
>
> Juan C. Munoz-Garcia, PhD
> Biomembrane Structure Unit
> University of Oxford
> South Parks Road
> Oxford OX1 3QU
> juan.munoz-garcia at bioch.ox.ac.uk
>
>
>
> --
> 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.
>