[gmx-users] "Justifying" 4fs production runs after 1fs equilibrations?

[Mark Abraham]

On Tue, Feb 11, 2014 at 12:11 PM, unitALX <alec.zander at gmail.com> wrote:
>
> Helllo all!
>
> In my general situation, I have a batch of homology models that I would
like
> to assess for stability by molecular dynamics.
>
> I am working with a postdoc in my lab who was extensive experience with
> NAMD, but does not use GROMACS. We have been developing protocol for these
> MD simulations, but we had some different views on equilibration and
> production.
>
> Equlibration
> My friend is recommending that I do equilibration for 500ps @ 1fs with no
> constraints. I have no real objection to that, but he made it seem like
the
> results might be quite different than an equilibration for 200ps @ 2fs
with
> LINCS all-bonds. Would it make a critical difference?

Highly unlikely. The point of equilibration is to move from some point near
the ensemble, to some point in the ensemble, perhaps without perturbing the
starting structure significantly. The last phase of equilibration should
match the production ensemble (or simply be subtracted from the production
run), but this is not likely to be critical, either.

>
> Production
> I was excited by the performance possible with LINCS all-bonds, -vsite
> aromatics @ 4fs, but my friend looks at 4fs with disgust. I responded with
> arguments from the GROMACS 4 paper, and a paper called "Improving
efficiency
> of large time-scale molecular dynamics simulations of hydrogen-rich
systems"
> re: the relationship between the period of the fastest vibration and
largest
> allowable timestep, but he seems quite convinced that for publication,
only
> 1fs and possibly 2fs production runs (preferably without barostat and
> thermostat??) are acceptable.

Historically, people used thermostats to cover up inadequacies in the model
physics (e.g. no long-ranged electrostatics), but it was bad practice. If
you can show energy conservation in the NVE version of the NPT/NVT/whatever
ensemble you propose, then most of the issues are covered. IIRC there's a
conserved quantity for these other ensembles too, but I'm not sure GROMACS
reports it in all cases. Using NVE, 1fs (and maybe double precision)
because historically that was the only observably rigorous ensemble
implementation in code X is not a relevant argument now.

> The suggestion to drop the barostat and
> thermostat for production follows the philosophy that if your minimization
> and equilibration is properly done, then the system should be stable in
> production without pressure & temperature control.

Stable, yes. But stability is more of a precondition these days, not an
objective ;-) In particular, if the conserved quantity gets conserved, then
you can be pretty sure of stability. Sampling the right ensemble is the
objective. The larger the system, the less the ensemble details matter.

> I'm having a hard time
> finding better literature "justifications" for usage of LINCS all-bonds @
> 2fs, and certainly for -vsites @ 4fs; P&T control seems quite standard in
> publications using GROMACS.  Can you suggest any papers or calculations I
> can do to show the validity of these options?

As Thomas has suggested, your simulation observables and their convergence
are the relevant things to consider. Supposing that there are minima of
relevant interactions "seen" at 1fs that are not seen at 4fs is all very
well, but worth testing rather than assuming. Why are there not relevant
minima seen at 0.5fs? ;-) The timescales of the system are relevant, and
only experience and measurement can settle that. I would argue that "I've
always done it with 1fs and been happy" is not a useful argument against
conservative 4fs, unless there's also evidence of femto- second scale
kinetics that matter.

Also, if such a "missed" configuration is interesting, it will be seen if
the ensemble converges, by definition. Worrying about it seems to give a
lot of (undue?) weight to the order in which states are visited, and by
implication the starting configuration... But since you are sampling in the
unconverged limit of N time steps, one subset is not necessarily any better
than another. It seems likely to me that a longer time step is more likely
to access new areas of phase space when in the unconverged N-step limit;
this is one reason for coarse-graining, of course.

Mark

>
>
> --
> View this message in context:
http://gromacs.5086.x6.nabble.com/Justifying-4fs-production-runs-after-1fs-equilibrations-tp5014461.html
> Sent from the GROMACS Users Forum mailing list archive at Nabble.com.
> --
> 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.
 Helllo all!

In my general situation, I have a batch of homology models that I would like
to assess for stability by molecular dynamics.

I am working with a postdoc in my lab who was extensive experience with
NAMD, but does not use GROMACS. We have been developing protocol for these
MD simulations, but we had some different views on equilibration and
production.

Equlibration
My friend is recommending that I do equilibration for 500ps @ 1fs with no
constraints. I have no real objection to that, but he made it seem like the
results might be quite different than an equilibration for 200ps @ 2fs with
LINCS all-bonds. Would it make a critical difference?

Production
I was excited by the performance possible with LINCS all-bonds, -vsite
aromatics @ 4fs, but my friend looks at 4fs with disgust. I responded with
arguments from the GROMACS 4 paper, and a paper called "Improving efficiency
of large time-scale molecular dynamics simulations of hydrogen-rich systems"
re: the relationship between the period of the fastest vibration and largest
allowable timestep, but he seems quite convinced that for publication, only
1fs and possibly 2fs production runs (preferably without barostat and
thermostat??) are acceptable. The suggestion to drop the barostat and
thermostat for production follows the philosophy that if your minimization
and equilibration is properly done, then the system should be stable in
production without pressure & temperature control. I'm having a hard time
finding better literature "justifications" for usage of LINCS all-bonds @
2fs, and certainly for -vsites @ 4fs; P&T control seems quite standard in
publications using GROMACS.  Can you suggest any papers or calculations I
can do to show the validity of these options?

--
View this message in context:
http://gromacs.5086.x6.nabble.com/Justifying-4fs-production-runs-after-1fs-equilibrations-tp5014461.html
Sent from the GROMACS Users Forum mailing list archive at Nabble.com.
--
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.