[gmx-developers] Question on -rerun

[Shirts, Michael (mrs5pt)]

> as the temperature coupling should
> only have a minor effect within one step. I don't think a user would care
> about the last few decimals being different.

It's not necessarily that small.  I'm seeing up to 1-3 kJ/mol differences
for a relatively small system.

Also, things are a bit more complicated than originally described -- the
descriptions of what works and doesn't were for Nose-Hoover -- for
v-rescale, it's actually md that gives matched energies, and md-vv doesn't.
Presumably some order of operations issues that can be fixed for md-vv,
since the ordering -- I'll keep poking.

Best,
~~~~~~~~~~~~
Michael Shirts
Assistant Professor
Department of Chemical Engineering
University of Virginia
michael.shirts at virginia.edu
(434)-243-1821


> From: Berk Hess <hess at kth.se>
> Date: Mon, 31 Dec 2012 13:31:08 +0100
> To: "michael.shirts at virginia.edu" <michael.shirts at virginia.edu>, Discussion
> list for GROMACS development <gmx-developers at gromacs.org>
> Subject: Re: [gmx-developers] Question on -rerun
> 
> Hi,
> 
> My opinion is that all potential energies and forces should be reproduced.
> With velocities in the trr the velocities, kinetic energy, temperature,
> virial
> and pressure should not be far off, as the temperature coupling should
> only have a minor effect within one step. I don't think a user would care
> about the last few decimals being different.
> We can simply note in the manual that all velocity dependent quantities
> are not present (or unreliable) without velocities and can be slightly off
> by as much as the T-coupling effect in one step when v is present.
> 
> Cheers,
> 
> Berk
> 
> On 12/31/2012 05:59 AM, Shirts, Michael (mrs5pt) wrote:
>> Hi all-
>> 
>> I'm double checking free energy behavior under -rerun; more a
>> clarification/discussion  question than a bug, so I'm emailing the
>> developers list.
>> 
>> What's the official policy on what the -rerun flag can reproduce or not from
>> the original trajectory if both x and v are specificed in the trajectory?  I
>> poked around on the mailing list a bit, and couldn't get a consistent
>> answer. The manual says "forces and energies" I've talked a bit with Berk
>> about this, but that was a couple of years ago, and I'm not sure if
>> everything was really finalized.  It could be that this has all been
>> discussed before, but it's not really talked about in the manual.
>> 
>> So:
>> 
>> Things that should be definitely be reproduced (and appear to be, once I
>> check in a couple of bug fixes for md-vv)
>> 
>> * potential energies and forces
>> * velocities and temperature (any NVE integrator, also md-vv for NVT and
>> NPT)
>> * pressure and virial (any NVE integrator)
>> 
>> Things that are not clear are necessarily reproduced (and definitely are not
>> always reproduced in the current code)
>> 
>> * velocities and temperature (md for NPT and NVT)
>> * pressure, virial (any NPT or NVT integrator)
>> 
>> For example, currently, if NoseHoover thermostat is used with the md
>> integrator, then the kinetic energies will not be correct, as the velocity
>> rescaling history is lost in the -trr rerun.
>> 
>> There are two main issues:
>> 1) There are auxiliary integrator variables that are not saved in .trrr and
>> 2) for md, an integration step taken in between when the coordinates are
>> printed out and when the .edr file is generated, so any integrator auxiliary
>> variables are eliminated. I wonder if we should think about reorganization
>> of this in 5.0, though I'm not sure it's even possible, since for md, you
>> need both half-step velocities, and hence you have to take an integration
>> step.
>> 
>> So for 4.6, it seems like the best thing may be to put more warnings in the
>> manuals, as well as in the output -- right now it only warns about kinetic
>> energies, temperatures and pressures if there are velocities frames missing
>> from the trajectory.
>> 
>> Best,
>> ~~~~~~~~~~~~
>> Michael Shirts
>> Assistant Professor
>> Department of Chemical Engineering
>> University of Virginia
>> michael.shirts at virginia.edu
>> (434)-243-1821
>> 
>