[gmx-users] Re: mdrun -rerun and box size bug not fixed?

[David Mobley]

Hi,

> By reactivating the barostat as mentioned below, I managed to recover LJ
> energies within 0.1% for my test system, which was a box of 1136 waters.
> So I applied the same thing to my system of interest (a 605 residue
> protein with around 25000 waters). The only difference is that the run
> was done with PME, and the rerun is done with RF, so I cannot compare
> electrostatics. For the other energies I get a systematic bias with
> respect to the original run, for example :
>
>    - LJ :  70 +-50 kJ/mol higher than original  (0.05% of total)
>    - g96 bonds : 160 +-60 kJ/mol higher than original (2.44% of total)
>
> These biases are very reproducible over many trajectories with this
> system. The LJ bias might be considered negligible as it is a small
> percentage of the total LJ energy. However, the g96 bond bias is
> surprising (bonds are present only in the protein). We expect that the
> inaccuracy of the xtc coordinates generates some random noise. But I
> don't understand how it could generate a systematic error of that
> magnitude (2.44%) ...
>
> I tried to change different parameters to trace the origin of this bias
> :  version of gmx, platform, huge compressibility, turning off the
> thermostats (for the g96 bond part), etc... Only turning the constraints
> off influenced the results, yielding an even higher bias towards higher
> g96 energies (210 +-60 kJ/mol).
>
> Have such inaccuracies been observed before with -rerun? Am I doing
> something wrong or is there no hope of recovering better energies with a
> rerun?

We routinely do reprocessing with the *same* settings we run with
(this is in the NVT ensemble, usually) and recover energies that are
very close (i.e. what one would expect from the trajectory precision).
Note this is using full precision trajectories, though. See for
example http://dx.doi.org/10.1021/jp0735987.

So I guess what I'm saying is that in principle you should be able to
get good energies with rerun. Maybe you should try and narrow down the
problem by doing some testing:
(a) Do some reprocessing with PME rather than RF to see if that helps.
(b) Do some reprocessing on full precision trajectories rather than
xtcs to rule out precision issues
etc.

Best wishes,
David


> Thanks,
> Michel
>
>
> -------- Original Message --------
> Subject:        Re: mdrun -rerun and box size bug not fixed?
> Date:   Tue, 06 Nov 2007 14:12:29 +0100
> From:   Michel Cuendet <michel.cuendet at isb-sib.ch>
> To:     gmx-users at gromacs.org
> References:     <20071102152757.0FAFD2435D at xray.bmc.uu.se>
>
> Hi,
>
> I think I got it :-) Using gmx 3.3.1 with the same thermostat and
> barostat settings as the original run, I get the right LJ energies
> within 0.1%. I used tcoupl and pcoupl = "no" because I somehow
> remembered that thermostats could do weird things when rerunning (see
> the old thread :
> http://www.gromacs.org/pipermail/gmx-developers/2005-August/001290.html )
>
> But apparently the box is not reread form the trajectory if the pcoupl
> option is deactivated. It would be much safer for users if the
> appropriate options were automatically set when -rerun in invoked:
>
> nstlist = 1
> nstenergy = 1
> NO rescaling of velocities
> NO rescaling of box size
> Do read the box sizes from the trajectory
>
> Cheers,
>
> Michel
>
>
>
> ==========================================================
> Michel Cuendet, Ph.D
> Molecular Modeling Group
> Swiss Institute of Bioinformatics
> CH-1015 Lausanne, Switzerland
> www.isb-sib.ch/groups/Molecular_Modeling.htm
> ==========================================================
>
>
>
>
>
>
> --
>
> ==========================================================
> Michel Cuendet, Ph.D
> Molecular Modeling Group
> Swiss Institute of Bioinformatics
> CH-1015 Lausanne, Switzerland
> www.isb-sib.ch/groups/Molecular_Modeling.htm
> ==========================================================
>
>
> _______________________________________________
> gmx-users mailing list    gmx-users at gromacs.org
> http://www.gromacs.org/mailman/listinfo/gmx-users
> Please search the archive at http://www.gromacs.org/search before posting!
> Please don't post (un)subscribe requests to the list. Use the
> www interface or send it to gmx-users-request at gromacs.org.
> Can't post? Read http://www.gromacs.org/mailing_lists/users.php
>