[gmx-users] mdrun -rerun with gmx3.2 and gmx3.3

[Xavier Periole]

David,

I'll try to explain a bit better and I attached a gziped pdf file 
containing a reduced set of energies
that I sent at my first post.

> I, too, am slightly confused about what exactly the differences are 
> you're seeing. Could you perhaps make a little table or something? 
> Partly, I just don't get what you mean about the difference between 
> the two versions... Are you taking trajectories you originally ran 
> with 3.2 and reprocessing them with 3.3? I think I had tried this at 
> one point and concluded that it wasn't a good idea since there are 
> some things the two versions do differently, so I shouldn't 
> necessarily get the same energies even using the same (full precision) 
> trajectory and tpr files.


The gmx3.2 and 3.3 version are the ones dowloaded for gromacs.org and 
installed without
special specifications. The gmx3.3 was compiled with the modified md.c, 
which includes
the correction on the rerun option !! The bug you identified !!

I do rerun trajectories generated by gmx3.2. The original 10ns 
trajectory has been cut down
to the first 200ps, and writen in xtc and trr format, from the 
corresponding original xtc and trr
trajectories.

The tpr files are generated with gmx3.2 including nstlist=1 and 
nstenergy=1. Two are generated,
one with the inital gro file and the second one with the final gro file 
(after 10ns). This is to
reproduce your observation that the energies are dependant on the gro 
file used to produce the
tpr file ... the box story.
Then I extract the energies from the ener.edr file generated with all 
combinaisons possibles of
the tpr, trajectories and gmx versions. They are given in the attached 
pdf file.

> The other issue is using mdrun -rerun with xtc files. Since these are 
> reduced precision, you should expect to get different energies even 
> reprocessing a 3.3 trajectory with 3.3 using the same tpr file, as 
> some of the energies are quite sensitive to distances. How much 
> difference you will see depends on the precision of your xtc files. I 
> *never* use xtc files for reprocessing for this reason.


The only *significant* differences I see are between the rerun on xtc 
and trr. Of course the precision
of the coordinates should introduce a difference. However I was not 
expected the differences to be
so important (up to 400 kJ/mol) even though they represent a small %  of 
the total energy. I would also
have expected a compensation of errors, which is not happenning.

Also I do not see any difference between the rerun using two different 
tpr files !! Neither
with  gmx3.2 nor gmx3.3-corrected. And that goes against the bug you 
found. You can imagine
that I checked the mdp, xtc, trr and tpr files (they differ), rerun 
again, and same result.
I am confused too !!

It is possible that all this is system dependent. My system is 
relatively big (>100K atom) in a
rectangular box, using RF for elect.

My conclusion so far is that trr file must be used for rerun (we learn 
every day, isn't it fabulous !)
and I do not see the bug you pointed out but probably due to the system 
I run !!

If you have more comment please ...
Best wishes
XAvier

-- 
----------------------------------
Xavier Periole - Ph.D.

Dept. of Biophysical Chemistry / MD Group   Univ. of Groningen
Nijenborgh 4
9747 AG Groningen
The Netherlands

Tel: +31-503634329
Fax: +31-503634398
email: x.periole at rug.nl
web-page: http://md.chem.rug.nl/~periole
----------------------------------
-------------- next part --------------
A non-text attachment was scrubbed...
Name: reduced.pdf.gz
Type: application/x-gzip
Size: 8404 bytes
Desc: not available
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20060322/90b176c1/attachment.gz>