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

[X.Periole]

Dear Mark,

I basically rerun a simulation, both in xtc and trr 
format,
using slightly different tpr file, varying the gro file
used to construct it, and different gmx version, 3.2 and
3.3. From those rerun I extracted and compared the bonded
and non-bonded energies.

The most stricking is that I do not see differences 
between
the two tpr files, only small differences between the two
gmx versions, and more significant changes when reruning
on trr and xtc files.

Those results fit with the already identified bug in 
gmx3.2
with rerun fixed for gmx3.3. The only problem is the lack 
of
differences between the two gmx versions, and the fact 
that
xtc and trr file give such differences from the "correct"
energies extracted from the original energy file.

Best
XAvier

>>
>> The energies indicate that:
>> 0) I see a systematic difference with the original 
>>energies (not 
>> significant), but using
>> the trr file helps getting better.
>> 1) I do not see ANY difference between the rerun using 
>>the two tpr file 
>> (Unexpected ?!?!)
>> 2) I do not see significant differences between gmx3.2 
>>and gmx3.3 (New 
>> ?!?!) in all cases.
>> 3) I see significant differences in energies when 
>>generated with xtc and 
>> trr files (New ?!?!).
>> trr files help.
>> 
>> All this seems in agreement with the past remarks of 
>>David and Mickael 
>> until I compare
>> the energies obtained with the twp tpr files ?! Am I 
>>missing something ?
>> Note that my system is quite big (in cubic (!) box about 
>>12.0, 12.0, 
>> 8.0, in nm).
>> 
>> Does this make sense to any of you ?
> 
> I'm having difficulty understanding what you've done, 
>but some things that might help you understand what 
>you've done are
> 
> 1) The .trr and .xtc versions of the trajectories will 
>correspond to slightly different positions and so you 
>should expect slightly different energies to result
> 2) You need to use .edr files with grompp/tpbconv for 
>exact restarts to be exact
> 3) Different versions of the mdrun code should be 
>expected to be slightly different. A default value of a 
>parameter may have changed, of course, but in general the 
>execution order of the two versions need not be the same. 
>If you add up a few thousand floating point numbers in a 
>slightly different order you will get different results. 
>Evaluation of LJ and Coulomb terms requires this.
> 4) Different versions may produce identical .tpr files, 
>or might not. You can check this with gmxdump.
> 
> Mark
> _______________________________________________
> gmx-users mailing list    gmx-users at gromacs.org
> http://www.gromacs.org/mailman/listinfo/gmx-users
> 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