[gmx-developers] More about the Ekin issue.

Xavier Periole x.periole at rug.nl
Thu Aug 11 16:32:36 CEST 2005


>>temperature average : 303.47 K
>>pressure average : 1687 bar
>>potential energy average difference : ~3500 KJ/mol     (between two 
>>different runs from same restart)
>Apart from the problems rerunning that you just reported, the energy
>difference should be related to the system size, e.g divide by the
>number of molecules, since you have ~25000 water molecules the
>difference is ~ 0.14 kJ/mol, not extreme, but incorrect nevertheless.
>T and P are of course absolute.

I like to share another view of the problem stressed out by Michel,
and would really appreciate your comments as I think that it is a
serious problem. Let me explain.

I also observe a small difference on the temperature extracted
from ener.edr generated by gmx3.1 and by gmx-3.3b(cvs) after rerunning
the frames and tpr of gmx3.1.
gmx3.1 : T=300.5K; P= 0.6 bar             averaged over 100ns
gmx3.3b: T=303.7K; P= 17.bar            averaged over 100ns

This is not so bad ... The bad news is what follows.

I have different systems that I have run at different temperatures
and plotted the Potential Energy of the whole system in function of
the temperature. (This is a very important aspect of REMD simulations.
The overlap between the Epot distributions at two temperatures defines
the exchange ratio between the two temperatures). When I saw the
difference between the Epot reported by Michel I got worried ...
I use gmx 3.1 for this.

What I did was to run again (with the cvs code) the same systems at one
temperature extracted from the previous ensembles. The way I rerun
is simple: grompp a conformation with the same topol.top, and mdp.
So same systems in different gmx version. (NPT)

The idea is to see at which temperature the old Pot Ener surface
corresponds. You will find for two systems a plot of the Epot obtained
with gmx3.1 and with the cvs code (gmx3.3b). The plots were too large to
go through the list. You can get them at :

small.jpg : http://md.chem.rug.nl/~periole/TMP/small.jpg
big.jpg : http://md.chem.rug.nl/~periole/TMP/big.jpg

Both systems are peptides in explicit solvent. small.jpg is an histogram 
of the Epot, whereas the big.jpg is just the plot in time of Epot. For
both systems the difference in Epot is about 2000 to 2500kJ/mol; divided
by the number of solvent molecules we get to relative small values. The
most important is that the difference in Epot from the gmx3.1 and
gmx3.3b corresponds to a change of about 30 K for the smaller system 
(3000 atoms) and 15 K for the bigger system (12000 at). This is not
negligeable !?

I do not really understand the reason for those huge differences !!!
The good thing is that for really big systems the error might be
not extreme !!

The main contributor to the potential energy and the temperature is
the solvent, thus I can approximate this observation by saying that
for the small system the solvent at 270K was exploring the potential
energy surface equivalent to 300K. Thus the peptide is interacting
with a solvent at 300K and not 270K. You can imagine that this has a
strong effect on its folding properties !!! This difference allows
my data (generated with gmx3.1) to fit the data produced by GROMOS,
which was impossible till now ... for a mysterious reason.

My concern is that the calculated Temp is certainly different but the
effect on the potential energy surface that the system is exploring is
much more than just the value of the temperature.

That is affecting all simulations run using gmx 3.1 and 3.2 ???


  Xavier Periole - Ph.D.

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

  Tel: +31-503634329
  Fax: +31-503634800
  email: x.periole at rug.nl
  web-page: http://md.chem.rug.nl/~periole

More information about the gromacs.org_gmx-developers mailing list