[gmx-users] Unexpected results arising from T- and P-coupling methods
dommert at icp.uni-stuttgart.de
Tue Apr 12 15:37:36 CEST 2011
I have also realized similar behaviour:
NpT simulations performed with a Berendsen barostat give lower densities
than simulations coupled with P-R. However I think this is due to the
fact that I never achieve a pressure of 1bar (it is always around
1.2bar) with P-R even after a previous equilibration of 40ns. I am aware
that pressure is a strongly fluctuating quantity and this topic has been
discussed for several times on the list, but the equations of motion for
the box size given by the Lagrangian for P-R should finally give an
average pressure of 1bar even if the fluctuations are large. So what I
assume is, that the Leap-Frog integrator is not suitable for this
coupling scheme, BUT THIS IS JUST AN ASSUMPTION I DO NOT KNOW, and
unfortunately I had no time to try and validate this, yet.
On Tue, 2011-04-12 at 09:13 -0400, chris.neale at utoronto.ca wrote:
> I can't tell you if there is a problem or not.
> The only intended difference that I can see is that for 4.5.X:
> # grompp by default sets the new nsttcouple parameter equal to
> nstlist, this means T-coupling is done less frequently; grompp checks
> if tau_t is large enough
> # grompp by default sets the new nstpcouple parameter equal to
> nstlist, this means P-coupling is done less frequently; grompp checks
> if tau_p is large enough
> (see http://www.gromacs.org/About_Gromacs/Release_Notes/Versions_4.5.x )
> Note that there were some fixes to P-R scaling in the 4.0.X series --
> se http://www.gromacs.org/About_Gromacs/Release_Notes/Revisions_in_4.0
> To test this further, I suggest that you need to define a good
> procedure to ensure that what you are seeing is (b) from a well
> equilibrated system and is also (b) statistically significant.
> I suggest that one way to do this is to cycle through your T- and P-
> coupling options at a given temperature. For example, at T=500 K, run
> X ns of Berendsen, then X ns of Parrinello-Rahman, then back to
> Berendsen, and so on for a few cycles. Each time you switch coupling
> method, be sure to use the structure output from the previous run.
> Also be sure that X ns is as long as you can afford. This way, you
> will be able to pick out systematic changes as temperature/density
> oscillations with periods that are related to your changes of
> algorithm. This also ensures that what you are seeing is not simply an
> artifact of having a poorly equilibrated density in your initial
> You could also run the same cycle with Berendsen and V-rescale.
> -- original message --
> Dear gmxers,
> According to my recent practice, we find that the Berensen methods
> for T- and P- coupling can yield reasonable averaged density as a
> function of temperature, but when the v-rescale method and the
> Parrinello-Rahman method are employed for T- and P- coupling, somewhat
> unexpected results (i.e. density at higher temperature is bigger than
> that at lower tempearature) are obtained. Generally, the latter setup
> is considered to be prefered to the former one in simulating realistic
> ensemble. I am using gmx-4.5.3, and previously I have also performed
> one similar work using 4.0 which can generate expected results using
> the latter setup. I wonder if this version 4.5.3 has some bugs in
> calculating T and P, and are they dealt with in 4.5.4? Please give me
> some hints.
> Yours sincerely,
> Chaofu Wu, Dr.
> Department of Chemistry and Materials Science, Hunan University of
> Humanities, Science and Technology, Loudi 417000, the People?s
> Republic of China (P.R. China)
> -------------- next part --------------
> An HTML attachment was scrubbed...
Dipl. - Phys.
Institute for Computational Physics
EMail: dommert at icp.uni-stuttgart.de
Tel.: +49 - (0)711 - 68563613
Fax.: +49 - (0)711 - 68563658
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 198 bytes
Desc: This is a digitally signed message part
More information about the gromacs.org_gmx-users