[gmx-users] Re:overcome the integration problem (installing CVS)

[Berk Hess]

>From: "Michael Shirts" <mrshirts at gmail.com>
>Reply-To: Discussion list for GROMACS users <gmx-users at gromacs.org>
>To: gmx-users at gromacs.org
>Subject: [gmx-users] Re:overcome the integration problem (installing CVS)
>Date: Wed, 11 Apr 2007 16:39:20 -0400
>
>>How about total energy? In tends to be constant in NVE simulations. I
>>don't know what the conserved property would be in an NVT simulation.
>>David.
>
>For most algorithms that give correct ensembles, there are
>pseudo-energy properties that are conserved.  For NVE, it's obviously
>the energy.  For Nose-Hoover NPT, there's a term involving the "mass"
>of the coupling to the heat reservoir that, when added to the scaled
>energy term produces a quantity that will be conserved.  For
>Parrinello-Rahman dynamics, there's a similar term involving the
>coupling to the pressure bath.
>
>>It would.
>>Although I would still argue that when using a thermostat there
>>is in most cases no "useful" quantity that would be affected
>>by the current, very small, velocity inaccuracies.
>>Berk
>
>I think algorithmic self-consistency could generally be justified as
>useful.  If one is applying a certain algorithm, it's good to have a
>measure of how well that algorithm is being implemented, so one can be
>sure that any errors are negligible.  I agree with Berk that such
>inaccuracies are small.  But it's good to have a measure of how small
>they are built in, just to be sure! :)
>
>Best,
>Michael

I have now programmed in the conserved quantity (not committed yet).
SPC water gives the same energy conservation in NVT Nose-Hoover as
in NVE.
I did a run of 100 ps for 216 SPC with dt=0.002 ps and SETTLE at 300 K,
tau_t=2 ps:
NVE: drift per degree of freedom -0.026 kJ/mol /ns
NVT: drift per degree of freedom -0.024 kJ/mol /ns
The difference seems to be within the noise.

That is reassuring.
Micheal, thanks for the suggestion!

Berk.

_________________________________________________________________
Live Search, for accurate results! http://www.live.nl