[gmx-users] NVE simulations

[Sidney Elmer]

Here is a summary in regards to the problems that I experienced with energy
drift in the microcanonical ensemble.  The problem was due to SHAKE.  When
I reduced the time step down to 1fs, the total energy of the system
*increased*.  So now the behavior of my system in the microcanonical
ensemble turned 180 degrees.  I assume from this that it is not a good idea
to run SHAKE at all in the microcanonical ensemble, as heat may be
transferred unexpectedly.  At least when it is coupled to a heat bath, the
problem goes away, since you want heat transfer anyway.  The dummy atom
method suggested by Anton is a better approach to increasing the time step
in NVE simulations.

Finally, when I implemented a flexible model of CCL4 (no constraints at
all), there was no energy drift as expected.

Sid


Anton Feenstra wrote:

> Hi Sidney & David,
>
> If I understood Sydney correctly, he is currently only interested
> in identifying the maximum timestep that he can push in his system.
>
> This is the same as I and Berk Hess have done in one of my papers:
>
> K. Anton Feenstra, Berk Hess and Herman J. C. Berendsen. Improving
> Efficiency
>   of Large Time-scale Molecular Dynamics Simulations of Hydrogen-rich
>   Systems. (1999) J. Comput. Chem. 20 (8), 786-798
> http://www.chem.vu.nl/staf/feenstra/articles/jcc-20-8-789-798-(1999).pdf
>
> To determine this maximum time step, you indeed want to monitor energy
> conservation. But that also means you need to eliminate any other
> possible sources of energy drift. The worst of these is cut-offs, but
> there are several others. If I am not mistaken (if so, someone should
> correct me), even when using Ewald, there are residual cut-off like
> effects on the boundary of explicit and long range interactions.
>
> To my knowledge the best method of elminitating all cut-off related
> effects is to use a so-called shifted potential, which decays to zero
> before the cut-off distance. While this potential introduces other
> artefacts, these are not relevant in this case, and it is exactly for
> this purpose that there is a shifted potential still implemented in
> Gromacs.  This also is the method we used in the paper.
>
> Another energy-drifting source can be the use of constraints (Shake
> sometimes cools the system), but these you cannot readily eliminate
> if you want to push to large time steps. Although I have not actually
> tested this, I would expect Shake-related drift to become worse with
> longer time steps. Generally speaking, the Lincs constraints algorithm
> will be more stable, and definitely is faster, but it is unable to
> handle the highly connected network of constraints in CCl4.
>
> If you really want to push the limit, it would be a good idea to
> make a combined dummy mass/dummy atom construction, similar to that
> concieved for -NH3 groups in the paper, that would allow Lincs to
> be used.
>
> Lastly, as detailed in one of AK Mazur's papers that I cite,
> energy drift is diffusive. To estimate it properly, you need a
> fair number of independent runs and average the individual drifts.
>
> --
> Groetjes,
>
> Anton
>  ________ ___________________________________________________________
> |        | Anton Feenstra                                            |
> | .      | Dept. of Pharmacochemistry - Free University Amsterdam    |
> | |----  | De Boelelaan 1083 - 1081 HV Amsterdam - The Netherlands   |
> | |----  | Tel: +31 20 44 47608 - Fax: +31 20 44 47610               |
> | ' __   | Feenstra at chem.vu.nl - http://www.chem.vu.nl/afdelingen/FAR|
> |  /  \  |-----------------------------------------------------------|
> | (    ) | Dept. of Biophysical Chemistry - University of Groningen  |
> |  \__/  | Nijenborgh 4 - 9747 AG Groningen - The Netherlands        |
> |   __   | Tel +31 50 363 4327 - Fax +31 50 363 4800                 |
> |  /  \  | K.A.Feenstra at chem.rug.nl - http://md.chem.rug.nl/~anton   |
> | (    ) |-----------------------------------------------------------|
> |  \__/  | "We Have No Money" (E. Clementi)                          |
> |________|___________________________________________________________|
> _______________________________________________
> 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.