[gmx-users] Minimal box vectors chosen by the pressure integration

[Mark Abraham]

On Tue, Nov 25, 2014 at 7:36 PM, Mirco Wahab <
mirco.wahab at chemie.tu-freiberg.de> wrote:

> On 25.11.2014 17:13, Mirco Wahab wrote:
>
>> ...
>> I'm somehow able to drive a 1000-water (mixed phase-)system,
>> starting from a ~50nm^3 box into a gas phase (0.001 bar,
>> resulting in a 230nm^3 box. This does work so far. When
>> increasing the applied pressure (Parrinello-Rahman) into
>> the liquid zone (eg. 0.2 bar), the system shrinks eventually
>> until a box size of 7.39492nm is reached. Then, the shrinking
>> stops, the box contains a water drop + vacuum.
>>
>> How does gromacs determine the minimal box size
>> that it  allows the pressure algorithm to induce?
>>
>
It shouldn't. Cutoff sizes are implementation artefacts that should only
affect the simulation through the minimum-image convention requiring a
minimum cell size.

Is this based on the (in this case) PME grid or
>> on the domain decomposition?
>>
>
> After inspecting the log-file, I have the suspicion it
> depends on th PME grid. The large box w/dilute vapor
> started with a PME grid after optimization:
>
> [md.log]
>  ...
>  step  650: the PME grid restriction limits the PME load balancing to a
> coulomb cut-off of 7.311
>  ...
>  step 1050: timed with pme grid 6 6 6, coulomb cutoff 7.311: 32.4 M-cycles
>                optimal pme grid 6 6 6, coulomb cutoff 7.311
>
>
> If I take a snapshot of this system (p=0.2bar, stuck at a=7.39nm)
> and *start it again* allowing gromacs to optimize a new PME grid,
> the system ends up after very short time in it's correct (liquid)
> phase (box size a = ~ 3.7 nm).
>
> Maybe running into a volume-change restriction would reward a warning
> in the log-file?


Any implementation-detail limit on rcoulomb should never feed back into
imposing a minimum cell size upon the simulation; if that's what's
happening, then it is a bug. What happens with mdrun -notunepme? mdrun with
a single thread and single MPI rank?

That said, mdrun with PME is not engineered for large changes in cell size,
because the usefulness of .mdp parameters like fourierspacing is tied to
the simulation volume, but the values are set by the .mdp (modulo
auto-tuning). You need more/less Fourier grid points for iso-accurate PME
with a larger/smaller volume. There'd be ways to make such stuff work, but
making it work fast and/or in parallel is more of a problem.

Mark


>
> Thanks & regards
>
> M.
>
>
> --
> Gromacs Users mailing list
>
> * Please search the archive at http://www.gromacs.org/
> Support/Mailing_Lists/GMX-Users_List before posting!
>
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>
> * For (un)subscribe requests visit
> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
> send a mail to gmx-users-request at gromacs.org.
>