[ESPResSo-devel] Re: constraints

[Ulf Schiller]

Hi,

I moved the thread to the developers list because I think constraints have to be
considered still being under development and at the moment they can only be
used safely if one knows quite some detail about implementation details of
ESPResSo.

> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: RIPEMD160
>
> Hi!
>
> If you sent a copy of your email to the ESPResSo user's mailing list,
> please write it in English!
>
> Nils Binz wrote:
> > Ich wollte mal nachfragen, aus welchem Grund du in der
"constraint_collision"
> > in polymer.c die folded-positions der Teilchen ueberpruefst, wir hatten hier
> > naemlich folgendes Problem:
> > Eine Wand (constraint) wird ganz an den Rand der Box gesetzt und nun wird
ein
> > Polymer aufgesetzt. Wenn das Polymer die Wand ueberschreiten wuerde,
> > ueberschreitet es gleichzeitig auch die Box, die folded-positions der beiden
> > kritischen Teilchen sind also auf der gleichen Seite der wall obwohl sie die
> > constraint verletzen.
>
> Probably it is useful to rehearse the discussion on the developers list
> on "Nsquare cellsystem and folded positions": The problem is, that the
> constraints so far used the "internally folded" coordinates to check
> whether a constraint was violated or not.  The internally folded
> coordinates depend on the used cellsystem.
>
> Consequently, the different constraints worked differently for different
> cellsystem, which definitely can not be a good solution.
> In your case of a wall on the border of the system, this had the effect
> that the constraints worked in the way you described: when a particle
> walks over the wall, the constraint was violated. This was, however,
> more or less a bug. The system should behave periodically: when a

One can also consider this a feature: when a particle walks over a wall, the
constraint is violated, at least if it is supposed to be impermeable. In that
case, I would like to have at least a warning message issued. My interpretation
is that exactly for this reason walls have been implemented as one sided.

> particle is moved over the limits of the box, it is folded, and the
> coordinates have to be evaluated according to the folded position, and
> not according to the absolute position - and even less according to the
> internally folded position.
>
> Also, think of the strange behaviour of your system: when you put a wall
> directly on the border of the system, particles can get pretty close to
> the border, which means that they effectively interact with the
> particles close to the border on the other side, i.e. you get periodic
> interactions!

I agree. The problem is that a wall that lies directly on the border of a
periodic system makes the system somewhat ill-defined. Theoretically, one might
even want to have interactions between particles on different sides of a wall,
maybe for semi-permeable constraints or so. At least, I would not completely
exclude that possibility.

> I think the correct solution would be not to put the wall on the border
> of the system, but to leave some safety margin to the border to catch
> the particles that jump over the wall.

That's similar to what I do in my lattice Boltzmann simulations of Poiseuille
flow using two walls (although I don't have any MD particles in the fluid yet).
This is however something that is done on the TCL level. From my point of view,
this reflects the fact that constraints can have a lot of side effects which
are difficult to cope with uniquely on the implementation level.

Concerning the original problem: Would it help to add an option to the
constraints which specifies what kind of position (folded or non-folded) should
be used to check constraint violation?

Best wishes,
Ulf

--
Ulf D. Schiller   *   Room 1.404   *   Phone +49 6131 379-481
Max Planck Institute for Polymer Research
Theory Group
D-55128 Mainz, Germany


----------------------------------------------------------------
This message was sent using IMP, the Internet Messaging Program.