[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [ESPResSo-users] Lattice Boltzmann boundaries in Espresso 3.1

From: Stefan Kesselheim
Subject: Re: [ESPResSo-users] Lattice Boltzmann boundaries in Espresso 3.1
Date: Tue, 13 Mar 2012 18:18:31 +0100

Hi everybody,

Am 13.03.2012 um 17:18 schrieb Ulf Schiller:

> Hi,
> I don't intend to create any confusion here, but the fact that the lattice 
> nodes now lie at half integer positions means that any boundary aligned with 
> the main lattice directions has effectively an integer position. So the 
> implementation should now be able to generate stick boundary conditions at 
> positions 0.0 and 32.0 in Wolfgang's example. The shift of the lattice nodes 
> has the rather nice feature that the half-grid offset of boundaries can be 
> hidden under the hood and, in my opinion, the lbboundary command should 
> produce boundaries at integer positions as specified by the user.

The lbboundary wall now produces boundaries and integer positions no matter 
what you put int. This is probably more desirable than haven BCs at 
half-integer positions :-).

However we did not notice that the UG was not corrected. We'll add that to the 
next release of \es and ASAP to the UG. 

> Interpolating the velocity field between fluid and wall nodes does not make 
> sense, since the velocity at a bounce-back node is ill-defined. To linear 
> order, it could be extrapolated to -u where u is the velocity of the fluid 
> node closest to the wall. This would generate a smooth extrapolation for 
> Couette flow but not for Poiseuille. The location of the boundary at 
> half-grid spacing holds for both though.
> So I'd suggest to revisit the implementation such that in the example the 
> wall nodes lie at y=-0.5 and y=32.5, and the zero-velocity is located 
> effectively at y=0.0 and y=32.0. If the user specifies a non-integer 
> position, there is a choice to put the wall location to ceil() or floor(), 
> possibly depending on the orientation of the wall. In this case, 
> multi-reflection BCs would be better anyway.

There was a slight confusion about that: What the particles feel in ES 3.1 is 
exactly what you wrote: The velocity is interpolated linearly between fluid 
nodes and the boundary to the actual boundary velocity. 
Unfortunately a preliminary version of lbfluid print_interpolated_velocity went 
into the release, that uses less meaningful standard interpolation method also 
next to walls.
We are fixing that also ASAP and it will go into ESP 3.1.1. (Hopefully I'll 
have time tomorrow.).
The GPU version still uses only the standard interpolation method also for 
particle coupling next to the wall. Dominic, we should improve that, too :-).


Stefan Kesselheim
Institute for Computational Physics
University of Stuttgart

reply via email to

[Prev in Thread] Current Thread [Next in Thread]