[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
RE: [ff3d-users] Periodic BCs
From: |
Stephane Del Pino |
Subject: |
RE: [ff3d-users] Periodic BCs |
Date: |
Thu, 6 Jan 2005 00:44:59 +0100 |
User-agent: |
KMail/1.7.1 |
Le Mercredi 5 Janvier 2005 17:33, Bharat Rangan a écrit :
> To start with a simple case, I want to solve the Navier Stioes in a box.(
> Something like flow through a rectangular channel) . I want to apply a
> periodic BC on the left and right faces so that whatever leaves the left
> face enters the right face and the otherway.
>
> Please give me a rough sketch as to how to implement this.
Ok. First, I need to implement the periodic boundary conditions. There are
several aspects:
- first, make it available in the ff3d language (this is in progress)
- second, implement the periodic conditions for linear operators (quite easy)
- third, implement it for convection operator (the characteristics method is
used to treat convection operator). This is not a big deal, but the link with
ff3d's language is not clear to me at the moment I am writting these lines: I
have a few ideas and have to find out which is the best.
Since I am not working at full time on ff3d, this might take some time (I hope
less than two weeks).
During this time, you can test the Navier-Stokes solver without periodic
conditions and give some feed back. Here are some technical details about the
solver:
- an example of an incompressible Navier-Stokes solver is provided with ff3d's
sources. This uses a projection method to preserve the divergence free of the
velocity.
- I have a version of a non projection algorithm. So if you are interessed ...
- ff3d's only uses Q1-Q1 finite element for both velocity and pressure. This
is not going to change in a near future, this requires penalization on the
pressure for stability reasons. By the way, in the special case of a
projection method, one can also use Q1-isoQ2-Q1 method to remove spurious
modes (the velocity grid is two times finer than the pressure's one).
- Finally, you have to know that the current implementation of the language
function evaluator is quite slow (the first target was genericity). This
makes the discretization of the convection operator slow. I know exactly what
to do to speed-up this a lot, but I am not sure about the time I need to do
so. If requested, this can be put at the top of my todo list.
Well, there might be lots of details that I forgot. I have not run NS
computations for a while, I just checked recently that the example was
running ok ...
I hope you are not discouraged by all this ;-)
Best regards,
Stephane.