[Top][All Lists]

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

Re: [Getfem-users] WG: `Coupling Equations

From: Roman Putanowicz
Subject: Re: [Getfem-users] WG: `Coupling Equations
Date: Thu, 26 Jan 2012 18:56:10 +0100
User-agent: Mutt/1.5.20 (2009-06-14)

On Thu, Jan 26, 2012 at 02:08:30PM +0100, Pichler, Franz wrote:
> Hello, 
> I posted this question months ago and I still would be quite interested in an 
> answer.
> Thanks
> Franz
> >Hello everybody,
> >I just get started with getfem and have some questions. What I want to
> >do is to couple several equations. 
I did the coupling of heat transfer and elasticity equations 
for thermo-mechanical problems. It required writing special purpose
brick. If you are interested in the source code you can browse 
the repository of our project:

The coupling brick is in thermomech.* and the sample application 
is in annulus_thm.cxx. Unfortunately the code is sparse on comments
but I can answer some questions directly.

> >1) Is there a brick that I can tell that in my equation there are two
> >variables with a laplacian (elliptic) term?
As far as I know the answer is no.
> >2) If I would try to write this brick myself, would I have to do it in
> >c++ (I am using the python interface)
Yes, at the moment you can add new bricks only at C++ level.
I have a proof of concept of a brick that can call arbitrary Python
scripts but it still requires some work (especially on returning the
tensors from Python (NumPy matrices) to C++). You would need some
experience in extending and embedding Python. If you are interested, 
this is also in our project.

> >3) Is there a way of assembling this without the model framework and
> >without putting the matrices together "by hand"
> >
> >I would like to do all of this as time efficient as possible because in
> >the end I want to solve time dependent problems with nonlinear terms.

My feeling of this issue is that using the model framework do not 
incur overhead so big that we should dismiss it in the first 
instance. Contrary, this framework has several advantages. 
Without profiling and investigation of the performance all 
recommendations are highly speculative.
For performance the whole system needs fine tuning.


Roman Putanowicz, PhD  < address@hidden  >
Institute for Computational Civil Engng (L-5)
Dept. of Civil Engng, Cracow Univ. of Technology, tel. +48 12 628 2569, fax 2034

reply via email to

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