On Mon, Mar 3, 2008 at 11:12 AM, Robert Cimrman cimr...@ntc.zcu.cz wrote:
Hi Sergey!
Ondrej Certik wrote:
Hi Sergey!
On Sat, Mar 1, 2008 at 9:57 PM, Sergey Dolgov sol...@gmail.com wrote:
Hi Robert, Ondrej, others.
For some time now I've been planing to learn finite element method with the purpose of applying it to electrostatic problems arising in field emission calculations. I'm also fond of free software tools. I've been thinking of using libmesh, which seems to be rather powerful, but currently I would say that it is too hardcore for me, given how weak my knowledge of C++ is; plus I don't yet need advanced capabilities like parallel computation.
So while reading about sympy, I've stumbled upon sfepy, and I feel optimistic about it because 1) it looks like a still relatively simple implementation, and 2) it is wrtten mostly in Python (I've got some experience with Python, numpy, scipy, matplotlib).
Yes, sfepy is very simple comparing to e.g. libmesh. The released package has about 350kB including example meshes and other data. However, it can already be used to solve various kinds of non-trivial problems, as you have probably seen on the Examples page. The web was not updated for some time, but we are now preparing new pages, so wait a bit for more up-to-date stuff.
What sfepy is lacking is of course documentation. I've read some introductory texts on FEM, but certainly don't feel knowledgeable enough to figure how to use sfepy right away. I wold say I need to
Yes this definitely needs to be addressed. Until recently I have developed sfepy alone, and as I tried to keep the code clear and simple, I did not need to write the documentation. Now that other people start joining or showing interest, this will change of course. I have to thank Ondrej for his feedback.
close the gap between pure theory and sfepy implementation. Thus I've got two questions:
- Do you think that using sfepy to learn FEM in general is a good idea? (By "using sfepy to learn FEM" I mean 1) learning how to describe the problem so that sfepy can solve it and 2) learning how sfepy actually works so that I can actually make modifications myself)
SfePy uses problem description files (input files) that resemble in structure to a mathematical description of a weak FE formulation on paper - you define fields of certain FE spaces (e.g. linear P1), variables, equations etc. But you have to know already some basics of FEM to understand the notions at least. The notes Ondrej pointed you to below are a good start. Look also at
http://ui505p06-mbs.ntc.zcu.cz/sfe/Documentation?action=AttachFile&do=get&target=terms.pdf
where a very brief information on how to solve a Poisson equation in sfepy. I will improve this soon and put it on the web.
A general info on FEM can be found on many places, e.g. here: http://en.wikipedia.org/wiki/Finite_element_method
- What book would you recommend as a guide to studying FEM (is there one that sfepy implementation follows more closely than the others)?
Well, I have no real preference here. Any book about FEM is ok.
Sorry if my questions look silly -- they certainly do to me :)
No, these are good questions - nobody starts as an expert :-)
If you have some specific questions, just ask - I can help you get running.
Nonetheless, thanks for the great work!
Thanks for the feedback. I am glad that sympy is also useful in a way to build the community of people doing scientific calculations. Together with scipy + numpy + matplotlib, we want to have a nice environment for doing basically anything. The community around all these these tools is wonderful.
+1.
And then you find out that the guy doing similar things lives next town...
To learn about FEM, you can try my notes here:
http://libmesh.sourceforge.net/wiki/index.php/Main_Page#Examples
mainly this:
http://ondrej.certik.cz/site_media/libmesh/fem.ps
Let me know if it is useful and if you don't understand something, just ask.
From my own experience, sfepy is much easier to use than libmesh, but
As to doing electrostatics, that is perfectly possible with sfepy. But the documentation is missing, you need to read the sources directly. that's because I know already Python and also because Robert helped me write the solver for things that I needed. I think you won't loose time learning it. If you decided you don't like it, you can always write something on your own.
But I would like to have more people working with sfepy, giving feedback and helping improving it!
Right now I am actually in Austin at the Enthought headquarters, it's awesome to discuss things directly with people like Fernando (ipython), Travis (numpy). So while I am here I will visit the libmesh guys here and definitely discuss ways of doing FEM with them.
Say hello to the *py* people!
I did and I also advertized sfepy a lot. Did you experience any difference in site statistics? :)
When I find some time, I'll write all my experiences here to my blog.
Ondrej