Re: 3D version of ETesla

["Tesla list" <tesla@xxxxxxxxxx>]

Original poster: Jim Lux <jimlux@xxxxxxxxxxxxx>

At 03:42 PM 7/7/2005, you wrote:


> Jim (?) mentioned a couple weeks ago that there are highly-optimized 
> array-handling codes available, though I don't know if they are free.....

ATLAS is what you're looking for.
http://math-atlas.sourceforge.net/
The ATLAS (Automatically Tuned Linear Algebra Software) project is an 
ongoing research effort focusing on applying empirical techniques in order 
to provide portable performance. At present, it provides C and Fortran77 
interfaces to a portably efficient <http://www.netlib.org/blas>BLAS 
implementation, as well as a few routines from 
<http://www.netlib.org/lapack>LAPACK.

One might also want to look at BLAS
http://www.netlib.org/blas/

Intel's optimized libraries: (free non-commercial download)
http://www.intel.com/cd/software/products/asmo-na/eng/perflib/mkl/219823.htm

An AMD optimized library:
http://www.cs.utexas.edu/users/kgoto/



> In my opinion, though, the most difficult aspect is displaying the results 
> - "slices" would have to be created to create the nice 2D surface plots 
> we're used to.  It would almost certainly be better to have the 
> application generate the plots than to load them into mathcad (Excel can't 
> handle more than 255 columns).  I was working on that particular issue a 
> number of years ago, but "ran into trouble" and gave up. :o


Octave is probably your answer.  Or Matlab ($100 for the academic edition, 
if you qualify). Decent 3D graphics visualization stuff (transparency, 
vectors, all the mod cons)

> As an afterthought I had just as I was about to click the send button, 
> this really is just a standard FEA problem.  Although I don't know of any 
> free FEA applications out there, there are discounted student copies (and 
> illegal copies) of many of the commercial ones.  Has anyone thought of 
> and/or looked into this?

Try: http://emlib.jpl.nasa.gov/ for a directory.  The Applied Computational 
Electromagnetics Society (ACES) might also be a resource.

It IS a standard FEA kind of problem.  What a commercial code gets you is:
1) Easier model entry, often with a graphical front end, or a translator 
from standard drawing packages like AutoCad, IGES, Mechanical Desktop, etc.
2) More sophisticated gridding (which is how you get decent performance.. 
grid finely where you need to, less fine farther away).  There are 
"free-ish" variable gridders out there, but they're usually someone's PhD 
dissertation, and often not very turnkey.
3) Well integrated visualization of the results.
4) Cluster computing awareness

The last is important, because these kinds of problems, particularly if 
you're doing a dynamic model over time steps, are quite compute intensive.

The good news is, though, that if you're willing to develop the modeling 
code for modeling dynamic spark growth in a non-uniform field, I might be 
able to find you a BIG cluster computer to develop and test on.