[SciPy-user] stopping criterion for integrate.odeint?

[Robert Kern]

Ryan Gutenkunst wrote:

> We'd really love to be able to use SUNDIALS 
> (http://www.llnl.gov/CASC/sundials/). The sensitivity capabilities and 
> ability to handle DAE systems are very attractive. (We've implemented 
> some of those capabilities using odeint, but our implementations are 
> pretty problem-specific and not that efficient.) The code is BSD, but 
> the suite itself is a 7 meg download (!).

It's actually not so bad. Most of that is redundant PDF and PS 
documentation. Strip those out, the gzipped tarball is only 716k and the 
uncompressed directory is 4M (compared to, say 3.4M for xplt) and 1M for 
  the installed static libraries on OS X (15M for a complete ATLAS).

It's really just a matter of someone wanting it enough to put the effort 
in. And it's going to be a fairly large job to do it thoroughly.

> In a more realistic vein, LSODAR 
> (http://www.netlib.org/odepack/opkd-sum) would be a nice addition. It 
> would add root-finding capabilities that would be useful to Hans and 
> myself. Since odeint already uses LSODA, I imagine wrapping LSODAR 
> wouldn't be that hard.

Probably easier than wrapping LSODA was. Poor Travis had to do it 
without f2py to help him!

If I used ODEs more (e.g. at all), I might do it myself. Alas, MacEnthon 
sucks away my Python-hours. (A testing release is coming! Soon! As soon 
as I write some readme-type docs and I get access to Enthought's 
download server.)

-- 
Robert Kern
rkern at ucsd.edu

"In the fields of hell where the grass grows high
  Are the graves of dreams allowed to die."
   -- Richard Harter