[SciPy-dev] sparse eigensolvers

[Ondrej Certik]

> what is the current status of scipy support for algorithms such as eigenvalue decomposition on (very large) sparse matrices? I remember looking at scipy some time ago and it seemed to be in the state of 'if you do it, we will include it'. By 'it', I mean wrapping some existing libraries such as PROPACK, ARPACK etc. to work transparently with scipy (and scipy.sparse in particular).
>
> The reason I'm asking is that I have a student who would be interested in working on this topic (as his bachelor thesis). Please let me know how things stand from a broader perspective/planning point of view and what could still be done. I found some earlier replies in this mailing list which concerned the topic, but they seemed to deal more with details&particulars. Still, I'll be thankful for any links,

I'd be also interested of opinions on this issue. Since I think the
best library for sparse matrices in Python is petsc4py at the moment:

http://code.google.com/p/petsc4py/

(python-petsc4py package in Debian)

So I think it's better to support just one way of doing sparse
matrices in Python and do it well, not two or three. So I find it nice
to support all the solvers from petsc and just provide python bindings
to petsc. What are the disadvantages of this approach?

Ondrej