[SciPy-dev] sparse eigensolvers

[Ondrej Certik]

> There is an ARPACK wrapper in the scipy sandbox.  Right now the Python
> interface only handles the standard eigenvalue problem Ax=wx
> (symmetric and nonsymmetric) but the f2py wrapper should be able to
> handle the generalized eigenproblem and all of the various shifted modes.
> There are some design decisions to be made regarding the interface
> and some more code to be written to accomplish that.
>
> There is a description of the current state and some ideas for
> an interface at
> http://projects.scipy.org/scipy/scipy/ticket/231
> http://projects.scipy.org/scipy/scipy/wiki/ArpackWrapper

Oh, I misread you are interested in eigensolvers, sorry about that. :)

Yes, I don't know about any opensource systematic approach, in the
same spirit as petsc4py. And I need this as well, for my master
thesis. There is slepc4py:

http://t2.unl.edu/documentation/slepc4py

which is free, but slepc itself is not free. Arpack is unfortunately
too archaic (doesn't perform well on my problems), but I have a very
good experience with blzpack[1] and pysparse[2], also primme[3] is
good.

However, I'd be interested in some kind of unified interface in
Python, the same way like one can use petsc4py for Ax=b problems.
There are many libraries for eigensolvers and linear solvers though,
so I am not sure if they all should be in SciPy? Maybe it's better to
create a separater project for a unified python interface to
eigensolvers, in a similar way like petsc4py is a unified python
interface for almost all opensource linear solvers available?

Ondrej

[1] http://crd.lbl.gov/~osni/
[2] http://people.web.psi.ch/geus/pyfemax/pysparse.html,
http://pysparse.sourceforge.net/
[3] http://www.cs.wm.edu/~andreas/software/