[SciPy-dev] lobpcg: removed symeig dependence

[Robert Cimrman]

Nathan Bell wrote:
> On Tue, Apr 8, 2008 at 8:58 AM, Robert Cimrman <cimrman3 at ntc.zcu.cz> wrote:
>> Hi,
>>
>>  so it is done, thanks again, Pearu, for your hint.
>>
>>  could someone try on the fresh SVN version (4114):
>>
>>  $ cd <your_path_to_scipy_sources>/scipy/sparse/linalg/eigen/lobpcg
>>  $ python lobpcg.py
>>
>>  the output should end with:
>>  [ 4.  5.  6.]
> 
> Works for me!

But note that I just run the single example that was already in the 
file... No power/time to make more :)

> Thanks for tackling this issue.  I was worried that lobpcg wouldn't be
> ready for the next release.
> 
> Robert, could we switch the order of the A and BlockVectorX
> parameters?  Having A first matches the convention used by other
> iterative solvers in SciPy.

Sure, go ahead, please.
I propose even: A, B = None, M = None, blockVectorX = <something>, ...

> Also, we should provide a short description of what the method
> computes in terms of A, B, lambda, etc.  A small example would be nice
> too (since the interface is a little different than, e.g. ARPACK).

There are already some examples contributed by Nils in the tests 
directory. Concerning the method, I coded it following a paper by Andrew 
Knyazev (see info.py) some not so small time ago, and remember now zero 
about what is going on :].

Thanks for testing it!

cheers,
r.