An advanced QR-decomposition for sparse matrices in Octave or Octave-Forge.

classic Classic list List threaded Threaded
2 messages Options
Reply | Threaded
Open this post in threaded view
|

An advanced QR-decomposition for sparse matrices in Octave or Octave-Forge.

Ole Jacob Hagen-2
Hi, Octavers


The present implemented QR-routine in Octave uses
LAPACK (Fortran) with a C++ interface.
The usage is, say: [Q,R,P] = qr(A), which makes the
following decomposition: A*P = Q*R.
 
Is it possible to extend it so we get this
decomposition, Pr*Q'*A*Pc' = [R; 0]? If so, how can
this be achieved?

The best alternative is if anyone out there has a
function ready for usage. This might even be a good
update to the present QR-implementation?

It might be a solution to integrate the code of Thomas
Robey; SPARSEQR, which I've tried out.  This routine
does a QR-factorization, which is described above. The
original code is written in C and C++ and is under
Library GPL license.

It has been used for rather large problems for Finite
Elements.
I've had some correspondance with Mr. Robey and it's
OK with him if I decided to integrate the sparseqr
routine of his in my RFSQP-routine in Octave-forge,
but only if I gave him credits.....And that's no
problem.

Is this routine (SPARSEQR) something for Octave of
Octave-forge?

Why I need it:
"I could integrate the code of his in my routine, but
since it would be dumb to implement this routine in
RFSQP, the SPARSEQR routine should be implemented in
main/sparse in octave-forge?

The reason why I'm interested in sparseqr, is that
RFSQP requires the Q matrix from the QR-routine, which
gives the possibility to solve a equality constrained
problem, using a "NULL SPACE" strategy. The reason is
that the Q-matrix gives an matrix with orthogonal
columns."


Any comments?

Cheers,
 
Ole J.
 


______________________________________________________
Få den nye Yahoo! Messenger på http://no.messenger.yahoo.com/
Nye ikoner og bakgrunner, webkamera med superkvalitet og dobbelt så morsom


Reply | Threaded
Open this post in threaded view
|

Re: An advanced QR-decomposition for sparse matrices in Octave or Octave-Forge.

Andy Adler
It would be nice to add this functionality to the octave
sparse functions. It would be best if this is done in
a compatible way to the sparse that's already in octave-forge.

Take a look at octave-forge/main/sparse - If you have difficulty
with the code, then ask me.

Note: there has been some discussion of moving away from
SuperLU. I have no problem with this, because it is fairly
incomplete and I had to write lots of extra code (about 30% of
SuperLU).

If we switch, it should be a well though out decision.

Andy
--
Andy Adler,
Assistant Professor, School of Information Technology and Engineering,
University of Ottawa, Tel:1(613)562-5800 X 2345, Email:[hidden email]


On Mon, 9 Dec 2002, Ole Jacob Hagen wrote:

> Hi, Octavers
>
>
> The present implemented QR-routine in Octave uses
> LAPACK (Fortran) with a C++ interface.
> The usage is, say: [Q,R,P] = qr(A), which makes the
> following decomposition: A*P = Q*R.
>
> Is it possible to extend it so we get this
> decomposition, Pr*Q'*A*Pc' = [R; 0]? If so, how can
> this be achieved?
>
> The best alternative is if anyone out there has a
> function ready for usage. This might even be a good
> update to the present QR-implementation?
>
> It might be a solution to integrate the code of Thomas
> Robey; SPARSEQR, which I've tried out.  This routine
> does a QR-factorization, which is described above. The
> original code is written in C and C++ and is under
> Library GPL license.
>
> It has been used for rather large problems for Finite
> Elements.
> I've had some correspondance with Mr. Robey and it's
> OK with him if I decided to integrate the sparseqr
> routine of his in my RFSQP-routine in Octave-forge,
> but only if I gave him credits.....And that's no
> problem.
>
> Is this routine (SPARSEQR) something for Octave of
> Octave-forge?
>
> Why I need it:
> "I could integrate the code of his in my routine, but
> since it would be dumb to implement this routine in
> RFSQP, the SPARSEQR routine should be implemented in
> main/sparse in octave-forge?
>
> The reason why I'm interested in sparseqr, is that
> RFSQP requires the Q matrix from the QR-routine, which
> gives the possibility to solve a equality constrained
> problem, using a "NULL SPACE" strategy. The reason is
> that the Q-matrix gives an matrix with orthogonal
> columns."
>
>
> Any comments?
>
> Cheers,
>
> Ole J.
>
>
>
> ______________________________________________________
> Få den nye Yahoo! Messenger på http://no.messenger.yahoo.com/
> Nye ikoner og bakgrunner, webkamera med superkvalitet og dobbelt så morsom
>
>