
Hello.
I am very interested in using ARPACK (large sparse matrix eigenvalue problems) together with Math.NET Numerics. Would this be feasible to integrate? I am considering implementing this myself, as I have some experience using ARPACK from managed code. Perhaps
you could recommend how it is best to go about doing this? Would a managed wrapper along with the other native wrappers (e.g. ATLAS) be most appropiate?
Many thanks, Alex


Coordinator
Aug 14, 2011 at 10:19 AM

Hi Alex,
I think this would be a great contribution to Math.Net. Math.Net already has a sparse matrix class. However, any of the factorization code in LinearAlgebra/Double/Factorization does not have any special code to treat sparse matrices. You'd want to start
by adding sparse factorization classes there. Each factorization class in turn calls an appropriate method in Algorithms/LinearAlgebra. There you could code to call the ARPACK methods.
I think it would be a really good contribution to Math.Net. Let us know how you get on!
Cheers, Jurgen



Hi Jurgen,
I've recently found some time in my crazy schedule to make an effort at this... The good news is that it doesn't seem too hard in concept. However, I still have a few other things occupying me. I'll try to keep you updated on progress  will you be available
(via this form) for potential questions, can I ask?
Cheers,
Alex



Any news, folks? I do hope this project is still active. :)


Coordinator
Nov 4, 2011 at 12:20 AM

Hi Alex,
Just to confirm, yes, the project is active, just moving slowly (see our
mainline commit history for recent changes, and our
recent releases).
Btw, concerning sparse linear algebra, I'm currently completely reworking the managed sparse implementation (I've discovered some serious issues there), a new fixed release v2.2 will soon be available.
Thanks,
Christoph



Hi Cristoph,
Thanks for the update on things.
It's good to hear that you're rewriting the sparse linear algebra parts. I take it the new implementation will include proper treatment for factorisation/decomposition of sparse matrices, as Jurgen referred to?
If this is the case, perhaps I can have a go at integration the ARPACK funcitonality into Math.NET Numerics once the 2.2 release is made. Or sooner even, if the necessary functionality is there in mainline?
Cheers, Alex



So, any estimated date for the v2.2 release (or at least when the functionality will appear in the dev branch)?


Coordinator
Nov 15, 2011 at 1:38 AM

Hi Alex,
I certainly hope to get v2.2 out in November, but I cannot give any guarantees on that (our free time is very limited and I can only contribute sporadically).
However, the mentioned serious issues around managed sparse vectors & matrices to be fixed in v2.2 are more about basic arithmetic algorithms on arbitrarily sparse data  not about managed sparse decompositions.
Thanks,
Chris



Hi Chris,
Indeed, we all have paid jobs that (unfortunately?) take up most of our time! That's great that your continuing solid work on the project though, and I look forward to the 2.2 release.
From what you say, I presume I wouldn't be going too wrong if I started trying to integrate the ARPACK source into the main development branch right now? I'll try and make an effort on that soon, in that case... alongside my other projects,
Cheers, Alex


Coordinator
Nov 15, 2011 at 8:18 PM

Hi Alex,
Yes, that would be great. (Provided you do not require managed sparse arithmetics to work correctly at this stage.)
Thanks,
Chris



Do you mean correctly or just efficiently in fact? All I need is a decent working implementation of sparse matrix multiplication and all should be good.
Thanks, Alex


Coordinator
Nov 16, 2011 at 11:01 PM

Correctly, unfortunately.
Thanks,
Chris



Hey folks... any update on this? Am I potentially in a position to do work on it soon? Cheers.

