[Fwd: Supporting sparse matrices]

[Ron Boisvert <boisvert@nist.gov>]

Sender: Daniel Winterstein <d.winterstein@thinktankmaths.co.uk>
Subject: Supporting sparse matrices



Firstly, let me say thank you for making Jama available. The interface 
is clear and gives me exactly the functionality I wanted. I've found it 
very useful. And forgive me if this topic has already been discussed to 
death.

Re. having JAMA as part of Java ("the primary linear algebra package for 
Java" as the website proposes), I think it would be good to add in a 
layer of abstraction, in order to support the use of sparse matrices.

Jama itself is tied to working with dense matrices, represented by 
arrays. This is inappropriate for sparse matrices. Sparse matrices occur 
naturally in a lot of important problems. Specialised representations 
and codes can give _orders of magnitude_ speed ups for sparse problems. 
Anyone who is unfamiliar with the subject, I recommend 
http://citeseer.ist.psu.edu/duff98matrix.html for an excellent intro to 
computing with sparse matrices.

I would like a linear algebra system which allowed different matrix 
representations and codes. High-level code should be able to work with 
any of these without worrying abut what is going on 'under the hood'.

This shouldn't take a great deal of modification. Perhaps as simple as 
extracting an IMatrix interface  from Matrix, then implementing a simple 
sparse-matrix to test that this makes sense. It can then be left for 
others to implement efficient codes for different types of sparse matrix.

Your thoughts?
- Daniel

-----------------------------------------------------------
"I know nothing, understand less, & most of that is wrong."

   Daniel Winterstein
   http://thinktankmaths.co.uk