Just my 2 bits: a "sparse" matrix may or may not have a specific form which makes a particular algorithm more or less rapid or stable, or both. It'd be more help to smp if we found out that his/her sparse matrix was actually, say, tridiagonal or banded, as is common for matrices arising from applications, as opposed to randomly distributed sparse.
So a "general purpose" sparse matrix solver isn't really the best solution. There's also a distinction between storing the sparse matrix efficiently, which is relatively easy, and solving the system efficiently...
Last edited by mostlyharmless; 06-25-2010 at 03:55 PM.