>If you are going to work on PDE solvers have a look at a some of those already available before you start coding to avoid duplication [3,4,5].
I should clarify that the PDE setup part is already written, in fact, the whole code is already written and optimized fairly nicely. I am just looking to update and improve the code by using a new iterative solver. I'm a summer student working for the professor who wrote the code.
>
Do you already know which iterative methods (eventually, with which>preconditioning) solve your problem(s) ?
I did not at the time, I've since looked into it. I have a non-symmetric sparse matrix that is diagonally dominant, but not positive. This limits the solution choices a bit, although some iterative methods like GMRES, OrthoMin, maybe Biconjugate Gradient methods would be able to find a solution. I may end up trying several different solution methods and then choosing the one that seems to provide the fastest solution.
>There are many free solvers. Octave is already using the best of them, as others have noted.
Thanks Jordi. Best is relative though... and I kinda need to know the parameters that make you think it is the best before I can see if that fits what I think is the best. :)
I'm not sure how SuiteSparse's direct solutions would compare to other packages iterative solutions in terms of memory or creation of very small terms. I couldn't find very readable or helpful documentation on the SuiteSparse package, and what there is of it seems to be focused on Matlab user instructions.
I may end up trying that package. Thanks.