forum.alglib.net

I have a 2D-FEM-simulation and want to solve the linear equation system Ax = b for it.
Hence, my matrix is
- quadratic
- symmetric
- hermitian
- real
- sparse as on https://en.wikipedia.org/wiki/Sparse_matrix

Which solver is the best for my matrix?
And could you maybe give me a codesnippet of how to code it in C# beginning with:
sparsesolvesks(...) works, but is not pretty fast ...

1. How many variables do you have?
2. How many non-zero elements per entire matrix do you have?

1. around 5000x5000 matrix
2. around 10 non-zero-elements per row

Did you try sparselu() function?

Ideally, you should use Cholesky for positive definite matrices - but presently we do not have sparse Cholesky in ALGLIB, it is planned to be included in version 3.17.0 (we have a lot of stuff in upcoming 3.16.0 which will benefit from sparse Cholesky, so it is of high priority for us).

Yes, Cholesky would be very nice and I am very interested in implementing this to ALGLIB :)

How do i get a row permutation matrix and a column permutation matrix?

alglib::sparselu returns them in parameters p and q, as product of individual permutations (see subroutine comments for exact format)