Eigenvalues of sparse matrix

[Natalie]

Hi!

I'm working with partial differential equations, especially with general elliptic equations. Discretizing those with a finite Element scheme leads to huge matrices (depending on my amount of grid points, but n >100 usually). I solely store the non-zero entries of the matrix and I want to compute the eigenvalues of it. Is there a lapack routine which does not require the full matrix as input?

Thanks and greetings,
Natalie

[Julien Langou]

Is there a lapack routine which does not require the full matrix as input?

If your matrix is symmetric (or Hermitian) and with a narrow bandwidth we have some direct eigensolver for this.
Otherwise => nope.

You probably need to turn to iterative methods package.
If the matrix is symmetric or Hermitian, check out PRIMME or BLOPEX.
If the matrix is nonsymmetric, checkout ARPACK.
You have some larger software like Trilinos or Petsc that might be worth checking out.

[Natalie]

A narrow bandwidth - well, that depends on the definition of "narrow" ;-) . What's the routine called which does not require the full matrix as input?

Thanks for answering,
Natalie

[Julien Langou]

DSBEV (or SSBEV, CHBEV, ZHBEV)
Julien.