I need to compute the eigenvalues of a lagre scale sparse matrixs(10000^2 or more),but i do not need all the eigenvalues.For example ,I just need about 200 eigenvalues around a given value 3.5.In matlab,I can achieve this goal by :eigs(A,speye(size(A)),200,3.5,options).A is a sparse matrix.
Can I use Eigen to accomplish this task?
Thanks~~~~

Hi, that's not possible yet since we don't have a built-in sparse eigenvalue decomposition. You might try the redsvd lib which use Eigen. In this case you'll have to compute for more eigenvalues because this lib implements a randomized approach: http://code.google.com/p/redsvd/

Please, if you try this lib, get back to us with your results, I'm very interested to know how this approach performs on real cases.

I tried using redsvd for a large FEM matrix and it did not work well at all.

I ended up reading the original paper to figure out why and noticed that none of their examples are large (I think most are around 100x100). I wonder if this was the source of my problems?

I have a wrapper I wrote for arpack that I would like to clean up and share as part of eigen (I copied the dense eigensolver interface). I have some questions first. Should I post them in this forum, on the mailing list, or in the bug tracker?

I noticed that there is a "bug" report for arpack++ support. I was unsuccessful in using arpack++, and even if I could get it to compile, it doesn't support 64-bit, which is a deal-breaker for many people. I could pick up that bug and use arpack instead of arpack++.

an arpack wrapper would be nice. You can indeed propose a patch on the arpack++ entry of the bug tracker.