Eigenvalues and Eigenvectors Computation in Intel Math Kernel

Question

Does any know what is the algorithm used for eigenvalues and eigenvectors computation in Intel Math Kernel library? From the link I can find is that it seems to use pdsyev algorithm, which uses QR algorithm, that would deteriorate in the face of large matrix.

If this were true, then the mkl isn't a suitable library for computing large scale eigenvalue problem.

What do you think? Or is it that I misunderstand the above text?

Answer 1

To my knowledge the MKL represents a "normal" Lapack packaged with an optimized Blas. Maybe with some small extras here and there. However, the Lapack contains only O(N^3)-scaling algorithms for the computation of eigenvalues and -vectors. For large-scale problems -- if only a limited number of eigenvalues is required -- other libraries like e.g. ARPACK can perform much better.