topic Hi Losif, in Intel® oneAPI Math Kernel Library

Sparse eigensolver

Iosif_M_ — Tue, 12 Apr 2016 08:16:54 GMT

Dear colleagues, is it possible to find eigenvalues and eigenvectors by means of MKL routines in the following case: Ax = kx, where A is a sparse, general, complex matrix. I found the routines for sparse symmetric/hermitian matrices but not for general sparse matrices.

Hi Losif,

Ying_H_Intel — Tue, 12 Apr 2016 09:16:00 GMT

Hi Losif,

For general sparse matrix, The Pardiso or DSS solver may be you wanted.

Best Regards,

Ying

Dear Ying,

Iosif_M_ — Tue, 12 Apr 2016 09:24:21 GMT

Dear Ying,

Thank you for the response. As far as I understand PARDISO can solve sparse SLAE, but it is not eigensolver.

Best regards,

Iosif

Hi Losif,

Ying_H_Intel — Tue, 12 Apr 2016 13:01:24 GMT

Hi Losif,

then other option may be :

The Extended Eigensolver functionality offers:
• Real/Complex and Single/Double precisions: double precision is recommended to provide better accuracy
of eigenpairs.
• Reverse communication interfaces (RCI) provide maximum flexibility for specific applications. RCI are
independent of matrix format and inner system solvers, so you must provide your own linear system
solvers (direct or iterative) and matrix-matrix multiply routines.
• Predefined driver interfaces for dense, LAPACK banded, and sparse (CSR) formats are less flexible but are
optimized and easy to use:
• The Extended Eigensolver interfaces for dense matrices are likely to be slower than the comparable
LAPACK routines because the FEAST algorithm has a higher computational cost.
7 Intel® Math Kernel Library Reference Manual
1628
• The Extended Eigensolver interfaces for banded matrices support banded LAPACK-type storage.
• The Extended Eigensolver sparse interfaces support compressed sparse row format and use the Intel
MKL PARDISO solver.

Best wishes,

Ying

Iosif, that's not possible

Gennady_F_Intel — Wed, 13 Apr 2016 05:44:28 GMT

Iosif, that's not possible for general sparse cases for now. one of the existing solution is feast 3.0 which is on our radar but without some timeline at this moment. Do you have some specific timeline when you need to have such functionality available?

Dear Gennady,

Iosif_M_ — Wed, 13 Apr 2016 14:24:07 GMT

Dear Gennady,

I have also found that functionality in the FEAST solver. Just now it is enough for us to use FEAST and switch to MKL/FEAST later.

Iosif