topic Eigenvalue problem for sparse matrices in Intel® oneAPI Math Kernel Library

Eigenvalue problem for sparse matrices

pourmatin85 — Mon, 08 Jun 2009 07:17:04 GMT

Hi guys;

I'm trying to compute "dynamic mode shapes" in my fortran code, so I need to find either eigenvalues of a symmetric sparse matrix or the inverse of that matrix (the first method is prefered).

Is there any routine in BLAS or any spesific procedure that could help me?

thanks in advance

Hossein

Re: Eigenvalue problem for sparse matrices

Gennady_F_Intel — Mon, 08 Jun 2009 13:03:29 GMT

Hossein, there is no such functionality for the sparse routines at all.
We have the similar functionality in the current version for the dense matrix only.
--Gennady

Re: Eigenvalue problem for sparse matrices

pourmatin85 — Mon, 08 Jun 2009 14:50:59 GMT

Quoting - Gennady Fedorov (Intel)

Hossein, there is no such functionality for the sparse routines at all.
We have the similar functionality in the current version for the dense matrix only.
--Gennady

Dear Gennady; thanks for your quick reply. I found a way to generate inverse of my sparse matrix. Hopefully, it will do the job.

Regards;

Hossein

Re: Eigenvalue problem for sparse matrices

Gennady_F_Intel — Mon, 08 Jun 2009 14:58:05 GMT

Quoting - pourmatin85

Dear Gennady; thanks for your quick reply. I found a way to generate inverse of my sparse matrix. Hopefully, it will do the job.

Regards;

Hossein

Hossein, may I ask you, how did you solve this problem?
--Gennady

Re: Eigenvalue problem for sparse matrices

pourmatin85 — Tue, 09 Jun 2009 06:15:58 GMT

Quoting - Gennady Fedorov (Intel)

Hossein, may I ask you, how did you solve this problem?
--Gennady

Well, suppose B is a (N,1) vector. Every components of B is set to zero but it's Ith component, which is 1. The resulting vector of the product of a matrix, say K, and B is the Ith column of K.

So, all you need to do, to find the inverse of K, is to solve N equation systems with PARDISO. It means that in the Ith iteration out of N iterations, the Ith component of B is 1, and the result is the Ith column of the inverse of K:

x*K=B -> x=inv(K)*B -> x=inv(K(1:N,I))

By the way, MKL doesn't have any routine for matrix-matrix product, in which both matrices are sparse (is that right?). Isn't it weird!! I have to do the same trick to find the answer of mymatrix-matrix products.

Hossein

Re: Eigenvalue problem for sparse matrices

ArturGuzik — Tue, 09 Jun 2009 06:38:40 GMT

Quoting - pourmatin85

see this thread.

A.