0

My objective is to find Eigen values and vectors of an input matrix of mxn size. Since it a rectangular matrix, converted the same to square-symmetric matrix by doing a transpose and then matrix-multiplication with source matrix.

After this, i am using Intel MKL library to determine Eigen values and vectors of square-symmetric matrix. Unfortunately, for square symmetric matrix size of 1000000 x 100000 the time taken is ~35 to ~40 secs, which does not suit my application.

Any suggestions on improving the performance would be really helpful.

Thanks in advance

Sravan
  • 131
  • 1
  • 2
  • 10
  • Computing eigenvalues of big matrices just is expensive. You either modify your algorithm to not need the eigenvalues or to produce matrices whose eigenvalues are easier to compute or you pay the runtime price to find the eigenvalues. No way around that. – Baum mit Augen Nov 29 '16 at 12:46

1 Answers1

0

Large eigenvalue problems are expensive. I'd suggest using the FEAST eigenvalue solver, which is available via the MKL library. I think that FEAST 3.0 is not available with MKL, but you can download and compile it separately.

It offers a tunable search space, if you only need a subset of Eigen pairs, and distributed/threaded parallelization.

Gavin Portwood
  • 1,217
  • 8
  • 9