Hi all,

I'm planning to use the Eigenvalues module to compute a full eigenvalue decomposition of a symmetric matrix but I need precisely the divide-and-conquer algorithm.
Does anyone knows if there is an implementation of the Lapack subroutine "dsyevd" in Eigen ?

http://www.netlib.org/lapack/lapack-3.1 ... evd.f.html

Thanks in advance,
Ryo

Eigen has a D&C algorithm for SVD (class BDCSVD), but not for EVD. For symmetric eigenvalue problem, the implemented algorithm uses a symmetric QR algorithm: "The matrix is first reduced to tridiagonal form using the Tridiagonalization class. The tridiagonal matrix is then brought to diagonal form with implicit symmetric QR steps with Wilkinson shift."

btw, why do you *precisely* need the D&C algorithm?

Thanks ggael for your reply.

I'm working on the implementation of an algorithm for the "Nearest Correlation Matrix" problem:
http://www.maths.manchester.ac.uk/~high ... bohi10.pdf

This algorithm requires a full eigenvalue decomposition of a symmetric matrix at each iteration.
And the reason for the "precise" use of the DC algorithm is that it is the one recommended by the authors (Cf. section 3.5 page 101 and section 5 page 105) .

Maybe I will test with the QR algorithm for a first approach.

Regards,
Ryo

ok, so the choice is motivated by speed. Depending on your application and problem sizes, this might not be critical...