Re: [eigen] generalized eigenvalue problem with complex matrices

[Manav Bhatia <bhatiamanav@xxxxxxxxx>]

I have coded generalized eigensolvers only for real matrices in the past, but I don't expect the algorithm for complex valued matrices to be too different.

 

Out of curiosity, is there a performance comparison of the generalized eigensolver coded in Eigen to that of a system optimized LAPACK? I understand the motivation behind not keeping too many external dependencies, but something like MKL or system optimized LAPACK are prevelant enough to consider providing an interface for?

-Manav

 

On Tue, Aug 13, 2013 at 12:20 PM, Christoph Hertzberg <chtz@xxxxxxxxxxxxxxxxxxxxxxxx> wrote:

On 13.08.2013 16:49, Manav Bhatia wrote:

I am not familiar with Eigen enough to know if it makes use of LAPACK
routines. But if it does, would this class be of interest for inclusion in
the library? If not is the expectation for someone to write an eigensolver
in Eigen instead?

Generally, we don't want to have any (mandatory) external dependencies (other than the C++ std-library). We could think about making Eigen call external LAPACK routines on request (similar to optional MKL support).

So the preferred solution is to actually implement the GeneralizedEigenSolver for complex matrices in Eigen itself. (I have no idea how much work that is, i.e. how much different from the real version this is)

Christoph



--
----------------------------------------------
Dipl.-Inf., Dipl.-Math. Christoph Hertzberg
Cartesium 0.049
Universität Bremen
Enrique-Schmidt-Straße 5
28359 Bremen

Tel: +49 (421) 218-64252
----------------------------------------------