Re: [eigen] Inversion of complex

[ Thread Index | Date Index | More Archives ]

The matrix is neither symmetric nor self-adjoint, but it is diagonizable (physical boundary conditions require this to be the case), if this is what you mean.


----- Original Message ----- From: "Norman Goldstein" <normvcr@xxxxxxxxx>
To: <eigen@xxxxxxxxxxxxxxxxxxx>
Sent: Friday, September 28, 2012 7:09 PM
Subject: Re: [eigen] Inversion of complex

Which class of matrices are your working with?
( What does your code do when Z is not diagonalizable? )

On 09/28/2012 09:34 AM, Claas H. Koehler wrote:
Hi everybody!

I want to port some of my code from LAPACK to eigen. The original code performs an eigenvalue decomposition of a matrix Z= T D T^{-1} into the diagonal matrix D and the transformation matrix T
consisting of the eigenvectors of Z.

Since both T and its inverse T^{-1} are required, T^{-1} is calculated by LU decomposition followed by an inversion. As far as I can see, there exists a class ComplexEigenSolver for complex matrices
for the first part of the problem.

For the LU-decomosition and inversion I would like to use FullPivLU. Has anybody of you made experiences with this method for complex matrices and knows how it performs compared to the LAPACK
routines ZGETRF and ZGETRI?

Kind regards

Mail converted by MHonArc 2.6.19+