Re: [eigen] GeneralizedEigenSolver eigenvalues different from lapack dggev

[ Thread Index | Date Index | More Archives ]

Hello Gael,

Thanks for looking at the complex eigenvalues. I missed your replies to the bug #645 thread as I had a problem with my other e-mail account in the last couple of weeks. I’ll try to look at the eigenvectors again when I get a chance.


On 04/08/2016, 16:51, "Bill Greene" <w.h.greene@xxxxxxxxx> wrote:

    Thanks for the quick reply!
    With 3.3-beta2, all 9 eigenvalues in my simple test agree between GeneralizedEigenSolver and lapack.
    For what its worth, the alpha and beta components for the complex conjugate pairs differ from those
    returned by dggev but, as I say, the quotients are the same.
    On Thu, Aug 4, 2016 at 11:17 AM, Gael Guennebaud 
    <gael.guennebaud@xxxxxxxxx> wrote:
    I've recently applied some changes in GeneralizedEigenSolver, so it might be worth trying 3.3-beta2. Then, it could also be the case that your pair of matrices is not diagonalizable, in which case both implementation returns garbage. Finally, sharing your
     matrices would help to identify the issue.
    On Thu, Aug 4, 2016 at 3:36 PM, Bill Greene 
    <w.h.greene@xxxxxxxxx> wrote:
    I have a simple test case with n=9 where alpha(i) and beta(i) for five eigenvalues
    agree but four do not. Only one pair of three of the complex conjugate pairs agree.
    For the two pair that do not agree, the alpha values are very large (i.e. 10^61).
    I took a look at bug 645 and it is unclear to me if the problem I'm seeing is related
    to that issue. I'm using Eigen 3.2.9 but could certainly try 3.3-beta if appropriate.
    Bill Greene

Mail converted by MHonArc 2.6.19+