[
https://issues.apache.org/jira/browse/MATH-1557?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17213993#comment-17213993
]
Randy Strauss edited comment on MATH-1557 at 10/14/20, 3:44 PM:
----------------------------------------------------------------
No. That gives complex eigenvalues, but not complex eigenvectors.
My linear algebra has rusted for about 40 years. But in our MatLab code, if I
use only the real part of the eigenvectors, I get different results. I need
the complex eigenvectors and then must partition them into 2 complex matrices
and do some right-solve and left-solve operations with them.
I guess it's all happening in a file next to the one you linked to,
SchurTransformer.java
If I look at [https://en.wikipedia.org/wiki/Schur_decomposition]
the code is transforming the matrix into Q * U * Qinverse
where U is upper-diagonal and its diagonal contains the eigenvalues.
Since these have complex values, it seems likely that Q will, too.
On your link, EigenDecomposition.java, line 893 looks like matrixT is actually
complex?
I don't know...
was (Author: rastr):
No. That gives complex eigenvalues, but not complex eigenvectors.
My linear algebra has rusted for about 40 years. But in our MatLab code, if I
use only the real part of the eigenvectors, I get different results. I need
the complex eigenvectors and then must partition them into 2 complex matrices
and do some right-solve and left-solve operations with them.
> Enhancement request: Complex Eigen Vectors
> ------------------------------------------
>
> Key: MATH-1557
> URL: https://issues.apache.org/jira/browse/MATH-1557
> Project: Commons Math
> Issue Type: Improvement
> Affects Versions: 3.6.1
> Reporter: Randy Strauss
> Priority: Major
> Attachments: real-eigen-vectors-test-case.txt
>
>
> I have a real hamiltonian matrix and need to get complex eigenvalues and
> eigenvectors, just like MatLab produces (I'm moving code from MatLab to Java).
> I put a test case in the real-eigen-vectors-test-case.txt file, attached.
> (Meanwhile, I'll try the ojAlg library... The only other one I haven't
> investigated is JEigen, a wrapper around the C++ eigen library...)
> Thanks.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
