[Insight-users] Bug in SymmetricEigenAnalysis?

[Andreas Keil]

Dear ITK users,

I'd say, I found a bug in Code/Common/itkSymmetricEigenAnalysis.txx. The
two methods

ComputeEigenValuesUsingQL
ComputeEigenValuesAndVectorsUsingQL

return the absolute values of the eigenvalues when OrderByMagnitude is
chosen. This makes absolutely no sense, since they are no longer the input
matrix' eigenvalues if made positive. Additionally, the documentation only
talks about eigenvalues ordered like |lambda_1| < |lambda_2| < ... which
doesn't suggest that they are made positive.

I would strongly favor having the *real* eigenvalues returned which should
be easily possible by slightly changing itkSymmetricEigenAnalysis.txx.

According to my research, changing from magnitudes to the signed
eigenvalues would not affect any other code in ITK. And even if it affects
others' code, it should be changed. Right now, the naming is just
incorrect (it's not eigenvalues what is currently returned).

If I don't get objections by you, I'd file a bug report soon.

Andreas