Failure of test cases with spectra-0.2.0 and eigen-3.2.8

[yixuan]

Failure of test cases with spectra-0.2.0 and eigen-3.2.8

[skyrulz]

A simple solution, apart from moving to a newer version of the Spectra and Eigen would be more helpful as it would force me to move to C++ 11.

[skyrulz]

Hi yixuan,

Any suggestions or this issue?

Thanks

[yixuan]

Hi @skyrulz , sorry for the delay. Both Spectra and Eigen require only C++98. C++ 11 is only used by the Catch library for the testing. If you just want to run some examples, you do not need C++ 11 at all.

[skyrulz]

Hi yixuan, Thanks for your response. Since c++11 is needed only for catch, I don't understand why the test cases fail for GenEigsSolver. The eigen values don't converge, based on the output I shared. Any particular reason for the failures. Following is my development environment: - SUSE Linux - g++ (SUSE Linux) 4.3.4 What are prerequisites for the package? Are there any environment settings that i am missing ? Thanks

…

On Mon, Apr 9, 2018, 7:34 PM Yixuan Qiu ***@***.***> wrote: Hi @skyrulz <https://github.com/skyrulz> , sorry for the delay. Both Spectra and Eigen require only C++98. C++ 11 is only used by the Catch library for the testing. If you just want to run some examples, you do not need C++ 11 at all. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#44 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AkVImbREXDn7feUlIiSGS8rbK1VpGyKnks5tm2pkgaJpZM4TH6ly> .

[yixuan]

Actually the failures in your shared log are "expected". If you take a closer look you will find that those are cases of "Smallest Magnitude" or "Smallest Imaginary Part". That is the nature of the algorithm -- it is good at computing largest eigenvalues rather than smallest eigenvalues. And that's why in newer versions of Spectra I allow these cases to fail in the test suite.

[yixuan]

If smallest eigenvalues are needed, the shift-and-invert mode should be used, see the README section.

[skyrulz]

Thanks a lot for the explanation yixuan. It helped a lot.