Page 2 of 2

Re: eigenvalues with pzlahqr

PostPosted: Thu Jul 12, 2012 11:08 am
by anto_n
Ok, thank you. I installed the last version of SCALAPACK (2.0.2) but I have always the same problem. The spectrum of eigenvalues obtained with pzlahqr doesn't match fully with the one obtained in the scalar case with zgees (there are few eigenvalues found with pzlahqr that aren't correct). Moreover changing the dimension of blocking the spectrum changes slightly finding, however, any eigenvalues that are wrong. Can you help me?

Re: eigenvalues with pzlahqr

PostPosted: Thu Jul 12, 2012 11:40 am
by anto_n
I want report another aspect. Reading the notes of pzlahqr I have seen that "If the parameter WANTZ is true on entry Z must contain the current matrix Z of trasformations accumulated by PCHSEQR" but this routine is not present in the SCALAPACK routines. Furthermore is said that "This code has only been tested for WANTZ =.TRUE. and may behave unpredictably for WANTZ set to .FALSE.". But how is possible using the routine pzlahqr with parameter wantz set to true if doesn't exist the routine pchseqr? Thank you in advance for your help.

Re: eigenvalues with pzlahqr

PostPosted: Fri Jul 13, 2012 6:53 am
by Julien Langou
The Nonsymmetric Eigenvalue test suite is not completed. Some routines are missing indeed. This is the case for P[C/Z]HSEQR. Julien.

Re: eigenvalues with pzlahqr

PostPosted: Fri Jul 13, 2012 8:40 am
by anto_n
Thank you Julien for the information. Then my problem with pzlahqr is also related to not completeness of test suite?

Re: eigenvalues with pzlahqr

PostPosted: Fri Jul 13, 2012 8:45 am
by Julien Langou
Indeed.

Re: eigenvalues with pzlahqr

PostPosted: Fri Jul 13, 2012 8:54 am
by anto_n
Ok, I got it. Thanks again. Now I know that is not a problem of my code but a more general problem.
Bye,
Antonio.