LAPACK Archives

[Lapack] dsyevd fails to converge

Hello Duncan,
I see my name in the original message but don't recall seeing it before, 
sorry. In any case, I have reproduced the bug. The problem is in the 
computation of the roots of a secular equation in subroutine dlaed6.f, 
and I am a bit suspicious of the test 'IF ( ( DSCALE( I )-TAU ).NE.ZERO 
) THEN' there. I will have to discuss with other folks around though.

Hi LAPACK developers,

I reported this issue about dsyevd failing to converge on the attached matrix 
in the LAPACK forum about a year ago, but I think it still has not been 
addressed. The original posting is here:

Can you take another look at this problem? Thanks.

Duncan Po

-----Original Message-----
From: Duncan Po
Sent: Wednesday, November 27, 2013 10:43 AM
To: Langou, Julien
Subject: RE:

Hi Julien,

Attached is the reproduction program and test matrix. The returned INFO is 
15529 on my machine.

Previously I have reported this issue to the Intel MKL team. I am not sure if 
they have already filed the same problem report to you.

Happy Thanksgiving to you too!


-----Original Message-----
From: Langou, Julien [mailto:Julien.Langou@Domain.Removed]
Sent: Tuesday, November 26, 2013 2:05 PM
To: Duncan Po

Hi Duncan,

I hope this email finds you well. We would like to have a look at your
matrix for which LAPACK DSYEVD fails while DSYEV converges.
See your post:

Would you mind sending me the incriminated matrix?

I will likely forward it to Osni Marques and we will take it from there.

Sorry for the delayed response on our side.

Wishing you a happy thanksgiving.


Lapack mailing list

-------------- next part --------------
An HTML attachment was scrubbed...

<Prev in Thread] Current Thread [Next in Thread>

For additional information you may use the LAPACK/ScaLAPACK Forum.
Or one of the mailing lists, or