LAPACK Archives

[Lapack] Bug: xSTEDC eigenvalues not sorted

Dear Christoph,
we just applied your patch to our repository (revision 1601).
The fix will be included in the coming LAPACK release.
Thank you very much for reporting the issue and submitting a patch
Sincerely
Julie
On Oct 7, 2015, at 5:39 PM, julie <julie@Domain.Removed> wrote:

Dear Christoph
Thank you for the bug report.
Your bug is listed as bug 139 in 
http://www.netlib.org/lapack/Errata/index2.html
We will take a look at it very shortly.
Julie
On Oct 6, 2015, at 4:20 AM, Christoph Conrads <cconrads@Domain.Removed> 
wrote:

Hi,

according to the LAPACK Doxygen documentation, xSTEDC guarantees to
return the eigenvalues in ascending order (parameter D on exit). To
this end, it sorts the eigenvalues if necessary.

If there is only a single subproblem of size n, no sorting algorithm is
called (cf. {s,d}tedc.f, line 450). Furthermore, xLAED0
({s,d}tedc.f:400) does not guarantee to return eigenvalues in ascending
order. Thus, xSTEDC may return eigenvalues that are not in ascending
order.

As a fix, I propose always sorting the eigenvalues (patch attached).

The bug can be triggered in practice (see compressed file in matrix
market format).

Sincerely
Christoph Conrads
<lapack-reference-svn-xSTEDC_always_sort.patch><sstedc-sort-bug.mtx.gz>_______________________________________________
Lapack mailing list
Lapack@Domain.Removed
http://lists.eecs.utk.edu/mailman/listinfo/lapack



<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