Thank you for your problem report. I can confirm that I get the same
failure.
Ok, I have also found some other 3x3 and many 4x4 integer matrix
systems which fail, if you'd like more examples.
It is hoped to have improved, more efficient, versions of the
generalized nonsymmetric solvers in a future release, but in the
meantime we aim to look at this specific problem. It does seem possible
that the exceptional shift strategy needs some work.
Does this mean that the small bulge aggressive deflation method
will be implemented for generalized eigensystems? If so, do you know when
it will be done? Even with a more efficient multi-shift solver, its
still important to fix dhgeqz, since the multi-shift solvers will have
to call it to get the shifts.
Patrick Alken
|