[Octave-bug-tracker] [bug #31479] Crash & bugs in eigs

[Rik]

Follow-up Comment #7, bug #31479 (project octave):

I had no idea that there was such a long history to this problem; this gives
me no confidence that the problem will be fixed in ARPACK since they have
already ignored it for 4 years.

You and I did come independently to the same conclusion as to the source of
the problem and that is good.  As part of this most recent investigation, jwe
extracted a self-contained example which is attached to this bug report (.cc
files, .f files, and Makefile).  Would this be enough evidence to convince
either the ARPACK maintainers or the Linux distro packagers?

My opinion now is that we should apply for getting ARPACK fixed, but
separately solve the issue in Octave.  It seems unnecessary to force users to
wait another 4 years for this issue to be resolved, when we can suppress it
ourselves.  I went back and using the problem matrix from the first bug report
in 2006 proved that my solution #3 is insufficient.  We really will need to
scale all the matrices to the size of p.  I re-ran a test of solution #2 and
it works for both the 2006 and 2010 problem matrices.  According to your
earlier work, this change would only need to be made for the NonSymmetric
cases in eigs-base.cc, correct?

    _______________________________________________________

Reply to this item at:

  <http://savannah.gnu.org/bugs/?31479>

_______________________________________________
  Message sent via/by Savannah
  http://savannah.gnu.org/