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

[John W. Eaton]

Follow-up Comment #11, bug #31479 (project octave):

I also started a discussion about this topic on the maintainers list, which
is archived starting here:

https://mailman.cae.wisc.edu/pipermail/octave-maintainers/2011-January/022538.html

Rik, in the debian package, it looks like they compile with -O3 but no other
special options.  I doubt there would be too much of a difference between that
and the default -O2 when using gfortran, but I haven't tried to do any
testing.  ARPACK already uses the BLAS and LAPACK for linear algebra, so what
is used for that will probably have more of an effect than the compiler
options for ARPACK itself.

In any case, the response so far has been to include ARPACK in Octave, so
I'll go ahead with that.

Note that the patch from David does not solve the problem of sometimes
getting different results.  But as Rik has pointed out, "The ARPACK library is
an iterative method and it's results are absolutely going to depend on where
in the solution space the search begins".

Marco, do you always get the exact same value when you use eigs(A,1), or is
it sometimes a complex value?  If it is sometimes complex, then I think you
really should be getting a complex pair.

    _______________________________________________________

Reply to this item at:

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

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