[Octave-bug-tracker] [bug #56582] [octave forge] (optim) lsqlin produces unexpected error on linearly dependent inputs

[Olaf Till]

Follow-up Comment #2, bug #56582 (project octave):

The original optimization problem has no single solution. According to the
reported output, Matlab indicates success with the exit flag but returns NaNs
as solution. For a similar problem, one element of the solution is not NaN,
according to the report. None of these outputs seems right. The exit flag is
supposed to indicate such problems. Returning NaNs is not necessary then.
Returning a single solution out of a solution space with many solutions is not
usually done by such functions, and if it is done, it makes no sense to set
some elements to NaN. User code relying on some solution elements being NaN
and some not is probably buggy.

The patch
https://sourceforge.net/p/octave/optim/ci/b29564bac152d65d3443b6da2b45f44d55b0b0f5/
sets an error exit flag for the condition previously causing the reported
error message.

The problem with different behaviour of Matlab for rank-deficient equality
constraint matrices is unrelated.

    _______________________________________________________

Reply to this item at:

  <https://savannah.gnu.org/bugs/?56582>

_______________________________________________
  Message sent via Savannah
  https://savannah.gnu.org/