[Octave-bug-tracker] [bug #57986] qp() reports full convergence while actually vioalating lbound/ubound:

[John W. Eaton]

Follow-up Comment #1, bug #57986 (project octave):

It's usually better to attach code and data files or use a verbatim block to
avoid the bug tracker comment processing.  For example, the first column of
zeros in your H matrix were transformed to 1., 2., etc. (an enumerated list).

I'm attaching a single file that I think has the inputs the way you intended.

For me, with Octave 5.2.0 and a copy built from the current development
sources, it stops at the max number of iterations (200) with the result


INFO =

  scalar structure containing the fields:

    solveiter =  200
    info =  3

ans =

   0.33333
   0.33333

octave:2> X
X =

   0.00000
   0.00000
   0.33333
  -0.00000
   0.00000
   0.33333
   0.29246
   0.00000
   0.00000
   0.00000
   0.00000
  -0.00000
   0.04087
   0.00000
   0.00000


What lapack and blas libraries (and versions) are you using?

As we discussed on IRC, I think this is a bad problem because there are
decision variables that have no effect on the objective function.  I haven't
tried to trace the iterations, but I suspect QP is cycling around the solution
it has found until it hits the iteration limit.  Maybe we could do a better
job of detecting problems like this and issue a meaningful warning?

(file #48582)
    _______________________________________________________

Additional Item Attachment:

File name: qptst.m                        Size:2 KB
    <https://savannah.gnu.org/file/qptst.m?file_id=48582>



    _______________________________________________________

Reply to this item at:

  <https://savannah.gnu.org/bugs/?57986>

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