Re: sqp behavior for functions with small values

[Juan Pablo Carbajal]

Hi Benjamin,

I am happy to hear that the -log solved this particular issue. The
problem you are facing is not a problem of sqp, but of qp.
If you inspect the results of line 412 of sqp, you see that the search
direction is 0,0, qp did 2 iterations and thinks it got a global
solution.
qp.m is just calling __qp__ defined in qp.cc [1]

in line 97 qp uses a fixed tolerance for the search given by sqrt(eps).

This is a bug in qp and one would need to pass the tolerance as a
parameter (or compute the correct tolerance for qp based on the
arguments of sqp). I am adding the author fo the __qp__.cc in CC to
see if he can fix this quickly. I also created a new bug report [2]
(thanks!).

[1]: http://octave.org/doxygen/4.3/da/dbd/____qp_____8cc_source.html
[2]: https://savannah.gnu.org/bugs/index.php?53506