[Octave-bug-tracker] [bug #31731] Test failed for sqp

[Rik]

Follow-up Comment #5, bug #31731 (project octave):


with 
[x, obj, info, iter, nf, lambda] = sqp (x0, @phi, @g,
[],-realmax,realmax,100,eps);

x =

  -1.71714348089298e+00
   1.59570958654167e+00
   1.82724591922751e+00
  -7.63643091793199e-01
  -7.63643084476455e-01

max(abs(x-x_opt))

ans =  3.91303500713036e-08

So, it is better than when using the default of sqrt(eps).  I re-ran the
command with sqrt(eps) as the tolerance and it reproduces exactly the same
result as the shorter form of the command where the bounds, iterations, and
tolerance were not specified.

As a side note, what is the return value of info?  Is it 101 which indicates
success?  Even if it is 101, there are two ways to get that return value. 
Either, all the constraints are met which is tested at line 428.  OR, the
stepsize has become too small which is tested at line 483.  Maybe try changing
line 484 to "info = 104;" and re-running the basic test.  This would help
isolate why the algorithm is stopping.

    _______________________________________________________

Reply to this item at:

  <http://savannah.gnu.org/bugs/?31731>

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