I have a noconvex, nonlinear model I’m attempting to solve repeatedly
using KNITRO at different random starting points. I’m collecting data
in each run on objective function, number of infeasibilities in the
solution, model solution status, solver termination conditions, and
run time… just getting familiar with my model and how it behaves.
I have some runs which teriminate to what GAMS says is a local optimal
solution (Modelstat = 2) and a normal termination (Solvestat = 1), but
which has a positive number of infeasibilities at the solution
(Numinfes > 0). I’m trying to figure out how to interpret that. Do
those reported infeasibiilties represent constraints that aren’t quite
satisfied, but which nonetheless meet the feasibility tolerances
specified? That seems like the obvious interpretation, but I just
wanted to confirm. What exactly does it mean when an infeasibility is
reported in Numinfes?
If that interpretation is correct, then that brings up another
question. In most of my runs, there are no reported infeasibilities.
But I have a number of equality constraints. Does the lack of
reported infeasibilities really mean that my equality constraints are
satisfied perfectly? That seems weird, but maybe possible. In some
other nonlinear solvers I’ve tried, I always had small residuals on my
equality constraints (which was fine because they were within the
tolerances I had specified).
Finally, I have some runs which terminate to a point which GAMS
reports is “feasible but not yet optimal” (Modelstat = 6) and where
the solver “quit with a problem (see LST file)” (Solvestat = 4). I
don’t see any interesting details in my LST file to enlighten me on
the error hinted at in the Solvestat status. But the LST file reports
a model status at termiation of “6 Intermediate infeasible.”
Infeasible? Actually, I’m guessing it has to be infeasible (sounds
too good to be true that there is this magical high objective function
point out in the middle of nowhere), but just want to confirm, given
the apparent Solvestat and LST report conflict. Is there maybe a
disconnect between the report statuses of KNITRO and GAMS?
I’ve asked a few other people about these issues, and they didn’t know
but were very interested in the answer too. So any help is much
appreciated!
Thanks,
Emily
\
I’d still appreciate any insight on this question.
Thanks,
Emily
On Sep 10, 5:08 pm, emily wrote:
I have a noconvex, nonlinear model I’m attempting to solve repeatedly
using KNITRO at different random starting points. I’m collecting data
in each run on objective function, number of infeasibilities in the
solution, model solution status, solver termination conditions, and
run time… just getting familiar with my model and how it behaves.
I have some runs which teriminate to what GAMS says is a local optimal
solution (Modelstat = 2) and a normal termination (Solvestat = 1), but
which has a positive number of infeasibilities at the solution
(Numinfes > 0). I’m trying to figure out how to interpret that. Do
those reported infeasibiilties represent constraints that aren’t quite
satisfied, but which nonetheless meet the feasibility tolerances
specified? That seems like the obvious interpretation, but I just
wanted to confirm. What exactly does it mean when an infeasibility is
reported in Numinfes?
If that interpretation is correct, then that brings up another
question. In most of my runs, there are no reported infeasibilities.
But I have a number of equality constraints. Does the lack of
reported infeasibilities really mean that my equality constraints are
satisfied perfectly? That seems weird, but maybe possible. In some
other nonlinear solvers I’ve tried, I always had small residuals on my
equality constraints (which was fine because they were within the
tolerances I had specified).
Finally, I have some runs which terminate to a point which GAMS
reports is “feasible but not yet optimal” (Modelstat = 6) and where
the solver “quit with a problem (see LST file)” (Solvestat = 4). I
don’t see any interesting details in my LST file to enlighten me on
the error hinted at in the Solvestat status. But the LST file reports
a model status at termiation of “6 Intermediate infeasible.”
Infeasible? Actually, I’m guessing it has to be infeasible (sounds
too good to be true that there is this magical high objective function
point out in the middle of nowhere), but just want to confirm, given
the apparent Solvestat and LST report conflict. Is there maybe a
disconnect between the report statuses of KNITRO and GAMS?
I’ve asked a few other people about these issues, and they didn’t know
but were very interested in the answer too. So any help is much
appreciated!
Thanks,
Emily
\