[Ipopt] objective value

[Andreas Waechter]

Hi Sakib,

> I am running a maximization version of IP on ipopt (3.9.2) through AMPL. 
> However I am getting negative values (while the optimum should be 
> positive). I was reading the documentation (i.e. 
> 09061.WaechterAndreas.Paper.2089.pdf).  It says for the minimization 
> version, Hessian regularization helps the solver to avoid maximum and 
> saddle points.  If I am not wrong, for the maximization version 
> similarly the minimum and saddle points would be avoided. I am wondering 
> if I am missing something.  I shall appreciate any suggestion/pointer.

The Ipopt algorithm always solves a minimization problem.  If you specify 
in the AMPL file that you want to maximize the objective function, the 
interface between AMPL and Ipopt will multiply the objective function from 
the AMPL model with -1 so that Ipopt can still solve a minimization 
problem.  Therefore, the output you see in Ipopt has the opposite sign of 
what you expect from your model description, and Ipopt will try to 
minimize that function.  Once Ipopt is done and the results are returned 
to AMPL, you will see the correct value of the objective within AMPL, 
e.g., using the "display" command.

I hope this explains what you observed.

Andreas