[Coin-ipopt] Feasibility

[Andreas Waechter]

Hi Christophe

It is hard to say what exactly the problem is, but here is a try:

What seems to happen is that Ipopt is converging to a feasible point, but
then it has trouble to continue.  Therefore it wants to switch to the
feasibility restoration phase, but the current iterate is already
feasible, there is no feasibility to restore.

Usually, this can happen when the derivative information is incorrect.
Also, you are using the quasi-Newton option, which is not so robust, but
it is funny that Ipopt quits already in the fifth iteration.  (Also, of
course, your objective and constraint functions must be smooth - if they
have nondifferentiable points like abs(x), Ipopt will get stuck)

Why are you not providing second derivatives, but use the quasi-Newton
option?  If it is because you are using some numerical methods to compute
the objective function or constraints, and only get first derivatives that
way, it might be that the first derivatives obtained that way are not very
exact.  If you actually have formulas for all functions,  I suggest to
implement second derivatives.

In your position, I would check if indeed all the derivatives are provided
to Ipopt correctly.  Also, one problem could always be that the problem is
very badly scaled.  I think ideally, the non-zero first partial
derivatives of the objective and constraint functions should typically be
on the order to 1-10 or so...

Good luck,

Andreas



On Tue, 25 Oct 2005, Ullrich Christophe wrote:

> Dear Andreas,
>
>
> I use the Fortran version of Ipopt to solve a small problem of 22
> variables and 16 constraints. I use the following options :
> 	Dtol = 1d-3
> 	Iquasi = 6
> Ipopt returns the error code 18 which is “Ipopt can not improve the
> feasibility of the problem”. I don’t understand where the problem is.
> In the output file, I notice that the objective function at the initial
> point is 0.3d-1 and the constraints are around 0.1d-2. But at the 5th
> iteration the error message appears and the objective function is 0.18 and
> the constraints about 1d-14. And the message “Restoration started with
> CNRM =   8.570921750106208E-014” appears. I have already tried different
> parameters but I was not able to solve my problem. What can I do?
>
> I can forward you the output file of the optimization.
>
>
> Thank you very much,
>
> Christophe
>
>
> _______________________________________________
> Coin-ipopt mailing list
> Coin-ipopt at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/coin-ipopt
>