[Ipopt] infeasible problem

[Ali Baharev]

Dear Andreas,

Thank you for the reply.

The attached two models differ only in the starting point. When the
prob2.mod (it was named infeas.mod earlier because Minos stated it was
infeasible) is solved Ipopt states that the solution is:

ampl: display x1, x2, x3, lambda;
x1 = 0.143564
x2 = 0.049858
x3 = 0.806578
lambda = 0.0100001

This is not the solution. (Otherwise, lambda is at its lower bound.)

In prob3.mod i supplied the solver with the true solution vector as
initial point, and now it concludes that it is indeed a solution:

ampl: display x1, x2, x3, lambda;
x1 = 0.116727
x2 = 0.0370726
x3 = 0.8462
lambda = 0.171504

> that option).  This final violation of the constraints is very small, and
> within the typical tolerances of NLP solvers and therefore declared
> feasible.

My question is: how can i make Ipopt to find the true solution and not
to stop at the approximate (incorrect) solution?

It seemed to me from the beginning that this numerically troublesome
problem confuses the solver when checking the stopping criteria.

Thank you for your help,

Ali
-------------- next part --------------
A non-text attachment was scrubbed...
Name: prob2.mod
Type: video/mpeg
Size: 5273 bytes
Desc: not available
Url : http://list.coin-or.org/pipermail/ipopt/attachments/20081119/309c1b60/attachment.m1v 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: prob3.mod
Type: video/mpeg
Size: 6426 bytes
Desc: not available
Url : http://list.coin-or.org/pipermail/ipopt/attachments/20081119/309c1b60/attachment-0001.m1v