[Ipopt] infeasible starting point

[Stefan Vigerske]

Hi,

Ipopt does not need a feasible solution as starting point.
Ipopt judges a model as feasible if it finds a feasible point.

Usually, Ipopt should converge in much less than 20000 iterations. If it 
does not do so, then maybe your problem is badly scaled or violates some 
constraint qualifications.
If you implemented your own Ipopt interface, then wrong gradient or 
Hessian information is also a popular source for convergence problems. 
Use the derivative checker (see docu) to check this.

Stefan

On 07/31/2013 12:57 AM, Chunhua Men wrote:
> Hi,
>
> I am new to IPOPT. I need to solve a constrained NLP problem, but I don't
> have a feasible initial solution.
>
> IPOPT ran a very long time (I set the max iteration as 20,000) and could
> not find a feasible solution (the "inf_pr" is still around 1000 after
> 20,000 iterations).
>
> Then I changed "start _with_resto" to "yes", then within 20000 iterations,
> still no feasible solution found.
>
> My model is not very big (~2000 decision variables with ~10 constraints).
>
>
> My question are:
> 1) Do I have to provide a feasible solution as the starting point?
> 2)  how does IPOPT judge a model  feasible or not?
>
> Thanks!
>
>
>
> _______________________________________________
> Ipopt mailing list
> Ipopt at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/ipopt
>