[Coin-ipopt] Problems finding optimal solution to a rather small problem

[Stefan Vigerske]

Hi,

> i am working with IPOPT to solve a rather small optimization problem of
> the following form:
 > [...]
> I know that there is an optimal solution very close to the starting
> point where x3=X4=x5 holds (coded an algorithm which tests for all
> values of x1,x2,x3 in combination in the given bounds with a width of
> 0.01). Nevertheless the algorithm of IPOPT always converges to a point
> where x1=x2=x3 as a local solution holds. I tested it for different
> values of starting points and a variation of  [A B C] and always it runs
> for x1=x2=x3. Even if  i change the objective function it comes up with
> the same behaviour and always with an optimal solution. I add the
> ipopt.out.

Does the solution with x1=x2=x3 have a worse objective function value 
than the one you expect?

> Has anyone an idea of what i did not consider? Or is the problem
> statement wrong or in wrong formulation? Maybe some options i should
> use?

What come in my mind to try are:
- What happens when you add x3=x4=x5 as a constraint to the problem? 
Does it then find the solution that you would expect?
- I haven't checked your starting point. There are some options which 
make Ipopt push the starting point a bit away from the bounds or relax 
the bounds a bit before starting. This might move your starting point. 
You can consider to switch this behaviour off.

How do you access Ipopt? Per callable library, AMPL, or GAMS?

Best,
Stefan

-- 
Stefan Vigerske
Humboldt University Berlin, Numerical Mathematics
http://www.math.hu-berlin.de/~stefan