[Ipopt] ipopt for linear filter coefficients

[Andreas Waechter]

Dear Tilman,

In the future, please send questions regarding Ipopt to the mailing list.

> I tried using IpOpt for fitting linear filter coefficients with some unity root
> constraints and I wonder that IpOpt does not converge.
> Can you imagine, that such problems are incompatible with IpOpts
> search strategy?  Or do you know people who apply IpOpt for digital filtering problems?
> It works only if my problem is one-dimensional, i have played
> around with all options and formulation of the constraints, and that was a lot work - but with no success.

I don't know the particular problem you mention, but maybe someone else 
on the mailing list does.

On immediate thing to verify is that your functions are smooth, ie., at 
least the functions and their first derivatives should be continuous.  If 
that is not the case, Ipopt (or any nonlinear optimization method relying 
on smooth functions) is not the method to use.

Secondly, if the derivative information given to Ipopt is incorrect, 
the algorithms also easily fails.  There is a simple derivative checker 
implemented that you can use (see the documentation).

If this doesn't help, you can post another message to this mailing list 
(you will have to subscribe first - we avoid spam that way).  In that 
case, please describe the problem formulation (formulas?) in more detail. 
Also, are you using AMPL or your own code?  What is the output?

Please don't send very large files to the mailing list.  Alternatively, 
you could file a ticket at the Ipopt trac page 
(projects.coin-or.org/Ipopt) where you can attach large files.

Regards,

Andreas