[Ipopt] Questions about IPOPT

[Andreas Waechter]

Hi,

There is also a new (inefficient) option to compute first derivatives for 
the constraints by finite differences.  The options are

jacobian_approximation        ("exact")
    Specifies technique to compute constraint Jacobian
    Possible values:
     - exact                   [user-provided derivatives]
     - finite-difference-values [user-provided structure, values by finite
                                differences]

findiff_perturbation                   0 <  (      1e-07) <  +inf
    Size of the finite difference perturbation for derivative approximation.
      This determines the relative perturbation of the variable entries.

You will still have to provide the sparcity structure of the constraint 
Jacobian.

Andreas


On Sun, 27 Jul 2008, Stefan Vigerske wrote:

> Hi,
>
>> it seems that IPOPT need the information about the gradient, Jaccobian
>> matrix and Hessian matrix of the optimization problem.
>> but it my problem is way too large to get the explicit matrices, is the
>> IPOPT going to work properly if I set all the matrix to be zero?
>
> No.
> You can let it approximate the hessians, but it needs at least exact
> first derivatives.
> Maybe the following can be helpful for you:
> http://www.coin-or.org/CppAD/Doc/ipopt_cppad.cpp.htm
>
> Best,
> Stefan
> _______________________________________________
> Ipopt mailing list
> Ipopt at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/ipopt
>