[Coin-ipopt] Opt. with only first derivatives

[Stefan Vigerske]

Hi,

> But, has the IPOPT the first order methods that don't need essentially
> Hessian?

I'm not sure if I understand your question right.
IPOPT is using Hessian information in its algorithm.
But it does not require the user to provide the Hessian. If the user do
not give Hessian information, IPOPT can approximate the Hessian by a
quasi-newton method. For this approximation, only derivatives are needed.
The approximation might be not as good as the original Hessian, but at
least it allows you to skip the implementation of 2nd derivatives.

I don't know where your optimization problem comes from. If it is
something static, I mean not something that is dynamically generated as
a subproblem from somewhere else, you can also think about formulating
your problem in AMPL, GAMS, or MATLAB, and use the corresponding
Ipopt-interfaces. Then you do not even need to implement first derivatives.

Stefan

> 
> Regards, ... rt
> 
> 
> On 2/24/07, Stefan Vigerske <stefan at vigerske.de> wrote:
>>
>> Hi,
>>
>> > If we have only first derivative of Objective function and Consts., is
>> > IPOPT
>> > suitable choice as optimization module?
>> >
>> > How we can exploit IPOPT with such problems? (in the presented examples
>> in
>> > package, Hessian were taken by IPOPT).
>>
>> You can tell Ipopt to approximate the hessian so that you do not need to
>> implement 2nd derivates.
>> Please have a look at the Ipopt documentation:
>> http://www.coin-or.org/Ipopt/documentation/node49.html
>>
>> Stefan
>>
>