[Coin-ipopt] Dense Hessian approximation in IPOPT?

[Andreas Waechter]

Hi Duraid,

Please send messages regarding Ipopt to coin-ipopt mailing list...

> 	I am a PhD student here in Tokyo, and I have been happily using
> IpOpt for some time now. Until now, I have always been making use of the
> limited-memory BFGS Hessian approximation code, but poking around today I
> discovered that there are mentions of an SR1 implementation also.

Glad to hear Ipopt is working for you!

> 	Well, I have just a quick question: is there any possibility that a
> *dense* BFGS or SR1 routine will be added to IpOpt in the near future?

Ipopt targets large-scale problems (thousands to millions of variables), 
and for those a dense quasi-Newton implementation would not make sense. 
(Well, if the nonlinearity is only in a few variables, and all other ones 
appear only linearly in the problem statement, it would actually me 
useful.)

It is not on my list of things to do for Ipopt to implement something like 
this.  Of course, if someone else writes it and would contribute it to the 
Ipopt project, there could be a dense quasi-Newtop option in the 
future.... ;-)

Cheers,

Andreas


>
> 	Many thanks in advance,
>
> 	Duraid Madina
> 	Tokyo University
>
>