[Coin-ipopt] Using Ipopt for special problems

[Andreas Waechter]

Hi,

My last response to your question was under the assumption that you are 
using the Matlab interface... If you are, then also the derivatives of the 
constraints can be estimated by the MEX interface (probably not very 
efficiently...)

Otherwise, the algorithm behind Ipopt assumes the EXACT irst derivatives 
are given.  You could try to write your own finite difference estimaion 
for the constraint Jacobian (if you can, you should not provide those 
elements that you know are always zero).  But using loose tolerances for 
the constraint evaluation to save time in early iterations would not give 
good estimates of the Jacobian, and Ipopt might fail to find a new 
iterate.

Of course, if your problem functions (for which you use the iterative 
procedure) are defined by some system of (nonlinear) equations and you are 
using some Newton-type method to solve those and get the values for the 
objective or constraints, you could just formulate your system so that it 
has this system of equations as constraints, and have Ipopt solve the 
nonlinear system defining your functions when it is optimizing, even if 
the problem becomes big.

Regards,

Andreas

On Tue, 11 Apr 2006 zwerhu at graduate.shu.edu.cn wrote:

> --===============2130531699==
> Content-type: text/html;
> 	charset="gb2312"
> Content-Transfer-Encoding: base64
>
> PGh0bWw+PGhlYWQ+PG1ldGEgaHR0cC1lcXVpdj0nQ29udGVudC1UeXBlJyBj
> b250ZW50PSd0ZXh0L2h0bWw7IGNoYXJzZXQ9Z2IyMzEyJz48c3R5bGUgdHlw
> ZT0ndGV4dC9jc3MnPjwhLS0uc3R5bGUzIHtjb2xvcjogIzAwNzJBOH0tLT48
> L3N0eWxlPjwvaGVhZD48Ym9keSBiZ2NvbG9yPScjRkZGRkZGJyB0b3BtYXJn
> aW49JzUnIGxlZnRtYXJnaW49JzUnIHJpZ2h0bWFyZ2luPSc1JyBib3R0b21t
> YXJnaW49JzUnIG1hcmdpbndpZHRoPScwJyBtYXJnaW5oZWlnaHQ9JzAnPjxQ
> PkhpLCBhbGw8L1A+DQo8UD5MYXN0IHRpbWUsIEkgYXNrIGEgcXVlc3Rpb24g
> Zm9yIHRoZSBvYmplY3RpdmUgZnVuY3Rpb25zIHRoYXQgaGF2ZSBubyBrbm93
> biBmb3Jtcy4gQnV0IGluIHNvbWUgY2FzZXMsIHRoZSBjb25zdHJhaW50cyBh
> bHNvIGNhbid0IGJlIHdyaXR0ZW4gZGlyZWN0bHkuIEkgbWVhbiB0aGUgY29u
> c3RyYWludHMgbXVzdCBiZSBleHByZXNzZWQgYnkgc29tZSBvdGhlciBpdGVy
> YXRpb25zIGp1c3QgYXMgbXkgbGFzdCBxdWVzdGlvbiB0aGF0IGFib3V0IG9i
> amVjdGl2ZSBmdW5jdGlvbnMuIFRoZXJlZm9yZSBldmFsdWF0aW9uIG9mIEph
> Y29iaWFuIG9mIGNvbnN0cmFpbnRzIGNhbid0IGJlIGdpdmVuLCB0b28uIEkg
> d29uZGVyIHdoZXRoZXIgdGhpcyBraW5kIG9mIE5MUCBxdWVzdGlvbnMgY2Fu
> IGJlIHNvbHZlZCBieSBJUE9QVC4gRm9yIElwb3B0IGlzIGJhc2VkIG9uIHRo
> ZSBpdGVyYXRpb24gbWV0aG9kLCBJIHRoaW5rLCBpZiB0aGUgaXRlcmF0aW9u
> IG9mIHRoZSBjb25zdHJhaW50cyBjYW4gYmUgY29tYmluZWQgaW50byB0aGUg
> aXRlcmF0aW9uIG9mIElwb3B0IGl0c2VsZiA/IFRoYW5rIHlvdS48L1A+PC9i
> b2R5PjwvaHRtbD4=
>
> --===============2130531699==
> Content-Type: text/plain; charset="us-ascii"
> MIME-Version: 1.0
> Content-Transfer-Encoding: 7bit
> Content-Disposition: inline
>
> _______________________________________________
> Coin-ipopt mailing list
> Coin-ipopt at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/coin-ipopt
>
> --===============2130531699==--
>