[Ipopt] Problems in models in C++

Kelly, Jeff (ON0F) jeff.kelly at honeywell.com
Wed Feb 9 07:49:11 EST 2011



Concerning the dense Jacobian you are seeing in the derivative checker,
my understanding is that IPOPT is not performing what is known as a
partitioning or graph-coloring problem (TOMS618 - DSM()) on the supplied
sparsity-pattern to only perturb variables in "groups".  Therefore the
it will look dense.


Also, the derivative checker only uses finite-difference derivative
evaluations which if your Jacobian values are 1's and -1's then you may
see more than expected truncation errors in the values.


I personally use the complex-step method to compute the derivative
values and I use the TOMS618 to group my variables so that fewer model
function calls are required.




From: ipopt-bounces at list.coin-or.org
[mailto:ipopt-bounces at list.coin-or.org] On Behalf Of Dan
Sent: Wednesday, February 09, 2011 4:57 AM
To: ipopt at list.coin-or.org
Subject: [Ipopt] Problems in models in C++


Dear all,


I have a problem with the C++ Interface. I am currently coding some
network models. I tried to model a small instance like the hs_071
example and it went well.


I coded my network in term of functions that my TNLP class will called
if needed.  

When I run it, it does not find a solution and it goes into a loop.I
caught this message run by IPOPT:

"WARNING : Problem in step computation; switching to emergency code."


In order to solve my problem, I tried several thing including the
derivative checker and two weird things appears:

*	my jacobian is not dense. I can not figure why. I checked at
almost step and I can not find why the derivative checker of my jacobian
is dense.
*	the value found by the derivative checker are eccentric. It says
mine are wrong while they are not. It is a adjacency matrix so it is
only 1 and -1.

If you have any ideas to solve my problem of ways to find my errors, I
would greatly appreciate.


Best Regards,



-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://list.coin-or.org/pipermail/ipopt/attachments/20110209/acb2f4b4/attachment.html 

More information about the Ipopt mailing list