[Ipopt] Iterations number changed after using sparse matrix ma27, reasonable?

Andreas Waechter andreasw at watson.ibm.com
Fri Dec 11 10:59:50 EST 2009


If you have an ill-condtioned or otherwise numerially unstable problem, 
small differences in the floating point operations can lead to a different 
number of iterations (in the worst case, even different solutions) of 
Ipopt.  But maybe something else had changed.  If you wanted to go to the 
bottom of this, you could compare the output files for the two cases with 
a high enough print level.



On Fri, 4 Dec 2009, lqc234 wrote:

> Hi everyone, hi Andreas! I use Ipopt 3.6.1 in my application and it 
> works well.At the first,it iterates 12 times and optimal solution found. 
> However I want to use sparse matrix technology which ma27 supports in 
> jacobi and hessian to improve my application. The optimal solution also 
> be founded,but the iteration number changed to 19. Is this situation 
> reasonable? I have thought the results is the same whether or not using 
> the sparse matrix technique. Appreciate for any suggestion!
>  Thank you in advance!

More information about the Ipopt mailing list