[Couenne] Errors while solving a (convex quadratic) model

Pietro Belotti pbelott at clemson.edu
Sat Feb 11 09:19:19 EST 2012


David,

it was indeed a subtle bug in the implied bound processing of powers of 
the form y >= x^k. It is now fixed in both trunk and stable/0.4, and it 
also in the case of expressions of the form y >= x*x. This behavior should 
not occur in previous versions as the bug was in a feature introduced with 
stable/0.4.

Regards,
Pietro

--
Pietro Belotti
Dept. of Mathematical Sciences
Clemson University
email: pbelott at clemson.edu
phone: 864-656-6765
web:   http://myweb.clemson.edu/~pbelott

On Fri, 10 Feb 2012, David Papp wrote:

> Hi,
> 
> I ran into a problem using Couenne (with AMPL), getting either the wrong 
> answer (problem is infeasible), or an exception, depending on the 
> version I use. I tried both the precompiled binaries, the latest version 
> compiled on my computer, and also the version on NEOS, all with default 
> settings. The minimal example is really simple, the AMPL model file is:
> 
> var X;
> var Y;
> minimize obj: X;
> subject to c: X-Y^2 >= 0;
> 
> I also get different wrong results if I replace Y^2 by Y*Y. (Some 
> versions crash on one, and "solve" the other, claiming it's infeasible.) 
> The problems seem to be on the solver side, as the same model files work 
> with Ipopt. The original example was non-convex, which is why I'm trying 
> to get Couenne work.
> 
> Thanks,
> David
> 
> 
>


More information about the Couenne mailing list