[Couenne] Debugging help with relation to CouenneCutGenerator::generateCuts

[Pete Janes]

Hi All,

Never mind, the previous email I have found the problem. I was generating
global cuts
when I was not supposed to (I was generating cuts for a non-convex curve).


Pete


On Thu, Apr 22, 2010 at 12:45 AM, Pete Janes <ppjanes at gmail.com> wrote:

> Hi All,
>
> I need some debugging help with Couenne. I am coding another specialized
> function
> for Couenne (The Two-Electron Repulsion Integral) which I plan to use for
> quantum
> chemistry problems.
>
> I am getting the wrong results, and I have narrowed the problem down to cut
> generation.
> Here is the background. In Cbc, the main function which processes a node is
> doOneNode().
> This step first solves the linear programming problem associated with the
> node given the
> new variable bounds, then if everything is still feasible it adds more cuts
> to the LP by calling
> CouenneCutGenerator::generateCuts.
>
> void CouenneCutGenerator::generateCuts (const OsiSolverInterface &si,
>                     OsiCuts &cs, const CglTreeInfo info)
>
> The problem I am having is not that my code generates the wrong cuts, but
> the initial cuts inside
> OsiCuts &cs are wrong. These cuts come from a prior cut generation step and
> it should NOT be
> included for the current step (if you can visualize an auxiliary expression
> as a curve, this prior cut
> cuts straight through it).
>
> Does anyone know the likely cause of this problem? I am sure this has been
> encountered before
> during debugging of Couenne.
>
> Thanks a lot in advance,
>
> Pete
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://list.coin-or.org/pipermail/couenne/attachments/20100422/4a3e8f54/attachment.html