coin-or
Potential calculation error by MibS
Dear MibS developer team, after checking the results for a particular problem instance multiple times, I believe that MibS may be producing an incorrect solution. Specifically, it seems that the lower-level problem is not being optimally solved for the returned upper-level solution, which results in a solution that is not bilevel-feasible.
I have attached the corresponding MPS and AUX files (both old and new format). In my assessment, the optimal objective value should be 32, but MibS returns 31.
Do you have any idea what might be causing this issue? If necessary, I can also provide the mathematical model as a PDF along with the parameter values.
Kind regards Julius
modell_auxNewFormat.txt modell_mps.txt
I was able to replicate the optimal solution value of 31. Indeed, this looks like it could be a problem with the sub-solver. With a version of MibS I built last year that uses CPLEX, I got the expected result of 32, but even switching to using SYMPHONY with that version, the result is still 32. We'll see if we can track down what is going on.
Thank you very much for checking. This insight already helps me a lot.
I was able to replicate the optimal solution value of 31. Indeed, this looks like it could be a problem with the sub-solver. With a version of MibS I built last year that uses CPLEX, I got the expected result of 32, but even switching to using SYMPHONY with that version, the result is still 32. We'll see if we can track down what is going on.
We've been able to isolate the issue, but it's probably a very subtle numerical issue that will be difficult to track down. We've extracted the MPS file of the lower-level subproblem that is incorrectly solved and when solved with SYMPHONY on the command-line, it solves correctly using a variety of parameter settings. So this may take come digging. There doesn't seem to be an obvious "smoking gun."