Originally posted by: CPLEX12USER

Hi,
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: CPLEX12USER

Hi,

I have a LP model that is infeasible. Currently I am using Feasopt to find the cause of infeasibility (using concert). I use FeasoptMode as 5 for finding the infeasibility. I have not set any specific time limit (TiLim) for the model. The problem runs for very long time and never gives the infeasibilities in the model. However, when I change feasoptmode anything from (1-3), I get the results. I would like to know how much of the retrieved values will be optimal ie. I want the least number of infeasibilities (constraints and bounds) that could help me to change the constraints or bounds and make the model feasible.

The above problem happens for specific data set and not for all.

Note: On the same note, I tried to set TiLim to 2 minutes for the model. The model exists after giving me the infeasibilities.

Apologize for me previous post. Mistakenly pressed "Post Message" before typing the full contents.

regards
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

If your original problem is an LP then you have the following

• feasopt modes 0 and 1 will only solve LPs.
• feasopt modes 2 and 3 will solve MIPs that are usually more difficult (and therefore more time consuming) than LPs.
• feasopt modes 4 and 5 will solve quadratic programs that might be even harder (and therefore even more time consuming) than the MIPs.

So, it is not surprising that using mode 0-3 is faster than using mode 5.

Moreover, modes 1, 3 and 5 perform a two-stage approach. First they find the minimal relaxation and then the optimal solution for the model to which the minimal relaxation was applied. Do you need that optimal solution constructed in the second phase? If not, you can safe time by restricting yourself to modes 0, 2 and 4.

If you terminate feasopt prematurely (for example by using a time limit) you may get relaxation values from CPLEX but you have no guarantee that they are minimal or that the model is indeed infeasible.

As far as I understand you want the minimal number of bounds/constraints that you must change to render your model feasible? That would be mode 2.

Did you consider the conflict refiner? This will produce a minimal set of constraints that are in conflict. Sometimes this makes it easier to detect modeling errors for example.
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: CPLEX12USER

Thanks a lot for the detailed explanation.

We require the minimal number of bounds/conflicts that could correct the model. With that as the objective, will feasoptMode of 5 would be better than that of feasoptMode of 2. If the quality of problem would be same as 5, I can try with feasoptmode of 2 itself.
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

If you are only interested in the minimal number of changes required then use feasopt modes 2 or 4.
Only if you want to have a solution that is feasible/optimal with the minimal number of changes applied use modes 3 or 5.
As far as I understand your problem you should be using modes 2 or 4, i.e. only due "phase 1" of feasopt.
#CPLEXOptimizers
#DecisionOptimization