First of all, I would like to apologize if this question does not fit into the "soft" category. I am quite a newbie around here, and maybe I can fail to get the feeling of what exactly is a "soft" question.
I have a problem with a business model we are trying to optimize. We are using LINGO for the task. The problem is quadratic in its constraints, and is quite large (~22000 variables and ~12000 constraints).
We want to do a sensitivity analysis, so we first solve the problem as a quadratic one, and then, in a second step, fix the values corresponding to the quadratic variables and constraints to its optimal values and then execute a sensitivity analysis on the resulting linear problem.
We are having some problems with the sensitivity analysis, as we have found that sometimes the "allowable increase" and "allowable decrease" values of the constraints optimal ranges appear somewhat interchanged.
My questions are: May the procedure of fixing the non linear parts of the problem and then solving the LP cause the posterior sensitivity analysis to fail? Could it be more related to the degenerate nature of the resulting LP? As far as I have seen, degeneracy can make commercial solvers to give partial information on optimal ranges, but I have not found any mention to interchanged increase and decrease values.
Any hint or suggestion would be greatly appreciated.
Thanks in advance.