Originally posted by: SystemAdmin
[prubin said:]
I don't think you can, since for one thing it results in an improperly posed optimization model. You want a convex feasible region (or, for discrete problems, a continuous relaxation with a convex feasible region), and A B does not result in one. For a continuous problem, you need something like
A - B >= z*epsilon - (1-z)*M
and
B - A >= (1-z)*epsilon - z*M
where z is a 0-1 variable, M is a sufficiently large positive parameter (an a priori bound on |A-B|), and epsilon is a small but strictly positive parameter. If A and B are integers, then epsilon = 1.
/Paul
#DecisionOptimization#MathematicalProgramming-General