Operations Research Stack Exchange is a question and answer site for operations research and analytics professionals, educators, and students. It only takes a minute to sign up.
Sign up to join this community
The problem is described as follows: considering $n$ variables which are continuous and bounded such that $$L_i \le x_i \le U_i\quad \forall i=1,2,\dots,n.$$
How can i define a set of mixed integer linear inequalities such that the feasible solution will be
$$y = \max(x_1,x_2,\dots,x_n).$$
I am not certain of my approach but this is what i have so far
constraint 1: $y \ge x_i$;
constraint 2: $y \ge x_i + (U_i-L_i)w$;
where $w\in\{0,1\}$
Constraint 1 is fine. It imposes $y \ge \max(x_1,x_2,\dots,x_n)$. For constraint 2, you need to reverse the inequality and also introduce binary variable $w_i$ (with the interpretation that $w_i=1$ implies that $x_i$ is the maximum): $$y \le x_i + M_i (1 - w_i).$$ We want the constraint to be redundant if $w_i = 0$, so take $M_i = \max(U_1,\dots,U_n) - L_i$.
Now also impose $$\sum_{i=1}^n w_i=1.$$ (Alternatively, you can relax this to $\ge 1$.)
If $n=2$, this formulation reduces to the one given here.
