# Expressing a chain of boolean ORs using ILP involving different variables

How can I express a chain of OR operations in an ILP, given that each operand is an inequality between two binary variables? I have asked a similar question here: Chain of Boolean ORs. In that question, I asked specifically about the scenario where the variable on the left-hand sides is always the same.

For this question, I am interested in the more general case. For example, $$(x_1 \leq x_3) \textrm{ OR } (x_2 \leq x_3) \textrm{ OR } (x_1 \leq x_4),$$ in which the left-hand sides nor the right-hand sides are always the same.

A boolean OR of just two variables can be expressed by introducing an additional variable and modifying the constraints appropriately. But how can we model the general case described above?

• Can we assume the following generic form: $x_1 \le x_2$ OR $x_3 \le x_4$ OR $\dots$ OR $x_{n-1} \le x_n$, with $x_1, \dots, x_n$ all binary? You can consider editing your question to add this if this is the case, or to add a general form that does work for you if not. Aug 20 '19 at 8:29
• No, there maybe repetitions of variables, however the repetition is restricted only on the LHS or on the RHS, I have edited the example to include such a scenario @KevinDalmeijer Aug 20 '19 at 10:19
• Thank you for the clarification. I also did an edit to make sure that your question is understood correctly. If you disagree, feel free to roll back my edit, or edit your question further. Aug 20 '19 at 10:58

Let $$P$$ be the set of $$(i,j)$$ pairs. Here’s a derivation via conjunctive normal form: $$$$\bigvee_{(i,j)\in P} \left(x_i \implies x_j\right) \\ \bigvee_{(i,j)\in P} \left(\neg x_i \vee x_j\right) \\ \sum_{(i,j)\in P} \left(1-x_i + x_j\right) \ge 1 \\ \sum_{(i,j)\in P} (x_i - x_j) \le |P| - 1$$$$ For your example, this yields $$2x_1+x_2 - 2x_3 - x_4 \le 2.$$