Computing the clique number of tournaments
Guillaume Aubian
TL;DR
The paper defines the tournament clique-number parameter $\\overrightarrow{\\omega}(T)$ as the minimum, over all vertex orderings, of the clique number of the backedge graph $T^{\\prec}$, and proves that for any $k \ge 3$, deciding whether $\\overrightarrow{\\omega}(T) \le k$ is NP-complete. It introduces gadget constructions and a lifting technique (via a $T_k$-based framework and $\\Delta$ compositions) to encode a $3$-SAT instance, establishing NP-hardness, and thereby NP-completeness, of computing the clique number of tournaments. A key methodological contribution is a tool that preserves the clique-number while enabling systematic control of subgraph occurrences, facilitating the gadget-based reduction. The paper also provides a counterexample to a Gyárfás–Sumner-type conjecture in the directed setting by constructing a forest backedge graph tournament that is not contained in any $D_k$, despite having $\\overrightarrow{\\omega}=2$ and unbounded dichromatic complexity, highlighting limits of χ-boundedness analogues for tournaments. Together, these results settle the NP-completeness of the fixed-$k$ problem, sharpen the understanding of tournament cliques, and identify new directions in directed analogues of classical graph-theoretic conjectures.
Abstract
The clique number of a tournament is the maximum clique number of a graph formed by keeping backwards arcs in an ordering of its vertices. We study the time complexity of computing the clique number of a tournament and prove that, for any integer $k \geq 3$, deciding whether a tournament has clique number at most $k$ is NP-complete. This answers an interrogation of Nguyen, Scott and Seymour. To do so, we make use of a construction which we then modify to provide a counterexample to a conjecture of Aboulker, Aubian, Charbit and Lopes.
