Hardness of Hypergraph Edge Modification Problems

Abstract

For a fixed graph $F$, let $ex_F(G)$ denote the size of the largest $F$-free subgraph of $G$. Computing or estimating $ex_F(G)$ for various pairs $F,G$ is one of the central problems in extremal combinatorics. It is thus natural to ask how hard is it to compute this function. Motivated by an old problem of Yannakakis from the 80's, Alon, Shapira and Sudakov [ASS'09] proved that for every non-bipartite graph $F$, computing $ex_F(G)$ is NP-hard. Addressing a conjecture of Ailon and Alon (2007), we prove a hypergraph analogue of this theorem, showing that for every $k \geq 3$ and every non-$k$-partite $k$-graph $F$, computing $ex_F(G)$ is NP-hard. Furthermore, we conjecture that our hardness result can be extended to all $k$-graphs $F$ other than a matching of fixed size. If true, this would give a precise characterization of the $k$-graphs $F$ for which computing $ex_F(G)$ is NP-hard, since we also prove that when $F$ is a matching of fixed size, $ex_F(G)$ is computable in polynomial time. This last result can be considered an algorithmic version of the celebrated Erdős-Ko-Rado Theorem. The proof of [ASS'09] relied on a variety of tools from extremal graph theory, one of them being Turán's theorem. One of the main challenges we have to overcome in order to prove our hypergraph extension is the lack of a Turán-type theorem for $k$-graphs. To circumvent this, we develop a completely new graph theoretic approach for proving such hardness results.