Generalized Graph Packing Problems Parameterized by Treewidth

Abstract: $H$-Packing is the problem of finding a maximum number of vertex-disjoint copies of $H$ in a given graph $G$. $H$-Partition is the special case of finding a set of vertex-disjoint copies that cover each vertex of $G$ exactly once. Our goal is to study these problems and some generalizations on bounded-treewidth graphs. The case of $H$ being a triangle is well understood: given a tree decomposition of $G$ having treewidth $tw$, the $K_3$-Packing problem can be solved in time $2^{tw} \cdot n^{O(1)}$, while Lokshtanov et al.~[{\it ACM Transactions on Algorithms} 2018] showed, under the Strong Exponential-Time Hypothesis (SETH), that there is no $(2-\epsilon)^{tw}\cdot n^{O(1)}$ algorithm for any $\epsilon>0$ even for $K_3$-Partition. Similar results can be obtained for any other clique $K_d$ for $d\ge 3$. We provide generalizations in two directions: - We consider a generalization of the problem where every vertex can be used at most $c$ times for some $c\ge 1$. When $H$ is any clique $K_d$ with $d\ge 3$, then we give upper and lower bounds showing that the optimal running time increases to $(c+1)^{tw}\cdot n^{O(1)}$. We consider two variants depending on whether a copy of $H$ can be used multiple times in the packing. - If $H$ is not a clique, then the dependence of the running time on treewidth may not be even single exponential. Specifically, we show that if $H$ is any fixed graph where not every 2-connected component is a clique, then there is no $2^{o({tw}\log {tw})}\cdot n^{O(1)}$ algorithm for \textsc{$H$-Partition}, assuming the Exponential-Time Hypothesis (ETH).