Linear Kernels for Separating a Graph into Components of Bounded Size

Abstract: Graph separation and partitioning are fundamental problems that have been extensively studied both in theory and practice. The \textsc{$p$-Size Separator} problem, closely related to the \textsc{Balanced Separator} problem, is to check whether we can delete at most $k$ vertices in a given graph $G$ such that each connected component of the remaining graph has at most $p$ vertices. This problem is NP-hard for each fixed integer $p\geq 1$ and it becomes the famous \textsc{Vertex Cover} problem when $p=1$. It is known that the problem with parameter $k$ is W[1]-hard for unfixed $p$. In this paper, we prove a kernel of $O(pk)$ vertices for this problem, i.e., a linear vertex kernel for each fixed $p \geq 1$. In fact, we first obtain an $O(p^2k)$ vertex kernel by using a nontrivial extension of the expansion lemma. Then we further reduce the kernel size to $O(pk)$ by using some `local adjustment' techniques. Our proofs are based on extremal combinatorial arguments and the main result can be regarded as a generalization of the Nemhauser and Trotter's theorem for the \textsc{Vertex Cover} problem. These techniques are possible to be used to improve kernel sizes for more problems, especially problems with kernelization algorithms based on techniques similar to the expansion lemma or crown decompositions.