Parameterized Complexity of Vertex Deletion into Perfect Graph Classes

Abstract

Vertex deletion problems are at the heart of parameterized complexity. For a graph class  $\ensuremath{\mathcal{F}}$, the $\ensuremath{\mathcal{F}}$-Deletion problem takes as input a graph G and an integer k. The question is whether it is possible to delete at most k vertices from G such that the resulting graph belongs to $\ensuremath{\mathcal{F}}$. Whether Perfect Deletion is fixed-parameter tractable, and whether Chordal Deletion admits a polynomial kernel, when parameterized by k, have been stated as open questions in previous work. We show that Perfect Deletion (k) and Weakly Chordal Deletion (k) are W[2]-hard. In search of positive results, we study restricted variants such that the deleted vertices must be taken from a specified set X, which we parameterize by |X|. We show that for Perfect Deletion and Weakly Chordal Deletion, although this restriction immediately ensures fixed parameter tractability, it is not enough to yield polynomial kernels, unless NP ⊆ coNP/poly. On the positive side, for Chordal Deletion, the restriction enables us to obtain a kernel with $\mathcal{O}(|X|^4)$ vertices.

Supported by the Netherlands Organization for Scientific Research, project “KERNELS: Combinatorial Analysis of Data Reduction”, and by the Research Council of Norway, project “SCOPE: Exploiting Structure to Cope with Hard Problems”.