Polynomial Kernels for lambda-extendible Properties Parameterized Above the Poljak-Turzik Bound

Poljak and Turzík (Discrete Mathematics 1986) introduced the notion of λ-extendible properties of graphs as a generalization of the property of being bipartite. They showed that for any 0 < λ < 1 and λ-extendible property Π, any connected graph G on n vertices and m edges contains a spanning subgraph H ∈ Π with at least λm+ 1−λ 2 (n−1) edges. The property of being bipartite is λ-extendible for λ = 1/2, and so the Poljak-Turzík bound generalizes the well-known Edwards-Erdős bound for Max-Cut. Other examples of λ-extendible properties include: being an acyclic oriented graph, a balanced signed graph, or a q-colorable graph for some q ∈ N. Mnich et. al. (FSTTCS 2012) defined the closely related notion of strong λ-extendibility. They showed that the problem of finding a subgraph satisfying a given strongly λ-extendible property Π is fixed-parameter tractable (FPT) when parameterized above the Poljak-Turzík bound—does there exist a spanning subgraph H of a connected graph G such that H ∈ Π and H has at least λm+ 1−λ 2 (n− 1) + k edges?—subject to the condition that the problem is FPT on a certain simple class of graphs called almost-forests of cliques. This generalized an earlier result of Crowston et al. (ICALP 2012) for Max-Cut, to all strongly λ-extendible properties which satisfy the additional criterion. In this paper we settle the kernelization complexity of nearly all problems parameterized above Poljak-Turzík bounds, in the affirmative. We show that these problems admit quadratic kernels (cubic when λ = 1/2), without using the assumption that the problem is FPT on almost-forests of cliques. Thus our results not only remove the technical condition of being FPT on almost-forests of cliques from previous results, but also unify and extend previously known kernelization results in this direction. Our results add to the select list of generic kernelization results known in the literature.