Determining whether a parameterized problem is kernelizable and has a small kernel size has recently become one of the most interesting topics of research in the area of parameterized complexity and algorithms. Theoretically, it has been proved that a parameterized problem is kernelizable if and only if it is fixed-parameter tractable. Practically, applying a data reduction algorithm to reduce an instance of a parameterized problem to an equivalent smaller instance (i.e., a kernel) has led to very efficient algorithms and now goes hand-in-hand with the design of practical algorithms for solving $\mathcal{NP}$-hard problems. Well-known examples of such parameterized problems include the vertex cover problem, which is kernelizable to a kernel of size bounded by $2k$, and the planar dominating set problem, which is kernelizable to a kernel of size bounded by $335k$. In this paper we develop new techniques to derive upper and lower bounds on the kernel size for certain parameterized problems. In terms of our lower bound results, we show, for example, that unless $\mathcal{P} = \mathcal{NP}$, planar vertex cover does not have a problem kernel of size smaller than $4k/3$, and planar independent set and planar dominating set do not have kernels of size smaller than $2k$. In terms of our upper bound results, we further reduce the upper bound on the kernel size for the planar dominating set problem to $67 k$, improving significantly the $335 k$ previous upper bound given by Alber, Fellows, and Niedermeier [J. ACM, 51 (2004), pp. 363–384]. This latter result is obtained by introducing a new set of reduction and coloring rules, which allows the derivation of nice combinatorial properties in the kernelized graph leading to a tighter bound on the size of the kernel. The paper also shows how this improved upper bound yields a simple and competitive algorithm for the planar dominating set problem.

The k -Leaf Out-Branching problem is to find an out-branching, that is a rooted oriented spanning tree, with at least k leaves in a given digraph. The problem has recently received much attention from the viewpoint of parameterized algorithms. Here, we take a kernelization based approach to the k -Leaf-Out-Branching problem. We give the first polynomial kernel for Rooted k -Leaf-Out-Branching, a variant of k -Leaf-Out-Branching where the root of the tree searched for is also a part of the input. Our kernel with O ( k ³ ) vertices is obtained using extremal combinatorics. For the k -Leaf-Out-Branching problem, we show that no polynomial-sized kernel is possible unless coNP is in NP/poly . However, our positive results for Rooted k -Leaf-Out-Branching immediately imply that the seemingly intractable k -Leaf-Out-Branching problem admits a data reduction to n independent polynomial-sized kernels. These two results, tractability and intractability side by side, are the first ones separating Karp kernelization from Turing kernelization . This answers affirmatively an open problem regarding “cheat kernelization” raised by Mike Fellows and Jiong Guo independently.

Let G = (V,E) be a graph of order n and let B(S) be the set of vertices in V\S that have a neighbor in the vertex set S. The differential of a vertex set S is defined as ?(S) = |B(S)| - |S| and the maximum value of ?(S) for any subset S of V is the differential of G. A Roman dominating function of G is a function f : V ? {0,1,2} such that every vertex u with f(u) = 0 is adjacent to a vertex v with f(v) = 2. The weight of a Roman dominating function is the value f(V) = ?u?V f(u). The minimum weight of a Roman dominating function of a graph G is the Roman domination number of G, written ?R(G). We prove that ?R(G) = n - ?(G) and present several combinatorial, algorithmic and complexity-theoretic consequences thereof.