Better Algorithms and Bounds for Directed Maximum Leaf Problems

The {\sc Directed Maximum Leaf Out-Branching} problem is to find an out-branching (i.e. a rooted oriented spanning tree) in a given digraph with the maximum number of leaves. In this paper, we improve known parameterized algorithms and combinatorial bounds on the number of leaves in out-branchings. We show that \begin{itemize} \item every strongly connected digraph $D$ of order $n$ with minimum in-degree at least 3 has an out-branching with at least $(n/4)^{1/3}-1$ leaves; \item if a strongly connected digraph $D$ does not contain an out-branching with $k$ leaves, then the pathwidth of its underlying graph is $O(k\log k)$; \item it can be decided in time $2^{O(k\log^2 k)}\cdot n^{O(1)}$ whether a strongly connected digraph on $n$ vertices has an out-branching with at least $k$ leaves. \end{itemize} All improvements use properties of extremal structures obtained after applying local search and of some out-branching decompositions.

💡 Research Summary

The paper addresses the Directed Maximum Leaf Out‑Branching (DMLOB) problem, which asks for an out‑branching (a rooted spanning arborescence) in a directed graph that maximizes the number of leaves. While previous work has offered approximation algorithms and fixed‑parameter tractable (FPT) procedures with running times roughly 2^{O(k log k)}·n^{O(1)}, the authors obtain three new results that tighten both combinatorial bounds and algorithmic complexity.

First, they consider strongly connected digraphs D of order n with minimum in‑degree δ⁻(D) ≥ 3. By applying a local‑search framework to an “extremal” out‑branching (one that cannot be improved by a single local operation), they prove that such a digraph always contains an out‑branching with at least (n/4)^{1/3} − 1 leaves. The proof hinges on the observation that if an out‑branching has too few leaves, many internal vertices must have three distinct incoming arcs, which forces a structural contradiction under the strong‑connectivity assumption. This yields a polynomial‑in‑n lower bound that improves earlier √n‑type estimates for the same degree condition.

Second, they examine the complementary situation: a strongly connected digraph D that does not admit an out‑branching with k leaves. They introduce a novel “out‑branching decomposition,” which partitions the underlying undirected graph G(D) into blocks aligned with a candidate out‑branching. By analyzing the interaction between blocks, they show that G(D) has pathwidth O(k log k). This result refines earlier O(k²) or O(k log k) bounds and demonstrates that the absence of a large‑leaf out‑branching forces the graph to be “thin” in the sense of having a small pathwidth.

Third, leveraging the pathwidth bound, they design an FPT algorithm that decides whether a strongly connected digraph on n vertices possesses an out‑branching with at least k leaves. The algorithm proceeds as follows: (i) compute a path decomposition of width O(k log k) (possible because of the previous theorem); (ii) run a dynamic programming (DP) routine over the decomposition, where each DP state records the set of vertices already assigned to the partial out‑branching and the current leaf count. To keep the state space manageable, they incorporate the same local‑search improvements used in the extremal analysis, which prune redundant configurations. The number of DP states is bounded by 2^{O(k log k)}, and each transition incurs an additional O(log k) factor, leading to an overall running time of 2^{O(k log² k)}·n^{O(1)}.

The paper also discusses the methodological significance of combining extremal combinatorial arguments (via local search) with structural graph decompositions. The minimum‑in‑degree condition is crucial for the leaf‑lower‑bound; relaxing it to δ⁻(D) ≥ 2 remains an open challenge. Moreover, the out‑branching decomposition technique may be adaptable to other directed‑graph optimization problems such as maximum directed path or minimum feedback arc set.

In summary, the authors deliver (1) a new polynomial leaf‑lower bound for strongly connected digraphs with δ⁻ ≥ 3, (2) a tight O(k log k) pathwidth bound for digraphs lacking a k‑leaf out‑branching, and (3) an improved FPT algorithm with running time 2^{O(k log² k)}·n^{O(1)}. These contributions advance both the theoretical understanding of directed spanning trees and the practical tractability of the DMLOB problem.