[Paper Review] On Directed Steiner Trees with Multiple Roots

Ondřej Suchý|arXiv (Cornell University)|Apr 18, 2016

Advanced Graph Theory Research23 references4 citations

TL;DR

This paper introduces the $q$-Root Steiner Tree ($q$-RST) problem, a generalization of Directed Steiner Tree and Strongly Connected Steiner Subgraph, and presents an FPT algorithm running in $O(2^{2q+4|T|} imes n^{2q+O(1)})$ time for the restricted case where $R \subseteq T$. It further extends this to planar graphs with a runtime of $2^{O(q\log q + |T|\log q)} \cdot n^{O(\sqrt{q})}$, achieving fixed-parameter tractability under key structural constraints.

ABSTRACT

We introduce a new Steiner-type problem for directed graphs named extsc{$q$-Root Steiner Tree}. Here one is given a directed graph $G=(V,A)$ and two subsets of its vertices, $R$ of size $q$ and $T$, and the task is to find a minimum size subgraph of $G$ that contains a path from each vertex of $R$ to each vertex of $T$. The special case of this problem with $q=1$ is the well known extsc{Directed Steiner Tree} problem, while the special case with $T=R$ is the extsc{Strongly Connected Steiner Subgraph} problem. We first show that the problem is W[1]-hard with respect to $|T|$ for any $q \ge 2$. Then we restrict ourselves to instances with $R \subseteq T$. Generalizing the methods of Feldman and Ruhl [SIAM J. Comput. 2006], we present an algorithm for this restriction with running time $O(2^{2q+4|T|}\cdot n^{2q+O(1)})$, i.e., this restriction is FPT with respect to $|T|$ for any constant $q$. We further show that we can, without significantly affecting the achievable running time, loosen the restriction to only requiring that in the solution there are a vertex $v$ and a path from each vertex of $R$ to $v$ and from $v$ to each vertex of~$T$. Finally, we use the methods of Chitnis et al. [SODA 2014] to show that the restricted version can be solved in planar graphs in $O(2^{O(q \log q+|T|\log q)}\cdot n^{O(\sqrt{q})})$ time.

Motivation & Objective

• To study the parameterized complexity of a new Steiner-type problem in directed graphs, $q$-Root Steiner Tree ($q$-RST), which generalizes both Directed Steiner Tree and Strongly Connected Steiner Subgraph.
• To determine the parameterized complexity of $q$-RST with respect to key parameters: $|T|$ (number of terminals), $q$ (number of roots), and solution size.
• To identify structural constraints under which $q$-RST becomes fixed-parameter tractable (FPT), particularly when $R \subseteq T$ or when a common vertex exists that all paths can route through.
• To extend FPT algorithms from standard DST and SCSS to $q$-RST by leveraging game-theoretic acceleration and treewidth-based optimization.
• To investigate the complexity of $q$-RST in planar graphs, aiming to match the efficiency of known algorithms for DST and SCSS in such settings.

Proposed method

• Proposes a new problem, $q$-RST, defined as finding a minimum subgraph that provides a directed path from each root in $R$ to each terminal in $T$.
• Introduces a restricted variant where $R \subseteq T$, enabling the use of dynamic programming and tree decomposition techniques.
• Applies an accelerated game framework inspired by Chitnis et al. to model path construction as token movements, reducing the search space.
• Uses a labeling scheme for edges and vertices in a derived digraph $D_X$, assigning sets of $D$-tokens to track which roots and terminals are processed in each move.
• Employs cost functions $cv$ and $ce$ to compute the minimum cost of supermoves (e.g., MultipleFlips), where $ce$ is redefined using a shortest path oracle $ST(x,Y)$ to account for path overlaps.
• Applies the Klein and Marx algorithm for minimum-cost mapping to a treewidth-bounded underlying graph, achieving $n^{O(\sqrt{q})}$ time via dynamic programming on tree decompositions.

Experimental results

Research questions

• RQ1 Is the $q$-RST problem fixed-parameter tractable with respect to $|T|$ when $q$ is constant, particularly under the constraint $R \subseteq T$?
• RQ2 Can the $q$-RST problem be solved efficiently in planar directed graphs, and what is the best possible parameterized runtime?
• RQ3 What structural property of the solution enables efficient algorithm design—specifically, does the existence of a common vertex through which all required paths factorize allow FPT algorithms?
• RQ4 Can the algorithmic techniques used for $q$-RST be extended to the unrestricted $q$-RST variant, or is it fundamentally harder?
• RQ5 What is the parameterized complexity of $q$-RST with respect to the number of nonterminals in the solution, and how does it compare to known hardness results?

Key findings

• The $q$-RST problem is W[1]-hard with respect to $|T|$ for any $q \geq 2$, indicating that the problem is unlikely to be FPT in general without structural restrictions.
• For the restricted case where $R \subseteq T$, the problem admits an FPT algorithm with running time $O(2^{2q+4|T|} \cdot n^{2q+O(1)})$, making it fixed-parameter tractable in $|T|$ for constant $q$.
• The restriction $R \subseteq T$ can be loosened to require only a common vertex $v$ such that all roots in $R$ can reach $v$ and $v$ can reach all terminals in $T$, without significantly increasing the runtime.
• In planar graphs, the restricted $q$-RST problem can be solved in $2^{O(q\log q + |T|\log q)} \cdot n^{O(\sqrt{q})}$ time, matching the efficiency of known algorithms for DST and SCSS in planar settings.
• The algorithm uses a novel combination of game-theoretic acceleration, token-based labeling, and dynamic programming on tree decompositions to achieve the best-known runtime for this class of problems.
• The correctness of the algorithm is grounded in the representability of the accelerated game and established lemmata from prior work on SCSS, ensuring the solution is both valid and optimal.