The Directed Steiner Tree (DST) problem is defined on a directed graph $G=(V,E)$, where we are given a designated root vertex $r$ and a set of $k$ terminals $K \subseteq V \setminus {r}$. The goal is to find a minimum-cost subgraph that provides directed $r \rightarrow t$ paths for all terminals $t \in K$. The approximability of DST has long been a central open problem in network design. Although there exist polylogarithmic-approximation algorithms with quasi-polynomial running times (Charikar et al. 1998; Grandoni, Laekhanukit, and Li 2019; Ghuge and Nagarajan 2020), the best-known polynomial-time approximation until now has remained at $k^\epsilon$ for any constant $\epsilon > 0$. Whether a polynomial-time algorithm achieving a polylogarithmic approximation exists has been a longstanding mystery. In this paper, we resolve this question by presenting a polynomial-time algorithm that achieves an $O(\log^3 k)$-approximation for DST on arbitrary directed graphs. This result nearly matches the state-of-the-art $O(\log^2 k / \log\log k)$ approximations known only via quasi-polynomial-time algorithms. The resulting gap -- $O(\log^3 k)$ versus $O(\log^2 k / \log\log k)$ -- mirrors the known complexity landscape for the Group Steiner Tree problem. This parallel suggests intriguing new directions: Is there a hardness result that provably separates the power of polynomial-time and quasi-polynomial-time algorithms for DST?
翻译:有向斯坦纳树(DST)问题定义在一个有向图 $G=(V,E)$ 上,其中给定一个指定的根顶点 $r$ 和一个包含 $k$ 个终端的集合 $K \subseteq V \setminus {r}$。目标是找到一个最小成本的子图,该子图为所有终端 $t \in K$ 提供有向的 $r \rightarrow t$ 路径。DST 的近似性长期以来一直是网络设计中的一个核心开放问题。尽管存在具有拟多项式运行时间的多对数近似算法(Charikar 等人,1998;Grandoni、Laekhanukit 和 Li,2019;Ghuge 和 Nagarajan,2020),但迄今为止已知的最佳多项式时间近似比对于任意常数 $\epsilon > 0$ 仍为 $k^\epsilon$。是否存在一种多项式时间算法能够实现多对数近似,一直是一个长期悬而未决的谜题。在本文中,我们通过提出一种多项式时间算法解决了这个问题,该算法在任意有向图上为 DST 实现了 $O(\log^3 k)$ 的近似比。这一结果几乎与目前仅通过拟多项式时间算法已知的最优 $O(\log^2 k / \log\log k)$ 近似比相匹配。由此产生的差距——$O(\log^3 k)$ 对比 $O(\log^2 k / \log\log k)$——反映了群斯坦纳树问题已知的复杂性图景。这种相似性暗示了新的研究方向:是否存在一个硬度结果,能够严格区分 DST 问题在多项式时间算法与拟多项式时间算法上的能力界限?