Breaking a Long-Standing Barrier: 2-$\varepsilon$ Approximation for Steiner Forest

TL;DR

This work breaks the long-standing 2-approximation barrier for Steiner Forest by presenting a deterministic, polynomial-time algorithm that achieves a factor of $2 - 10^{-11}$. The approach centers on a novel moat-growing framework augmented by a dual-based local search, extensions, and autarkic-pair techniques to tightly couple Steiner Forest demands with Steiner Tree insights. A key auxiliary contribution is a $1.943$-approximation for the Steiner Tree problem, derived via a refined claw-like bound and a novel prefix-time/assignment analysis. By introducing Shadow Moat Growing, extension steps with epsilon-potential, and autarkic pairs, the authors effectively bound the maximum component-wise deviations from OPT, enabling a global improvement beyond 2 for Steiner Forest and informing broader strategy for related forest problems. The results mark a significant, deterministic advance with potential implications for related combinatorial optimization problems and LP-relaxation gaps.

Abstract

The Steiner Forest problem, also known as the Generalized Steiner Tree problem, is a fundamental optimization problem on edge-weighted graphs where, given a set of vertex pairs, the goal is to select a minimum-cost subgraph such that each pair is connected. This problem generalizes the Steiner Tree problem, first introduced in 1811, for which the best known approximation factor is 1.39 [Byrka, Grandoni, Rothvoß, and Sanità, 2010] (Best Paper award, STOC 2010). The celebrated work of [Agrawal, Klein, and Ravi, 1989] (30-Year Test-of-Time award, STOC 2023), along with refinements by [Goemans and Williamson, 1992] (SICOMP'95), established a 2-approximation for Steiner Forest over 35 years ago. Jain's (FOCS'98) pioneering iterative rounding techniques later extended these results to higher connectivity settings. Despite the long-standing importance of this problem, breaking the approximation factor of 2 has remained a major challenge, raising suspicions that achieving a better factor -- similar to Vertex Cover -- might indeed be hard. Notably, fundamental works, including those by Gupta and Kumar (STOC'15) and Groß et al. (ITCS'18), introduced 96- and 69-approximation algorithms, possibly with the hope of paving the way for a breakthrough in achieving a constant-factor approximation below 2 for the Steiner Forest problem. In this paper, we break the approximation barrier of 2 by designing a novel deterministic algorithm that achieves a $2 - 10^{-11}$ approximation for this fundamental problem. As a key component of our approach, we also introduce a novel dual-based local search algorithm for the Steiner Tree problem with an approximation guarantee of $1.943$, which is of independent interest.

Figures (13)

• Figure 1: A Steiner Tree instance. (a) Initial configuration with four terminal vertices and a central vertex $v$, where $\xi > 0$ is sufficiently small. (b) After Legacy Moat Growing, moats centered on terminals expand, fully coloring the outer edges, resulting in a total growth of 4 and a solution cost of 8. (c) A boost action on $v$ connects moats earlier via central edges, reducing the total growth to $\frac{5}{2}(1+\xi)$ and the solution cost to $4(1+\xi)$.
• Figure 2: A Steiner Forest instance. (a) Initial configuration with $n$ rows (here, $n = 3$), and small $\xi \in (0, 1/n]$. Each row contains three demand pairs, and vertices $v$ and $u$ also form a required pair. A vertical path through the middle rows costs $n + 1$ and connects $v$ and $u$, while an edge of cost 2 offers an alternative. (b) After Legacy Moat Growing, vertices in each row quickly form two groups, but require significant growth to connect, resulting in coloring the vertical path. The total cost is $2n + 1 + O(1)$. (c) We detect vertices in each row as autarkic pairs and directly connect one vertex pair per row using their shortest paths, assuming a zero-cost edge connects them from this point on. (d) Running Legacy Moat Growing on the modified graph, the row vertices connect quickly via the new links, avoiding further growth and preventing coloring of the vertical path. Vertices $u$ and $v$ now connect via the edge of cost 2, and the total cost becomes $n + 2 + O(1)$.
• Figure 3: Here, there are two demand pairs: star pairs and square pairs. (a) Initially, all vertices form active sets and grow until the square pairs reach each other. (b) Once the connected component containing squares becomes inactive, the remaining connected components continue to grow until one of them reaches the square component. (c) At this point, there are two connected components that are both active and grow until they meet, causing the edge between the stars to become fully colored. Note that while all edges are added to $F$ at this moment, the edge between a star and a square will be removed during the pruning phase since there was an inactive set containing squares that cut only this edge.
• Figure 4: Flow of Algorithm \ref{['alg:main']}, showing the different modules and their order. Each module results in an "execution" of a monotonic moat growing algorithm, with the name of each execution shown below the corresponding circle. The symbol inside each circle is used to refer to variables from that execution by placing it as a superscript. The three solutions compared at the end of the algorithm are also identified, along with their positions in the flow and how they are obtained.
• Figure 5: Connected components, active sets (black dashed lines), and their superactive sets (solid red lines) at the same moment in two monotonic moat growing executions. The fingerprint in (b) is larger than in (a). As shown, every connected component, active set, and superactive set in (a) is a subset of its counterpart in (b). The execution in (b) exhibits greater connectivity: any two vertices connected in (a) are also connected in (b), some vertices that are inactive in (a) are active in (b), and some vertices not included in any superactive set in (a) belong to one in (b).

Theorems & Definitions (295)

• Theorem 1
• Theorem 2
• Definition 3: $\textsc{Unsatisfied}$
• Corollary 4
• Definition 5: Monotonic Moat Growing Algorithm
• Definition 6: Fingerprint
• Lemma 7
• proof
• Lemma 8
• proof