[论文解读] A Better-Than-2 Approximation for Weighted Tree Augmentation

Bang-Jensen, Jørgen, Klinkby, Kristine Vitting+2 more|arXiv (Cornell University)|Apr 14, 2021

Numerical Methods and Algorithms被引用 1

一句话总结

本论文提出了首个能够突破长期存在的2-近似瓶颈的加权树增强问题（WTAP）近似算法，实现了(1 + ln 2 + ε)的近似比——当ε较小时约为1.7。该方法采用一种新颖的相对贪心算法，通过迭代收缩一类经过精心选择的、大小为超常数的组件集合，利用近似分解定理确保向最优解的逐步推进。

ABSTRACT

In the Vertex Connectivity Survivable Network Design (VC-SNDP) problem, the input is a graph G and a function d: V(G) × V(G) → ℕ that encodes the vertex-connectivity demands between pairs of vertices. The objective is to find the smallest subgraph H of G that satisfies all these demands. It is a well-studied NP-complete problem that generalizes several network design problems. We consider the case of uniform demands, where for every vertex pair (u,v) the connectivity demand d(u,v) is a fixed integer κ. It is an important problem with wide applications. We study this problem in the realm of Parameterized Complexity. In this setting, in addition to G and d we are given an integer 𝓁 as the parameter and the objective is to determine if we can remove at least 𝓁 edges from G without violating any connectivity constraints. This was posed as an open problem by Bang-Jansen et.al. [SODA 2018], who studied the edge-connectivity variant of the problem under the same settings. Using a powerful classification result of Lokshtanov et al. [ICALP 2018], Gutin et al. [JCSS 2019] recently showed that this problem admits a (non-uniform) FPT algorithm where the running time was unspecified. Further they also gave an (uniform) FPT algorithm for the case of κ = 2. In this paper we present a (uniform) FPT algorithm any κ that runs in time 2^{O(κ² 𝓁⁴ log 𝓁)}⋅ |V(G)|^O(1). Our algorithm is built upon new insights on vertex connectivity in graphs. Our main conceptual contribution is a novel graph decomposition called the Wheel decomposition. Informally, it is a partition of the edge set of a graph G, E(G) = X₁ ∪ X₂ … ∪ X_r, with the parts arranged in a cyclic order, such that each vertex v ∈ V(G) either has edges in at most two consecutive parts, or has edges in every part of this partition. The first kind of vertices can be thought of as the rim of the wheel, while the second kind form the hub. Additionally, the vertex cuts induced by these edge-sets in G have highly symmetric properties. Our main technical result, informally speaking, establishes that "nearly edge-minimal’’ κ-vertex connected graphs admit a wheel decomposition - a fact that can be exploited for designing algorithms. We believe that this decomposition is of independent interest and it could be a useful tool in resolving other open problems.

研究动机与目标

• 为自1980年代以来长期存在的加权树增强问题（WTAP）的2-近似瓶颈提供突破，尽管该问题历经广泛研究仍未被突破。
• 设计一种多项式时间近似算法，即使在任意边权的一般情况下，也能实现严格优于2的近似比。
• 设计一种作用于新型超常数大小组件类的相对贪心算法，从而在以往方法失效的情况下实现优于2的近似。
• 证明一个近似分解定理，保证在任意最优WTAP解中均存在适合收缩的良好组件，从而确保算法的正确性与性能分析。

提出的方法

• 引入一种受Zelikovsky针对Steiner树方法启发的相对贪心算法框架，将其适配至WTAP，并采用一种新颖的组件收缩策略。
• 定义一类k-薄组件（k为常数），其大小为超常数，且允许对结构进行高效枚举与优化。
• 对树中每个节点v，枚举所有大小至多为k的边集Y ⊆ L，以及所有可能的上边状态x ∈ {−, +}，形成三元组(v, Y, x)以指导组件选择。
• 对每个此类三元组，为v的每个子节点vi计算最优边集Ci ⊆ L[Dvi]，使得结果集合CY = (Y \ Y) ∪ ⋃i Ci 最大化松弛度ρ(CY, Y, v)，同时保持k-薄性。
• 使用动态规划，基于上边ui ∈ δU(Dvi) ∩ δU(v) 的存在性与权重，通过在C−i与C+i（若可行）之间进行选择，以最大化松弛度并保持k-薄性。
• 在组件构造过程中检测不可行性：若对给定的(v, Y, x)不存在有效Ci，则将该三元组标记为不可行，从而避免无效解的产生。

实验结果

研究问题

• RQ1 能否在具有任意边权的一般加权树增强问题（WTAP）中实现优于2的近似？
• RQ2 是否可能设计一种作用于超常数大小组件的WTAP相对贪心算法，从而避免先前工作对常数大小组件的限制？
• RQ3 是否存在一个近似分解定理，可保证在任意最优WTAP解中均存在适合收缩的良好组件？
• RQ4 尽管组件类的规模呈指数级增长，该算法能否高效地在每一步找到最优收缩组件？

主要发现

• 本论文实现了WTAP的(1 + ln 2 + ε)近似比，当ε > 0足够小时该值小于1.7，标志着40多年来首次突破2-近似界限。
• 该算法采用一种新颖的相对贪心方法，通过收缩超常数大小的组件，克服了以往方法必须依赖常数大小组件的关键限制。
• 证明了一个近似分解定理，表明任意最优WTAP解均可被划分为适合收缩的组件，从而确保了算法的正确性。
• 该算法在多项式时间内运行，具体为O(|V|^{O(k)})，通过高效枚举大小至多为k的Y集合，并为每个子树计算最优边集。
• 该方法在组件构造过程中能正确检测不可行三元组，确保仅考虑有效且k-薄的组件。
• 该方法的适用范围超越了有界直径树与无权情况，与以往依赖有界权比或常数树直径的进展不同。