[论文解读] A Combinatorial Cut-Toggling Algorithm for Solving Laplacian Linear Systems

Henzinger, Monika, Daniel A. Spielman+2 more|arXiv (Cornell University)|Nov 13, 2013

Matrix Theory and Algorithms参考文献 35被引用 7

一句话总结

本论文提出了首个针对对称对角占优（SDD）线性系统的并行算法，其工作量接近线性，深度为对数多对数。该算法通过谱图稀疏化构造稀疏近似逆链，实现通过重复矩阵-向量乘法高效求解，可在 O(log n log κ log 1/ǫ) 深度和 O((m + n log³κ) log 1/ǫ) 工作量内实现 ǫ-近似解。

ABSTRACT

Over the last two decades, a significant line of work in theoretical algorithms has made progress in solving linear systems of the form 𝐋𝐱 = 𝐛, where 𝐋 is the Laplacian matrix of a weighted graph with weights w(i,j) > 0 on the edges. The solution 𝐱 of the linear system can be interpreted as the potentials of an electrical flow in which the resistance on edge (i,j) is 1/w(i,j). Kelner, Orrechia, Sidford, and Zhu [Kelner et al., 2013] give a combinatorial, near-linear time algorithm that maintains the Kirchoff Current Law, and gradually enforces the Kirchoff Potential Law by updating flows around cycles (cycle toggling). In this paper, we consider a dual version of the algorithm that maintains the Kirchoff Potential Law, and gradually enforces the Kirchoff Current Law by cut toggling: each iteration updates all potentials on one side of a fundamental cut of a spanning tree by the same amount. We prove that this dual algorithm also runs in a near-linear number of iterations. We show, however, that if we abstract cut toggling as a natural data structure problem, this problem can be reduced to the online vector-matrix-vector problem (OMv), which has been conjectured to be difficult for dynamic algorithms [Henzinger et al., 2015]. The conjecture implies that the data structure does not have an O(n^{1-ε}) time algorithm for any ε > 0, and thus a straightforward implementation of the cut-toggling algorithm requires essentially linear time per iteration. To circumvent the lower bound, we batch update steps, and perform them simultaneously instead of sequentially. An appropriate choice of batching leads to an Õ(m^{1.5}) time cut-toggling algorithm for solving Laplacian systems. Furthermore, we show that if we sparsify the graph and call our algorithm recursively on the Laplacian system implied by batching and sparsifying, we can reduce the running time to O(m^{1 + ε}) for any ε > 0. Thus, the dual cut-toggling algorithm can achieve (almost) the same running time as its primal cycle-toggling counterpart.

研究动机与目标

• 设计一种并行算法，用于求解SDD线性系统，实现接近线性工作量与对数多对数深度。
• 通过改用谱图稀疏化而非低拉伸生成树，克服先前接近线性时间求解器对低拉伸生成树的依赖。
• 构建一个稀疏近似逆链——一系列稀疏矩阵的序列，其乘积近似表示输入矩阵的逆。
• 通过重复对链中矩阵进行矩阵-向量乘法，实现高效且对称的求解。
• 对任意期望精度 ǫ，实现接近线性工作量与对数多对数深度，且不依赖于低拉伸树假设。

提出的方法

• 使用谱图稀疏化而非低拉伸生成树，为SDDM矩阵 M = D − A 构建稀疏近似逆链。
• 利用Orecchia和Vishnoi的BalCut算法，在接近线性工作量与对数多对数深度下实现图的稀疏化。
• 通过外积的稀疏采样近似矩阵 D − A D⁻¹ A，利用有效电阻进行边采样。
• 应用定理6.2（基于Rudelson和Vershynin的浓度不等式），以权重与有效电阻成比例的方式采样 O(n log n / ǫ²) 条边。
• 构建数据结构，使每顶点的边采样时间复杂度为 O(log n)，从而实现稀疏化矩阵的高效并行构造。
• 通过递归应用稀疏化与矩阵近似，逐层降低条件数，从而迭代构建逆链。

实验结果

研究问题

• RQ1 能否设计一种并行SDD线性系统求解器，同时实现接近线性工作量与对数多对数深度？
• RQ2 是否可能在不依赖低拉伸生成树的前提下构建稀疏近似逆链？
• RQ3 仅使用谱图稀疏化是否足以构建SDD矩阵的稳定且高效的逆链？
• RQ4 使用此类链求解SDD系统的工作量与深度复杂度是多少，且能保证近似误差？
• RQ5 逆链的构建能否在保持接近线性工作量与对数深度的前提下实现高效并行化？

主要发现

• 该算法在 O(m logᶜ¹n log³κ) 工作量与 O(logᶜ²n log κ) 深度内构建稀疏近似逆链，其中 c₁、c₂ 为某些常数。
• 对任意 ǫ > 0，该算法在 O((m + n log³κ) log 1/ǫ) 工作量与 O(log n log κ log 1/ǫ) 深度内计算出 Mx = b 的 ǫ-近似解。
• 稀疏化器的构造成功概率至少为 1 − 1/n²，逆链构造的成功概率至少为 1/2。
• 该方法避免了对低拉伸生成树的需求，完全依赖谱稀疏化与有效电阻采样。
• 逆链支持通过每层两次矩阵-向量乘法求解线性系统，确保对称性与稳定性。
• 输入矩阵的条件数 κ 在深度与工作量中以对数形式出现，反映了其在解的稳定性和收敛性中的作用。