English

A Sketching Algorithm for Spectral Graph Sparsification

Data Structures and Algorithms 2014-12-30 v1

Abstract

We study the problem of compressing a weighted graph GG on nn vertices, building a "sketch" HH of GG, so that given any vector xRnx \in \mathbb{R}^n, the value xTLGxx^T L_G x can be approximated up to a multiplicative 1+ϵ1+\epsilon factor from only HH and xx, where LGL_G denotes the Laplacian of GG. One solution to this problem is to build a spectral sparsifier HH of GG, which, using the result of Batson, Spielman, and Srivastava, consists of O(nϵ2)O(n \epsilon^{-2}) reweighted edges of GG and has the property that simultaneously for all xRnx \in \mathbb{R}^n, xTLHx=(1±ϵ)xTLGxx^T L_H x = (1 \pm \epsilon) x^T L_G x. The O(nϵ2)O(n \epsilon^{-2}) bound is optimal for spectral sparsifiers. We show that if one is interested in only preserving the value of xTLGxx^T L_G x for a {\it fixed} xRnx \in \mathbb{R}^n (specified at query time) with high probability, then there is a sketch HH using only O~(nϵ1.6)\tilde{O}(n \epsilon^{-1.6}) bits of space. This is the first data structure achieving a sub-quadratic dependence on ϵ\epsilon. Our work builds upon recent work of Andoni, Krauthgamer, and Woodruff who showed that O~(nϵ1)\tilde{O}(n \epsilon^{-1}) bits of space is possible for preserving a fixed {\it cut query} (i.e., x{0,1}nx\in \{0,1\}^n) with high probability; here we show that even for a general query vector xRnx \in \mathbb{R}^n, a sub-quadratic dependence on ϵ\epsilon is possible. Our result for Laplacians is in sharp contrast to sketches for general n×nn \times n positive semidefinite matrices AA with O(logn)O(\log n) bit entries, for which even to preserve the value of xTAxx^T A x for a fixed xRnx \in \mathbb{R}^n (specified at query time) up to a 1+ϵ1+\epsilon factor with constant probability, we show an Ω(nϵ2)\Omega(n \epsilon^{-2}) lower bound.

Keywords

Cite

@article{arxiv.1412.8225,
  title  = {A Sketching Algorithm for Spectral Graph Sparsification},
  author = {Jiecao Chen and Bo Qin and David P. Woodruff and Qin Zhang},
  journal= {arXiv preprint arXiv:1412.8225},
  year   = {2014}
}

Comments

18 pages

R2 v1 2026-06-22T07:45:23.199Z