English

Towards (1+\epsilon)-Approximate Flow Sparsifiers

Data Structures and Algorithms 2013-10-14 v1 Combinatorics

Abstract

A useful approach to "compress" a large network GG is to represent it with a {\em flow-sparsifier}, i.e., a small network HH that supports the same flows as GG, up to a factor q1q \geq 1 called the quality of sparsifier. Specifically, we assume the network GG contains a set of kk terminals TT, shared with the network HH, i.e., TV(G)V(H)T\subseteq V(G)\cap V(H), and we want HH to preserve all multicommodity flows that can be routed between the terminals TT. The challenge is to construct HH that is small. These questions have received a lot of attention in recent years, leading to some known tradeoffs between the sparsifier's quality qq and its size V(H)|V(H)|. Nevertheless, it remains an outstanding question whether every GG admits a flow-sparsifier HH with quality q=1+ϵq=1+\epsilon, or even q=O(1)q=O(1), and size V(H)f(k,ϵ)|V(H)|\leq f(k,\epsilon) (in particular, independent of V(G)|V(G)| and the edge capacities). Making a first step in this direction, we present new constructions for several scenarios: * Our main result is that for quasi-bipartite networks GG, one can construct a (1+ϵ)(1+\epsilon)-flow-sparsifier of size \poly(k/\eps)\poly(k/\eps). In contrast, exact (q=1q=1) sparsifiers for this family of networks are known to require size 2Ω(k)2^{\Omega(k)}. * For networks GG of bounded treewidth ww, we construct a flow-sparsifier with quality q=O(logw/loglogw)q=O(\log w / \log\log w) and size O(w\poly(k))O(w\cdot \poly(k)). * For general networks GG, we construct a {\em sketch} sk(G)sk(G), that stores all the feasible multicommodity flows up to factor q=1+\epsq=1+\eps, and its size (storage requirement) is f(k,ϵ)f(k,\epsilon).

Keywords

Cite

@article{arxiv.1310.3252,
  title  = {Towards (1+\epsilon)-Approximate Flow Sparsifiers},
  author = {Alexandr Andoni and Anupam Gupta and Robert Krauthgamer},
  journal= {arXiv preprint arXiv:1310.3252},
  year   = {2013}
}

Comments

Full version of a paper accepted to SODA 2014

R2 v1 2026-06-22T01:45:21.316Z