Near-optimal light spanners

Abstract

A spanner H of a weighted undirected graph G is a sparse subgraph that approximately preserves distances between every pair of vertices in G. We refer to H as a δ-spanner of G for some parameter δ ≥ 1 if the distance in H between every vertex pair is at most a factor δ bigger than in G. In this case, we say that H has stretch δ. Two main measures of the sparseness of a spanner are the size (number of edges) and the total weight (the sum of weights of the edges in the spanner).It is well-known that for any positive integer k, one can efficiently construct a (2k -- 1)-spanner of G with O(n1+1/k) edges where n is the number of vertices [2]. This size-stretch tradeoff is conjectured to be optimal based on a girth conjecture of Erdos [17]. However, the current state of the art for the second measure is not yet optimal.Recently Elkin, Neiman and Solomon [ICALP 14] presented an improved analysis of the greedy algorithm, proving that the greedy algorithm admits (2k -- 1) · (1 + e) stretch and total edge weight of Oe((k/log k) · ω(MST(G)) · [EQUATION]1/k), where ω(MST(G)) is the weight of a minimum spanning tree of G. The previous analysis by Chandra et al. [SOCG 92] admitted (2k -- 1) · (1 + e) stretch and total edge weight of Oe(kω(MST(G))n1/k). Hence, Elkin et al. improved the weight of the spanner by a log k factor.In this work, we complectly remove the k factor from the weight, presenting a spanner with (2k -- 1) · (1 + e) stretch, Oe(ω(MST(G))n1/k) total weight, and O(n1+1/k) edges. Up to a (1 + e) factor in the stretch this matches the girth conjecture of Erdos [17].