Strand Life Sciences

Abstract

Given a weighted graph G and an error parameter ε> 0, the graph sparsification problem requires sampling edges in G and giving the sampled edges appropriate weights to obtain a sparse graph Gε with the following property: the weight of every cut in Gε is within a factor of (1 ± ε) of the weight of the corresponding cut in G. Benczúr and Karger [2] showed how to obtain Gε with O(nlogn/ε 2) edges in time O(mlog 3 n) for weighted graphs and O(mlog 2 n) for unweighted graphs using a combinatorial approach based on strong connectivity. Spielman et al [22] showed how to obtain Gε with O(nlogn/ε 2) edges in time O(mlog c n) for some (large) constant c using an algebraic approach based on effective resistances. Our contributions are as below (all for weighted graphs G with n vertices and m edges having polynomial-sized weights, unless otherwise stated): • Benczúr and Karger [2] conjectured that using standard connectivity instead of strong connectivity for sampling would simplify the result substantially, and posed this as an open question. In this correspondence, we resolve this question by showing that sampling using standard connectivities also preserves cut weights and yields a Gε with O(nlog 2 n/ε 2) edges. • We provide a very simple strictly linear time algorithm (i.e. O(m) time) for graph sparsification