8
32

Deterministic Distributed Expander Decomposition and Routing with Applications in Distributed Derandomization

Abstract

There is a recent exciting line of work in distributed graph algorithms in the CONGEST\mathsf{CONGEST} model that exploit expanders. All these algorithms so far are based on two tools: expander decomposition and expander routing. An (ϵ,ϕ)(\epsilon,\phi)-expander decomposition removes ϵ\epsilon-fraction of the edges so that the remaining connected components have conductance at least ϕ\phi, i.e., they are ϕ\phi-expanders, and expander routing allows each vertex vv in a ϕ\phi-expander to very quickly exchange deg(v)\text{deg}(v) messages with any other vertices, not just its local neighbors. In this paper, we give the first efficient deterministic distributed algorithms for both tools. We show that an (ϵ,ϕ)(\epsilon,\phi)-expander decomposition can be deterministically computed in poly(ϵ1)no(1)\text{poly}(\epsilon^{-1}) n^{o(1)} rounds for ϕ=poly(ϵ)no(1)\phi = \text{poly}(\epsilon) n^{-o(1)}, and that expander routing can be performed deterministically in poly(ϕ1)no(1)\text{poly}(\phi^{-1})n^{o(1)} rounds. Both results match previous bounds of randomized algorithms by [Chang and Saranurak, PODC 2019] and [Ghaffari, Kuhn, and Su, PODC 2017] up to subpolynomial factors. Consequently, we derandomize existing distributed algorithms that exploit expanders. We show that a minimum spanning tree on no(1)n^{o(1)}-expanders can be constructed deterministically in no(1)n^{o(1)} rounds, and triangle detection and enumeration on general graphs can be solved deterministically in O(n0.58)O(n^{0.58}) and n2/3+o(1)n^{2/3 + o(1)} rounds, respectively. We also give the first polylogarithmic-round randomized algorithm for constructing an (ϵ,ϕ)(\epsilon,\phi)-expander decomposition in poly(ϵ1,logn)\text{poly}(\epsilon^{-1}, \log n) rounds for ϕ=1/poly(ϵ1,logn)\phi = 1 / \text{poly}(\epsilon^{-1}, \log n). The previous algorithm by [Chang and Saranurak, PODC 2019] needs nΩ(1)n^{\Omega(1)} rounds for any ϕ1/polylogn\phi\ge 1/\text{poly}\log n.

View on arXiv
Comments on this paper