Voting algorithms for unique games on complete graphs

Abstract

An approximation algorithm for a constraint satisfaction problem is called robust if it outputs an assignment satisfying a $(1 - f(ε))$-fraction of the constraints on any $(1-ε)$-satisfiable instance, where the loss function $f$ is such that $f(ε) \rightarrow 0$ as $ε\rightarrow 0$. Moreover, the runtime of a robust algorithm should not depend in any way on $ε$. In this paper, we present such an algorithm for Min-Unique-Games on complete graphs with $q$ labels. Specifically, the loss function is $f(ε) = (ε+ c_ε ε^2)$, where $c_ε$ is a constant depending on $ε$ such that $\lim_{ε\rightarrow 0} c_ε = 16$. The runtime of our algorithm is $O(qn^3)$ (with no dependence on $ε$) and can run in time $O(qn^2)$ using a randomized implementation with a slightly larger constant $c_ε$. Our algorithm is combinatorial and uses voting to find an assignment. It can furthermore be used to provide a PTAS for Min-Unique-Games on complete graphs, recovering a result of Karpinski and Schudy with a simpler algorithm and proof. We also prove NP-hardness for Min-Unique-Games on complete graphs and (using a randomized reduction) even in the case where the constraints form a cyclic permutation, which is also known as Min-Linear-Equations-mod-$q$ on complete graphs.