Towards an $O(\sqrt[3]{\log n})$-Approximation Algorithm for {\sc Balanced Separator}

TL;DR

This paper introduces a new family of mathematical programs that, if solved efficiently, could improve approximation algorithms for the Balanced Separator problem beyond current best-known bounds.

Contribution

It extends the ARV framework by proposing a parameterized family of programs that potentially achieve better approximation ratios, linking their solvability to improved graph partitioning algorithms.

Findings

Proposes a new family of mathematical programs for graph partitioning.

Shows that solving these programs could improve approximation ratios.

Connects the solvability of these programs to advanced optimization techniques.

Abstract

The {\sc $c$-Balanced Separator} problem is a graph-partitioning problem in which given a graph $G$, one aims to find a cut of minimum size such that both the sides of the cut have at least $cn$ vertices. In this paper, we present new directions of progress in the {\sc $c$-Balanced Separator} problem. More specifically, we propose a new family of mathematical programs, which depends upon a parameter $\epsilon > 0$, and extend the seminal work of Arora-Rao-Vazirani ({\sf ARV}) \cite{ARV} to show that the polynomial time solvability of the proposed family of programs implies an improvement in the approximation factor to $O(\log^{{1/3} + \epsilon} n)$ from the best-known factor of $O(\sqrt{\log n})$ due to {\sf ARV}. In fact, for $\epsilon = 1/3$, the program we get is the SDP proposed by {\sf ARV}. For $\epsilon < 1/3$, this family of programs is not convex but one can transform them into so called \emph{\textbf{concave programs}} in which one optimizes a concave function over a convex feasible set. The properties of concave programs allows one to apply techniques due to Hoffman \cite{H81} or Tuy \emph{et al} \cite{TTT85} to solve such problems with arbitrary accuracy. But the problem of finding of a method to solve these programs that converges in polynomial time still remains open. Our result, although conditional, introduces a new family of programs which is more powerful than semi-definite programming in the context of approximation algorithms and hence it will of interest to investigate this family both in the direction of designing efficient algorithms and proving hardness results.