Cut Sparsification of the Clique Beyond the Ramanujan Bound: A Separation of Cut Versus Spectral Sparsification

TL;DR

This paper demonstrates that random regular graphs can serve as effective cut sparsifiers of the clique with approximation errors below the Ramanujan spectral bound, highlighting a fundamental separation between cut and spectral sparsification.

Contribution

It establishes new bounds for cut sparsification of the clique beyond the Ramanujan bound and proves lower bounds for spectral sparsification, revealing a separation between the two.

Findings

Random regular graphs are cut sparsifiers with approximation error below Ramanujan bound.

Spectral sparsifiers of the clique have approximation error at least the Ramanujan bound.

There is a fundamental separation between spectral and cut sparsification for dense graphs.

Abstract

We prove that a random $d$-regular graph, with high probability, is a cut sparsifier of the clique with approximation error at most $\left(2\sqrt{\frac 2 \pi} + o_{n,d}(1)\right)/\sqrt d$, where $2\sqrt{\frac 2 \pi} = 1.595\ldots$ and $o_{n,d}(1)$ denotes an error term that depends on $n$ and $d$ and goes to zero if we first take the limit $n\rightarrow \infty$ and then the limit $d \rightarrow \infty$. This is established by analyzing linear-size cuts using techniques of Jagannath and Sen derived from ideas in statistical physics, and analyzing small cuts via martingale inequalities. We also prove new lower bounds on spectral sparsification of the clique. If $G$ is a spectral sparsifier of the clique and $G$ has average degree $d$, we prove that the approximation error is at least the "Ramanujan bound'' $(2-o_{n,d}(1))/\sqrt d$, which is met by $d$-regular Ramanujan graphs, provided that either the weighted adjacency matrix of $G$ is a (multiple of) a doubly stochastic matrix, or that $G$ satisfies a certain high "odd pseudo-girth" property. The first case can be seen as an "Alon-Boppana theorem for symmetric doubly stochastic matrices," showing that a symmetric doubly stochastic matrix with $dn$ non-zero entries has a non-trivial eigenvalue of magnitude at least $(2-o_{n,d}(1))/\sqrt d$; the second case generalizes a lower bound of Srivastava and Trevisan, which requires a large girth assumption. Together, these results imply a separation between spectral sparsification and cut sparsification. If $G$ is a random $\log n$-regular graph on $n$ vertices, we show that, with high probability, $G$ admits a (weighted subgraph) cut sparsifier of average degree $d$ and approximation error at most $(1.595\ldots + o_{n,d}(1))/\sqrt d$, while every (weighted subgraph) spectral sparsifier of $G$ having average degree $d$ has approximation error at least $(2-o_{n,d}(1))/\sqrt d$.