Vertex Percolation on Expander Graphs

Abstract: We say that a graph $G=(V,E)$ on $n$ vertices is a $\beta$-expander for some constant $\beta>0$ if every $U\subseteq V$ of cardinality $|U|\leq \frac{n}{2}$ satisfies $|N_G(U)|\geq \beta|U|$ where $N_G(U)$ denotes the neighborhood of $U$. In this work we explore the process of deleting vertices of a $\beta$-expander independently at random with probability $n^{-\alpha}$ for some constant $\alpha>0$, and study the properties of the resulting graph. Our main result states that as $n$ tends to infinity, the deletion process performed on a $\beta$-expander graph of bounded degree will result with high probability in a graph composed of a giant component containing $n-o(n)$ vertices that is in itself an expander graph, and constant size components. We proceed by applying the main result to expander graphs with a positive spectral gap. In the particular case of $(n,d,\lambda)$-graphs, that are such expanders, we compute the values of $\alpha$, under additional constraints on the graph, for which with high probability the resulting graph will stay connected, or will be composed of a giant component and isolated vertices. As a graph sampled from the uniform probability space of $d$-regular graphs with high probability is an expander and meets the additional constraints, this result strengthens a recent result due to Greenhill, Holt and Wormald about vertex percolation on random $d$-regular graphs. We conclude by showing that performing the above described deletion process on graphs that expand sub-linear sets by an unbounded expansion ratio, with high probability results in a connected expander graph.