Upper tail of the spectral radius of sparse Erdos-R\'enyi graphs
Abstract
We consider an Erdos-R\'enyi graph G(n,p) on n vertices with edge probability p such that \[ n n np n1/2-o(1), eq:abs \] and derive the upper tail large deviations of λ(G(n,p)), the largest eigenvalue of its adjacency matrix. Within this regime we show that, for p n-2/3 the -probability of the upper tail event of λ(G(n,p)) equals to that of planting a clique of an appropriate size (upon ignoring smaller order terms), while for p n-2/3 the same is given by that of the existence of a high degree vertex. This, in particular, shows an emergence of non-planted localized structure in the latter regime. We also confirm that in the entire regime eq:abs the large deviation probability is asymptotically approximated by the solution of the mean-field variational problem, and further identify the typical structure of G(n,p) conditioned on the upper tail event of λ(G(n,p)) in a certain sub-regime of p. For p such that (np) n the large deviations of λ(G(n,p)) is deduced from those of Hom(C2t, G(n,p)), the homomorphism counts of cycle of length 2t, for t 3 and p such that n1/2-o(1) np n1/t. In this latter regime the typical structure of G(n,p) conditioned on the upper tail of Hom(C2t, G(n,p)) is identified and asymptotic tightness of the mean-field approximation is also established.
Turn this paper into a lesson
ArcXiv compiles a structured reading guide from this paper's metadata: plain-English importance, contributions, prerequisite concepts, which sections to read first, flashcards, and a quiz. Grounded in the abstract, never invented.