New results on eigenvalues and degree deviation

TL;DR

This paper introduces a new upper bound for the spectral deviation of a graph from regularity, improving on previous bounds by employing numerical analysis techniques involving Rayleigh quotients.

Contribution

It presents a novel upper bound on spectral deviation that outperforms existing bounds, using methods from numerical analysis to analyze eigenvalues.

Findings

New upper bound on spectral deviation $\\epsilon(G)$

Outperforms previous bounds by Nikiforov

Uses numerical analysis techniques involving Rayleigh quotients

Abstract

Let $G$ be a graph. In a famous paper Collatz and Sinogowitz had proposed to measure its deviation from regularity by the difference of the (adjacency) spectral radius and the average degree: $\epsilon(G)=\rho(G)-\frac{2m}{n}$. We obtain here a new upper bound on $\epsilon(G)$ which seems to consistently outperform the best known upper bound to date, due to Nikiforov. The method of proof may also be of independent interest, as we use notions from numerical analysis to re-cast the estimation of $\epsilon(G)$ as a special case of the estimation of the difference between Rayleigh quotients of proximal vectors.