Strengthened upper bound on the third eigenvalue of graphs
Sida Li
TL;DR
This paper resolves a longstanding conjecture-like claim by proving the strengthened upper bound for the third eigenvalue proportion in graphs, specifically establishing that no graph sequence can achieve $(\lambda_{n-1}+\lambda_n)/n$ approaching $-\frac{\sqrt{2}}{2}$. The authors shift focus from $\lambda_3$ directly to the sum $\lambda_{n-1}+\lambda_n$, introducing a novel graph operation $^*$ that constrains minimising graphs to have limited clique and chromatic numbers and to resemble invariant, structured forms. They then restrict analysis to $n/2$-regular graphs within this invariant framework, deriving a detailed, multi-stage inequality argument (involving front/middle/back decompositions and Type 1/Type 2 eigenvectors) to show the desired bound strictly improves beyond the conjectured limit. A key consequence is supporting the plausibility of $c_3=\tfrac{1}{3}$ and highlighting a deep connection between extremal spectral properties and invariant graph families such as $H_{a,b}$ and $\Pi_n$, with potential alternative proofs via graphon theory. The work thus advances spectral extremal graph theory by both tightening a precise bound and clarifying the structural landscape of near-worst-case graphs.
Abstract
Let $G$ be a graph on $n \ge 3$ vertices, whose adjacency matrix has eigenvalues $λ_1 \ge λ_2 \ge \dots \ge λ_n$. The problem of bounding $λ_k$ in terms of $n$ was first proposed by Hong and was studied by Nikiforov, who demonstrated strong upper and lower bounds for arbitrary $k$. Nikiforov also claimed a strengthened upper bound for $k \ge 3$, namely that $\frac{λ_k}{n} < \frac{1}{2\sqrt{k-1}} - \varepsilon_k$ for some positive $\varepsilon_k$, but omitted the proof due to its length. In this paper, we give a proof of this bound for $k = 3$. We achieve this by instead looking at $λ_{n-1} + λ_n$ and introducing a new graph operation which provides structure to minimising graphs, including $ω\le 3$ and $χ\le 4$. Then we reduce the hypothetical worst case to a graph that is $n/2$-regular and invariant under said operation. By considering a series of inequalities on the restricted eigenvector components, we prove that a sequence of graphs with $\frac{λ_{n-1} + λ_n}{n}$ converging to $-\frac{\sqrt{2}}{2}$ cannot exist.
