[Paper Review] Graphs characterized by the second distance eigenvalue

Rundan Xing, Bo Zhou|arXiv (Cornell University)|Jun 14, 2014

Graph theory and applications13 references4 citations

TL;DR

This paper characterizes all connected graphs with second distance eigenvalue less than −0.5858, proving that such graphs are precisely the complete graph $K_n$ for $n \geq 2$, or graphs of the form $K_1 \vee (K_{n_1} \cup \cdots \cup K_{n_r})$ with $r \leq 4$ and specific constraints on the sizes $n_i$, including cases where the second eigenvalue lies in $(-1, -0.5858)$ based on combinatorial and spectral conditions.

ABSTRACT

We characterize all connected graphs with second distance eigenvalue less than $-0.5858$.

Motivation & Objective

• To characterize all connected graphs whose second distance eigenvalue is less than −0.5858.
• To determine the exact family of graphs satisfying this spectral condition using spectral graph theory and forbidden subgraph analysis.
• To establish a complete classification of such graphs based on their structure and eigenvalue behavior.
• To resolve the extremal spectral problem for the second distance eigenvalue in connected graphs.
• To extend prior work on first and last distance eigenvalues by analyzing the second largest eigenvalue.

Proposed method

• Uses the interlacing theorem for symmetric matrices to relate the second distance eigenvalue of a graph to that of its induced subgraphs.
• Applies the distance matrix eigenvalue interlacing to subgraphs like $P_3$, $P_4$, $C_5$, and $K_4 - e$ to derive spectral lower bounds.
• Employs spectral analysis of graphs of the form $K_1 \vee (K_{n_1} \cup \cdots \cup K_{n_r})$ with $r \leq 4$, computing characteristic polynomials and analyzing roots.
• Uses the complement graph structure to deduce that $\overline{G}$ must be disconnected, leading to the $K_1$-join structure.
• Applies forbidden subgraph arguments: if $\overline{G}$ is connected, then $G$ contains $P_4$ or $C_5$, forcing $\lambda_2(G) > -0.5858$, a contradiction.
• Employs case analysis on the number of components in the complement and their sizes, using numerical checks on eigenvalue bounds.

Experimental results

Research questions

• RQ1 Which connected graphs have second distance eigenvalue strictly less than −0.5858?
• RQ2 What structural properties must a graph possess to achieve such a low second distance eigenvalue?
• RQ3 How do the eigenvalues of $K_1 \vee (K_{n_1} \cup \cdots \cup K_{n_r})$ behave as functions of the component sizes $n_i$?
• RQ4 Can the second distance eigenvalue be bounded below by −0.5858 in graphs with certain induced subgraphs, such as $P_4$ or $C_5$?
• RQ5 What is the role of the complement graph in determining the spectral properties of the distance matrix?

Key findings

• The only connected graphs with $\lambda_2(G) < -0.5858$ are the complete graphs $K_n$ for $n \geq 2$.
• Graphs of the form $K_1 \vee (K_{n_1} \cup K_{n_2})$ have $\lambda_2 \in (-1, -0.5858)$ for all $n_1, n_2 \geq 1$.
• For $K_1 \vee (K_{n_1} \cup K_{n_2} \cup K_{n_3})$, the second eigenvalue lies in $(-1, -0.5858)$ for all $n_1, n_2, n_3 \geq 1$.
• For $K_1 \vee (K_{n_1} \cup K_{n_2} \cup K_{n_3} \cup K_{n_4})$ with $1 \leq n_1 \leq n_2 \leq n_3 \leq n_4$, $\lambda_2 \in (-1, -0.5858)$ if and only if one of five specific size conditions holds.
• The threshold value −0.5858 is sharp: $\lambda_2(K_1 \vee (K_1 \cup K_3 \cup K_4 \cup K_5)) = -0.58575 > -0.5858$, showing that the bound is tight.
• The second eigenvalue of $K_1 \vee (3K_2 \cup K_5)$ is $-0.5859$, which is still less than −0.5858, confirming the classification's precision.