Spectral Insights in Cubic Graphs

• The paper elucidates that the adjacency spectrum of cubic graphs captures invariants like regularity and order while failing to determine NP-hard properties such as edge-chromatic number and Hamiltonicity.
• It demonstrates methods including vertex truncation and local structure replacement to construct infinite families of cospectral cubic graphs with distinct edge-colorability and Hamiltonicity traits.
• Advanced spectral bounds offer sufficient criteria for perfect matchings, refining structural insights but also underscoring the inherent limitations of spectral techniques in completely characterizing cubic graphs.

A cubic graph is a connected, simple, undirected graph in which every vertex has degree exactly 3. The study of spectral characterizations of cubic graphs revolves around understanding which combinatorial properties of such graphs can be determined from the spectrum of their adjacency matrix, and which cannot. This is a central question in algebraic graph theory, with ramifications for graph isomorphism, colorability, matchings, and construction of cospectral pairs. Recent advances establish new infinite families of cospectral cubic graphs that differ in key invariants and sharpen spectral criteria for structural properties such as the existence of perfect matchings.

1. Spectral Invariants and Cubic Graphs

Given a cubic graph $G=(V,E)$ on $n$ vertices, its adjacency matrix $A(G)$ is an $n \times n$ matrix where $A_{ij}=1$ if $\{i,j\} \in E$, and $0$ otherwise. The eigenvalues $$\lambda_1 \geq \lambda_2 \geq \cdots \geq \lambda_n$$ (counted with multiplicity) form the adjacency spectrum of $G$. Many properties are determined by the spectrum: total number of edges, regularity, number of triangles, and bipartiteness. For cubic graphs, regularity and the order of the graph, bipartiteness, and vertex-chromatic number are spectrally determined. In contrast, other properties such as edge-chromatic number, perfect matchings, and Hamiltonicity are not fully characterized by the adjacency spectrum (Haemers, 7 Jan 2026).

Cospectral graphs $G$ and $H$ are graphs with coinciding spectra but that need not be isomorphic; such examples demonstrate that the spectrum is not a complete graph invariant.

2. Edge-Chromatic Number and Cospectral Non-Isomorphic Cubic Graphs

By Vizing’s theorem a cubic graph has edge-chromatic number $\chi'(G)$ equal to either 3 or 4. If $\chi'(G)=3$, its edges can be colored by three colors such that no two adjacent edges share a color, equivalently, its edges can be partitioned into three perfect matchings. Whether the property $\chi'(G)=3$ is determined by the spectrum has long been open.

Exhaustive computational search established a unique pair of connected cubic graphs on 16 vertices, $G$ and $H$, that are cospectral yet have different chromatic indices: $\chi'(G) = 3$ and $\chi'(H) = 4$ (Yan et al., 2021). The characteristic polynomial and eigenvalues coincide exactly; $G$ is 3-edge-colorable (has a Hamilton cycle), while $H$ is not and can be constructed from the Petersen graph by replacing each vertex by a triangle. This provided the smallest explicit example that the chromatic index (edge-colorability) is not spectrally determined, even within cubic graphs.

3. Infinite Families of Cospectral Cubic Graphs with Distinct Edge-Colorability

Haemers extended the isolated counterexample on 16 vertices to an infinite family using the operation of vertex truncation (Haemers, 7 Jan 2026). The truncation of $G$ at vertex $v$ replaces $v$ by a triangle and attaches each former neighbor to a distinct vertex of that triangle; full truncation applies this to every vertex. Notably, for a cubic graph $G$ with spectrum $\{\lambda_1=3,\lambda_2,\dots,\lambda_n\}$, the spectrum of $T(G)$ (the full truncation) is:

$$\{\tfrac12\pm\tfrac12\sqrt{13+4\lambda_i} \;\mid\; i=1,\dots,n\} \cup \{-2^{n/2},\,0^{n/2}\}.$$

Truncation preserves cospectrality and chromatic index: iterated truncations of an initial cospectral pair $(G,H)$ with $\chi'(G)\ne\chi'(H)$ yield infinite families of cospectral pairs $(G_k,H_k)$ with $\chi'(G_k)=3$ and $\chi'(H_k)=4$ for all $k$. This construction also applies to Hamiltonicity, yielding infinite cospectral cubic pairs differing in Hamiltonicity. Thus, chromatic index and Hamiltonicity remain spectrally undetectable for infinite families of cubic graphs (Haemers, 7 Jan 2026).

4. Sufficient Spectral Criteria for Perfect Matchings

While not all matching properties are spectrally characterized, new spectral bounds provide sufficient—but not necessary—criteria for the existence of perfect matchings in cubic graphs. Previous work established that if the third largest eigenvalue $\lambda_3\leq\theta$ for $\theta\approx 2.85577$, then a cubic graph has a perfect matching. Haemers improved this by showing that if the second largest eigenvalue $\lambda_2 < \theta' \approx 2.94272$ for $n>76$, the graph must have a perfect matching (Haemers, 7 Jan 2026). The proof leverages the interlacing property and Tutte’s theorem, excluding eigenvalue configurations incompatible with the absence of a perfect matching.

5. Construction Frameworks and Computational Prevalence

Systematic algorithms for constructing cospectral cubic graphs typically rely on local removals and neighborhood replacement (the Schwenk-Godsil property). Haythorpe and Newcombe formalized vertex composition: if $G_1, G_2$ (respectively $H_1, H_2$) are cospectral cubic graphs with replaceable vertices, different ways of gluing, via matching neighborhoods, yield quadruples of pairwise cospectral cubic graphs (Haythorpe et al., 2021). Computational data for orders $N \le 22$ suggest a rapidly growing proportion of cospectral cubic graphs arise by iterated use of these local operations. However, this scheme only generates graphs whose cyclic-edge-connectivity is at most three.

Table: Prevalence of Replaceable Structures in Cubic Graphs (from (Haythorpe et al., 2021))

Order $N$	Total cubic graphs	Fraction with replaceable vertex (\%)	Fraction with replaceable edge (\%)
12	85	2.4	3.5
14	509	1.6	3.1
16	4060	1.2	2.8
18	41301	0.9	1.6

A conjecture, supported by increasing computational evidence, posits that "almost all" cospectral cubic graphs can be generated by such constructions as $N\to\infty$.

6. Spectral Rigidity in Cayley Cubic Graphs

For certain classes the spectrum does determine the graph. The cubic Cayley graphs on dihedral groups $D_{2p}$ (with $p$ odd prime) provide such an example: two cubic Cayley graphs on $D_{2p}$ are isomorphic if and only if they are cospectral. The spectral expressions derive from the character table of $D_{2p}$, and cospectrality aligns with equivalence of the Cayley connection sets up to automorphism. Thus, no two non-isomorphic cubic Cayley graphs on $D_{2p}$ are cospectral (Huang et al., 2016). This phenomenon is rare and highlights that spectral indistinguishability is not a universal feature among cubic graphs.

7. Limitations of Spectral Methods and Open Questions

Despite the spectrum determining regularity, order, bipartiteness, and vertex chromatic number in cubic graphs, it cannot distinguish edge-chromatic number, Hamiltonicity, or the existence of three edge-disjoint perfect matchings. The construction of cospectral pairs differing in these invariants demonstrates the inherent limitations of spectral methods for cubic graphs (Haemers, 7 Jan 2026). A further restriction arises in structural cospectral constructions: the smallest cospectral cubic graphs with different chromatic index cannot be obtained using classical GM-switching (which uses rational orthogonal matrices), as no rational orthogonal similarity exists in these cases (Yan et al., 2021). Open problems remain in identifying further properties not spectrally determined and in characterizing which cospectral cubic graphs escape known constructible families.

---

Spectral characterizations of cubic graphs illuminate a spectrum of detectability and indistinguishability among core graph invariants. The adjacency spectrum encodes significant structural information but fails to detect certain NP-hard properties, underlining its power and its limits for the combinatorial analysis of cubic graphs.