Cospectral Covering Graphs

• Cospectral covering graphs are graph lifts that share identical adjacency spectra yet may be non-isomorphic, defined through voltage assignments and group representations.
• They are constructed by lifting cospectral base graphs or highly symmetric graphs like Paley graphs, using algebraic tools such as character theory and twisted matrices.
• These constructions offer applications in network design and molecular chemistry, challenging the limits of spectral invariants in distinguishing graph structures.

Cospectral covering graphs are pairs or families of graph covers (lifts) whose adjacency spectra coincide as multisets, yet the covers may be non-isomorphic as combinatorial objects. This phenomenon arises naturally in the study of regular covers via voltage assignment and group representations and has implications for spectral graph theory, network design, and mathematical chemistry. Key constructions proceed by lifting cospectral graphs, or by lifting highly regular graphs such as Paley graphs, where symmetry permits detailed spectral analysis. The cospectrality of lifts – especially non-trivial covers – illuminates fundamental limits of spectral invariants and motivates systematic exploration of algebraic graph operations.

1. Preliminaries: Graph Covers and Voltage Assignments

Let $G = (V, E)$ be a finite simple graph. A $k$-fold cover (or lift) of $G$ is a graph $G' = (V', E')$ equipped with a surjective map $\pi : V' \to V$ such that for each $v \in V$, $|\pi^{-1}(v)| = k$, and for every edge $\{u, v\} \in E$, and $v' \in \pi^{-1}(v)$, there is exactly one edge in $E'$ joining $v'$ to some $u' \in \pi^{-1}(u)$, ensuring local bijections of neighborhoods. The algebraic machinery encoding lifts employs group-valued voltage assignments: given a finite group $G_r$ of order $k$ and a function $s : \{\text{oriented edges of }G\} \to G_r$ subject to $s((v, u)) = s((u, v))^{-1}$, the derived graph $G(s)$ has vertex set $V \times G_r$, with edges determined by the voltage function, typically $h = g \cdot s((u, v))$ for each oriented edge. When $G_r$ is Abelian, vertices and adjacency matrices of the lift admit block-decomposition via the group’s characters, which is fundamental for spectral analysis (Ramezani, 2016).

2. Main Existence Results for Cospectral Covering Graphs

The cospectrality of lifts is governed by the behavior of their adjacency matrices under block construction. Given graphs $G$ and $H$ that are cospectral (i.e., $\text{spec}\,G = \text{spec}\,H$), and a finite group $G_r$ of order $k$ with symmetric permutation matrix $P_g$ for $g \in G_r$ satisfying $g^2 = e$, constant voltage assignments $s_G(e) = s_H(e) = g$ for all edges yield lifted graphs $G_g = G(s_G)$ and $H_g = H(s_H)$ that are cospectral non-trivial covers. Formally, $A(G_g) = A(G) \otimes P_g$, $A(H_g) = A(H) \otimes P_g$. The spectrum of the lift is the multiset product $$\{\lambda_i \omega_j : i = 1 \ldots n, j = 1 \ldots k\}$$ where $\text{spec}\,G = \{\lambda_i\}$ and $\omega_j$ are eigenvalues of $P_g$ (Ramezani, 2016).

3. Spectral Analysis via Group Representations and Character Theory

General voltage assignments permit refined spectral decompositions. For Abelian $G_r$, all irreducible representations are one-dimensional. In a shared basis, for each character $\chi_\ell : G_r \to \mathbb{C}^\times$, the twisted adjacency matrix $$A_{\chi_\ell} = \sum_{\{u,v\} \in E} \chi_\ell(s(u,v)) E_{u,v} + \chi_\ell(s(u,v))^{-1} E_{v,u}$$ encodes adjacency in the lift with weights from the character. The spectrum of $G(s)$ is the multiset-union of the spectra of these twisted adjacency matrices. If for each character $\chi$ the matrices $A_{\chi}^G$ and $A_{\chi}^H$ are cospectral, then the lifts $G(s)$ and $H(s')$ are cospectral as well. For non-Abelian $G_r$, higher-dimensional representations lead to block-matrix constructions $\sum_g A_g \otimes \rho(g)$ (Ramezani, 2016).

For Paley graphs, cyclic covers of prime order $m \ne p$ over finite fields $F_q$ are given by translation-invariant voltage assignments $\alpha : F_q^{\boxtimes} \rightarrow \mathbb{Z}/m$. The eigenvalues are computed by twisted character sums: $\theta_{a, k}^\alpha = \sum_{s \in F_q^{\boxtimes}} \zeta_p^{\Tr(a s)} \zeta_m^{k \alpha(s)}$ where $a \in F_q$ and $k \in \mathbb{Z}/m$, and the spectrum of the cover is the collection of all $\theta_{a, k}^\alpha$ (Dinin et al., 17 Jan 2026).

4. Construction and Analysis of Cospectral Non-Isomorphic Covers

Cospectral non-isomorphic covers are found explicitly for base graphs of sufficient complexity or symmetry, such as pairs of cospectral graphs on six vertices or lifted Paley graphs when $q = p^r > p$. In the Abelian cover construction, signatures $s$ and $s'$ are chosen on $G$ and $H$ so that all character-twisted adjacency matrices match in spectrum, even if their combinatorics do not coincide.

A fundamental result for Paley graphs states that for $q = p$, the spectrum of a translation-invariant cyclic cover determines the graph isomorphism class. However, for $q = p^r > p$, trace-preserving involutive permutations of orbits can be used to construct voltage assignments $\alpha, \beta$ such that the resulting covers $X^\alpha$ and $X^\beta$ are cospectral but non-isomorphic. The construction is made explicit via permutation polynomials $f(T)$ with $f(-T) = -f(T)$ relating the spectra through additive characters and an orthogonal change-of-basis matrix (Dinin et al., 17 Jan 2026).

Concrete examples are tabulated below:

Base Graph	Cover Parameters	Cospectrality	Isomorphism Class
Paley, $q=p$	$m\ne p$	Yes	Determined by spectrum
Paley, $q=p^r>p$	$m=3$, $q=25$	Yes	Non-isomorphic
$6$-vertex pair	$3$-lift, $\mathbb{Z}/2 \times \mathbb{Z}/2$	Yes	Non-isomorphic

5. Applications and Extensions

Iterated lifts produce infinite families of mutually cospectral, pairwise non-isomorphic graphs. These constructions serve applications in network design, particularly for creating large graphs with specified spectral gap, and in chemistry, in the design of isospectral molecules -- graphs modeling molecular structures with identical vibrational spectra. The algebraic approach, involving block decompositions, character theory, and permutation polynomials over finite fields, reveals deep connections between spectral graph theory, geometric group theory, and Galois theory.

A plausible implication is that the limits of spectral invariants in discriminating graph isomorphism are governed by the group-theoretic and combinatorial intricacy of the covering transformations and voltage assignments. The interaction of cospectral covers with other graph operations such as Godsil–McKay switching and graph products is largely unexplored, suggesting promising new directions for constructing broader families of cospectral objects (Ramezani, 2016).

6. Open Problems and Theoretical Significance

Current research seeks a full characterization of the signatures on a base graph $G$ that admit "twin signatures" on a cospectral $H$ such that all character-twisted adjacency spectra coincide. The behavior for non-Abelian cover groups, involving higher multiplicities and richer representation theory, remains an active area. The discovery that prime-cyclic covers of strongly regular graphs such as Paley graphs can be cospectral but non-isomorphic demonstrates the subtlety of the "hearing the shape of a graph" question – the spectrum does not always determine the graph's structure.

Cospectral covering graphs thus represent fundamental objects in spectral graph theory, providing test cases for conjectures about the determinacy of spectral invariants, and supplying a methodological bridge between combinatorial, algebraic, and geometric properties of graphs (Ramezani, 2016, Dinin et al., 17 Jan 2026).