---
title: Graph Automorphism Representation
url: https://www.emergentmind.com/topics/graph-automorphism-representation
type: topic
---

# Graph Automorphism Representation

Searching arXiv for recent and foundational papers on graph automorphism representation.
arxiv.search({"query":"all:(graph automorphism representation) OR ti:(automorphism group of a graph) OR abs:(automorphism representation of a graph)", "max_results": 10, "sort_by": "relevance"})
Reviewing the search results for papers directly relevant to representation-theoretic and realization-based notions of graph automorphisms.
Graph automorphism representation denotes a family of constructions in which graph automorphisms are encoded, analyzed, or realized through explicit mathematical models. In the literature surveyed here, the term covers at least four recurrent settings: the permutation action of $\operatorname{Aut}(G)$ on vertices and associated linear actions on $\mathbb{C}^n$ or $\mathbb{R}^n$; homological and Jacobian actions attached to a graph; auxiliary combinatorial objects that record automorphism orbits or endomorphisms; and realization theorems in which prescribed groups occur as full automorphism groups of graphs or graph-derived structures [1312.2778; 2102.04239; 2206.01469; 2505.21299]. The common theme is that automorphisms are not treated merely as abstract group elements, but as concrete transformations acting on a labeled vertex set, on eigenspaces, on cycle spaces, or on canonical representation data.

## 1. Permutation actions and the meaning of “representation”

For a graph $G=(V,E)$ with $n$ vertices, an automorphism is a bijection of $V$ preserving adjacency. Identifying $V$ with $[n]=\{1,\dots,n\}$ turns every automorphism into a permutation, hence into a permutation matrix acting on a vector space. One standard action is
\[
\sigma\boldsymbol{u}=(u_{\sigma^{-1}1},u_{\sigma^{-1}2},\dots,u_{\sigma^{-1}n})^t,
\]
so $\operatorname{Aut}(G)$ acquires a faithful linear representation on $\mathbb{C}^n$ or $\mathbb{R}^n$ [1312.2778; 1607.00547].

A more refined notion appears in distinguishing theory. Given a labeling $l:V(G)\to S$, the automorphism representation of $G$ under $l$ is
\[
Aut(G,V(G)_l):=\{\,l(\alpha):\alpha\in Aut(G)\,\}.
\]
This records not just the abstract isomorphism type of $\operatorname{Aut}(G)$, but the actual permutation set after labeling. The distinction is essential: the literature gives graphs $G_1,G_2$ with
\[
Aut(G_1)\cong Aut(G_2)\cong S_3
\]
but
\[
D(G_1)=3,\qquad D(G_2)=2,
\]
showing that isomorphic automorphism groups do not determine the same distinguishing behavior [2505.21299].

This suggests a basic conceptual separation. The abstract group $\operatorname{Aut}(G)$ answers which symmetries exist; an automorphism representation answers how those symmetries act on vertices, coordinates, cycles, or other structured carriers. Much of the subsequent theory depends on the latter.

## 2. Spectral, geometric, and block-theoretic representations

A central spectral characterization states that a permutation $\sigma$ is an automorphism of $G$ if and only if every eigenspace of the adjacency matrix $\mathbf{A}(G)$ is $\sigma$-invariant. Equivalently, automorphisms are exactly the permutations whose permutation matrices commute with $\mathbf{A}(G)$:
\[
P_\sigma^t A P_\sigma = A,
\qquad\text{equivalently}\qquad
P_\sigma A = A P_\sigma.
\]
Thus every eigenspace of $\mathbf{A}(G)$ is $\operatorname{Aut}(G)$-invariant, and the span of the orbit of an eigenvector under $\operatorname{Aut}(G)$ remains inside the corresponding eigenspace [1312.2778].

This spectral viewpoint leads to a representation-theoretic analysis of orbit spans. If
\[
\mathbb{C}^n=\bigoplus_{i=1}^h V_i
\]
is the canonical decomposition into isotypic components, with
\[
V_i=\bigoplus_{j=1}^{m_i} W_{ij},
\]
the maximal possible value of $\dim \operatorname{span}(\mathfrak{G}\boldsymbol{v})$ is
\[
\sum_{i=1}^h \min\{\dim V_i,\; (\dim W_i)^2\},
\]
and the paper further gives an exact formula for arbitrary $\boldsymbol{v}$ in terms of the irreducible decomposition [1312.2778]. This turns eigenspaces into explicit representation spaces for graph symmetries.

A parallel development connects permutation representations with block systems. The action of $\mathfrak{G}=\operatorname{Aut}(G)$ on $V(G)$ yields orbits and blocks, and the paper “On the Automorphism Group of a Graph” studies how block systems are encoded by irreducible representations of $\mathfrak{G}$ inside the permutation representation on $\mathbb{R}^n$ [1607.00547]. Equitable partitions are used as an intermediate structure:
\[
\mathbf{A}(G)\mathbf{R}=\mathbf{R}\mathbf{A}(G/\Pi),
\]
and the column space of $\mathbf{R}$ is $\mathbf{A}$-invariant precisely when $\Pi$ is equitable. In this framework, projections of characteristic vectors onto irreducible invariant subspaces recover the orbit partition of block stabilizers. The same paper culminates in an algorithm solving the structure problem of an automorphism group in time
\[
n^{C\log n}
\]
for some constant $C$, producing a generating set, all block systems, and decompositions of eigenspaces into irreducible representations [1607.00547].

A plausible implication is that graph automorphism representation is not merely a language for symmetry; it is also a compression device. Spectral subspaces, equitable partitions, and irreducible constituents isolate the parts of vertex symmetry that are visible to linear algebra.

## 3. Homology, cycle space, and the Jacobian

Another major line of work represents graph automorphisms on homological invariants. For a finite connected simple graph $X$, with Betti number
\[
\beta(X)=e(X)-v(X)+1,
\]
a spanning tree $T$ determines a basis of fundamental cycles in
\[
H_1(X,\mathbb{Z})\cong \mathbb{Z}^{\beta(X)}.
\]
This yields a homomorphism
\[
\Theta_T:\mathrm{Aut}(X)\to U_\beta,
\qquad
U_\beta=\{A\in M_\beta(\mathbb{Z}) : \det(A)=\pm 1\},
\]
obtained by expressing the action of an automorphism on the fundamental-cycle basis as a unimodular integer matrix [2102.04239].

The faithfulness problem is completely classified. The representation $\Theta_T$ is not injective if and only if at least one of the following holds: $X$ is a tree and $X\neq 1$; $X$ contains a pendant tree $S$ such that $S\neq 1$; or $X$ is periodic unicyclic [2102.04239]. In particular, if $X$ has no pendant vertices and is not a simple cycle, then $\Theta_T$ is faithful, so $\operatorname{Aut}(X)$ acts faithfully on $H_1(X,\mathbb{Z})$. The paper presents this as a discrete analogue of the classical faithfulness of automorphism actions on homology for Riemann surfaces of genus greater than one [2102.04239].

A related but distinct construction uses the Jacobian $\mathrm{Jac}(X)$, also called the critical group or Picard group. For a connected graph $X$,
\[
|\mathrm{Jac}(X)|=\tau(X),
\]
the number of spanning trees. Fixing a harmonic $\mathrm{Jac}(X)$-flow $\xi$, one obtains a homomorphism
\[
\Theta:\mathrm{Aut}(X)\to \mathrm{Aut}(\mathrm{Jac}(X))
\]
by transport of structure on darts [2206.01469]. The main faithfulness theorem states that if $X$ is simple, connected, and $3$-edge-connected, and if $G\le \mathrm{Aut}(X)$ acts semiregularly on darts and vertices, then $\Theta(G)\cong G$ [2206.01469].

This has immediate structural consequences. If a connected, $3$-edge-connected graph admits a nonabelian semiregular group of automorphisms, then $\mathrm{Jac}(X)$ has rank at least $2$; in particular, a Cayley graph arising from a nonabelian group and of degree at least three has non-cyclic Jacobian [2206.01469]. The semiregularity hypothesis is essential: the paper gives an example with
\[
|\mathrm{Aut}(X)|=72,\qquad \mathrm{Jac}(X)\cong \mathbb{Z}_3\times \mathbb{Z}_3,
\]
so
\[
\mathrm{Aut}(\mathrm{Jac}(X))\cong \mathrm{GL}_2(3)
\]
of order $48$, and therefore $\mathrm{Aut}(X)$ does not embed into $\mathrm{Aut}(\mathrm{Jac}(X))$ [2206.01469].

Together, these results show that graph automorphism representation on cycle space and on the Jacobian detects different layers of symmetry. Homology captures cycle-level rigidity; the Jacobian constrains arithmetic structure such as cyclicity and rank.

## 4. Orbit encodings and derived combinatorial models

A different use of automorphism representation is to build an auxiliary graph or labeled structure whose adjacency records automorphism orbits. For a commutative ring $R$ with identity, the graph
\[
\Gamma_{\aut}(R)
\]
has vertex set $R$, and two distinct vertices $x,y$ are adjacent if and only if there exists $\sigma\in \aut(R)$ such that $\sigma(x)=y$ [1003.0174]. In this representation, each orbit $O(x)$ is exactly the clique containing $x$, every clique is some orbit, and the degree of $x$ is $|O(x)|-1$ [1003.0174]. The construction translates orbit-space questions into graph-theoretic invariants such as connectivity and planarity. For instance,
\[
\Gamma_{\aut}(R)\text{ is totally disconnected} \iff \Aut(R)=\mathrm{id},
\]
and for a finite ring,
\[
\Gamma_{\aut}(R)\text{ planar} \iff R\text{ is of type }\le 3
\]
[1003.0174].

In graph $C^\ast$-algebras, the representation object is a labeled directed multigraph. Given an endpoint-fixing permutation $\tau$ of length-$k$ paths, the associated permutation graph $(E_\tau,L_\tau)$ has
\[
E_\tau^0 = E^{k-1}, \qquad E_\tau^1 = E^k,
\]
and labels
\[
L_1(\mu)=\mu_1,\qquad L_2(\mu)=\tau(\mu)_k,\qquad L_\tau(\mu)=[L_1(\mu),L_2(\mu)].
\]
This graph gives a visual representation of the endomorphism $\lambda_\tau$ and turns automorphism recognition into a synchronization problem [1401.4274]. The decisive criterion is
\[
(E_\tau,L_\tau)\text{ synchronizing in both labels} \iff T_\tau \text{ is two-sided 1--1} \iff \lambda_\tau \text{ is an automorphism of }C^\ast(E)
\]
[1401.4274].

The distinguishing-number literature pushes the labeled-permutation viewpoint even further. If
\[
Aut(G_1,V(G_1)) = Aut(G_2,V(G_2)),
\]
then
\[
D(G_1)=D(G_2),
\]
but the converse statement with only
\[
Aut(G_1)\cong Aut(G_2)
\]
is false [2505.21299]. This corrects a common misconception: for distinguishing colorings, the relevant datum is not the abstract group but the permutation representation induced by a labeling. The same paper develops this idea to study the cost number $\rho(G)$ and proves that when
\[
D(G)=2,\qquad Det(G)=2,
\]
one has
\[
\rho(G)=2,\ 3\text{ or }4
\]
[2505.21299].

These constructions share a common strategy. Rather than attacking automorphisms directly, they externalize the action into a derived graph, multigraph, or labeled permutation system where orbit structure becomes locally readable.

## 5. Realization theorems and representable classes

A classical realization problem asks which groups occur as automorphism groups of graphs. The adversarial vertex-deletion framework generalizes Frucht-type realization by showing that one can encode not only one group but an entire deletion process. Given finite groups $T_0,T_1,\dots,T_k$ and $\ell>1$, there exists a graph $G_0$ with
\[
\mathrm{Aut}(G_0)\cong T_0
\]
such that, regardless of the adversary’s sequence of $\ell$ choices from $\{T_1,\dots,T_k\}$, vertices can be deleted one by one so that after round $j$ the automorphism group is the requested $T_{i_j}$ [1201.4367]. The construction relies on a gadget $H$ with trivial automorphism group but
\[
\mathrm{Aut}(H-x)\cong T,
\]
so deletion reveals a prescribed symmetry [1201.4367].

For restricted graph classes, the representable groups can often be classified exactly. For bicyclic graphs, if $\mathcal{S}$ denotes the class of groups representable as automorphism groups of bicyclic graphs, then
\[
\mathcal{S}=\mathcal{T}\cup \mathcal{B}_1\cup \mathcal{B}_2,
\]
where $\mathcal{T}$ is Jordan’s class of automorphism groups of trees and $\mathcal{B}_1,\mathcal{B}_2$ are explicit families built from groups in $\mathcal{T}$ by direct products, wreath products, and semidirect products with $\mathbb{Z}_2\times \mathbb{Z}_2$ [2104.02325]. Every bicyclic graph has automorphism group in $\mathcal{S}$, and every group in $\mathcal{S}$ is realizable by a bicyclic graph [2104.02325].

Geometric graph classes support another kind of representation theory, based on canonical data structures encoding all geometric realizations. For interval graphs, PQ-trees show that interval graphs have exactly the same automorphism groups as trees, and for each interval graph one can construct a tree with the same automorphism group [1407.2136]. For permutation graphs and circle graphs, modular trees and split trees yield inductive descriptions of automorphism groups via products, homomorphisms, group actions, semidirect products, and wreath products [1407.2136]. At the opposite extreme of restriction, every abstract group can be realized as the automorphism group of a comparability graph, or of a poset of dimension at most four [1407.2136].

A more specialized realization notion is Frobenius graphical representation. A group $G$ has a GFR if there exists a simple graph $\Gamma$ with
\[
\operatorname{Aut}(\Gamma)\cong G
\]
and $G$ acts on vertices as a Frobenius group; in that case $\Gamma$ is necessarily a Cayley graph on the Frobenius kernel [1909.03690]. The paper proves that infinitely many Higman groups $A(f,q_0)$ admit GFRs, providing an infinite family with non-abelian $2$-group Frobenius kernel [1909.03690].

A plausible implication is that “graph automorphism representation” has a dual meaning in the realization literature: not only representing automorphisms by matrices or actions, but representing abstract groups by graph symmetries under structural constraints.

## 6. Cayley graphs, local rigidity, and explicit automorphism-group determinations

Concrete case studies show how automorphism representations are computed in practice. For the Andrásfai graph
\[
And(k)=Cay(\mathbb{Z}_{3k-1};C),
\qquad
C=\{3t+1\mid 0\le t\le k-1\},
\]
the full automorphism group is
\[
Aut(And(k)) \cong \mathbb{D}_{2n},
\qquad n=3k-1.
\]
The proof identifies translations
\[
f_v(x)=x+v
\]
and inversion
\[
a(x)=-x
\]
as automorphisms, then shows via the stabilizer of $0$ and a connected bipartite subgraph $H$ that these generate all symmetries [2105.07594].

For Cayley graphs generated by transpositions, normality depends on the local cycle structure of the transposition graph. If $S$ is a set of transpositions whose transposition graph has girth at least $5$, then
\[
\operatorname{Aut}(\operatorname{Cay}(H,S)) = R(H)\rtimes \operatorname{Aut}(H,S),
\]
where $H=\langle S\rangle$ and $R(H)$ is the right regular representation [1303.5974]. The uniqueness of certain $4$- and $6$-cycles is what forces automorphisms fixing the identity to come from group automorphisms [1303.5974].

Two prominent counterexamples to normality are the complete transposition graph and the complete alternating group graph. For the complete transposition graph on $S_n$,
\[
\Aut(\Cay(S_n,S)) = (R(S_n) \rtimes \Inn(S_n)) \rtimes \mathbb{Z}_2,
\]
where the extra $\mathbb{Z}_2$ is generated by inversion
\[
h:\alpha\mapsto \alpha^{-1},
\]
and the graph is not normal for all $n\ge 3$ [1404.7363]. Likewise, for the complete alternating group graph
\[
CAG_n=\mathrm{Cay}(A_n,S)
\]
with $S$ the set of all $3$-cycles,
\[
\mathrm{Aut}(CAG_n)=\bigl(R(A_n)\rtimes \mathrm{Inn}(S_n)\bigr)\rtimes \mathbb{Z}_2
\]
for $n\ge 5$, again with inversion supplying the non-normal extra symmetry [1605.06664].

These examples clarify an important point. In Cayley settings, the right regular action gives a canonical representation of the vertex set, but the full graph automorphism group can exceed the expected semidirect product by additional symmetries such as inversion. Local neighborhood analysis, distance-layer arguments, and rigidity of stabilizers are the usual tools for proving that no further automorphisms exist [1404.7363; 1605.06664].

Taken together, these directions show that graph automorphism representation is not a single formalism but a research program. It includes permutation representations on vertices, invariant-subspace descriptions via adjacency eigenspaces, unimodular and Jacobian actions on cycle-theoretic invariants, orbit graphs and labeled multigraphs that externalize symmetry, and realization theorems identifying which groups occur as full automorphism groups in specified graph classes. The unifying principle is that automorphisms become mathematically tractable when they are represented on a structured carrier where orbit, stabilizer, and rigidity phenomena can be read off explicitly.

Source: https://www.emergentmind.com/topics/graph-automorphism-representation