--- title: Derangement Graph Overview url: https://www.emergentmind.com/topics/derangement-graph type: topic --- # Derangement Graph Overview A derangement graph most commonly denotes the Cayley graph associated with a permutation action: if a finite permutation group \(G\) acts on a finite set \(\Omega\), and \(D(G)=\{g\in G: g\text{ fixes no point of }\Omega\}\) is the set of derangements, then the derangement graph is \(\Gamma(G,D(G))\), with vertex set \(G\) and adjacency \(g\sim h\) if and only if \(g^{-1}h\in D(G)\) [1310.2571]. In the literature, however, closely related but distinct notions also occur: the derangement graph on \(S_n\) with adjacency “disagree in every position” [2207.06466], generalized \(k\)-derangement graphs on \(S_n\) [1106.5522], derangement action digraphs on an arbitrary set \(X\) generated by fixed-point-free permutations [1804.01384], and “graph derangements,” which are fixed-point-free adjacency-respecting permutations of the vertex set of a graph and are explicitly not the same object as the permutation-group derangement graph [1306.6436]. The term therefore denotes a family of constructions unified by fixed-point-free behavior, but differentiated by whether the vertices are group elements, permutations, tuples, perfect matchings, or vertices of an underlying graph. ## 1. Core definition in permutation group theory For a finite permutation group \(G\leq \mathrm{Sym}(\Omega)\), the standard derangement graph is the Cayley graph \[ \Gamma_G=\operatorname{Cay}(G,\operatorname{Der}(G)), \] where \(\operatorname{Der}(G)\) is the set of elements of \(G\) with no fixed points on \(\Omega\) [2102.05250]. Equivalently, \[ g\sim h \iff g^{-1}h\in \operatorname{Der}(G), \] so adjacency records whether the relative permutation between two group elements is a derangement [2311.05575]. This graph is regular of valency \(|\operatorname{Der}(G)|\), vertex-transitive by the right-regular action of \(G\), and normal because the derangement set is a union of conjugacy classes [2009.01086]. For a normal Cayley graph, the adjacency eigenvalues are given by irreducible characters: \[ \lambda_\chi=\frac{1}{\chi(1)}\sum_{g\in \operatorname{Der}(G)}\chi(g), \] a formula repeatedly used in spectral analyses of derangement graphs [2102.05250]. The independent sets of \(\Gamma_G\) are precisely the intersecting families in \(G\): a subset \(S\subseteq G\) is intersecting when every \(g,h\in S\) satisfy that \(g^{-1}h\) fixes some point of \(\Omega\), equivalently \(g^{-1}h\notin D(G)\) [1310.2571]. This identification makes the derangement graph a standard encoding device for Erdős–Ko–Rado-type problems in permutation groups. Dually, cliques are subsets \(C\subseteq G\) such that all pairwise quotients are derangements [2311.05575]. Two recurrent numerical invariants are the independence number \(\alpha(\Gamma_G)\) and the clique number \(\omega(\Gamma_G)\). The clique–coclique bound \[ \alpha(\Gamma(G,D(G)))\cdot \omega(\Gamma(G,D(G)))\le |G| \] links clique growth to upper bounds on intersecting families [2311.05575]. This suggests why both cocliques and cliques are central in the literature: cocliques control EKR-type extremal problems, while cliques force corresponding upper bounds. ## 2. The symmetric-group derangement graph and its cycle structure For the classical derangement graph on \(S_n\), the vertex set is \(S_n\), and two permutations \(\sigma,\tau\) are adjacent exactly when they disagree in every position: \[ \sigma\sim \tau \iff \forall i\in[n],\ \sigma(i)\neq \tau(i), \] equivalently \(\tau\sigma^{-1}\in D_n\), where \(D_n\) is the set of derangements in \(S_n\) [2207.06466]. In this formulation the graph is the normal Cayley graph \[ \Gamma_n=\operatorname{Cay}(S_n,D_n), \] with \(|V(\Gamma_n)|=n!\) and degree \(!n=n!\sum_{k=0}^n(-1)^k/k!\) [2508.12618]. Several basic structural facts are established explicitly. The graph is connected for \(n\ge 4\); \(\Gamma_2\) is a single edge; and \(\Gamma_3\) is disconnected, consisting of two 3-cycles [2508.12618]. For connected cases, the diameter is \(2\): if \(n\ge 4\), then for \(u,v\in S_n\), one has \(\operatorname{dist}(u,v)=1\) if \(vu^{-1}\in D_n\), and otherwise \(\operatorname{dist}(u,v)=2\) [1610.06735]. The same paper uses the diameter-two identity \[ D=2(J-I)-A \] for the distance matrix \(D\) in terms of the adjacency matrix \(A\), enabling a direct translation from adjacency spectra to distance spectra [1610.06735]. The graph also has a rich cycle structure. It was already known to be Hamiltonian and Hamilton-connected, and the stronger result now recorded is that \(\Gamma_n\) is edge pancyclic for \(n\ge 4\): every edge lies on cycles of all lengths \(3,4,\dots,n!\) [2207.06466]. The proof combines a partition of \(S_n\) into \(n\)-cliques \(A_\tau\), a fixed-point counting identity \[ \sum_{i=0}^{n-1}\Delta(\alpha,\sigma^i\beta)=n \] for \(\sigma\in C_n\), and the edge pancyclicity of the complement of \(\Gamma_{n-1}\) [2207.06466]. The adjacency spectrum is controlled by the representation theory of \(S_n\). If \(\lambda\vdash n\), then the adjacency eigenvalue indexed by \(\lambda\) is \[ \eta_\lambda=\frac{1}{f^\lambda}\sum_{g\in D_n}\chi_\lambda(g), \] with multiplicity \((f^\lambda)^2\) [1207.3878]. A central recurrence is \[ \eta_{\lambda}=(-1)^{r-1}\lambda_r\,\eta_{\lambda-\hat{c}}+(-1)^{\lambda_r}\eta_{\lambda-\hat{l}},\qquad \eta_{\emptyset}=1, \] where \(\lambda-\hat{l}\) deletes the last row and \(\lambda-\hat{c}\) removes the first column [1207.3878]. The same work proves the Alternating Sign Property \[ \operatorname{sign}(\eta_\lambda)=(-1)^{|\lambda|-\lambda_1}, \] and strict monotonicity of \(|\eta_\lambda|\) under dominance order, settling the Ku–Wales conjecture on extremal eigenvalue magnitudes [1207.3878]. These spectral facts feed directly into extremal combinatorics. In particular, the smallest adjacency eigenvalue is attained at \((n-1,1)\), \[ \eta_{(n-1,1)}=-\frac{!n}{n-1}, \] and Hoffman's bound yields \(\alpha(\Gamma_n)=(n-1)!\), the classical EKR value for permutations [1207.3878]. ## 3. Intersecting families, EKR theorems, and geometric classification The derangement graph formalism is particularly effective for classification theorems on maximum intersecting sets. A definitive example is \(G=\mathrm{PGL}(3,q)\) acting on the points of the projective plane \(\mathrm{PG}(2,q)\). In this setting, \[ \Gamma=\operatorname{Cay}(G,D),\qquad D=\{g\in \mathrm{PGL}(3,q): g\text{ has no fixed point on }\mathrm{PG}(2,q)\}, \] with \[ |G|=(q^2+q+1)q^3(q^2-1)(q-1),\qquad |D|=\frac{(q^2-1)^2q^4}{3} \] [1310.2571]. The main theorem in that case classifies all maximum independent sets. If \(S\subseteq G\) is intersecting of maximum size, then \(S\) is a coset of the stabilizer of a point or a coset of the stabilizer of a line, and \[ |S|=q^3(q^2-1)(q-1) \] [1310.2571]. The abstract of the paper contains the phrase “points of the projective line,” but the paper’s theorem concerns the action on \(\mathrm{PG}(2,q)\), and this is explicitly identified as a typo [1310.2571]. The proof combines two layers. First, a ratio-bound argument based on character theory shows that the independence number is exactly \[ \alpha(\Gamma)=q^3(q^2-1)(q-1), \] using the minimum eigenvalue \[ \tau=-\frac{(q-1)(q^2-1)q^3}{3} \] and the largest eigenvalue \(|D|=(q^2-1)^2q^4/3\) [1310.2571]. Second, a linear-algebraic classification analyzes a matrix \(A\) indexed by group elements and ordered pairs of points, together with a block \(M\) consisting of the derangement rows. The characteristic vector of a maximum independent set lies in the column space of \(A\), and the right kernel of \(M\) is described explicitly in terms of points and lines in projective geometry [1310.2571]. This geometric outcome is noteworthy because the maximum cocliques are not of a single type: both point stabilizers and line stabilizers occur. The paper presents this as the first confirmed case of a higher-dimensional projective conjecture in which two genuinely different geometric families of maximum intersecting sets appear [1310.2571]. A plausible implication is that the derangement graph does not merely encode fixed-point-free behavior abstractly; in geometric actions it can detect dual incidence structures with full rigidity at the extremal level. More broadly, the connection between cocliques and intersecting families underlies a sequence of results for transitive groups. The existence of large cliques forces upper bounds on the intersection density \(\rho(G)\le n/\omega(\Gamma(G,D(G)))\), and this mechanism motivates the study of cliques in derangement graphs for innately transitive and other structured permutation groups [2311.05575]. ## 4. Cliques, multipartite structure, and Latin-square correspondences Clique structure in derangement graphs has developed into an independent theme. One universal result is that if \(G\) is transitive of degree at least \(3\), then the derangement graph \(\Gamma_G\) contains a triangle [2009.01086]. A stronger recent theorem states that if a transitive permutation group has degree exceeding \(30\), then its derangement graph contains a \(K_4\); the only exceptions listed are degrees \(1,2,3,6,18,\) and \(30\) with specific groups [2502.01287]. This theorem is used to deduce a bound \(|G:U|\le 10\) in an index-\(3\) covering problem related to Kronecker classes [2502.01287]. A different line of work studies when the whole derangement graph is complete multipartite. For a transitive group \(G\), define \[ F(G):=\langle G_\omega:\omega\in\Omega\rangle =\big\langle \{g\in G:\pi(g)>0\}\big\rangle. \] Then \(\Gamma_G\) is complete multipartite if and only if \(F(G)\) is intersecting; in that case the parts are exactly the left cosets of \(F(G)\) [2102.05250]. Two infinite families are constructed explicitly. One uses a subgroup \(G_q(A)\le \mathrm{AGL}(2,q)\) acting transitively on the affine lines of \(\mathrm{AG}(2,q)\), producing a complete \((q+1)\)-partite derangement graph of degree \(q(q+1)\) [2102.05250]. The other gives, for odd \(\ell\), a transitive group of degree \(4\ell\) whose derangement graph is complete \(2\ell\)-partite [2102.05250]. The complete multipartite case is spectrally rigid. For a complete \(t\)-partite graph with equal part size \(m\), \[ (v,k,\lambda,\mu)=\bigl(tm,\ m(t-1),\ (t-2)m,\ (t-1)m\bigr), \] and the spectrum is \[ k,\quad -m\ \text{with multiplicity }t-1,\quad 0\ \text{with multiplicity }t(m-1) \] [2102.05250]. In the affine-line family, this yields \[ v=q^2(q^2-1),\qquad k=q^3(q-1), \] while the maximum cocliques are exactly the cosets of \(M_q=F(G_q(A))\) [2102.05250]. A separate but related clique theory emerges for the symmetric-group derangement graph \(X_N\), where every maximal clique has size exactly \(N\) [2407.14155]. Here a Latin square \(L\) of order \(N\) gives a maximal clique \[ C_L=\{\sigma_k:k\in[N]\}, \qquad \sigma_k(i)=j\ \text{iff}\ L_{ij}=k, \] and two orthogonal Latin squares produce two disconnected maximal cliques [2407.14155]. The paper proves a bijection between pairs of orthogonal Latin squares and pairs of disconnected ordered maximal cliques in \(X_N\) [2407.14155]. This bijection is then combined with modular obstructions derived from the natural representation. For \(B\subset S_N\), \[ P_{\mathrm{nat}}(1_B)(\tau)=\sum_{\sigma\in B}\operatorname{Fix}(\sigma\tau^{-1}), \] and in characteristic dividing \(N\) the dimension of the image drops from \((N-1)^2+1\) to \((N-1)^2-2N+4\) [2407.14155]. These modular dependencies are used to analyze small \(N\), culminating in a short proof that there do not exist two disconnected maximal cliques in \(X_6\), hence no pair of orthogonal Latin squares of order \(6\) [2407.14155]. ## 5. Generalizations on permutation spaces The ordinary derangement graph on \(S_n\) is only one member of a broader family of Cayley and association-scheme constructions. A first extension is the generalized \(k\)-derangement graph \(\mathcal{D}_{n,k}\) on \(S_n\), where a permutation is a \(k\)-derangement if it fixes no \(k\)-subset setwise [1106.5522]. Two permutations \(\sigma,\tau\) are adjacent when \(\sigma\tau^{-1}\) is a \(k\)-derangement [1106.5522]. The cycle criterion states that \(\sigma\) is a \(k\)-derangement if and only if its cycle decomposition contains no submultiset of cycle lengths summing to \(k\) [1106.5522]. For \(n>3\) and \(k