betaFeedback

WorksMathematics

Which groups are automorphism groups of racks?

Abstract

Every group is the automorphism group of a graph, of a magma, and even of a set-theoretic solution of the Yang-Baxter equation. We show that this fails for racks and quandles: the quaternion group of order eight is not the automorphism group of any rack, and it is the smallest group with this property. More generally, generalized quaternion groups and Hamiltonian groups are never automorphism groups of racks, and a dicyclic group is the automorphism group of a rack if and only if its order is not divisible by eight. On the positive side, dihedral, alternating and all finite abelian groups are automorphism groups of finite quandles.

Introduction

Which abstract groups can be realized as the full automorphism group of a structure of a prescribed kind? This old question, known as the representation problem or abstract group problem, goes back to Kőnig [34] and has accompanied the development of algebra and combinatorics ever since. In Babai’s terminology [4], an object OO of a category CC represents a group GG if Aut⁡(O)≅G\operatorname{Aut}(O) \cong G; the category CC is universal if every group is represented by an object of C\mathcal{C}, and finitely universal if every finite group is represented by a “finite” object of C\mathcal{C} [5 Section 4.1]. The first and best known answer is Frucht’s theorem [28]: the category of graphs is finitely universal, i.e. every finite group is the automorphism group of a finite graph. Two decades later, de Groot [19] and Sabidussi [45] proved, independently, that it is also universal. Under the name of the abstract group problem, the same question is the starting point of the theory of representations of groups and categories developed in [43]. For more details on this classical problem, which is still of current interest, see [4, 5, 39, 43] and the references therein.

Babai also pointed out that positive answers are the rule rather than the exception: even the category of fields is universal [5 Section 4.1]. The exceptions are therefore all the more interesting. The best known one concerns groups themselves: there is no group whose automorphism group is cyclic of order 3, so the category of groups is neither universal nor finitely universal. The reason is structural: the automorphism group of a group GG always contains the normal subgroup Inn⁡(G)≅G/Z(G)\operatorname{Inn}(G) \cong G/Z(G) of inner automorphisms, and this imposes strong restrictions on Aut⁡(G)\operatorname{Aut}(G). Another exception, of a different nature, is the category of all left non-degenerate solutions of the Frobenius–Separability equation, which is not universal either: the automorphism groups of these objects are precisely the centralizers of single maps, and not every group is of this form [2 Theorems 2.19 and 4.2].

In [39] we addressed the representation problem for the category YB\mathcal{YB} of all set-theoretic solutions of the Yang–Baxter equation, and we showed that this category is universal and finitely universal [39 Theorem 2.18]. In the same spirit, the centralizer of any finite family of maps is the automorphism group of an idempotent solution of the braid equation [41 Theorem 3.1].

In this paper we consider the same problem for the category Rack\mathsf{Rack} of all racks and the category Qnd\mathsf{Qnd} of all quandles. Via the derived solutions recalled below, Rack\mathsf{Rack} is a full subcategory of YB\mathcal{YB}; nevertheless, unlike YB\mathcal{YB}, neither Rack\mathsf{Rack} nor Qnd\mathsf{Qnd} is universal or finitely universal.

Racks and quandles are, by now, classical objects. Their axioms are the algebraic counterparts of the Reidemeister moves of knot diagrams: Joyce [33] and Matveev [38] associated, independently, a quandle with every knot and proved that it is an almost complete invariant of knots, while racks were developed by Fenn and Rourke [25] in connection with framed links. Racks are equally important in the theory of quantum groups: the braided vector spaces arising from groups, whose Nichols algebras are a key step in the classification of finite-dimensional pointed Hopf algebras, are described by racks and 2-cocycles [3 Theorem 4.14], and this led Andruskiewitsch and Graña to a classification of all finite simple racks [3 Theorems 3.9 and 3.12]. Quandles have recently been proposed also as a framework for the observables of physical theories [27]. For more details on the theory of racks and quandles, see the monographs [9, 23] and the references therein.

In view of their role in the theory of braided vector spaces, it is not surprising that racks also provide solutions of the celebrated quantum Yang–Baxter equation in its set-theoretic form. In fact, they are a natural and very rich source of bijective non-degenerate solutions of the braid equation: for instance, they are far more numerous than the solutions arising from skew left braces, as Table 1 below shows. If (X,⊲)(X,\lhd) is a rack, then

nn1234567
derived solutions of racks, r(n)r(n)12619743532080
derived solutions of quandles, q(n)q(n)11372273298
solutions arising from skew left braces, s(n)s(n)1112161

Table 1. Isomorphism classes of solutions of the braid equation on a set with nn elements: derived solutions of racks and of quandles, and solutions arising from skew left braces.

R ⁣:X×X→X×X,R(x,y)≔(y,x⊲y)R\colon X \times X \to X \times X,\qquad R(x,y) \coloneqq(y,x \lhd y)

is a bijective non-degenerate solution of the braid equation, as first observed in 1988 by Brieskorn [16] (see also [9 Remark 6.2]), and Aut⁡(X,R)=Aut⁡(X,⊲)\operatorname{Aut}(X,R)=\operatorname{Aut}(X,\lhd).

The automorphism groups of racks and quandles have been studied intensively, but in the opposite direction to the one considered in this paper: a rack or a quandle, usually constructed from a group or a module, is given, and its automorphism group is computed. This has been done for the quandles of small order [22, 30, 37, 46], the dihedral quandles [22], the Alexander quandles [11, 26, 31], the conjugation and Takasaki quandles of groups [8, 10, 47],1 the extensions of quandles [10, 12], and the quandles with one non-trivial column [17]; see also [9 Chapter 8] and the references therein. The representation problem asks the converse question:

Given a group GG, is there a rack or a quandle whose automorphism group is isomorphic to GG?

As far as we know, this question has not been considered before, and the answer turns out to be negative. As in the classical case of groups, the obstruction comes from the inner automorphisms: the inner automorphism group Inn⁡(X)\operatorname{Inn}(X) of a rack XX is a normal subgroup of Aut⁡(X)\operatorname{Aut}(X) [33]. The proof is based on a simple observation (Lemma 2.1): if G=Aut⁡(X)G=\operatorname{Aut}(X) and x∈Xx\in X, then every element of GG which normalizes the stabilizer GxG_x and commutes with the right translation rxr_x belongs to GxZ(G)G_x Z(G). Applied to a point xx which is moved by a suitable central element of GG, this gives a general obstruction (Theorem 2.2), and we obtain:

Theorem A. (1) Let GG be a non-abelian group and assume that there exists 1≠z∈G1\ne z\in G such that z∈⟨g⟩z\in\langle g\rangle, for all g∈G∖Z(G)g\in G\setminus Z(G). Then GG is not isomorphic to the automorphism group of any rack, finite or infinite. This is the case for the groups Q2m×E×AQ_{2^m}\times E\times A, where Q2mQ_{2^m} is a generalized quaternion group, EE is an elementary abelian 22-group and AA is an abelian group whose elements have odd order; in particular, no Hamiltonian group is the automorphism group of a rack.

(2) The dicyclic group Dic⁡n\operatorname{Dic}_n of order 4n4n, n≥2n\ge2, is isomorphic to the automorphism group of a rack if and only if nn is odd; in this case, Dic⁡n\operatorname{Dic}_n is the automorphism group of a quandle with n+4n+4 elements.

(3) The quaternion group Q8Q_8 is the smallest group which is not isomorphic to the automorphism group of a rack or of a quandle.

Consequently, the categories of racks and of quandles are neither universal nor finitely universal.

Theorem A is proved in Section 2 (Corollary 2.3, Corollary 2.4, Corollary 2.5 and Corollary 2.6) and Section 3 (Proposition 3.9 and Corollary 3.3). It also shows that a Frucht-type theorem for non-degenerate solutions of the braid equation, if it exists, cannot be obtained from racks. For comparison, Q8Q_8 is the automorphism group of a degenerate (bipartite) solution on a set with 11 elements [39 Example 2.22]. The obstruction of Theorem A is subtle: for instance, the groups Dic⁡6≅C3⋊Q8\operatorname{Dic}_6\cong C_3\rtimes Q_8 and SL⁡(2,3)≅Q8⋊C3\operatorname{SL}(2,3)\cong Q_8\rtimes C_3 both have order 24 and a unique involution, but only the second one is the automorphism group of a quandle. On the positive side, we prove in Section 3:

Theorem B. (1) Let g1,…,gkg_1,\ldots,g_k be pairwise commuting and pairwise non-conjugate permutations of a set ZZ, all different from Id⁡Z\operatorname{Id}_Z. Then the centralizer Aut⁡(g1,…,gk)\operatorname{Aut}(g_1,\ldots,g_k) of g1,…,gkg_1,\ldots,g_k in Sym⁡(Z)\operatorname{Sym}(Z) is the automorphism group of a quandle on a set with ∣Z∣+k|Z|+k elements. In particular, every finite abelian group Cn1×⋯×CnkC_{n_1}\times\cdots\times C_{n_k} is the automorphism group of a quandle with k+n1+⋯+nkk+n_1+\cdots+n_k elements.

(2) For any n≥3n\ge3, the dihedral group DnD_n of order 2n2n is the automorphism group of a quandle with n+2n+2 elements.

(3) Every alternating group AnA_n and every symmetric group SnS_n is the automorphism group of a finite quandle.

(4) The group SL⁡(2,3)\operatorname{SL}(2,3) is the automorphism group of a quandle with 8 elements.

Theorem B is proved in § 4: part (1) in Proposition 3.1 and Corollary 3.2, part (2) in Proposition 3.7, part (3) in Corollary 3.13 and part (4) in Proposition 3.10. Part (1) is the analogue for quandles of [41 Theorem 3.1], where the centralizer of an arbitrary finite family of maps is realized as the automorphism group of an idempotent solution of the braid equation. In view of Theorem A, such a result cannot hold for arbitrary finite families of permutations: for instance, Q8Q_8 is the centralizer in Sym⁡(Q8)\operatorname{Sym}(Q_8) of the left translations by two generators of Q8Q_8. Part (2) should be compared with [22 Theorem 2.1]: the automorphism group of the dihedral quandle with nn elements is the affine group Aff⁡(Zn)\operatorname{Aff}(\mathbb{Z}_n), which is isomorphic to DnD_n only for n∈{3,4,6}n \in\{3, 4, 6\}; adding two points to the dihedral quandle we get rid of all the extra automorphisms. The quandles of parts (2) and (4), as well as those of Theorem A(2), are augmented quandles in the sense of Joyce and Fenn–Rourke, for which we give a general criterion (Proposition 3.4); the quandles of part (3) are conjugation quandles of groups with trivial center (Proposition 3.11). We end the paper with some open questions (§ 5).

Preliminaries

Given a set XX, Sym⁡(X)\operatorname{Sym}(X) denotes the symmetric group on XX, i.e. the group of all bijections X→XX \to X, and τX:X×X→X×X\tau_X: X \times X \to X \times X, τX(x,y):=(y,x)\tau_X(x,y) := (y,x), is the flip map. For a map R:X×X→X×XR: X \times X \to X \times X we use the usual notation R12:=R×Id⁡XR^{12} := R \times\operatorname{Id}_X and R23:=Id⁡X×RR^{23} := \operatorname{Id}_X \times R. The map RR is called a solution of the braid equation if

R12R23R12=R23R12R23.(1)R^{12}R^{23}R^{12} = R^{23}R^{12}R^{23}. \tag*{(1)}

A morphism of solutions σ:(X,R)→(X′,R′)\sigma: (X, R) \to(X', R') is a map σ:X→X′\sigma: X \to X' such that (σ×σ)∘R=R′∘(σ×σ)(\sigma\times\sigma) \circ R = R' \circ(\sigma\times\sigma). We denote by BR\mathcal{BR} the category of all solutions of the braid equation, i.e. its objects are the pairs (X,R)(X, R) consisting of a set XX and a solution RR of the braid equation on XX. The automorphism group Aut⁡(X,R)\operatorname{Aut}(X, R) of (X,R)(X, R) is the group of all bijections σ:X→X\sigma: X \to X which are morphisms of solutions (X,R)→(X,R)(X, R) \to(X, R). A map RR is a solution of the braid equation if and only if τX∘R\tau_X \circ R is a solution of the Yang–Baxter equation, and (X,R)↦(X,τX∘R)(X, R) \mapsto(X, \tau_X \circ R) is an isomorphism between BR\mathcal{BR} and the category YB\mathcal{YB} of all set-theoretic solutions of the Yang–Baxter equation, which preserves the automorphism groups [39 Section 1]. For this reason, in what follows we shall work only with the braid equation.

Following [2 Definition 1.1], the centralizer of a family (fi)i∈I(f_i)_{i \in I} of maps fi:Z→Zf_i: Z \to Z is the subgroup of Sym⁡(Z)\operatorname{Sym}(Z) given by

Aut⁡(fi,i∈I):={β∈Sym⁡(Z)∣β∘fi=fi∘β, for all i∈I}.\operatorname{Aut}(f_i, i \in I) := \left\{\beta\in\operatorname{Sym}(Z) \mid\beta\circ f_i = f_i \circ\beta,\ \text{for all } i \in I\right\}.

We use the standard notation from group theory. Let GG be a group. If GG acts on a set XX, then GxG_x and GxGx denote the stabilizer and the orbit of x∈Xx \in X; a subgroup of Sym⁡(X)\operatorname{Sym}(X) acts on XX in the obvious way. For a subgroup K≤GK \leq G and a subset S⊆GS \subseteq G, we denote by NG(K)N_G(K), CG(S)C_G(S) and Z(G)Z(G) the normalizer of KK, the centralizer of SS and the center of GG. Moreover, core⁡G(K):=⋂g∈GgKg−1\operatorname{core}_G(K) := \bigcap_{g \in G} gKg^{-1} is the core of KK, i.e. the kernel of the action of GG on the set G/KG/K of left cosets, and κg:G→G\kappa_g: G \to G, κg(h):=ghg−1\kappa_g(h) := ghg^{-1}, is the inner automorphism of GG defined by g∈Gg \in G. An involution of GG is an element of order 22. As usual, CnC_n, DnD_n, SnS_n and AnA_n denote the cyclic group of order nn, the dihedral group of order 2n2n, the symmetric group and the alternating group of degree nn.

Racks and quandles. A magma [15 Definition 1] is a set XX endowed with a binary operation (x,y)↦x⊲y(x,y) \mapsto x \lhd y. For x∈Xx \in X we denote by ℓx,rx:X→X\ell_x, r_x: X \to X the left and right translations of the magma (X,⊲)(X, \lhd), i.e.

ℓx(y):=x⊲y,rx(y):=y⊲x,for all y∈X.\ell_x(y) := x \lhd y,\qquad r_x(y) := y \lhd x,\qquad\text{for all } y \in X.

A morphism of magmas (X,◃)→(X′,◃′)(X,\triangleleft) \to(X',\triangleleft') is a map φ:X→X′\varphi:X \to X' such that

φ(x◃y)=φ(x)◃′φ(y),for all x,y∈X.\varphi(x \triangleleft y)=\varphi(x)\triangleleft'\varphi(y),\qquad\text{for all }x,y\in X.

The bijective endomorphisms of a magma (X,◃)(X,\triangleleft) are its automorphisms; they form a subgroup of Sym⁡(X)\operatorname{Sym}(X), denoted by Aut⁡(X,◃)\operatorname{Aut}(X,\triangleleft), or simply by Aut⁡(X)\operatorname{Aut}(X). Since φ(x◃y)=φ(ry(x))\varphi(x\triangleleft y)=\varphi(r_y(x)) and φ(x)◃φ(y)=rφ(y)(φ(x))\varphi(x)\triangleleft\varphi(y)=r_{\varphi(y)}(\varphi(x)), a bijection φ:X→X\varphi:X\to X is an automorphism of (X,◃)(X,\triangleleft) if and only if

φ∘ry∘φ−1=rφ(y),for all y∈X.(2)\varphi\circ r_y\circ\varphi^{-1}=r_{\varphi(y)},\qquad\text{for all }y\in X. \tag*{(2)}

The category of magmas is universal and finitely universal [39 Remark 1.2]; the results of this paper show that this is no longer true for the magmas satisfying the identities which define racks or quandles.

A rack is a magma (X,◃)(X,\triangleleft) whose right translations are automorphisms of (X,◃)(X,\triangleleft); that is, for any z∈Xz\in X, the right translation rzr_z is a bijection and

(x◃y)◃z=(x◃z)◃(y◃z),for all x,y∈X.(3)(x\triangleleft y)\triangleleft z=(x\triangleleft z)\triangleleft(y\triangleleft z),\qquad\text{for all }x,y\in X. \tag*{(3)}

A rack (X,◃)(X,\triangleleft) is called a quandle if x◃x=xx\triangleleft x=x, for all x∈Xx\in X, and it is called trivial if x◃y=xx\triangleleft y=x, for all x,y∈Xx,y\in X; the automorphism group of a trivial rack is Sym⁡(X)\operatorname{Sym}(X). Applying (1.2) to the bijections φ=rz\varphi=r_z, we obtain that a magma (X,◃)(X,\triangleleft) whose right translations are bijective is a rack if and only if

rz∘ry∘rz−1=rrz(y),for all y,z∈X.(4)r_z\circ r_y\circ r_z^{-1}=r_{r_z(y)},\qquad\text{for all }y,z\in X. \tag*{(4)}

A morphism of racks is a morphism of the underlying magmas. We denote by Rack\mathsf{Rack} the category of racks and by Qnd\mathsf{Qnd} its full subcategory of quandles. For a rack (X,◃)(X,\triangleleft), the subgroup of Aut⁡(X)\operatorname{Aut}(X) generated by the right translations rzr_z, z∈Xz\in X, is called the inner automorphism group of XX and is denoted by Inn⁡(X)\operatorname{Inn}(X); by (1.2), Inn⁡(X)\operatorname{Inn}(X) is a normal subgroup of Aut⁡(X)\operatorname{Aut}(X). For more details on racks and quandles we refer to [9, 23].

Let (X,◃)(X,\triangleleft) be a rack. Then the map

R=R◃:X×X→X×X,R(x,y):=(y,x◃y),(5)R=R^\triangleleft:X\times X\to X\times X,\qquad R(x,y):=(y,x\triangleleft y), \tag*{(5)}

is a bijective non-degenerate [24 Definition 1.1] solution of the braid equation, called the derived solution of the rack (X,◃)(X,\triangleleft) [16] (see also [9 Remark 6.2]). Let (X′,◃′)(X',\triangleleft') be another rack and φ:X→X′\varphi:X\to X' a map. Since

(φ×φ)(R◃(x,y))=(φ(y),φ(x◃y))andR◃′(φ(x),φ(y))=(φ(y),φ(x)◃′φ(y)),(\varphi\times\varphi)(R^\triangleleft(x,y))=(\varphi(y),\varphi(x\triangleleft y))\qquad\text{and}\qquad R^{\triangleleft'}(\varphi(x),\varphi(y))=(\varphi(y),\varphi(x)\triangleleft'\varphi(y)),

φ\varphi is a morphism of solutions (X,R◃)→(X′,R◃′)(X,R^\triangleleft)\to(X',R^{\triangleleft'}) if and only if it is a morphism of racks. Hence (X,◃)↦(X,R◃)(X,\triangleleft)\mapsto(X,R^\triangleleft) is a fully faithful functor, injective on objects,

Rack↪BR,(6)\mathsf{Rack}\hookrightarrow\mathsf{BR}, \tag*{(6)}

which identifies Rack\mathsf{Rack} with a full subcategory of BR\mathsf{BR}, and hence of YB\mathsf{YB}. In particular, for any rack (X,◃)(X,\triangleleft) we have

Aut⁡(X,R◃)=Aut⁡(X,◃).(7)\operatorname{Aut}(X,R^\triangleleft)=\operatorname{Aut}(X,\triangleleft). \tag*{(7)}

Thus all the results below on automorphism groups of racks can be equivalently stated for the derived solutions of racks.

Remark 1.1. The embedding (1.6) has a left adjoint F ⁣:BR→Rack\mathcal{F}\colon\mathcal{BR} \to\mathsf{Rack}. Its existence follows from Freyd’s adjoint functor theorem [13 Corollary 8.17], exactly as in [1 Section 3]; it is the rack version of [36 Theorem 9], [48 Theorem 4.5] and [7 Theorem 3.20]. Explicitly, for a solution (X,R)(X,R) of the braid equation, written as R(x,y)=(x⋅y,x∗y)R(x,y)=(x\cdot y,x\ast y), F(X,R)\mathcal{F}(X,R) is the rack generated by the set XX with the relations

x⋅y=yandx∗y=x◃y,for all x,y∈X,x\cdot y=y \qquad\text{and}\qquad x\ast y=x\triangleleft y,\qquad\text{for all }x,y\in X,

i.e. the quotient of the free rack on XX [25] by the smallest congruence containing these relations. The canonical map η ⁣:X→F(X,R)\eta\colon X\to\mathcal{F}(X,R) is a morphism of solutions from (X,R)(X,R) to the derived solution of F(X,R)\mathcal{F}(X,R), and it is universal: for any rack (Y,◃)(Y,\triangleleft) and any morphism of solutions θ ⁣:(X,R)→(Y,R◃)\theta\colon(X,R)\to(Y,R^{\triangleleft}) there exists a unique morphism of racks u ⁣:F(X,R)→Yu\colon\mathcal{F}(X,R)\to Y such that u∘η=θu\circ\eta=\theta. Consequently, η\eta is injective if and only if (X,R)(X,R) is isomorphic to a subsolution of the derived solution of a rack. This is not always the case: for R=Id⁡X×XR=\operatorname{Id}_{X\times X} we have x⋅y=xx\cdot y=x, the relations give η(x)=η(y)\eta(x)=\eta(y), for all x,y∈Xx,y\in X, and F(X,Id⁡X×X)\mathcal{F}(X,\operatorname{Id}_{X\times X}) is the rack with one element.

Remark 1.2. Racks and skew left braces both form varieties of algebras which produce bijective non-degenerate solutions of the braid equation, so it is natural to ask which of them produces more solutions. By the above discussion, two racks are isomorphic if and only if their derived solutions are isomorphic; hence the number of isomorphism classes of derived solutions of racks (resp. of quandles) on a set with nn elements is the number r(n)r(n) of racks (resp. the number q(n)q(n) of quandles) of order nn up to isomorphism. These numbers were computed for n≤13n\le13 by Vojtěchovský and Yang [50 Tables 1 and 2]2; for n≤7n\le7 we confirmed them with an independent computer program, based on the Cauchy–Frobenius lemma. They are recorded in Table 1, together with the number s(n)s(n) of isomorphism classes of solutions of the braid equation on a set with nn elements which arise from skew left braces, i.e. which are isomorphic to the solution RBR_B associated to a skew left brace BB [29 Theorem 3.1]. For n≤6n\le6, the numbers s(n)s(n) were computed in [40 Tables 3 and 5], where the corresponding Yang–Baxter solutions τX∘RB\tau_X\circ R_B are counted, and s(7)=1s(7)=1, since there is only one skew left brace of order 77 up to isomorphism [29 Table 5.1].

Thus the derived solutions of racks are much more numerous than the solutions arising from skew left braces: for instance, for n=7n=7 there are 2080 isomorphism classes of derived solutions of racks, and only one isomorphism class of solutions arising from skew left braces.

Dicyclic groups. For n≥2n\ge2, the dicyclic group of order 4n4n, also called the binary dihedral group, is the group given by generators and relations

Dic⁡n=⟨a,b∣a2n=1, b2=an, bab−1=a−1⟩,\operatorname{Dic}_n=\langle a,b\mid a^{2n}=1,\ b^2=a^n,\ bab^{-1}=a^{-1}\rangle,

see [44 pp. 347–348]. For n=2m−2n=2^{m-2}, with m≥3m\ge3, Dic⁡n\operatorname{Dic}_n is the generalized quaternion group Q2mQ_{2^m} of order 2m2^m [18]; in particular, Dic⁡2=Q8\operatorname{Dic}_2=Q_8 is the quaternion group. The group Dic⁡n\operatorname{Dic}_n has 4n4n elements, the subgroup ⟨a⟩\langle a\rangle is cyclic of order 2n2n, and every element of Dic⁡n\operatorname{Dic}_{n} can be written uniquely as aka^{k} or akba^{k}b, with 0≤k<2n0 \le k < 2n [14 Proposition 1.1]; that is, Dic⁡n=⟨a⟩⊔⟨a⟩b\operatorname{Dic}_{n} = \langle a\rangle\sqcup\langle a\rangle b. The relation bab−1=a−1bab^{-1} = a^{-1} gives bak=a−kbba^{k} = a^{-k}b, for all k∈Zk \in\mathbb{Z}. The following elementary properties of the dicyclic groups are well known; see, for instance, [14 Section 1], where the center of Dic⁡n\operatorname{Dic}_{n} and the orders of its elements are computed. Since they will be used several times, we include a short proof for the reader’s convenience.

Lemma 1.3. Let G:=Dic⁡nG := \operatorname{Dic}_{n}, n≥2n \ge2, and z:=anz := a^{n}. Then:

(1) every element g∈G\⟨a⟩g \in G \backslash\langle a\rangle has order 44, g2=zg^{2} = z, and gg acts on ⟨a⟩\langle a\rangle by inversion, i.e. gakg−1=a−kga^{k}g^{-1} = a^{-k}, for all k∈Zk \in\mathbb{Z};

(2) zz is the unique involution of GG, and Z(G)=⟨z⟩Z(G) = \langle z\rangle;

(3) every subgroup of ⟨a⟩\langle a\rangle is a normal subgroup of GG, and CG(ak)=⟨a⟩C_{G}(a^{k}) = \langle a\rangle, for all ak∉⟨z⟩a^{k} \notin\langle z\rangle;

(4) CG(b)=⟨b⟩={1,z,b,zb}C_{G}(b) = \langle b\rangle= \{1, z, b, zb\}, and the conjugacy class of bb in GG is {a2jb∣j∈Z}\{a^{2j}b \mid j \in\mathbb{Z}\}.

Proof. (1) Let g=ajbg = a^{j}b. Using bak=a−kbba^{k} = a^{-k}b and b2=an=zb^{2} = a^{n} = z, we obtain

gakg−1=aj(bakb−1)a−j=a−kandg2=aj(bajb−1)b2=z.ga^{k}g^{-1} = a^{j}(ba^{k}b^{-1})a^{-j} = a^{-k} \qquad\text{and}\qquad g^{2} = a^{j}(ba^{j}b^{-1})b^{2} = z.

As zz has order 22, gg has order 44.

(2) The cyclic group ⟨a⟩\langle a\rangle of order 2n2n has a unique involution, namely z=anz = a^{n}, and by (1) the elements of G\⟨a⟩G \backslash\langle a\rangle are not involutions. An element aka^{k} is central if and only if it commutes with bb, i.e. a−k=aka^{-k} = a^{k}, i.e. ak∈⟨z⟩a^{k} \in\langle z\rangle. By (1), no element g∈G\⟨a⟩g \in G \backslash\langle a\rangle is central, since gag−1=a−1≠agag^{-1} = a^{-1} \ne a, as n≥2n \ge2. Hence Z(G)=⟨z⟩Z(G) = \langle z\rangle.

(3) By (1), g⟨ak⟩g−1=⟨a−k⟩=⟨ak⟩g\langle a^{k}\rangle g^{-1} = \langle a^{-k}\rangle= \langle a^{k}\rangle, for all g∈G\⟨a⟩g \in G \backslash\langle a\rangle; this proves the first statement, since ⟨a⟩\langle a\rangle is abelian. If ak∉⟨z⟩a^{k} \notin\langle z\rangle, then a−k≠aka^{-k} \ne a^{k}; hence, by (1), no element of G\⟨a⟩G \backslash\langle a\rangle commutes with aka^{k}, and CG(ak)=⟨a⟩C_{G}(a^{k}) = \langle a\rangle.

(4) As in the proof of (2), aja^{j} commutes with bb if and only if aj∈⟨z⟩a^{j} \in\langle z\rangle; moreover, ajba^{j}b commutes with bb if and only if aja^{j} does. Hence CG(b)={1,z,b,zb}C_{G}(b) = \{1, z, b, zb\}, which is the cyclic subgroup generated by bb, as b2=zb^{2} = z and b3=zbb^{3} = zb. Finally, for all j∈Zj \in\mathbb{Z} we have

ajba−j=a2jband(ajb)b(ajb)−1=ajba−j=a2jb,a^{j}ba^{-j} = a^{2j}b \qquad\text{and}\qquad (a^{j}b)b(a^{j}b)^{-1} = a^{j}ba^{-j} = a^{2j}b,

which proves the last statement. □\square

Racks and quandles are not universal

In this section we prove a general obstruction for a group to be the automorphism group of a rack (Theorem 2.2), and we apply it to the Hamiltonian groups and to the dicyclic groups. The key ingredient is the following lemma: if G=Aut⁡(X)G = \operatorname{Aut}(X), then the elements of GG which normalize the stabilizer of a point x∈Xx \in X and commute with the right translation rxr_{x} are central, up to an element of the stabilizer.

Lemma 2.1. Let (X,◃)(X, \triangleleft) be a rack, G:=Aut⁡(X)G := \operatorname{Aut}(X) and x∈Xx \in X. Then:

(1) rx∈CG(Gx)r_{x} \in C_{G}(G_{x}), i.e. the right translation rxr_{x} commutes with all elements of the stabilizer GxG_{x};

(2) for any m∈NG(Gx)∩CG(rx)m \in N_{G}(G_{x}) \cap C_{G}(r_{x}) there exists ζ∈Z(G)\zeta\in Z(G) such that m∈ζGxm \in\zeta G_{x} and ζ(y)=y\zeta(y) = y, for all y∈X\Gxy \in X \backslash Gx. In particular,

NG(Gx)∩CG(rx)⊆GxZ(G).(8)N_{G}(G_{x}) \cap C_{G}(r_{x}) \subseteq G_{x}Z(G). \tag*{(8)}

Proof. (1) By the definition of a rack, rx∈Gr_x \in G. If φ∈Gx\varphi\in G_x, then (1.2) gives φ∘rx∘φ−1=rφ(x)=rx\varphi\circ r_x \circ\varphi^{-1} = r_{\varphi(x)} = r_x.

(2) Let m∈NG(Gx)∩CG(rx)m \in N_G(G_x) \cap C_G(r_x) and let O:=GxO := Gx be the orbit of xx. We define ψ:X→X\psi: X \to X by

ψ(φ(x)):=φ(m(x)),for all φ∈G,andψ(y):=y,for all y∈X∖O.\psi(\varphi(x)) := \varphi(m(x)), \quad\text{for all } \varphi\in G, \qquad\text{and} \qquad\psi(y) := y, \quad\text{for all } y \in X \setminus O.

The map ψ\psi is well defined: if φ(x)=φ′(x)\varphi(x) = \varphi'(x), with φ,φ′∈G\varphi,\varphi' \in G, then φ−1φ′∈Gx\varphi^{-1}\varphi' \in G_x, hence m−1φ−1φ′m∈Gxm^{-1}\varphi^{-1}\varphi'm \in G_x, since mm normalizes GxG_x; that is, φ(m(x))=φ′(m(x))\varphi(m(x)) = \varphi'(m(x)). The same construction, applied to m−1∈NG(Gx)∩CG(rx)m^{-1} \in N_G(G_x) \cap C_G(r_x), gives the inverse of ψ\psi; thus ψ∈Sym⁡(X)\psi\in\operatorname{Sym}(X). Moreover, ψ\psi commutes with all elements of GG: for g,φ∈Gg,\varphi\in G we have

ψ(gφ(x))=gφ(m(x))=g(ψ(φ(x))).\psi(g\varphi(x)) = g\varphi(m(x)) = g(\psi(\varphi(x))).

and ψ\psi is the identity on X∖OX \setminus O, which is a GG-invariant subset of XX.

We prove now that ψ\psi is an automorphism of (X,◃)(X,\triangleleft), i.e. that ψ\psi satisfies (1.2). Let y∈Xy \in X. Since ry∈Gr_y \in G commutes with ψ\psi, we have ψ∘ry∘ψ−1=ry\psi\circ r_y \circ\psi^{-1} = r_y, so we have to show that rψ(y)=ryr_{\psi(y)} = r_y. This is obvious for y∈X∖Oy \in X \setminus O. For y=φ(x)∈Oy = \varphi(x) \in O, we apply (1.2) to the automorphisms φm\varphi m and φ\varphi, and we use that mm commutes with rxr_x:

rψ(φ(x))=rφm(x)=(φm)∘rx∘(φm)−1=φ∘rx∘φ−1=rφ(x).r_{\psi(\varphi(x))} = r_{\varphi m(x)} = (\varphi m) \circ r_x \circ(\varphi m)^{-1} = \varphi\circ r_x \circ\varphi^{-1} = r_{\varphi(x)}.

Thus ψ∈G\psi\in G and, since ψ\psi commutes with all elements of GG, we obtain that ζ:=ψ∈Z(G)\zeta:= \psi\in Z(G). By construction, ζ\zeta is the identity on X∖GxX \setminus Gx and ζ(x)=m(x)\zeta(x) = m(x), i.e. ζ−1m∈Gx\zeta^{-1}m \in G_x. Hence m∈ζGx⊆GxZ(G)m \in\zeta G_x \subseteq G_x Z(G), which proves (2.1).

We can now prove the main result of this section.

Theorem 2.2. Let GG be a group and 1≠z∈Z(G)1 \ne z \in Z(G) such that, for every subgroup K≤GK \leq G with z∉Kz \notin K and every c∈CG(K)c \in C_G(K), we have

NG(K)∩CG(c)⊈KZ(G).(9)N_G(K) \cap C_G(c) \not\subseteq K Z(G). \tag*{(9)}

Then there is no rack (X,◃)(X,\triangleleft), finite or infinite, such that Aut⁡(X)≅G\operatorname{Aut}(X) \cong G.

Proof. Assume that (X,◃)(X,\triangleleft) is a rack such that Aut⁡(X)≅G\operatorname{Aut}(X) \cong G; we identify G=Aut⁡(X)≤Sym⁡(X)G = \operatorname{Aut}(X) \leq\operatorname{Sym}(X). Since z≠Id⁡Xz \ne\operatorname{Id}_X, there exists x∈Xx \in X such that z(x)≠xz(x) \ne x, i.e. z∉K:=Gxz \notin K := G_x. By Lemma 2.1(1), c:=rx∈CG(K)c := r_x \in C_G(K), and by Lemma 2.1(2), NG(K)∩CG(c)⊆KZ(G)N_G(K) \cap C_G(c) \subseteq K Z(G). This contradicts (2.2).

The hypothesis of Theorem 2.2 can be easily checked in many cases. The first one is the following:

Corollary 2.3. Let GG be a non-abelian group and assume that there exists 1≠z∈G1 \ne z \in G such that z∈⟨g⟩z \in\langle g\rangle, for all g∈G∖Z(G)g \in G \setminus Z(G). Then GG is not isomorphic to the automorphism group of any rack, finite or infinite.

Proof. The element zz commutes with all g∈G∖Z(G)g \in G \setminus Z(G), since z∈⟨g⟩z \in\langle g\rangle, and with all central elements; hence z∈Z(G)z \in Z(G). Let K≤GK \leq G be a subgroup with z∉Kz \notin K, and let c∈CG(K)c \in C_G(K). By assumption, KK contains no element of G∖Z(G)G \setminus Z(G), i.e. K⊆Z(G)K \subseteq Z(G); hence NG(K)=GN_G(K) = G and KZ(G)=Z(G)K Z(G) = Z(G). Assume that (2.2) does not hold. Then

CG(c)=NG(K)∩CG(c)⊆Z(G).C_G(c) = N_G(K) \cap C_G(c) \subseteq Z(G).

In particular, c∈Z(G)c \in Z(G), hence G=CG(c)⊆Z(G)G = C_G(c) \subseteq Z(G), i.e. GG is abelian, a contradiction. Thus (2.2) holds, and the conclusion follows from Theorem 2.2.

Recall that a Hamiltonian group is a non-abelian group all of whose subgroups are normal. By the theorem of Dedekind [20] (for finite groups) and Baer [6], the Hamiltonian groups are precisely the groups Q8×E×AQ_8 \times E \times A, where EE is an elementary abelian 2-group and AA is an abelian group all of whose elements have odd order.

Corollary 2.4. Let m≥3m \ge3, let EE be an elementary abelian 2-group and AA an abelian group all of whose elements have odd order (E and A can be infinite). Then the group Q2m×E×AQ_{2^m} \times E \times A is not isomorphic to the automorphism group of any rack. In particular, no Hamiltonian group is isomorphic to the automorphism group of a rack.

Proof. Let Q:=Q2m=Dic⁡2m−2Q := Q_{2^m} = \operatorname{Dic}_{2^{m-2}} and G:=Q×E×AG := Q \times E \times A. By Lemma 1.3(2), QQ has a unique involution z0z_0 and Z(Q)=⟨z0⟩Z(Q) = \langle z_0 \rangle; hence Z(G)=⟨z0⟩×E×AZ(G) = \langle z_0 \rangle\times E \times A. We show that z:=(z0,1,1)z := (z_0, 1, 1) satisfies the hypothesis of Corollary 2.3. Let g=(q,e,t)∈G∖Z(G)g = (q, e, t) \in G \setminus Z(G), i.e. q∉⟨z0⟩q \notin\langle z_0 \rangle. Then qq is neither trivial nor an involution, so its order is 2s2^s, with s≥2s \ge2, and q2s−1=z0q^{2^{s-1}} = z_0, since z0z_0 is the unique involution of QQ. Let oo be the order of tt, which is odd. Since e2=1e^2 = 1, to=1t^o = 1 and 2s−12^{s-1} is even, we obtain

go2s−1=(z0,eo2s−1,to2s−1)=(z0,1,1)=z.g^{o2^{s-1}} = \left(z_0, e^{o2^{s-1}}, t^{o2^{s-1}}\right) = (z_0, 1, 1) = z.

Thus z∈⟨g⟩z \in\langle g \rangle, for all g∈G∖Z(G)g \in G \setminus Z(G), and the conclusion follows from Corollary 2.3. The last statement follows from the theorem of Dedekind and Baer recalled above.

If nn is even but not a power of 2, then the dicyclic group Dic⁡n\operatorname{Dic}_n does not satisfy the hypothesis of Corollary 2.3: if d>1d > 1 is the largest odd divisor of nn, then a2n/da^{2n/d} is a non-central element of odd order dd, and the cyclic subgroup that it generates does not contain the involution z=anz = a^n. Nevertheless, Theorem 2.2 can still be applied.

Corollary 2.5. Let n≥2n \ge2 be an even integer. Then the dicyclic group Dic⁡n\operatorname{Dic}_n is not isomorphic to the automorphism group of any rack, finite or infinite.

Proof. Let G:=Dic⁡nG := \operatorname{Dic}_n and z:=anz := a^n; by Lemma 1.3(2), zz is the unique involution of GG and Z(G)=⟨z⟩Z(G) = \langle z \rangle. Let K≤GK \le G be a subgroup with z∉Kz \notin K, and let c∈CG(K)c \in C_G(K). Then KK contains no involution, so it has odd order; by Lemma 1.3(1), K⊆⟨a⟩K \subseteq\langle a \rangle, hence KK is a normal subgroup of GG, by Lemma 1.3(3). Thus

NG(K)∩CG(c)=CG(c),N_G(K) \cap C_G(c) = C_G(c),

while KZ(G)=K×⟨z⟩K Z(G) = K \times\langle z \rangle has order 2∣K∣2|K|, with ∣K∣|K| odd, and therefore contains no element of order 4. On the other hand, CG(c)C_G(c) contains an element of order 4. Indeed, if c∉⟨a⟩c \notin\langle a \rangle, then cc itself has order 4, by Lemma 1.3(1); if c∈⟨a⟩c \in\langle a \rangle, then CG(c)C_G(c) contains the cyclic group ⟨a⟩\langle a \rangle, whose order 2n2n is divisible by 4, as nn is even. Hence (2.2) holds, and the conclusion follows from Theorem 2.2.

In the terminology of Babai recalled in the Introduction, we obtain:

Corollary 2.6. The category of racks and the category of quandles are neither universal nor finitely universal: the quaternion group Q8Q_8 is not represented by any rack, finite or infinite. The same holds for the class of derived solutions R(x,y)=(y,x▹y)R(x,y) = (y, x \triangleright y) of racks, regarded as a class of solutions of the braid equation.

Proof. By Corollary 2.5, applied for n=2n = 2, the group Dic⁡2=Q8\operatorname{Dic}_2 = Q_8 is not isomorphic to Aut⁡(X)\operatorname{Aut}(X), for any rack (X,◃)(X, \triangleleft); in particular, for any quandle. The last statement follows from (7). ▫

Remark 2.7. Theorem 2.2 and its corollaries also apply to a class of involutive solutions. Let

R:X×X→X×X,R(x,y)=(σx(y),σσx(y)−1(x)),R:X\times X\to X\times X,\qquad R(x,y)=(\sigma_x(y),\sigma_{\sigma_x(y)}^{-1}(x)),

be an involutive left non-degenerate solution of the braid equation [24], where σx∈Sym⁡(X)\sigma_x\in\operatorname{Sym}(X), for all x∈Xx\in X, and assume that σσx(y)=σy\sigma_{\sigma_x(y)}=\sigma_y, for all x,y∈Xx,y\in X. The braid equation gives

σx∘σy=σσx(y)∘σσσx(y)−1(x)−1,for all x,y∈X.\sigma_x\circ\sigma_y=\sigma_{\sigma_x(y)}\circ\sigma_{\sigma_{\sigma_x(y)}^{-1}(x)}^{-1}, \qquad\text{for all }x,y\in X.

The assumption, applied to the elements uu and σu−1(x)\sigma_u^{-1}(x), gives σx=σσu−1(x)\sigma_x=\sigma_{\sigma_u^{-1}(x)}, for all u,x∈Xu,x\in X; hence the right hand side above is σy∘σx\sigma_y\circ\sigma_x, i.e. the permutations σx\sigma_x commute pairwise. Therefore

σz∘σy∘σz−1=σy=σσz(y),for all y,z∈X,\sigma_z\circ\sigma_y\circ\sigma_z^{-1}=\sigma_y=\sigma_{\sigma_z(y)}, \qquad\text{for all }y,z\in X,

which is exactly (1.4) for the magma (X,◃)(X,\triangleleft) given by x◃y:=σy(x)x\triangleleft y:=\sigma_y(x); thus (X,◃)(X,\triangleleft) is a rack. Moreover, a bijection φ\varphi of XX is an automorphism of (X,R)(X,R) if and only if φ∘σx∘φ−1=σφ(x)\varphi\circ\sigma_x\circ\varphi^{-1}=\sigma_{\varphi(x)}, for all $x\in X (the condition on the second component of RR follows from this one), i.e. if and only if φ∈Aut⁡(X,◃)\varphi\in\operatorname{Aut}(X,\triangleleft), by (1.2). Hence Aut⁡(X,R)=Aut⁡(X,◃)\operatorname{Aut}(X,R)=\operatorname{Aut}(X,\triangleleft); for instance, the automorphism group of such a solution is never isomorphic to a Hamiltonian group, or to a dicyclic group Dic⁡n\operatorname{Dic}_n with nn even. This does not contradict [39 Theorem 2.18], where every group is realized as the automorphism group of an involutive solution of the braid equation: the bipartite solutions used there are degenerate.

On the other hand, for an arbitrary bijective non-degenerate solution (X,R)(X,R) of the braid equation, Aut⁡(X,R)\operatorname{Aut}(X,R) is only a subgroup of the automorphism group of its derived rack [49, 35], which in general does not contain the inner automorphisms of the derived rack; hence Theorem 2.2 does not apply (see § 5).

Groups which are automorphism groups of quandles

In this section we show that several classical families of groups are automorphism groups of quandles: the centralizers of commuting families of permutations, in particular all finite abelian groups (Subsection 3.1), the dihedral groups, the dicyclic groups Dic⁡n\operatorname{Dic}_n with nn odd and the group SL⁡(2,3)\operatorname{SL}(2,3) (Subsection 3.3), and the alternating groups (Subsection 3.4).

Centralizers of commuting families of permutations

The next result is the analogue for quandles of [41 Theorem 3.1]. For k=1k=1, Y={0}Y=\{0\} and Z={1,…,n}Z=\{1,\ldots,n\}, the quandle (X,◃)(X,\triangleleft) below is the quandle PnσP_n^\sigma with one non-trivial column of Cazet [17], with σ=g1\sigma=g_1, and Proposition 3.1 recovers [17 Theorem 3.4].

Proposition 3.1. Let ZZ be a set, Y={y1,…,yk}Y=\{y_1,\ldots,y_k\} a set with k≥1k\ge1 elements disjoint from ZZ, and g1,…,gk∈Sym⁡(Z)∖{Id⁡Z}g_1,\ldots,g_k\in\operatorname{Sym}(Z)\setminus\{\operatorname{Id}_Z\} pairwise commuting permutations. On X:=Y∪ZX:=Y\cup Z we define the binary operation

x◃yi:={x,if x∈Y,gi(x),if x∈Z,x◃z:=x,for all x∈X, z∈Z;x\triangleleft y_i:= \begin{cases} x, & \text{if }x\in Y,\\ g_i(x), & \text{if }x\in Z, \end{cases} \qquad x\triangleleft z:=x,\quad\text{for all }x\in X,\ z\in Z;

that is, the right translations are

ryi=Id⁡Y∪gi,rz=Id⁡X,for all i=1,…,k and z∈Z,r_{y_i}=\operatorname{Id}_Y\cup g_i,\qquad r_z=\operatorname{Id}_X,\qquad\text{for all }i=1,\ldots,k\text{ and }z\in Z,

where α⊔β\alpha\sqcup\beta denotes the bijection of X=Y⊔ZX=Y\sqcup Z which is equal to α∈Sym⁡(Y)\alpha\in\operatorname{Sym}(Y) on YY and to β∈Sym⁡(Z)\beta\in\operatorname{Sym}(Z) on ZZ. Then (X,⊲)(X,\lhd) is a quandle, and the map

Hg:={(α,β)∈Sym⁡(Y)×Sym⁡(Z)∣β∘gi∘β−1=gj whenever α(yi)=yj}→Aut⁡(X),H_g:=\left\{(\alpha,\beta)\in\operatorname{Sym}(Y)\times\operatorname{Sym}(Z)\mid\beta\circ g_i\circ\beta^{-1}=g_j\ \text{whenever }\alpha(y_i)=y_j\right\}\to\operatorname{Aut}(X),
(α,β)↦α⊔β,(\alpha,\beta)\mapsto\alpha\sqcup\beta,

is an isomorphism of groups. If, moreover, g1,…,gkg_1,\ldots,g_k are pairwise non-conjugate in Sym⁡(Z)\operatorname{Sym}(Z), then

Aut⁡(X)={Id⁡Y⊔β∣β∈Aut⁡(g1,…,gk)}≅Aut⁡(g1,…,gk).\operatorname{Aut}(X)=\left\{\operatorname{Id}_Y\sqcup\beta\mid\beta\in\operatorname{Aut}(g_1,\ldots,g_k)\right\}\cong\operatorname{Aut}(g_1,\ldots,g_k).

Proof. All the right translations are bijections, and rx(x)=xr_x(x)=x, for all x∈Xx\in X. We check (1.4), i.e. ru∘ry∘ru−1=rru(y)r_u\circ r_y\circ r_u^{-1}=r_{r_u(y)}, for all u,y∈Xu,y\in X:

  • if u∈Zu\in Z, then ru=Id⁡Xr_u=\operatorname{Id}_X, and both sides are equal to ryr_y;

  • if u=yiu=y_i and y=yjy=y_j, then ryi(yj)=yjr_{y_i}(y_j)=y_j and, since gig_i and gjg_j commute,

ryi∘ryj∘ryi−1=Id⁡Y⊔gigjgi−1=Id⁡Y⊔gj=ryj;r_{y_i}\circ r_{y_j}\circ r_{y_i}^{-1}=\operatorname{Id}_Y\sqcup g_i g_j g_i^{-1}=\operatorname{Id}_Y\sqcup g_j=r_{y_j};
  • if u=yiu=y_i and y∈Zy\in Z, then ryi(y)=gi(y)∈Zr_{y_i}(y)=g_i(y)\in Z, and both sides are equal to Id⁡X\operatorname{Id}_X.

Thus (X,⊲)(X,\lhd) is a quandle.

Since gi≠Id⁡Zg_i\ne\operatorname{Id}_Z, for all ii, we have

Z={x∈X∣rx=Id⁡X}.Z=\left\{x\in X\mid r_x=\operatorname{Id}_X\right\}.

Let φ∈Aut⁡(X)\varphi\in\operatorname{Aut}(X). By (1.2), rφ(x)=φ∘rx∘φ−1r_{\varphi(x)}=\varphi\circ r_x\circ\varphi^{-1}; hence rx=Id⁡Xr_x=\operatorname{Id}_X if and only if rφ(x)=Id⁡Xr_{\varphi(x)}=\operatorname{Id}_X, and therefore φ(Z)=Z\varphi(Z)=Z and φ(Y)=Y\varphi(Y)=Y. Thus any automorphism of XX is of the form φ=α⊔β\varphi=\alpha\sqcup\beta, for some α∈Sym⁡(Y)\alpha\in\operatorname{Sym}(Y) and β∈Sym⁡(Z)\beta\in\operatorname{Sym}(Z). For such a bijection φ\varphi, the condition (1.2) is trivially satisfied for y∈Zy\in Z, while for y=yiy=y_i, with α(yi)=yj\alpha(y_i)=y_j, it becomes

Id⁡Y⊔βgiβ−1=Id⁡Y⊔gj.\operatorname{Id}_Y\sqcup\beta g_i\beta^{-1}=\operatorname{Id}_Y\sqcup g_j.

Hence α⊔β∈Aut⁡(X)\alpha\sqcup\beta\in\operatorname{Aut}(X) if and only if (α,β)∈Hg(\alpha,\beta)\in H_g. Since (α,β)↦α⊔β(\alpha,\beta)\mapsto\alpha\sqcup\beta is an injective morphism of groups Sym⁡(Y)×Sym⁡(Z)→Sym⁡(X)\operatorname{Sym}(Y)\times\operatorname{Sym}(Z)\to\operatorname{Sym}(X), this proves the first statement.

If g1,…,gkg_1,\ldots,g_k are pairwise non-conjugate, then βgiβ−1=gj\beta g_i\beta^{-1}=g_j implies i=ji=j. Hence (α,β)∈Hg(\alpha,\beta)\in H_g if and only if α=Id⁡Y\alpha=\operatorname{Id}_Y and β∈Aut⁡(g1,…,gk)\beta\in\operatorname{Aut}(g_1,\ldots,g_k), and the proof is finished. □\square

Corollary 3.2. Let n1,…,nk≥2n_1,\ldots,n_k\ge2. Then the abelian group Cn1×⋯×CnkC_{n_1}\times\cdots\times C_{n_k} is isomorphic to the automorphism group of a quandle with k+n1+⋯+nkk+n_1+\cdots+n_k elements. In particular, every finite abelian group is isomorphic to the automorphism group of a finite quandle.

Proof. Let Z:=O1⊔⋯⊔OkZ:=O_1\sqcup\cdots\sqcup O_k, with ∣Oj∣=nj|O_j|=n_j, let cj∈Sym⁡(Z)c_j\in\operatorname{Sym}(Z) be a cycle of length njn_j with support OjO_j, and let

gi:=c1c2⋯ci,for all i=1,…,k.g_i:=c_1c_2\cdots c_i,\qquad\text{for all }i=1,\ldots,k.

The cycles cjc_j have disjoint supports, so the permutations gig_i commute pairwise. They are pairwise non-conjugate, since gig_i moves exactly n1+⋯+nin_1+\cdots+n_i points. As g1,…,gkg_1,\ldots,g_k and c1,…,ckc_1,\ldots,c_k generate the same subgroup of Sym⁡(Z)\operatorname{Sym}(Z), we have

Aut⁡(g1,…,gk)=Aut⁡(c1,…,ck).(10)\operatorname{Aut}(g_1,\ldots,g_k)=\operatorname{Aut}(c_1,\ldots,c_k). \tag*{(10)}

Let β∈Aut⁡(c1,…,ck)\beta\in\operatorname{Aut}(c_1,\ldots,c_k). Since β\beta commutes with cjc_j, it maps the support OjO_j of cjc_j onto itself, and its restriction to OjO_j commutes with the njn_j-cycle cj∣Ojc_j|_{O_j}; hence β∣Oj\beta|_{O_j} is a power of cj∣Ojc_j|_{O_j}. Conversely, every product of powers of the cycles cjc_j commutes with all of them. Therefore

Aut⁡(c1,…,ck)=⟨c1⟩×⋯×⟨ck⟩≅Cn1×⋯×Cnk,\operatorname{Aut}(c_1,\ldots,c_k)=\langle c_1\rangle\times\cdots\times\langle c_k\rangle\cong C_{n_1}\times\cdots\times C_{n_k},

and the conclusion follows from Proposition 3.1. Finally, the trivial group is the automorphism group of the quandle with one element.

Corollary 3.3. The quaternion group Q8Q_8 is the smallest group which is not isomorphic to the automorphism group of a rack, of a quandle, or of the derived solution of a rack.

Proof. By Corollary 2.6, Q8Q_8 is not the automorphism group of a rack. The groups of order at most 88, other than Q8Q_8, are the abelian ones, the symmetric group S3S_3 and the dihedral group D4D_4. The abelian groups are covered by Corollary 3.2, and S3S_3 is the automorphism group of the trivial quandle with 33 elements. Finally, applying Proposition 3.1 for k=1k=1, Z={1,2,3,4}Z=\{1,2,3,4\} and g1=(1 2)(3 4)g_1=(1\ 2)(3\ 4), we obtain a quandle with 55 elements whose automorphism group is isomorphic to the centralizer of (1 2)(3 4)(1\ 2)(3\ 4) in S4S_4, i.e. to D4D_4 [39 Example 2.7]. By (7), the same holds for the derived solutions of racks.

Augmented quandles

Every rack XX is determined by the action of Inn⁡(X)\operatorname{Inn}(X) on XX and by the equivariant map x↦rxx\mapsto r_x; this is the point of view of augmented racks and quandles [33 Section 9], [25], [9 Definition 7.66]. We use it in order to realize a given group GG as the automorphism group of a quandle on which GG acts by inner automorphisms.

Let GG be a group, H1,…,HsH_1,\ldots,H_s subgroups of GG, and ci∈Z(Hi)c_i\in Z(H_i), for all i=1,…,si=1,\ldots,s. Let

X:=∐i=1sG/HiX:=\coprod_{i=1}^{s}G/H_i

be the disjoint union of the sets of left cosets, with the natural action of GG, and let ∂:X→G\partial:X\to G be the map given by

∂(gHi):=gcig−1,for all g∈G, i=1,…,s.\partial(gH_i):=gc_i g^{-1},\qquad\text{for all }g\in G,\ i=1,\ldots,s.

The map ∂\partial is well defined, since cic_i commutes with all elements of HiH_i, and it is GG-equivariant, i.e. ∂(gx)=g∂(x)g−1\partial(gx)=g\partial(x)g^{-1}, for all g∈Gg\in G and x∈Xx\in X. We define

x◃y:=∂(y)x,for all x,y∈X.(11)x\triangleleft y:=\partial(y)x,\qquad\text{for all }x,y\in X. \tag*{(11)}

We say that two pairs (H,c)(H,c) and (H′,c′)(H',c'), consisting of a subgroup and an element of GG, are conjugate in GG if there exists g∈Gg\in G such that H′=gHg−1H'=gHg^{-1} and c′=gcg−1c'=gcg^{-1}.

Proposition 3.4. With the above notation, assume that ⋂i=1score⁡G(Hi)=1\bigcap_{i=1}^{s}\operatorname{core}_G(H_i)=1, i.e. GG acts faithfully on XX, and that GG is generated by the conjugates of c1,…,csc_1,\ldots,c_s. Then (X,◃)(X,\triangleleft) is a quandle, Inn⁡(X)=G\operatorname{Inn}(X)=G as subgroups of Sym⁡(X)\operatorname{Sym}(X), and

Aut⁡(X)={φ∈NSym⁡(X)(G)∣∂∘φ=θφ∘∂},where θφ(g):=φgφ−1.(12)\operatorname{Aut}(X)=\{\varphi\in N_{\operatorname{Sym}(X)}(G)\mid\partial\circ\varphi=\theta_\varphi\circ\partial\},\qquad\text{where }\theta_\varphi(g):=\varphi g\varphi^{-1}. \tag*{(12)}

Assume, moreover, that:

(a) if θ∈Aut⁡(G)\theta\in\operatorname{Aut}(G) and π∈Ss\pi\in S_s are such that the pair (θ(Hi),θ(ci))(\theta(H_i),\theta(c_i)) is conjugate in GG to (Hπ(i),cπ(i))(H_{\pi(i)},c_{\pi(i)}), for all i=1,…,si=1,\ldots,s, then θ\theta is an inner automorphism of GG;

(b) the only GG-equivariant bijections ψ:X→X\psi:X\to X such that ∂∘ψ=∂\partial\circ\psi=\partial are the actions of the elements of Z(G)Z(G).

Then Aut⁡(X)=G\operatorname{Aut}(X)=G.

Proof. The right translations are ry(x)=∂(y)xr_y(x)=\partial(y)x, i.e. ryr_y is the action of ∂(y)∈G\partial(y)\in G on XX; in particular, they are bijections. For x=gHix=gH_i we have

x◃x=∂(x)x=gcig−1gHi=gciHi=x,x\mathbin{\triangleleft}x=\partial(x)x=gc_ig^{-1}gH_i=gc_iH_i=x,

since ci∈Hic_i\in H_i. Moreover, for all y,z∈Xy,z\in X, the equivariance of ∂\partial gives

rrz(y)=∂(∂(z)y)=∂(z)∂(y)∂(z)−1=rz∘ry∘rz−1,r_{r_z}(y)=\partial(\partial(z)y)=\partial(z)\partial(y)\partial(z)^{-1}=r_z\circ r_y\circ r_z^{-1},

i.e. (1.4) holds. Thus (X,◃)(X,\mathbin{\triangleleft}) is a quandle, and Inn⁡(X)\operatorname{Inn}(X) is the subgroup of Sym⁡(X)\operatorname{Sym}(X) generated by the elements ∂(y)\partial(y), y∈Xy\in X, i.e. by the conjugates of c1,…,csc_1,\ldots,c_s; by assumption, this subgroup is GG.

Let φ∈Sym⁡(X)\varphi\in\operatorname{Sym}(X). By (1.2), φ∈Aut⁡(X)\varphi\in\operatorname{Aut}(X) if and only if

φ∂(y)φ−1=∂(φ(y)),for all y∈X.\varphi\partial(y)\varphi^{-1}=\partial(\varphi(y)),\qquad\text{for all }y\in X.

If this holds, then the conjugation by φ\varphi maps the generating set ∂(X)\partial(X) of GG into GG, hence φGφ−1⊆G\varphi G\varphi^{-1}\subseteq G; the same holds for the automorphism φ−1\varphi^{-1}, so φ∈NSym⁡(X)(G)\varphi\in N_{\operatorname{Sym}(X)}(G), and the above condition reads θφ∘∂=∂∘φ\theta_\varphi\circ\partial=\partial\circ\varphi. The converse is clear, and (3.2) is proved.

Assume now that (a) and (b) hold, and let φ∈Aut⁡(X)\varphi\in\operatorname{Aut}(X) and θ:=θφ∈Aut⁡(G)\theta:=\theta_\varphi\in\operatorname{Aut}(G). Since φ(gx)=θ(g)φ(x)\varphi(gx)=\theta(g)\varphi(x), for all g∈Gg\in G and x∈Xx\in X, the bijection φ\varphi maps GG-orbits onto GG-orbits. Let π∈Ss\pi\in S_s be such that φ(G/Hi)=G/Hπ(i)\varphi(G/H_i)=G/H_{\pi(i)}, and write φ(Hi)=giHπ(i)\varphi(H_i)=g_iH_{\pi(i)}, with gi∈Gg_i\in G. The stabilizer of the point φ(Hi)\varphi(H_i) is, on the one hand, θ(Hi)\theta(H_i), and, on the other hand, giHπ(i)gi−1g_iH_{\pi(i)}g_i^{-1}. Moreover, by (3.2),

θ(ci)=θ(∂(Hi))=∂(φ(Hi))=∂(giHπ(i))=gicπ(i)gi−1.\theta(c_i)=\theta(\partial(H_i))=\partial(\varphi(H_i))=\partial(g_iH_{\pi(i)})=g_ic_{\pi(i)}g_i^{-1}.

Thus the pair (θ(Hi),θ(ci))(\theta(H_i),\theta(c_i)) is conjugate to (Hπ(i),cπ(i))(H_{\pi(i)},c_{\pi(i)}), for all ii, and (a) gives θ=κg\theta=\kappa_g, for some g∈Gg\in G. Let ψ:=g−1φ\psi:=g^{-1}\varphi, where g−1g^{-1} acts on XX. Then ψ∈Aut⁡(X)\psi\in\operatorname{Aut}(X) and θψ=κg−1∘θ=Id⁡G\theta_\psi=\kappa_{g^{-1}}\circ\theta=\operatorname{Id}_G, i.e. ψ\psi is GG-equivariant, and ∂∘ψ=θψ∘∂=∂\partial\circ\psi=\theta_\psi\circ\partial=\partial. By (b), ψ\psi is the action of an element of Z(G)Z(G), hence φ=gψ∈G\varphi=g\psi\in G. Since G=Inn⁡(X)⊆Aut⁡(X)G=\operatorname{Inn}(X)\subseteq\operatorname{Aut}(X), we obtain Aut⁡(X)=G\operatorname{Aut}(X)=G.

The following lemma gives a convenient sufficient condition for (b).

Lemma 3.5. With the notation of Proposition 3.4, let GG be finite and assume that:

(1) the pairs (Hi,ci)(H_i,c_i), i=1,…,si=1,\ldots,s, are pairwise non-conjugate in GG;

(2) NG(Hi)∩CG(ci)⊆HiZ(G)N_G(H_i)\cap C_G(c_i)\subseteq H_iZ(G), for all i=1,…,si=1,\ldots,s;

(3) there exists i0i_0 such that Z(G)⊆HiZ(G)\subseteq H_i, for all i≠i0i\ne i_0.

Then condition (b) of Proposition 3.4 holds. Proof. Let ψ:X→X\psi: X \to X be a GG-equivariant bijection such that ∂∘ψ=∂\partial\circ\psi= \partial. Fix ii and write ψ(Hi)=gHj\psi(H_i) = gH_j. Since ψ\psi is GG-equivariant, for any h∈Hih \in H_i we have

gHj=ψ(Hi)=ψ(hHi)=hψ(Hi)=hgHj.gH_j = \psi(H_i) = \psi(hH_i) = h\psi(H_i) = hgH_j.

i.e. Hi⊆gHjg−1H_i \subseteq gH_jg^{-1}. Moreover, ψ\psi maps the orbit G/HiG/H_i bijectively onto the orbit G/HjG/H_j; as GG is finite, ∣Hi∣=∣Hj∣|H_i| = |H_j|, and therefore Hi=gHjg−1H_i = gH_jg^{-1}. Using ∂∘ψ=∂\partial\circ\psi= \partial, we also obtain

ci=∂(Hi)=∂(ψ(Hi))=∂(gHj)=gcjg−1.c_i = \partial(H_i) = \partial(\psi(H_i)) = \partial(gH_j) = gc_jg^{-1}.

Thus the pairs (Hi,ci)(H_i,c_i) and (Hj,cj)(H_j,c_j) are conjugate, and (1) gives j=ij = i. Then g∈NG(Hi)∩CG(ci)g \in N_G(H_i) \cap C_G(c_i) and, by (2), g=hζig = h\zeta_i, with h∈Hih \in H_i and ζi∈Z(G)\zeta_i \in Z(G). Hence, for all x∈Gx \in G,

ψ(xHi)=xψ(Hi)=xgHi=xζihHi=ζixHi,\psi(xH_i) = x\psi(H_i) = xgH_i = x\zeta_i hH_i = \zeta_i xH_i,

i.e., ψ\psi is, on the orbit G/HiG/H_i, the action of the element ζi∈Z(G)\zeta_i \in Z(G). By (3), every ζ∈Z(G)\zeta\in Z(G) acts trivially on G/HiG/H_i, for all i≠i0i \ne i_0, since ζxHi=xζHi=xHi\zeta xH_i = x\zeta H_i = xH_i. Hence ψ\psi is the action of ζi0\zeta_{i_0} on the whole of XX.

Remark 3.6. Condition (2) of Lemma 3.5 is also necessary for the equality Aut⁡(X)=G\operatorname{Aut}(X) = G in Proposition 3.4: if Aut⁡(X)=G\operatorname{Aut}(X) = G, then the stabilizer of the point Hi∈G/HiH_i \in G/H_i is HiH_i and rHir_{H_i} is the action of ∂(Hi)=ci\partial(H_i) = c_i, so Lemma 2.1(2) gives NG(Hi)∩CG(ci)⊆HiZ(G)N_G(H_i) \cap C_G(c_i) \subseteq H_i Z(G). For connected quandles, a description of Aut⁡(X)\operatorname{Aut}(X) in terms of the group generated by the right translations and of its automorphisms is given in [32 Proposition 5.8]; Proposition 3.4 also applies to quandles which are not connected, as in Proposition 3.9 below.

Dihedral groups, dicyclic groups and SL⁡(2,3)\operatorname{SL}(2,3)

For n≥3n \ge3, let RnR_n be the dihedral quandle, i.e. the set Zn\mathbb{Z}_n with the operation j⊲k:=2k−jj \lhd k := 2k - j. By [22 Theorem 2.1], Aut⁡(Rn)\operatorname{Aut}(R_n) is isomorphic to the affine group Aff⁡(Zn)=Zn⋊Zn×\operatorname{Aff}(\mathbb{Z}_n) = \mathbb{Z}_n \rtimes\mathbb{Z}_n^\times, which is isomorphic to DnD_n if and only if Zn×={±1}\mathbb{Z}_n^\times= \{\pm1\}, i.e. n∈{3,4,6}n \in\{3,4,6\}; for n=4n = 4, see also [10 Corollary 5.6]. Adding two points to RnR_n, we obtain a quandle whose automorphism group is DnD_n, for all n≥3n \ge3.

Proposition 3.7. Let n≥3n \ge3 and X:=Zn⊔PX := \mathbb{Z}_n \sqcup P, where P:={p1,p−1}P := \{p_1,p_{-1}\}. For all j,k∈Znj,k \in\mathbb{Z}_n and ε,δ∈{1,−1}\varepsilon,\delta\in\{1,-1\} we define

j⊲k:=2k−j,pε⊲k:=p−ε,j⊲pε:=j+ε,pε⊲pδ:=pε.j \lhd k := 2k-j,\qquad p_\varepsilon\lhd k := p_{-\varepsilon},\qquad j \lhd p_\varepsilon:= j+\varepsilon,\qquad p_\varepsilon\lhd p_\delta:= p_\varepsilon.

Then (X,⊲)(X,\lhd) is a quandle with n+2n+2 elements and Aut⁡(X)=Inn⁡(X)≅Dn\operatorname{Aut}(X) = \operatorname{Inn}(X) \cong D_n.

Proof. Let t,sk∈Sym⁡(Zn)t,s_k \in\operatorname{Sym}(\mathbb{Z}_n) be given by t(j):=j+1t(j) := j+1 and sk(j):=2k−js_k(j) := 2k-j, for all j,k∈Znj,k \in\mathbb{Z}_n, and let τ\tau be the transposition of PP. Then the right translations of (X,⊲)(X,\lhd) are

rk=sk⊔τandrpε=tε⊔Id⁡P,for all k∈Zn, ε∈{1,−1},r_k = s_k \sqcup\tau\qquad\text{and}\qquad r_{p_\varepsilon} = t^\varepsilon\sqcup\operatorname{Id}_P,\qquad\text{for all }k \in\mathbb{Z}_n,\ \varepsilon\in\{1,-1\},

where, as in Proposition 3.1, α⊔β\alpha\sqcup\beta denotes the bijection of X=Zn⊔PX = \mathbb{Z}_n \sqcup P which is equal to α\alpha on Zn\mathbb{Z}_n and to β\beta on PP. They are all bijections, and rx(x)=xr_x(x) = x, for all x∈Xx \in X. A straightforward computation shows that, for all k,l∈Znk,l \in\mathbb{Z}_n,

skslsk−1=s2k−l,sktsk−1=t−1,tskt−1=sk+1.(13)s_k s_l s_k^{-1} = s_{2k-l},\qquad s_k t s_k^{-1} = t^{-1},\qquad t s_k t^{-1} = s_{k+1}. \tag*{(13)}

We check (1.4), i.e. ru∘ry∘ru−1=rru(y)r_u \circ r_y \circ r_u^{-1} = r_{r_u(y)}, for all u,y∈Xu,y \in X. Let k,l∈Znk,l \in\mathbb{Z}_n and ε,δ∈{1,−1}\varepsilon,\delta\in\{1,-1\}. Using (3.3), we obtain:

  • rk∘rl∘rk−1=s2k−l⊔τ=r2k−l=rrk(l)r_k \circ r_l \circ r_k^{-1}=s_{2k-l}\sqcup\tau=r_{2k-l}=r_{r_k(l)};

  • rk∘rpε∘rk−1=t−ε⊔Id⁡P=rp−ε=rrk(pε)r_k \circ r_{p_\varepsilon}\circ r_k^{-1}=t^{-\varepsilon}\sqcup\operatorname{Id}_P=r_{p_{-\varepsilon}}=r_{r_k(p_\varepsilon)};

  • rpε∘rk∘rpε−1=sk+ε⊔τ=rk+ε=rrpε(k)r_{p_\varepsilon}\circ r_k\circ r_{p_\varepsilon}^{-1}=s_{k+\varepsilon}\sqcup\tau=r_{k+\varepsilon}=r_{r_{p_\varepsilon}(k)};

  • rpε∘rpδ∘rpε−1=tδ⊔Id⁡P=rpδ=rrpε(pδ)r_{p_\varepsilon}\circ r_{p_\delta}\circ r_{p_\varepsilon}^{-1}=t^\delta\sqcup\operatorname{Id}_P=r_{p_\delta}=r_{r_{p_\varepsilon}(p_\delta)}.

Thus (X,◃)(X,\triangleleft) is a quandle with n+2n+2 elements.

For c∈Znc\in\mathbb{Z}_n and ε∈{1,−1}\varepsilon\in\{1,-1\}, let φc,ε∈Sym⁡(X)\varphi_{c,\varepsilon}\in\operatorname{Sym}(X) be given by

φc,ε(j):=c+εj,φc,ε(pδ):=pεδ,for all j∈Zn, δ∈{1,−1}.\varphi_{c,\varepsilon}(j):=c+\varepsilon j,\qquad\varphi_{c,\varepsilon}(p_\delta):=p_{\varepsilon\delta},\qquad\text{for all }j\in\mathbb{Z}_n,\ \delta\in\{1,-1\}.

Then φc,1=rp1c\varphi_{c,1}=r_{p_1}^{c} and φc,−1=rp1c∘r0\varphi_{c,-1}=r_{p_1}^{c}\circ r_0; hence φc,ε∈Inn⁡(X)⊆Aut⁡(X)\varphi_{c,\varepsilon}\in\operatorname{Inn}(X)\subseteq\operatorname{Aut}(X), for all cc and ε\varepsilon. Since

φc,ε∘φc′,ε′=φc+εc′,εε′,\varphi_{c,\varepsilon}\circ\varphi_{c',\varepsilon'}=\varphi_{c+\varepsilon c',\varepsilon\varepsilon'},

the 2n2n bijections φc,ε\varphi_{c,\varepsilon} form a subgroup of Sym⁡(X)\operatorname{Sym}(X), which is isomorphic to DnD_n: by restriction to Zn\mathbb{Z}_n, it is identified with the group of the maps j↦c±jj\mapsto c\pm j, i.e. with the group of symmetries of a regular nn-gon whose vertices are labelled by Zn\mathbb{Z}_n.

Conversely, let φ∈Aut⁡(X)\varphi\in\operatorname{Aut}(X). By (2), rφ(x)=φ∘rx∘φ−1r_{\varphi(x)}=\varphi\circ r_x\circ\varphi^{-1}, so rxr_x and rφ(x)r_{\varphi(x)} have the same order, for all x∈Xx\in X. Since rkr_k has order 22, for all k∈Znk\in\mathbb{Z}_n, while rp1r_{p_1} and rp−1r_{p_{-1}} have order n≥3n\geq3, we obtain φ(P)=P\varphi(P)=P and φ(Zn)=Zn\varphi(\mathbb{Z}_n)=\mathbb{Z}_n. Let β\beta be the restriction of φ\varphi to Zn\mathbb{Z}_n, and let ε∈{1,−1}\varepsilon\in\{1,-1\} be such that φ(p1)=pε\varphi(p_1)=p_\varepsilon; then φ(p−1)=p−ε\varphi(p_{-1})=p_{-\varepsilon}. Restricting to Zn\mathbb{Z}_n the condition (2), written for y=p1y=p_1, we obtain β∘t∘β−1=tε\beta\circ t\circ\beta^{-1}=t^\varepsilon, i.e.

β(j+1)=β(j)+ε,for all j∈Zn.\beta(j+1)=\beta(j)+\varepsilon,\qquad\text{for all }j\in\mathbb{Z}_n.

By induction, β(j)=c+εj\beta(j)=c+\varepsilon j, where c:=β(0)c:=\beta(0); hence φ=φc,ε\varphi=\varphi_{c,\varepsilon}. Therefore

Aut⁡(X)={φc,ε∣c∈Zn, ε∈{1,−1}}=Inn⁡(X)≅Dn,\operatorname{Aut}(X)=\{\varphi_{c,\varepsilon}\mid c\in\mathbb{Z}_n,\ \varepsilon\in\{1,-1\}\}=\operatorname{Inn}(X)\cong D_n,

and the proof is finished.

Remark 3.8. The quandle of Proposition 3.7 is the quandle given by (11) for the group G=Dn=⟨ρ,σ∣ρn=σ2=1, σρσ=ρ−1⟩G=D_n=\langle\rho,\sigma\mid\rho^n=\sigma^2=1,\ \sigma\rho\sigma=\rho^{-1}\rangle, s=2s=2, H1=⟨σ⟩H_1=\langle\sigma\rangle, c1=σc_1=\sigma, H2=⟨ρ⟩H_2=\langle\rho\rangle and c2=ρc_2=\rho: the coset ρjH1\rho^jH_1 corresponds to j∈Znj\in\mathbb{Z}_n, and the cosets H2H_2 and σH2\sigma H_2 correspond to p1p_1 and p−1p_{-1}. Moreover, the quandles of Proposition 3.1 and Proposition 3.7 are particular cases of the quandle structures on the disjoint union of two quandles constructed in [10 Proposition 9.2]; their automorphism groups are not considered there.

Like Q8Q_8, the dicyclic groups have a unique involution. By Corollary 2.5, Dic⁡n\operatorname{Dic}_n is not the automorphism group of a rack if nn is even. We show now that the situation is completely different if nn is odd; for instance, for n=3n=3 we obtain a quandle with 77 elements whose automorphism group is the dicyclic group Dic⁡3\operatorname{Dic}_3 of order 1212.

Proposition 3.9. Let n≥3n\geq3 be an odd integer and G:=Dic⁡nG:=\operatorname{Dic}_n. Consider the quandle (X,◃)(X,\triangleleft) given by (11) for s=2s=2 and

H1:=⟨a2⟩,c1:=a2,H2:=⟨b⟩,c2:=b.H_1:=\langle a^2\rangle,\qquad c_1:=a^2,\qquad H_2:=\langle b\rangle,\qquad c_2:=b.

Then XX has n+4n+4 elements and Aut⁡(X,◃)=G≅Dic⁡n\operatorname{Aut}(X,\triangleleft)=G\cong\operatorname{Dic}_n. In particular, for n≥2n\geq2, the dicyclic group Dic⁡n\operatorname{Dic}_n is isomorphic to the automorphism group of a rack if and only if nn is odd.

Proof. Let z:=anz := a^n. Since nn is odd, a2a^2 has order nn; thus ∣H1∣=n|H_1| = n, while ∣H2∣=4|H_2| = 4, by Lemma 1.3(1), and XX has 4n/n+4n/4=n+44n/n + 4n/4 = n + 4 elements. We first check the assumptions of Proposition 3.4. As ∣H1∣|H_1| and ∣H2∣|H_2| are coprime, we have

core⁡G(H1)∩core⁡G(H2)⊆H1∩H2=1.\operatorname{core}_G(H_1) \cap\operatorname{core}_G(H_2) \subseteq H_1 \cap H_2 = 1.

The subgroup generated by a2a^2 and bb contains b2=anb^2 = a^n; as nn is odd, it also contains a=(a2)(n+1)/2a−na = (a^2)^{(n+1)/2}a^{-n}, hence it is equal to GG. Next, we check the conditions of Lemma 3.5:

  • (1) holds, since ∣H1∣=n≠4=∣H2∣|H_1| = n \ne4 = |H_2|;

  • (2) holds: as n≥3n \ge3, we have a2∉⟨z⟩a^2 \notin\langle z\rangle, hence CG(a2)=⟨a⟩C_G(a^2) = \langle a\rangle, by Lemma 1.3(3), and ⟨a⟩=⟨a2⟩⟨an⟩=H1Z(G)\langle a\rangle= \langle a^2\rangle\langle a^n\rangle= H_1 Z(G), since nn is odd; moreover, CG(b)=⟨b⟩=H2C_G(b) = \langle b\rangle= H_2, by Lemma 1.3(4);

  • (3) holds for i0=1i_0 = 1, since Z(G)=⟨z⟩⊆⟨b⟩=H2Z(G) = \langle z\rangle\subseteq\langle b\rangle= H_2.

Finally, we check (a) of Proposition 3.4. Let θ∈Aut⁡(G)\theta\in\operatorname{Aut}(G) and π∈S2\pi\in S_2 be as in (a). Since ∣H1∣≠∣H2∣|H_1| \ne|H_2|, π\pi is the identity, so θ(a2)\theta(a^2) is conjugate to a2a^2 and θ(b)\theta(b) is conjugate to bb. As 2n>42n > 4, the elements of order 2n2n of GG belong to ⟨a⟩\langle a\rangle, by Lemma 1.3(1); hence θ(a)=au\theta(a) = a^u, for some integer uu coprime with 2n2n, and θ(b)∉⟨a⟩\theta(b) \notin\langle a\rangle, i.e. θ(b)=avb\theta(b) = a^v b, for some v∈Zv \in\mathbb{Z}. By Lemma 1.3(4), v=2jv = 2j is even. By Lemma 1.3(1), the conjugacy class of a2a^2 is {a2,a−2}\{a^2, a^{-2}\}, hence a2u=a±2a^{2u} = a^{\pm2}, i.e. u≡±1(modn)u \equiv\pm1 \pmod n; since uu and nn are odd, this gives u≡±1(mod2n)u \equiv\pm1 \pmod{2n}. As κb(a)=a−1\kappa_b(a) = a^{-1} and κb(a2jb)=a−2jb\kappa_b(a^{2j}b) = a^{-2j}b, replacing θ\theta by κb∘θ\kappa_b \circ\theta if necessary, we may assume that θ(a)=a\theta(a) = a. Then θ\theta and κaj\kappa_{a^j} coincide on aa and on bb, since ajba−j=a2jba^j b a^{-j} = a^{2j}b; hence θ=κaj\theta= \kappa_{a^j} is an inner automorphism. By Lemma 3.5 and Proposition 3.4, Aut⁡(X)=G≅Dic⁡n\operatorname{Aut}(X) = G \cong\operatorname{Dic}_n. The last statement follows from Corollary 2.5.

The group SL(2,3)SL(2,3) also has a unique involution, and its Sylow 2-subgroup is the quaternion group Q8Q_8. Nevertheless, it is the automorphism group of a quandle with 8 elements, namely of a subquandle of the symplectic quandle on F32\mathbb{F}_3^2 [9 Example 6.12], [42].

Proposition 3.10. Let F3\mathbb{F}_3 be the field with three elements, X:=F32∖{0}X := \mathbb{F}_3^2 \setminus\{0\}, and let ω(x,y):=x1y2−x2y1\omega(x,y) := x_1y_2 - x_2y_1, for all x=(x1,x2)tx = (x_1,x_2)^t, y=(y1,y2)t∈F32y = (y_1,y_2)^t \in\mathbb{F}_3^2. Then XX is a quandle with 8 elements with respect to the operation

x⊲y:=x+ω(x,y)y,for all x,y∈X,x \lhd y := x + \omega(x,y)y, \qquad\text{for all } x,y \in X,

and Aut⁡(X,⊲)≅SL(2,3)\operatorname{Aut}(X,\lhd) \cong SL(2,3).

Proof. Let G:=SL(2,3)G := SL(2,3), acting on XX by matrix multiplication; this action is faithful and transitive, and the stabilizer of e1:=(1,0)te_1 := (1,0)^t is the abelian group

U:={(1λ01)∣λ∈F3}.U := \left\{ \begin{pmatrix} 1 & \lambda\\ 0 & 1 \end{pmatrix} \mathrel{\big|} \lambda\in\mathbb{F}_3 \right\}.

For y∈Xy \in X, the map ty:F32→F32t_y: \mathbb{F}_3^2 \to\mathbb{F}_3^2, ty(x):=x+ω(x,y)yt_y(x) := x + \omega(x,y)y, is a transvection; thus ty∈Gt_y \in G, and x⊲y=ty(x)∈Xx \lhd y = t_y(x) \in X, for all x∈Xx \in X. Since ω(gx,gy)=det⁡(g)ω(x,y)=ω(x,y)\omega(gx,gy) = \det(g)\omega(x,y) = \omega(x,y), for all g∈Gg \in G, we have

gtyg−1=tgy,for all g∈G, y∈X.g t_y g^{-1} = t_{gy}, \qquad\text{for all } g \in G,\ y \in X.

Hence, identifying XX with G/UG/U via ge1↦gUge_1 \mapsto gU, the operation ⊲\lhd is given by (3.1) for s=1s = 1, H1:=UH_1 := U and

c1:=u:=te1=(1−101)∈U.c_1 := u := t_{e_1} = \begin{pmatrix} 1 & -1 \\ 0 & 1 \end{pmatrix} \in U.

Moreover, GG is generated by the conjugates of uu: the group SL(2,3)SL(2,3) is generated by the elementary matrices, which are powers of u=te1u = t_{e_1} and of its conjugate te2=(1011)t_{e_2} = \begin{pmatrix} 1 & 0 \\ 1 & 1 \end{pmatrix}, where e2:=(0,1)te_2 := (0,1)^t. By Proposition 3.4, (X,◃)(X,\triangleleft) is a quandle. We check now the conditions of Lemma 3.5: (1) and (3) are trivially fulfilled for s=1s=1, and (2) holds since the matrices which commute with uu are the upper triangular matrices with equal diagonal entries, so that

CG(u)={±I2}U=UZ(G).C_G(u)=\{\pm I_2\}U=UZ(G).

It remains to check (a) of Proposition 3.4. We first show that every automorphism of GG is the conjugation by an element of GL(2,3)GL(2,3). The normal Sylow 22-subgroup of GG is the quaternion group Q8Q_8, generated by the matrices (0−110)\begin{pmatrix}0&-1\\1&0\end{pmatrix} and (111−1)\begin{pmatrix}1&1\\1&-1\end{pmatrix}, and a direct computation shows that the only matrices which commute with both of them are the scalar matrices. Consequently, CG(Q8)={±I2}=Z(G)C_G(Q_8)=\{\pm I_2\}=Z(G), and the conjugation by the elements of GL(2,3)GL(2,3) gives an injective morphism of groups PGL(2,3)→Aut⁡(G)PGL(2,3)\to\operatorname{Aut}(G), whose image has 24 elements. On the other hand, the restriction map Aut⁡(G)→Aut⁡(Q8)≅S4\operatorname{Aut}(G)\to\operatorname{Aut}(Q_8)\cong S_4 is injective. Indeed, if θ∈Aut⁡(G)\theta\in\operatorname{Aut}(G) is the identity on Q8Q_8, then, for all g∈Gg\in G and q∈Q8q\in Q_8,

θ(g)qθ(g)−1=θ(gqg−1)=gqg−1,\theta(g)q\theta(g)^{-1}=\theta(gqg^{-1})=gqg^{-1},

so g−1θ(g)∈CG(Q8)=Z(G)g^{-1}\theta(g)\in C_G(Q_8)=Z(G). Thus θ(g)=gμ(g)\theta(g)=g\mu(g), where μ:G→Z(G)≅C2\mu:G\to Z(G)\cong C_2 is a morphism of groups; since G/[G,G]≅C3G/[G,G]\cong C_3, μ\mu is trivial, and θ=Id⁡G\theta=\operatorname{Id}_G. Hence ∣Aut⁡(G)∣≤24|\operatorname{Aut}(G)|\leq24, and every automorphism of GG is the conjugation by some g∈GL(2,3)g\in GL(2,3).

Let now θ=κg\theta=\kappa_g, with g∈GL(2,3)g\in GL(2,3), be as in (a); in particular, θ(u)\theta(u) is conjugate to uu in GG. Assume that det⁡(g)=−1\det(g)=-1 and write g=g0dg=g_0d, with g0∈Gg_0\in G and d:=diag⁡(1,−1)d:=\operatorname{diag}(1,-1). Since dud−1=u−1dud^{-1}=u^{-1}, we obtain

θ(u)=g0(dud−1)g0−1=g0u−1g0−1.\theta(u)=g_0(dud^{-1})g_0^{-1}=g_0u^{-1}g_0^{-1}.

so u−1u^{-1} is conjugate to uu in GG. This is not possible: if huh−1=u−1=dud−1huh^{-1}=u^{-1}=dud^{-1}, with h∈Gh\in G, then d−1hd^{-1}h commutes with uu, hence it is an upper triangular matrix with equal diagonal entries and its determinant is 1, while det⁡(d−1h)=−1\det(d^{-1}h)=-1. Thus det⁡(g)=1\det(g)=1, i.e. θ\theta is an inner automorphism of GG. By Lemma 3.5 and Proposition 3.4, Aut⁡(X)=G=SL(2,3)\operatorname{Aut}(X)=G=SL(2,3). □

Conjugation quandles and alternating groups

Let GG be a group and C⊆GC\subseteq G a union of conjugacy classes of GG. Then CC is a quandle with respect to the operation x◃y:=yxy−1x\triangleleft y:=yxy^{-1}; it is a subquandle of the conjugation quandle of GG [9 Example 6.17]. The following result is a variant, for subquandles, of the description of the automorphism group of the conjugation quandle of a group with trivial center [10 Corollary 4.4].

Proposition 3.11. Let GG be a group with Z(G)=1Z(G)=1 and C⊆GC\subseteq G a union of conjugacy classes which generates GG. Then the map

{θ∈Aut⁡(G)∣θ(C)=C}→Aut⁡(C,◃),θ↦θ∣C,\{\theta\in\operatorname{Aut}(G)\mid\theta(C)=C\}\to\operatorname{Aut}(C,\triangleleft),\qquad\theta\mapsto\theta|_C,

is an isomorphism of groups. In particular, Aut⁡(C,◃)≅G\operatorname{Aut}(C,\triangleleft)\cong G if and only if every automorphism of GG which preserves CC is inner.

Proof. If θ∈Aut⁡(G)\theta\in\operatorname{Aut}(G) and θ(C)=C\theta(C)=C, then θ∣C\theta|_C is an automorphism of (C,◃)(C,\triangleleft), and θ↦θ∣C\theta\mapsto\theta|_C is an injective morphism of groups, since CC generates GG. It remains to prove that it is surjective. Consider the morphism of groups

G→Inn⁡(C),g↦κg∣C.(14)G\to\operatorname{Inn}(C),\qquad g\mapsto\kappa_g|_C. \tag*{(14)}

It is surjective, since ry=κy∣Cr_y=\kappa_y|_C, for all y∈Cy \in C, and CC generates GG; it is injective, since its kernel is CG(C)=Z(G)=1C_G(C)=Z(G)=1. Let φ∈Aut⁡(C,◃)\varphi\in\operatorname{Aut}(C,\triangleleft). By (1.2), the conjugation by φ\varphi maps Inn⁡(C)\operatorname{Inn}(C) onto itself; via the above isomorphism, it induces an automorphism θ\theta of GG such that

κθ(g)∣C=φ∘κg∣C∘φ−1,for all g∈G.\kappa_{\theta(g)}|_C=\varphi\circ\kappa_g|_C\circ\varphi^{-1},\qquad\text{for all }g\in G.

For c∈Cc \in C, using (1.2) once more, we obtain

κθ(c)∣C=φ∘rc∘φ−1=rφ(c)=κφ(c)∣C.\kappa_{\theta(c)}|_C=\varphi\circ r_c\circ\varphi^{-1}=r_{\varphi(c)}=\kappa_{\varphi(c)}|_C.

hence θ(c)=φ(c)\theta(c)=\varphi(c). Thus θ(C)=C\theta(C)=C and φ=θ∣C\varphi=\theta|_C. The last statement follows, since the inner automorphisms of GG preserve CC and form a subgroup of Aut⁡(G)\operatorname{Aut}(G) isomorphic to G/Z(G)=GG/Z(G)=G.

We apply Proposition 3.11 to the alternating groups. Recall that Z(An)=1Z(A_n)=1, for all n≥4n\geq4, and that every automorphism of AnA_n is the conjugation by an element of SnS_n, for all n≥4n\geq4, n≠6n\ne6 [21 Section 8.2].

Proposition 3.12. Let n≥4n\geq4 and let C⊆AnC\subseteq A_n be:

  1. the conjugacy class in AnA_n of the nn-cycle (1 2 ⋯ n)(1\ 2\ \cdots\ n), if nn is odd;

  2. the conjugacy class in AnA_n of the (n−1)(n-1)-cycle (1 2 ⋯ n−1)(1\ 2\ \cdots\ n-1), if nn is even and n≠6n\ne6;

  3. the union of the conjugacy classes in A6A_6 of (1 2 3)(1\ 2\ 3) and (1 2 3 4 5)(1\ 2\ 3\ 4\ 5), if n=6n=6.

Then Aut⁡(C,◃)≅An\operatorname{Aut}(C,\triangleleft)\cong A_n. The quandle CC has (n−1)!/2(n-1)!/2 elements in case (i), n(n−2)!/2n(n-2)!/2 elements in case (ii), and 112 elements in case (iii).

Proof. We start with a well known fact. Let g∈Ang\in A_n be an nn-cycle, if nn is odd, or an (n−1)(n-1)-cycle, if nn is even. The conjugacy class of gg in SnS_n has (n−1)!(n-1)!, respectively n(n−2)!n(n-2)! elements, hence CSn(g)=⟨g⟩≤AnC_{S_n}(g)=\langle g\rangle\leq A_n. If σ∈Sn\sigma\in S_n is an odd permutation, then σgσ−1\sigma g\sigma^{-1} is not conjugate to gg in AnA_n: otherwise, σgσ−1=hgh−1\sigma g\sigma^{-1}=hgh^{-1}, for some h∈Anh\in A_n, and h−1σ∈CSn(g)≤Anh^{-1}\sigma\in C_{S_n}(g)\leq A_n, a contradiction. Thus the conjugacy class of gg in SnS_n is the union of two conjugacy classes of AnA_n with the same number of elements, which are interchanged by the conjugation with any odd permutation.

This proves the statement on the number of elements of CC in the cases (i) and (ii). In the case (iii), the conjugacy class of (1 2 3)(1\ 2\ 3) in A6A_6 is the set TT of all the 40 cycles of length 3, since (1 2 3)(1\ 2\ 3) commutes with the odd permutation (4 5)(4\ 5), and the conjugacy class C′C' of (1 2 3 4 5)(1\ 2\ 3\ 4\ 5) in A6A_6 has 144/2=72144/2=72 elements; hence ∣C∣=112|C|=112.

The subgroup of AnA_n generated by CC is normal and non-trivial, so it is equal to AnA_n: this is clear for n≥5n\geq5, when AnA_n is simple, while for n=4n=4 the normal subgroups of A4A_4 are 11, the Klein group and A4A_4, and CC consists of elements of order 3. By Proposition 3.11, it is enough to prove that every automorphism θ\theta of AnA_n such that θ(C)=C\theta(C)=C is inner.

If n≠6n\ne6, then θ\theta is the conjugation by some σ∈Sn\sigma\in S_n, and σ\sigma is even, since the conjugation by an odd permutation maps CC onto a different conjugacy class of AnA_n. Hence θ\theta is inner.

Let now n=6n=6, so that C=T∪C′C=T\cup C'. Since θ\theta preserves the orders of the elements, we have θ(T)=T\theta(T)=T and θ(C′)=C′\theta(C')=C'. We show that θ\theta is the conjugation by a permutation. For i∈{1,…,6}i\in\{1,\ldots,6\}, let Mi≅A5M_i\cong A_5 be the stabilizer of ii in A6A_6, and let L:=θ(Mi)L:=\theta(M_i). Then L≅A5L\cong A_5 and LL contains cycles of length 3, since MiM_i does and θ(T)=T\theta(T)=T. The orbits of LL on {1,…,6}\{1,\ldots,6\} have 1, 5 or 6 elements, since A5A_5 has no subgroups of index 2, 3 or 4. If LL were transitive, then a cycle of length 3 of LL, which fixes three points, would belong to the stabilizer in LL of one of these points; but this stabilizer has order 60/6=1060/6 = 10, a contradiction. Hence LL fixes a point jj, and L=MjL = M_j, as ∣L∣=∣Mj∣|L| = |M_j|. Since θ\theta is bijective, we obtain a permutation π∈S6\pi\in S_6 such that θ(Mi)=Mπ(i)\theta(M_i) = M_{\pi(i)}, for all ii. Let θ′:=κπ−1∘θ\theta' := \kappa_{\pi}^{-1} \circ\theta, where κπ\kappa_{\pi} denotes the conjugation by π\pi, restricted to A6A_6; then θ′(Mi)=Mi\theta'(M_i) = M_i, for all ii. Since gMig−1=Mg(i)gM_i g^{-1} = M_{g(i)} for all g∈A6g \in A_6, we obtain

Mθ′(g)(i)=θ′(g)Miθ′(g)−1=θ′(gMig−1)=θ′(Mg(i))=Mg(i),M_{\theta'(g)(i)} = \theta'(g) M_i \theta'(g)^{-1} = \theta'(g M_i g^{-1}) = \theta'(M_{g(i)}) = M_{g(i)},

hence θ′(g)(i)=g(i)\theta'(g)(i) = g(i), for all ii, i.e. θ′(g)=g\theta'(g) = g. Thus θ=κπ\theta= \kappa_{\pi} and, as above, π\pi is even, since θ(C′)=C′\theta(C') = C'. Hence θ\theta is inner, and the proof is finished.

Corollary 3.13. Every alternating group AnA_n and every symmetric group SnS_n is isomorphic to the automorphism group of a finite quandle.

Proof. The symmetric group SnS_n is the automorphism group of the trivial quandle with nn elements. The groups A1A_1 and A2A_2 are trivial, A3≅C3A_3 \cong C_3 is covered by Corollary 3.2, and for n≥4n \ge4 we apply Proposition 3.12.

Remark 3.14. For n=6n = 6, the conjugacy class C′C' of (1 2 3 4 5)(1\ 2\ 3\ 4\ 5) in A6A_6 is not enough: a computer check shows that Aut⁡(C′,◃)\operatorname{Aut}(C',\triangleleft) has order 720, so, by Proposition 3.11, C′C' is preserved by an automorphism of A6A_6 which is not inner. More generally, by Proposition 3.11, if GG is a finite non-abelian simple group and CC is a union of non-trivial conjugacy classes of GG which is not preserved by any outer automorphism of GG, then Aut⁡(C,◃)≅G\operatorname{Aut}(C,\triangleleft) \cong G.

Concluding remarks and open questions

We have shown that, unlike the category of all set-theoretic solutions of the Yang–Baxter equation [39], the categories of racks and of quandles are neither universal nor finitely universal and that, as for groups, the reason is the presence of the inner automorphisms, which leads to the local condition (2.1). This raises the following natural problem: which finite groups are represented by finite racks, respectively by finite quandles? By Theorem 2.2, if a group GG is the automorphism group of a rack, then for every 1≠z∈Z(G)1 \ne z \in Z(G) there exist a subgroup K≤GK \le G with z∉Kz \notin K and an element c∈CG(K)c \in C_G(K) such that

NG(K)∩CG(c)⊆KZ(G).N_G(K) \cap C_G(c) \subseteq K Z(G).

We do not know whether this necessary condition is also sufficient for finite groups. For instance, a computer check shows that the groups Dic⁡3×C2\operatorname{Dic}_3 \times C_2 and Q8×C4Q_8 \times C_4 satisfy it, but we do not know whether they are automorphism groups of racks. The condition is empty for groups with trivial center: is every finite group with trivial center the automorphism group of a finite rack, or of a finite quandle? It is also natural to ask whether there exists a group which is the automorphism group of a rack, but not of a quandle; all the groups realized in Section 3 are automorphism groups of quandles.

For the quandles of Corollary 3.2 and of Propositions 3.7, 3.9, 3.10 and 3.12 we have Inn⁡(X)=Aut⁡(X)\operatorname{Inn}(X) = \operatorname{Aut}(X). This is no accident for simple groups: since Inn⁡(X)\operatorname{Inn}(X) is a normal subgroup of Aut⁡(X)\operatorname{Aut}(X), a finite non-abelian simple group GG can only be represented by a rack XX such that G=Aut⁡(X)=Inn⁡(X)G = \operatorname{Aut}(X) = \operatorname{Inn}(X). Indeed, if Inn⁡(X)=1\operatorname{Inn}(X) = 1, then XX is a trivial rack and Aut⁡(X)=Sym⁡(X)\operatorname{Aut}(X) = \operatorname{Sym}(X), which is not a non-abelian simple group. By Proposition 3.11, such a rack exists as soon as GG has a union of non-trivial conjugacy classes which is not preserved by any outer automorphism of GG, as for the alternating groups in Proposition 3.12. Is every finite non-abelian simple group the automorphism group of a finite quandle? Following Babai [5], one can also ask quantitative questions: given a finite group GG which is represented by a finite rack, what is the smallest size of such a rack? Since Aut⁡(X)\operatorname{Aut}(X) is a subgroup of Sym⁡(X)\operatorname{Sym}(X), this size is at least the minimal degree of a faithful permutation representation of GG; for instance, the quandles of Proposition 3.7 and Proposition 3.9 have n+2n+2, respectively n+4n+4 elements.

Finally, as explained in Remark 2.7, Theorem 2.2 does not apply to arbitrary bijective non-degenerate solutions of the braid equation, which leaves open the question raised in [39 Section 3]. The first test case is the following: is the quaternion group Q8Q_8 isomorphic to the automorphism group of a bijective non-degenerate solution of the braid equation? More generally, is the category of bijective non-degenerate solutions of the braid equation universal, or finitely universal? By Theorem 2.2 and Remark 2.7, a positive answer would require solutions which are neither derived solutions of racks nor involutive solutions with σσx(y)=σy\sigma_{\sigma_x(y)}=\sigma_y, for all x,y∈Xx,y\in X.

Acknowledgements

Claude Opus 5.5 (Anthropic) was used throughout the preparation of this paper: to find references, to carry out the computer verifications, to explore and develop constructions and proofs, and for drafting and proofreading. All statements, proofs and computational results were checked by the author, who takes full responsibility for them.

University of Bucharest, Faculty of Mathematics and Computer Science

Str. Academiei 14, RO-010014 Bucharest 1, Romania

E-mail address: [email protected]

References

  1. [1]A.L. Agore, A. Chirvasitu and G. Militaru, A categorical approach to the set-theoretic Yang–Baxter equation, arXiv:2610.15347. 6arxiv.org/abs/2610.15347
  2. [2]A.L. Agore, A. Chirvasitu and G. Militaru, The set-theoretic Yang–Baxter equation, Kimura semigroups and functional graphs, Res. Math. Sci. 12 (2025), Paper No. 34. 2, 4arxiv.org/abs/2303.06700
  3. [3]N. Andruskiewitsch and M. Graña, From racks to pointed Hopf algebras, Adv. Math. 178 (2003), 177–243. 2
  4. [4]L. Babai, On the abstract group of automorphisms, in: Combinatorics (Swansea, 1981), London Math. Soc. Lecture Note Ser. 52, Cambridge University Press, Cambridge, 1981, pp. 1–40. 1, 2
  5. [5]L. Babai, Automorphism groups, isomorphism, reconstruction, in: Handbook of Combinatorics, Vol. 2, Elsevier, Amsterdam, 1995, pp. 1447–1540. 2, 19
  6. [6]R. Baer, Situation der Untergruppen und Struktur der Gruppe, S.-B. Heidelberg. Akad. Wiss. 2 (1933), 12–17. 9DOI
  7. [7]C. Bai, L. Guo, Y. Sheng and R. Tang, Post-groups, (Lie-)Butcher groups and the Yang–Baxter equation, Math. Ann. 388 (2024), 3127–3167. 6DOI
  8. [8]V.G. Bardakov, P. Dey and M. Singh, Automorphism groups of quandles arising from groups, Monatsh. Math. 184 (2017), 519–530. 3DOI
  9. [9]V.G. Bardakov, M. Elhamdadi and M. Singh, Algebra of the Yang–Baxter equation. Skew braces, quandles, and cohomology, Springer Monographs in Mathematics, Springer, to appear. 2, 3, 5, 12, 16, 17
  10. [10]V.G. Bardakov, T.R. Nasybullov and M. Singh, Automorphism groups of quandles and related groups, Monatsh. Math. 189 (2019), 1–21. 3, 14, 15, 17arxiv.org/abs/1705.10607
  11. [11]V.G. Bardakov and B. Sangaré, Constructions and automorphisms of quandles, arXiv:2610.18676. 3
  12. [12]V.G. Bardakov and M. Singh, Quandale cohomology, extensions and automorphisms, J. Algebra 585 (2021), 558–591. 3DOI
  13. [13]G.M. Bergman and A.O. Hausknecht, Co-groups and co-rings in categories of associative rings, Mathematical Surveys and Monographs, American Mathematical Society, Providence, RI, 1996. 6DOI
  14. [14]A. Bishop, C. Cyr, J. Hutchens, C. May, N. Schwartz and B. Turner, On involutions and generalized symmetric spaces of dicyclic groups, arXiv:1310.0121. 7
  15. [15]N. Bourbaki, Algebra I. Chapters 1–3, Elements of Mathematics (Berlin). Springer-Verlag, Berlin, 1998. 4
  16. [16]E. Brieskorn, Automorphic sets and braids and singularities, in: Braids (Santa Cruz, CA, 1986), Contemp. Math. 78, Amer. Math. Soc., Providence, RI, 1988, pp. 45–115. 2, 5DOI
  17. [17]N. Cazet, Quandles with one non-trivial column, arXiv:2303.15815. 3, 10
  18. [18]K. Conrad, Generalized quaternions, expository note, https://kconrad.math.uconn.edu/blurbs/grouptheory/genquat.pdf. 6
  19. [19]J. de Groot, Groups represented by homeomorphism groups I, Math. Ann. 138 (1959), 80–102. 2DOI
  20. [20]R. Dedekind, Ueber Gruppen, deren sämtliche Theiler Normaltheiler sind, Math. Ann. 48 (1897), 548–561. 9DOI
  21. [21]J.D. Dixon and B. Mortimer, Permutation groups, Graduate Texts in Mathematics 163, Springer-Verlag, New York, 1996. 18
  22. [22]M. Elhamdadi, J. Macquarrie and R. Restrepo, Automorphism groups of quandles, J. Algebra Appl. 11 (2012), 1250008, 9 pp. 3, 4, 14DOI
  23. [23]M. Elhamdadi and S. Nelson, Quandles. An introduction to the algebra of knots, Student Mathematical Library 74, American Mathematical Society, Providence, RI, 2015. 2, 5
  24. [24]P. Etingof, T. Schedler and A. Soloviev, Set-theoretical solutions to the quantum Yang–Baxter equation, Duke Math. J. 100 (1999), 169–209. 5, 10arxiv.org/abs/math/9801047
  25. [25]R. Fenn and C. Rourke, Racks and links in codimension two, J. Knot Theory Ramifications 1 (1992), 343–406. 2, 6, 12DOI
  26. [26]A. Ferman, T. Nowik and M. Teicher, On the structure and automorphism group of finite Alexander quandles, J. Knot Theory Ramifications 20 (2011), 463–468. 3arxiv.org/abs/0811.4211
  27. [27]T. Fritz, Self-distributive structures in physics, arXiv:2403.14458. 2
  28. [28]R. Frucht, Herstellung von Graphen mit vorgegebener abstrakter Gruppe, Compositio Math. 6 (1939), 239–250. 2
  29. [29]L. Guarnieri and L. Vendramin, Skew braces and the Yang–Baxter equation, Math. Comp. 86 (2017), 2519–2534. 6DOI
  30. [30]B. Ho and S. Nelson, Matrices and finite quandles, Homology Homotopy Appl. 7 (2005), 197–208. 3DOI
  31. [31]X.D. Hou, Automorphism groups of Alexander quandles, J. Algebra 344 (2011), 373–385. 3DOI
  32. [32]A. Hulpke, D. Stanovský and P. Vojtěchovský, Connected quandles and transitive groups, J. Pure Appl. Algebra 220 (2016), 735–758. 14arxiv.org/abs/1409.2249
  33. [33]D. Joyce, A classifying invariant of knots, the knot quandle, J. Pure Appl. Algebra 23 (1982), 37–65. 2, 3, 12DOI
  34. [34]D. König, Theorie der endlichen und unendlichen Graphen, Akademische Verlagsgesellschaft, Leipzig, 1936. 1
  35. [35]V. Lebed and L. Vendramin, On structure groups of set-theoretic solutions to the Yang–Baxter equation, Proc. Edinb. Math. Soc. 62 (2019), 683–717. 10DOI
  36. [36]J.-H. Lu, M. Yan and Y.-C. Zhu, On the set-theoretical Yang–Baxter equation, Duke Math. J. 104 (2000), 1–18. 6DOI
  37. [37]J. Macquarrie, Automorphism groups of quandles, Master’s thesis, University of South Florida, 2011. 3
  38. [38]S.V. Matveev, Distributive groupoids in knot theory, Mat. Sb. (N.S.) 119(161) (1982), 78–88 (in Russian); English transl. in Math. USSR-Sb. 47 (1984), 73–83. 2
  39. [39]G. Militaru, Every group is the automorphism group of a set-theoretic solution of the Yang–Baxter equation, arXiv:2610.39228. 2, 3, 4, 5, 10, 12, 19, 20arxiv.org/abs/2610.39228
  40. [40]G. Militaru, Bipartite solutions of the set-theoretic Yang–Baxter equation, arXiv:2610.30845. 6
  41. [41]G. Militaru, Centralizers of maps as automorphism groups of Yang–Baxter solutions, arXiv:2611.03380. 2, 4, 10
  42. [42]E.A. Navas and S. Nelson, On symplectic quandles, Osaka J. Math. 45 (2008), 973–985. 16arxiv.org/abs/math/0703727
  43. [43]A. Pultr and V. Trnková, Combinatorial, algebraic and topological representations of groups, semigroups and categories, North-Holland Mathematical Library 22, North-Holland, Amsterdam–New York, 1980. 2DOI
  44. [44]S. Roman, Fundamentals of group theory. An advanced approach, Birkhäuser/Springer, New York, 2012. 6DOI
  45. [45]G. Sabidussi, Graphs with given infinite group, Monatsh. Math. 64 (1960), 64–67. 2DOI
  46. [46]Q.J.M. Salix and P.P. Wood, More automorphism groups of quandles, arXiv:2510.26673. 3
  47. [47]B. Sangare, Automorphisms and antiautomorphisms of quandles, arXiv:2602.03080. 3
  48. [48]A. Smoktunowicz and L. Vendramin, On skew braces (with an appendix by N. Byott and L. Vendramin), J. Comb. Algebra 2 (2018), 47–86. 6DOI
  49. [49]A. Soloviev, Non-unitary set-theoretical solutions to the quantum Yang–Baxter equation, Math. Res. Lett. 7 (2000), 577–596. 10DOI
  50. [50]P. Vojtěchovský and S.Y. Yang, Enumeration of racks and quandles up to isomorphism, Math. Comp. 88 (2019), 2523–2540. 6arxiv.org/abs/1911.04991

Paper details

vlnc:W_9da39jyw
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Claude Opus 5.5

View statement
Looking for related papers…
Contents