Groups with special presentations and star-graph K 3 , 3 K_{3,3} Bridgette Amoako, Ihechukwu Chinyere, Bernard Bainson Abstract We consider a question of Edjvet and Vdovina concerning which groups defined by special presentations are large. For each integer n ≥ 3 n\geq 3 , we construct an n n -generator one-relator presentation whose star graph is the complete bipartite graph K n , n K_{n,n} ; the resulting groups are large and hyperbolic. We also classify concise special presentations with star graph K 3 , 3 K_{3,3} , showing that they are one-relator presentations and that, up to Tietze equivalence, there are exactly twelve that define torsion-free groups. The torsion cases arise precisely as positive powers of the relators in the torsion-free cases, and define pairwise non-isomorphic groups that remain large and hyperbolic. Keywords: One-relator groups, special presentations, hyperbolic, large, star-graph. 1 Introduction Constrained incidence structures and group presentations serve as a meeting point between combinatorial group theory, incidence geometry, and geometric group theory. In this article, we study group presentations whose star-graphs (see Section 2 ) are the incidence graphs of generalised polygons. This connection was observed by Howie [ 12 ] in his study of the SQ-universality of T(6) -groups. In that work, Howie showed that these presentations form the only class of T(6) -groups for which SQ-universality could not be settled using his methods. (Recall that a countable group G is SQ-universal if every countable group embeds into a quotient of G .) Owing to this exceptional status, they became known as special presentations , that is, finite presentations with relators of length three whose star-graphs are incidence graphs of finite projective planes. Howie further posed the problem of whether any such group is SQ-universal [ 12 , Question 6.11] . The theory of special presentations was systematically developed by Edjvet and Howie [ 7 ] , who showed that many such groups contain non-abelian free subgroups. Howie’s question was subsequently answered in the negative by Edjvet and Vdovina [ 8 ] , who also introduced a broader class of presentations, called (m,k) -special presentations . These are finite presentations P=\langle X\mid R\rangle satisfying: 1. the star-graph of P is connected, bipartite, of diameter m and girth 2m , and each vertex has degree at least 3 ; 2. each relator in R has length k ; and 3. if m=2 , then k\geq 4 . By the classical theorem of Feit and Higman [ 9 ] , the diameter parameter of a generalised polygon is restricted to m\in\{2,3,4,6,8\}. A group defined by P is said to have a special presentation (or an (m,k) -special presentation, to emphasise the combinatorial parameters). Howie’s original special presentations correspond exactly to the (3,3) case. A further generalisation, the class of (m,k,\nu) -special presentations, was introduced in [ 4 , 5 ] . In this setting, the star graph is allowed to have \nu\geq 1 isomorphic connected components, each of which is the incidence graph of a generalised polygon. In particular, (m,k) -special presentations correspond precisely to the case \nu=1 , and we retain this notation when \nu=1 . In the case of cyclic presentations, the complete classification obtained in [ 4 ] shows that only m=2 or m=3 can occur. Moreover, all (3,k) -special presentations in this setting define hyperbolic groups, except for the (3,3) -special presentation, which defines a just-infinite group [ 8 ] , that is, an infinite group all of whose non-trivial quotients are finite. It remains an open problem (see [ 8 , Question 2] ) whether there exist general construction methods (“example machines”) for producing (m,k) -special presentations for m\in\{4,6,8\} . The geometric significance of groups defined by special presentations was explored in subsequent work of Vdovina and collaborators. In [ 17 , 18 ] , Vdovina constructed finite polygonal complexes whose links are prescribed bipartite graphs, giving hyperbolic complexes whose universal covers are two-dimensional buildings, and the fundamental groups act simply transitively on the vertices of the universal cover. Kangaslampi and Vdovina [ 13 ] and Carbone, Kangaslampi, and Vdovina [ 3 ] further developed this framework via triangle presentations associated to the smallest generalised quadrangle GQ(2,2) , obtaining and classifying groups acting cocompactly and simply transitively on the corresponding triangular hyperbolic buildings. In analogy with this setting, the interest of the present paper lies in the smallest thin case, associated with the incidence graph K_{3,3} . Recall that a group is said to be large if it has a finite-index subgroup admitting a surjection onto the free group of rank two. From the viewpoint of largeness and SQ-universality, the (3,3) -special presentations case stands apart since no group defined by such a presentation is SQ-universal [ 8 ] , so largeness does not occur in this setting. For all other (m,k) -special presentations with (m,k)\neq(3,3) , largeness becomes a natural and largely unexplored property. This motivates the following open problem posed by Edjvet and Vdovina [ 8 , Problem 2] . Question 1 . Which groups defined by (m,k) -special presentations are large? For the 23 torsion-free groups of Kangaslampi and Vdovina [ 13 ] , SQ-universality follows from [ 8 , Theorem 2(v)] , but largeness remains unknown. Also, despite the infinite families produced in [ 4 ] , no explicit example with \nu=1 , and in particular with m=2 , is known to be both large and hyperbolic. Our first main result provides such a family, thereby providing concrete evidence for to Question 1 . Theorem A . For each integers \alpha\geq 1 and n\geq 2 , let P(n,\alpha)=\langle x_{1},x_{2},\dots,x_{n}\mid w_{n}^{\alpha}\rangle be the one-relator presentation with relator defined recursively by w_{1}=x_{1},\quad w_{n}=w_{n-1}\,x_{n}\,\prod_{j=1}^{n-1}(x_{n}x_{n-j}). Then the star-graph of P(n,\alpha) is isomorphic to the complete bipartite graph K_{n,n} . By a standard curvature argument (see [ 8 , Proof of Theorem 2] and [ 11 ] ), any group defined by an (m,k,\nu) -special presentation satisfying \frac{1}{m}+\frac{2}{k}<1 (1) is hyperbolic. Moreover, by the theorem of Baumslag–Pride [ 2 ] , a group admitting a one-relator presentation on n generators has deficiency n-1 and is therefore large for n\geq 3 . In view of Theorem A , the presentations under consideration are (2,n^{2}) -special, and the inequality ( 1 ) holds for all n\geq 3 . Hence, both hyperbolicity and largeness follow immediately. Corollary B . For every integer n\geq 3 , the presentation P(n,1) is a (2,n^{2}) -special presentation and defines a group that is both large and hyperbolic. A group presentation is said to be redundant if it contains a freely redundant relator , that is, a relator which is freely trivial or freely conjugate to another relator (or to its inverse). A presentation containing no freely redundant relators is called concise , and deleting all freely redundant relators gives a concise refinement [ 6 ] . Next, we obtain a complete classification of concise (m,k) -special presentations (up to Tietze equivalence) whose star-graphs are isomorphic to K_{3,3} . These constitute the smallest special presentations in terms of the number of generators. Theorem C . A group G admits a concise (m,k) -special presentation whose star-graph is K_{3,3} if and only if there exists an integer n\geq 1 such that G has a presentation of the form \langle x,y,z\mid r^{n}\rangle, where r is one of the following words: \displaystyle x^{2}y^{2}z^{2}xzy, \displaystyle\qquad x^{2}yxz^{2}y^{2}z, \displaystyle x^{2}yx^{-1}zyz^{-1}yz, \displaystyle\qquad x^{2}yx^{-1}zyz^{-1}y^{-1}z, \displaystyle x^{2}yx^{-1}zy^{-1}zyz, \displaystyle\qquad x^{2}yx^{-1}zy^{-1}z^{-1}yz, \displaystyle x^{2}yz^{-1}xy^{-1}zyz, \displaystyle\qquad x^{2}yz^{-1}yzyx^{-1}z, \displaystyle x^{2}yz^{-1}yx^{-1}zyz, \displaystyle\qquad x^{2}y^{-1}zxyzyz^{-1}, \displaystyle x^{2}y^{-1}zyz^{-1}xyz, \displaystyle\qquad x^{2}y^{-1}x^{-1}zyz^{-1}yz. Theorem C enumerates all concise (m,k) -special presentations (up to equivalence) with star-graph K_{3,3} , while Corollary B guarantees that the corresponding groups are large and hyperbolic. Among these, the torsion-free cases are characterised as follows. Corollary D . Among the group presentations described in Theorem C , exactly twelve, corresponding to the case n=1 , define torsion-free groups. Corollary D follows immediately from [ 14 , Theorem 3] . 1.1 Outline The paper is organised as follows. Section 2 derives structural constraints on (2,9) -special relators. Section 3 enumerates all admissible relators and reduces them to distinct equivalence classes. Section 4 contains the proofs of the main Theorem A and Theorem C . The Appendix B describes the computational methods, all of which were implemented in SageMath [ 16 ] . 2 Preliminary results In this section, we introduce the basic notions and structural properties that will be used throughout the paper. We begin by recalling the definition of the star-graph associated with a group presentation. We then define (m,k,\nu) -special presentations and establish several elementary restrictions on the possible forms of their star-graphs. In particular, we show that any special presentation must involve at least three generators, and we determine the parameters of those whose star-graph is isomorphic to K_{3,3} . Definition 2 (star-graph) . Let P=\langle X\mid R\rangle be a group presentation. The star-graph of P is the undirected labelled graph \Gamma with vertex set V(\Gamma)=X^{\pm 1}. An edge joins two vertices x_{1},x_{2}\in V(\Gamma) whenever there exists a relator of the form x_{1}x_{2}^{-1}u or x_{2}x_{1}^{-1}u^{-1} in \widetilde{R} , where \widetilde{R} (also known as the symmetrised closure of R ) denotes the set consisting of all elements of R\cup R^{-1} together with all of their cyclic permutations. Definition 3 ( [ 4 , Definition 2.1] ) . Let m\geq 2 , k\geq 3 , and \nu\geq 1 be integers. A finite group presentation P=\langle X\mid R\rangle is said to be (m,k,\nu) –special if the following conditions hold: 1. The star-graph \Gamma of P has \nu isomorphic connected components, each of which is bipartite of diameter m and girth 2m , and in which every vertex has degree at least 3 . 2. Each relator r\in R has length k . 3. If m=2 , then k\geq 4 . We will refer to an (m,k,\nu) –special presentation simply as a special presentation whenever the parameters need not be specified. Lemma 4 . Suppose P=\langle X\mid R\rangle is an (m,k,\nu) -special presentation. Then |X|\geq 3 . Proof. Let \Gamma denote the star-graph of P , and let V(\Gamma) be its vertex set. Then V(\Gamma)=X\cup X^{-1},\quad\text{so }|V(\Gamma)|=2|X|. If |X|\leq 2 , then |V(\Gamma)|\leq 4 . Since \Gamma is bipartite, its vertex set admits a bipartition (A,B) with |B|\leq 2 . Every vertex of \Gamma is adjacent only to vertices in the opposite part of the bipartition, so any vertex in A has degree at most |B|\leq 2 . Hence \Gamma contains a vertex of degree at most 2 . This contradicts the assumption that every vertex has degree at least 3 . Therefore |X|\geq 3 . ∎ If a one–relator presentation P=\langle X\mid r\rangle has star-graph isomorphic to K_{3,3} , then for every n\in\mathbb{N} the presentation \langle X\mid r^{\,n}\rangle also has star-graph K_{3,3} . Indeed, replacing r by a positive power does not change the set of length–two subwords appearing in the symmetrised closure, and hence does not alter the edge set of the star-graph. Also, freely redundant relators contribute no new edges to the star-graph, so replacing a presentation by a concise refinement leaves the star-graph unchanged. Consequently, in classifying presentations whose star-graph is K_{3,3} , we may assume without loss of generality that the relator r is not a proper power and that the presentation is concise. Lemma 5 . Suppose P=\langle X\mid R\rangle is a concise (m,k,\nu) -special presentation whose star-graph is K_{3,3} . Then R=\{r\} , \ell(r)=k=9 , and \nu=1 . Proof. Let P=\langle X\mid R\rangle be a concise (m,k,\nu) –special presentation whose star-graph is K_{3,3} . Then, m=2 and so every relator has length k\geq 4 by Definition 3 . The graph K_{3,3} has exactly 9 edges. In the construction of the star-graph, each relator of length k contributes precisely k edges. Therefore k\cdot|R|\;=\;9. Since k\geq 4 , the only possible factorisation is k=9 and |R|=1 . Hence R=\{r\} with \ell(r)=k=9 . Finally, K_{3,3} is connected, so the star-graph has a single component and hence \nu=1 . ∎ A group G is called Hopfian if every surjective endomorphism of G is an automorphism. Before ending this section, we give a result about one-relator groups which will be useful later. Lemma 6 . Let H_{1}=\langle X\mid r\rangle and H_{2}=\langle Y\mid s\rangle be Hopfian groups with r,s not proper powers. For n\geq 2 , set G_{1}=\langle X\mid r^{n}\rangle and G_{2}=\langle Y\mid s^{n}\rangle. If H_{1}\not\cong H_{2} , then G_{1}\not\cong G_{2}. Proof. Assume for contradiction that H_{1}\not\cong H_{2} but G_{1}\cong G_{2} . Each G_{i} admits a natural epimorphism onto H_{i} by adding the relation r=1 or s=1 , respectively. Transporting the map G_{1}\twoheadrightarrow H_{1} across an isomorphism G_{1}\cong G_{2} gives an epimorphism \psi:G_{2}\twoheadrightarrow H_{1}. In G_{2} , the element s has finite order n , and every torsion element is conjugate to a power of s . Since H_{1} is torsion-free, \psi(s)=1 . Hence \ker(\phi) contains the normal closure of s , and therefore \psi factors through the quotient H_{2}=G_{2}/\!\langle\!\langle s\rangle\!\rangle, giving an epimorphism from H_{2} to H_{1} . By symmetry, we also obtain an epimorphism from H_{1} to H_{2} . Hence, H_{1} and H_{2} admit mutual epimorphisms. Since H_{1} and H_{2} are Hopfian, the compositions H_{1}\twoheadrightarrow H_{1},\quad H_{2}\twoheadrightarrow H_{2} are automorphisms. Hence H_{1}\cong H_{2} , contradicting the hypothesis. Therefore G_{1}\not\cong G_{2} . ∎ 3 Relators for (2,9) -special presentations Let F(x,y,z) denote the free group of rank 3 on the generators x,y,z . In this section, we classify all cyclically reduced words of length 9 in F(x,y,z) that define (2,9) -special presentations whose star-graphs are isomorphic to K_{3,3} . By Lemma 5 , each such presentation is determined by a cyclically-reduced relator of length 9 . Our objective is to enumerate all admissible relators and then refine this enumeration to obtain the isomorphism classes of the corresponding one-relator groups. The classification proceeds in two steps. First, we impose the combinatorial constraints coming from the K_{3,3} star-graph condition and use a computer search (Appendix B ) to generate all admissible relators. This gives a finite list of words, which is then reduced modulo cyclic permutation, re-labelling and inversion. Lemma 7 . The set of admissible relators R decomposes as R=\bigcup_{i=1}^{12}R_{i}, where each R_{i} is given explicitly below. \displaystyle R_{1} \displaystyle=\{x^{2}yxzy^{2}z^{2},\;x^{2}yzxz^{2}y^{2},\;x^{2}y^{2}xzyz^{2},\;x^{2}y^{2}zyxz^{2},\;x^{2}y^{2}z^{2}yxz,\;x^{2}y^{2}z^{2}xzy\}, \displaystyle R_{2} \displaystyle=\{x^{2}yzy^{2}xz^{2},\;x^{2}yz^{2}xzy^{2},\;x^{2}yz^{2}y^{2}xz,\;x^{2}y^{2}xz^{2}yz,\;x^{2}y^{2}zxz^{2}y,\;x^{2}yxz^{2}y^{2}z\}, \displaystyle R_{3} \displaystyle=\{x^{2}yx^{-1}zyz^{-1}yz,\;x^{2}yzy^{-1}zyx^{-1}z\},R_{4}=\{x^{2}yx^{-1}zyz^{-1}y^{-1}z,\;x^{2}yz^{-1}y^{-1}zyx^{-1}z\}, \displaystyle R_{5} \displaystyle=\{x^{2}yx^{-1}zy^{-1}zyz,\;x^{2}yzyz^{-1}yx^{-1}z\},R_{6}=\{x^{2}yx^{-1}zy^{-1}z^{-1}yz,\;x^{2}yzy^{-1}z^{-1}yx^{-1}z\}, \displaystyle R_{7} \displaystyle=\{x^{2}yz^{-1}xy^{-1}zyz,\;x^{2}yzyz^{-1}xy^{-1}z\},R_{8}=\{x^{2}yz^{-1}yzyx^{-1}z,\;x^{2}yx^{-1}zyzy^{-1}z\}, \displaystyle R_{9} \displaystyle=\{x^{2}yz^{-1}yx^{-1}zyz,\;x^{2}yzyx^{-1}zy^{-1}z\},R_{10}=\{x^{2}y^{-1}zxyzyz^{-1},\;x^{2}y^{-1}zyzxyz^{-1}\}, \displaystyle R_{11} \displaystyle=\{x^{2}y^{-1}zyz^{-1}xyz,\;x^{2}yzxy^{-1}zyz^{-1}\},R_{12}=\{x^{2}y^{-1}x^{-1}zyz^{-1}yz,\;x^{2}yzy^{-1}zyx^{-1}z^{-1}\}. The proof is computational and uses structural constraints imposed by the K_{3,3} star-graph condition. Each admissible relator must contain each of x^{\pm 1},y^{\pm 1},z^{\pm 1} exactly three times. By replacing a generator t\in\{x,y,z\} with t^{-1} if necessary, we may further assume that the number of occurrences of t is at least the number of occurrences of t^{-1} . Note that the K_{3,3} condition forces the occurrence of t^{2} , and not t^{3} , as a subword for some t\in\{x,y,z\} . We normalise by cyclic permutation, re-labelling, and inversion so that each word begins with x^{2}y^{\pm 1} . Indeed, after fixing an initial segment x^{2} , the next letter cannot be x^{\pm 1} without violating cyclic reduction of K_{3,3} condition, and hence must be either y^{\pm 1} or z^{\pm 1} . In the latter case, we apply the automorphism of the free group that swaps y and z , reducing to the former situation. The Sage code in Appendix B enumerates all cyclically reduced words of length 9 that satisfy these constraints, and then selects those whose star-graphs have girth 4 and diameter 2 . The output consists of 32 words, which form the 12 equivalence classes of Lemma 7 . To obtain a complete classification, we next introduce additional automorphisms of F(x,y,z) . For each t\in\{x,y,z\} , let \phi_{t} be the automorphism fixing t and swapping the other two generators, and let \rho_{t} be the automorphism inverting t while fixing the remaining generators. These automorphisms, together with cyclic permutation and inversion, generate all identifications needed to relate relators within each class R_{i} . Lemma 8 . Let R=\bigcup_{i=1}^{12}R_{i} be the decomposition given in Lemma 7 . For each i=1,\dots,12 and any w,w^{\prime}\in R_{i} , the one-relator groups defined by \langle x,y,z\mid w\rangle and \langle x,y,z\mid w^{\prime}\rangle are isomorphic via compositions of the automorphisms \phi_{t} , \rho_{t} , together with cyclic permutation and inversion. Proof. For each class R_{i} , Table 1 in the Appendix gives an explicit composition of automorphisms, inversion, and cyclic permutation sending any relator in R_{i} to a chosen representative w_{0} . This shows that all presentations within a given class define isomorphic one-relator groups. ∎ 4 Main In this section, we prove Theorem A and Theorem C . We begin with the proof of Theorem A , which constructs an explicit family of one-relator presentations and determines their star-graphs. Proof of Theorem A . For integers \alpha\geq 1 and n\geq 2 consider the one–relator presentation P(n,\alpha)=\langle x_{1},\dots,x_{n}\mid w_{n}^{\alpha}\rangle,\quad w_{1}=x_{1},\quad w_{n}=w_{n-1}\,x_{n}\prod_{j=1}^{n-1}(x_{n}x_{n-j})\ (n\geq 2). We show that the star-graph of P_{n} is the complete bipartite graph K_{n,n} . Recall that for a cyclically reduced relator w , each cyclic subword ab contributes an oriented edge recorded as (a,b^{-1}) . Since w_{n} is a positive word, all such pairs are of the form (x_{i},x_{j}^{-1}) . Hence, it suffices to prove that the cyclic adjacencies in w_{n} realise every ordered pair (x_{i},x_{j}^{-1}) with 1\leq i,j\leq n exactly once. The argument proceeds by induction. For n=2 we have w_{2}=x_{1}x_{2}x_{2}x_{1} , whose cyclic adjacencies give precisely the four pairs (x_{i},x_{j}^{-1}) with i,j\in\{1,2\} . Assume the statement holds for n-1\geq 2 . Write w_{n}=w_{n-1}\,T,\quad T=x_{n}\,(x_{n}x_{n-1})\,(x_{n}x_{n-2})\cdots(x_{n}x_{1}). The cyclic adjacencies contributed by w_{n-1} realise all pairs (x_{i},x_{j}^{-1}) with 1\leq i,j\leq n-1 . The new adjacencies arising from T and from the boundary between w_{n-1} and T contributes exactly the pairs (x_{n},x_{n}^{-1}),\quad(x_{n},x_{j}^{-1}),\quad(x_{j},x_{n}^{-1})\quad(1\leq j\leq n-1). These are disjoint from the previous (n-1)^{2} pairs and account for 2n-1 new pairs. Hence, the total number of distinct pairs realised by w_{n} is (n-1)^{2}+(2n-1)=n^{2}, which equals the total number of ordered pairs (x_{i},x_{j}^{-1}) with 1\leq i,j\leq n . Therefore the star-graph of P(n,\alpha) is K_{n,n} . ∎ Finally, we turn to Theorem C , where the goal is to distinguish the resulting groups up to isomorphism. This requires a finer invariant: for a finitely presented group, the collection of abelianisations of its finite-index subgroups is an isomorphism invariant. We use this computationally, via SageMath , to separate the groups arising in our classification. Proof of Theorem C . Let G_{i}=\langle x,y,z\mid r_{i}\rangle for 1\leq i\leq 12 , where r_{i}\in R_{i} as given in Lemma 7 . By Lemma 8 , the isomorphism type of G_{i} depends only on the equivalence class R_{i} and not on the particular choice of relator r_{i}\in R_{i} . To prove that these groups are pairwise non–isomorphic, we compare the abelianisations \mathrm{Ab}(X)=X/[X,X] of low–index subgroups. If G\cong H , then for each k the multisets \bigl\{\,\mathrm{Ab}(U)\mid U\leq G,\ [G:U]=k\,\bigr\}\quad\text{and}\quad\bigl\{\,\mathrm{Ab}(V)\mid V\leq H,\ [H:V]=k\,\bigr\} coincide. Hence, to show G_{i}\not\cong G_{j} , it suffices to find an index k such that one of these multisets differs. All computations were carried out in SageMath using low–index subgroup enumeration. We record the separating invariants used in each case. The group G_{1} . Among the index– 3 subgroups of G_{1} , there is one with abelianisation \mathbb{Z}^{4}\oplus\mathbb{Z}_{9} , which does not occur for any index– 3 subgroup of G_{j} with j>1 . Hence G_{1}\not\cong G_{j} for all j>1 . The group G_{2} . An index– 3 subgroup of G_{2} has abelianisation \mathbb{Z}^{4}\oplus\mathbb{Z}_{3}^{2} , which does not occur for any index– 3 subgroup of G_{j} with j>2 . Hence G_{2}\not\cong G_{j} for all j>2 . The group G_{3} . Among index– 3 subgroups, G_{3} has one with abelianisation \mathbb{Z}^{4}\oplus\mathbb{Z}_{2}^{2} , which does not occur for G_{j} with j\in\{4,5,6,7,8,10,11\} . At index 5 , the group G_{9} has a subgroup with abelianisation \mathbb{Z}^{6}\oplus\mathbb{Z}_{11} , which does not occur for G_{3} , so G_{3}\not\cong G_{9} . Finally, G_{12} has two index– 5 subgroups with abelianisation \mathbb{Z}^{6}\oplus\mathbb{Z}_{2} , whereas G_{3} has only one. Hence G_{3} is distinct from all G_{j} with j>3 . The group G_{4} . Among index– 3 subgroups, G_{4} has one with abelianisation \mathbb{Z}^{4}\oplus\mathbb{Z}_{7} , which does not occur for G_{j} with j\in\{5,6,7,8,9,10,12\} . Another index– 3 subgroup has abelianisation \mathbb{Z}^{4}\oplus\mathbb{Z}_{2} , which does not occur for G_{11} . Hence G_{4}\not\cong G_{j} for all j>4 . The group G_{5} . An index– 3 subgroup has abelianisation \mathbb{Z}^{5} , which does not occur for G_{j} with j\in\{6,7,9,10,12\} . At index 4 , G_{5} has a subgroup with abelianisation \mathbb{Z}^{5}\oplus\mathbb{Z}_{5} , which does not occur for G_{8} . Moreover, G_{11} has an index– 3 subgroup with abelianisation \mathbb{Z}^{4}\oplus\mathbb{Z}_{7} , which does not occur for G_{5} . Hence G_{5}\not\cong G_{j} for all j>5 . The group G_{6} . An index– 3 subgroup has abelianisation \mathbb{Z}^{4}\oplus\mathbb{Z}_{2} , which does not occur for G_{j} with j\in\{8,9,11,12\} . An index– 4 subgroup has abelianisation \mathbb{Z}^{5}\oplus\mathbb{Z}_{5} , which distinguishes G_{6} from G_{7} and G_{10} . Hence G_{6}\not\cong G_{j} for all j>6 . The group G_{7} . An index– 3 subgroup has abelianisation \mathbb{Z}^{4}\oplus\mathbb{Z}_{2} , which does not occur for G_{j} with j\in\{8,9,11,12\} . At index 5 , G_{7} has three subgroups with abelianisation \mathbb{Z}^{6}\oplus\mathbb{Z}_{2} , whereas G_{10} has only two. Hence G_{7}\not\cong G_{j} for all j>7 . The group G_{8} . An index– 3 subgroup has abelianisation \mathbb{Z}^{5} , which does not occur for G_{j} with j\in\{9,10,12\} . Moreover, G_{11} has an index– 3 subgroup with abelianisation \mathbb{Z}^{4}\oplus\mathbb{Z}_{7} , which does not occur for G_{8} . Hence G_{8}\not\cong G_{j} for all j>8 . The group G_{9} . An index– 3 subgroup has abelianisation \mathbb{Z}^{4}\oplus\mathbb{Z}_{2}^{2} , which distinguishes G_{9} from G_{10} and G_{11} . At index 5 , G_{9} has a subgroup with abelianisation \mathbb{Z}^{6}\oplus\mathbb{Z}_{11} , which does not occur for G_{12} . Hence G_{9}\not\cong G_{j} for all j>9 . The group G_{10} . An index– 3 subgroup has abelianisation \mathbb{Z}^{4}\oplus\mathbb{Z}_{2} , which distinguishes G_{10} from G_{11} and G_{12} . Hence G_{10}\not\cong G_{j} for all j>10 . The group G_{11} . An index– 3 subgroup has abelianisation \mathbb{Z}^{4}\oplus\mathbb{Z}_{7} , which does not occur for G_{12} . Hence G_{11}\not\cong G_{12} . Since every pair G_{i},G_{j} with i\neq j is separated by at least one low–index abelianisation invariant, the twelve groups are pairwise non–isomorphic. The rest of the proof follows from Lemma 6 since the groups G_{i} are hyperbolic, hence Hopfian by [ 10 , Corollary 2.9] (see also [ 19 , 15 ] ). ∎ Acknowledgments Much of the content of this article is based on the first author’s thesis completed at the African Institute for Mathematical Sciences (AIMS) in Ghana (see [ 1 ] ). The second author gratefully acknowledges the support of the London Mathematical Society through a Scheme 5 “Collaborations with Developing Countries” grant (Reference: 52107), which enabled travel to AIMS in 2023 for further discussions and collaboration on this work. Appendix A Automorphisms identifying relators R_{i} Word w Composition to w_{0} R_{1} x^{2}y^{2}z^{2}xzy w_{0} x^{2}yxzy^{2}z^{2} \phi_{x}\circ\phi_{y} , cyclic permutation x^{2}y^{2}xzyz^{2} \rho_{x}\circ\rho_{y}\circ\rho_{z} , invert, \phi_{z} , cyclic permutation x^{2}y^{2}zyxz^{2} \phi_{x}\circ\phi_{z} , cyclic permutation x^{2}y^{2}z^{2}yxz \phi_{y}\circ\rho_{x}\circ\rho_{y}\circ\rho_{z} , cyclic permutation R_{2} x^{2}yxz^{2}y^{2}z w_{0} x^{2}yzy^{2}xz^{2} \phi_{x}\circ\phi_{y}\circ\rho_{x}\circ\rho_{y}\circ\rho_{z} , cyclic permutation x^{2}yz^{2}xzy^{2} \phi_{x}\circ\phi_{z} , cyclic permutation x^{2}yz^{2}y^{2}xz \rho_{x}\circ\rho_{y}\circ\rho_{z} , invert, \phi_{x} , cyclic permutation x^{2}y^{2}xz^{2}yz \phi_{y} , cyclic permutation x^{2}y^{2}zxz^{2}y \phi_{x}\circ\phi_{z}\circ\rho_{x}\circ\rho_{y}\circ\rho_{z} , cyclic permutation R_{3} x^{2}yx^{-1}zyz^{-1}yz w_{0} x^{2}yzy^{-1}zyx^{-1}z \phi_{x}\circ\rho_{z}\circ\rho_{y}\circ\rho_{x} , invert, cyclic permutation R_{4} x^{2}yx^{-1}zyz^{-1}y^{-1}z w_{0} x^{2}yz^{-1}y^{-1}zyx^{-1}z \phi_{x}\circ\rho_{z}\circ\rho_{y}\circ\rho_{x} , invert, cyclic permutation R_{5} x^{2}yx^{-1}zy^{-1}zyz w_{0} x^{2}yzyz^{-1}yx^{-1}z \phi_{x}\circ\rho_{z}\circ\rho_{y}\circ\rho_{x} , invert, cyclic permutation R_{6} x^{2}yx^{-1}zy^{-1}z^{-1}yz w_{0} x^{2}yzy^{-1}z^{-1}yx^{-1}z \phi_{x}\circ\rho_{z}\circ\rho_{y}\circ\rho_{x} , invert, cyclic permutation R_{7} x^{2}yz^{-1}xy^{-1}zyz w_{0} x^{2}yzyz^{-1}xy^{-1}z \phi_{x}\circ\rho_{z}\circ\rho_{y}\circ\rho_{x} , invert, cyclic permutation R_{8} x^{2}yz^{-1}yzyx^{-1}z w_{0} x^{2}yx^{-1}zyzy^{-1}z \phi_{x}\circ\rho_{z}\circ\rho_{y}\circ\rho_{x} , invert, cyclic permutation R_{9} x^{2}yz^{-1}yx^{-1}zyz w_{0} x^{2}yzyx^{-1}zy^{-1}z \phi_{x}\circ\rho_{z}\circ\rho_{y}\circ\rho_{x} , invert, cyclic permutation R_{10} x^{2}y^{-1}zxyzyz^{-1} w_{0} x^{2}y^{-1}zyzxyz^{-1} \phi_{x}\circ\rho_{z}\circ\rho_{y}\circ\rho_{x} , invert, cyclic permutation R_{11} x^{2}y^{-1}zyz^{-1}xyz w_{0} x^{2}yzxy^{-1}zyz^{-1} \phi_{x}\circ\rho_{z}\circ\rho_{y}\circ\rho_{x} , invert, cyclic permutation R_{12} x^{2}y^{-1}x^{-1}zyz^{-1}yz w_{0} x^{2}yzy^{-1}zyx^{-1}z^{-1} \phi_{x}\circ\rho_{z}\circ\rho_{y}\circ\rho_{x} , invert, cyclic permutation Table 1: Automorphisms identifying the relators in each class R_{i} with the representative w_{0} . We use the convention that (f\circ g)(x)=g(f(x)) . Appendix B Computational enumeration of (2,9) -special relators ⬇ 1 from sage . graphs . graph import Graph 2 #Generates all words of length ‘l‘ from a given alphabet. 3 def words ( alphabet , l ): 4 if l == 0: 5 give [] 6 else : 7 for word in words ( alphabet , l - 1): 8 for letter in alphabet : 9 give word + [ letter ] 10 11 #Constructs a star-graph for a given word ‘g‘. 12 def stargraph ( g ): 13 G = Graph ( loops = True ) 14 G . add_vertices ([ ’x’ , ’y’ , ’z’ , ’X’ , ’Y’ , ’Z’ ]) 15 for i in range ( len ( g )): 16 prev = g [ i -1] 17 cur = g [ i ] 18 if cur . isupper (): 19 G . add_edge (( prev , cur . lower ())) 20 else : 21 G . add_edge (( prev , cur . upper ())) 22 return G 23 24 #Generates words of length 9 from the alphabet {x,y,z,X,Y,Z} & filters them 25 alphabet = [ ’x’ , ’y’ , ’z’ , ’X’ , ’Y’ , ’Z’ ] 26 word_length = 9 27 for word in words ( alphabet , word_length ): 28 if not ( word [0:3] == [ ’x’ , ’x’ , ’y’ ] or word [0:3] == [ ’x’ , ’x’ , ’Y’ ]): 29 continue 30 if not ( word . count ( ’x’ ) + word . count ( ’X’ ) == 3 and 31 word . count ( ’y’ ) + word . count ( ’Y’ ) == 3 and 32 word . count ( ’z’ ) + word . count ( ’Z’ ) == 3 and 33 word . count ( ’y’ ) > word . count ( ’Y’ ) and 34 word . count ( ’z’ ) > word . count ( ’Z’ )): 35 continue 36 SG = stargraph ( word ) 37 if SG . girth () == 4 and SG . diameter () == 2: 38 print ( ’’ . join ( word )) 39 40 #Running the code above gives this list of 32 words 41 words_list = [ 42 "xxyxzyyzz" , "xxyxzzyyz" , "xxyyxzyzz" , "xxyyxzzyz" , "xxyyzxzzy" , 43 "xxyyzyxzz" , "xxyyzzxzy" , "xxyyzzyxz" , "xxyzxzzyy" , "xxyzxYzyZ" , 44 "xxyzyyxzz" , "xxyzyXzYz" , "xxyzyZxYz" , "xxyzyZyXz" , "xxyzzxzyy" , 45 "xxyzzyyxz" , "xxyzYzyXz" , "xxyzYzyXZ" , "xxyzYZyXz" , "xxyXzyzYz" , 46 "xxyXzyZyz" , "xxyXzyZYz" , "xxyXzYzyz" , "xxyXzYZyz" , "xxyZxYzyz" , 47 "xxyZyzyXz" , "xxyZyXzyz" , "xxyZYzyXz" , "xxYzxyzyZ" , "xxYzyzxyZ" , 48 "xxYzyZxyz" , "xxYXzyZyz" 49 ]