betaFeedback

WorksMathematics

Flatness and Tightness in Hypertopes

Abstract

A hypertope is a thin, residually connected incidence geometry. For each pair of types {i,j}\{i,j\}, the residues of that type must be polygons, and if all residues of that type are the same type of polygon, then we can record this information in a diagram DD. Given such a diagram, we prove a lower bound on the number of chambers of a hypertope with that diagram. A hypertope that achieves this bound is called tight. We prove that a hypertope is tight if and only if its diagram DD is a forest and, for every pair of types {i,j}\{i,j\} that is not an edge of DD, every ii-element is incident to every jj-element. This generalizes a previously known result about abstract polytopes. Finally, we describe a combinatorial method and an algebraic method for constructing tight hypertopes.

Introduction

The theory of abstract polytopes is rich and by now well-established. It combines elements of geometry, combinatorics and group theory. Abstract polytopes are usually defined as posets satisfying a set of defining properties inspired by those of convex polytopes. From the point of view of incidence geometry, an abstract polytope is defined as a thin, residually connected geometry with a linear Buekenhout diagram [1]. The linearity of the diagram induces a partial order on the elements and allows for the use of many geometrical tools. Nonetheless, much of the underlying combinatorics and group theory remains valid regardless of the diagram. Hypertopes, first introduced in [9], are the objects obtained by dropping the linearity condition on the diagram of a polytope. They are thus thin, residually connected geometries.

A hypertope is called regular when it has the highest degree of symmetry, meaning that its automorphism group acts transitively on its chambers. It turns out that regular hypertopes can always be constructed from smooth quotients of Coxeter groups. From this point of view, allowing nonlinear diagrams is a natural step. It brings into consideration geometries associated with arbitrary Coxeter groups, allowing for new behaviours and symmetries while preserving a considerable amount of the structure of polytopes. This broader setting also connects the theory to other well-known objects,

E-mail addresses: [email protected], [email protected], [email protected].

Date: November 20, 2026. such as hypermaps and Coxeter complexes. Moreover, nonlinear diagrams allow for other interactions between the different types of elements, and in particular for symmetry phenomena that cannot occur in the linear setting, such as trialities or higher order correlations [11, 12].

Even when focusing on polytopes, it can be beneficial to distinguish between properties that are intrinsically related to the linearity of the diagram, and thus to the natural partial order on the faces, and those that are more combinatorial or group-theoretical in nature. We therefore believe that studying hypertopes can help deepen our understanding of thin, residually connected geometries and thus, in particular, of abstract polytopes.

As with hypertopes, the study of abstract polytopes is primarily concentrated on those that are regular or otherwise highly symmetric. A natural way to remove this symmetry requirement while still insisting on some degree of homogeneity is to study equivelar polytopes, which are precisely polytopes that have a Buekenhout diagram where every edge {i,j}\{i,j\} is labeled with a single number pijp_{ij}. In other words, these are the polytopes such that residues of each type {i,j}\{i,j\} (i≠ji \ne j) are pijp_{ij}-gons. Once we fix a diagram, a simple counting argument establishes a lower bound on the number of chambers ([3 Proposition 3.3]). A polytope that achieves this lower bound is called tight. Several papers have explored the structure of tight polytopes, whether in the presence of high symmetry [2, 4, 7, 8] or without it [3]. There are two important general results. The first is that the diagram of a tight polytope cannot have two odd labels in a row [3 Proposition 3.9]. The second is that an nn-polytope is tight if and only if it is (i,i+2)(i,i+2)-flat for each 0≤i≤n−30 \le i \le n-3 (that is, every ii-element is incident to every (i+2)(i+2)-element) [3 Theorem 4.4].

In this paper, we seek to extend the machinery from [3] to hypertopes. First, we give a lower bound on the number of chambers of a hypertope with a given diagram DD. A hypertope that meets this lower bound is called tight. We investigate two questions:

  1. If a hypertope is tight, what restrictions does that put on its diagram DD?

  2. Is there a relation between tightness and flatness, analogous to the situation with polytopes?

This culminates in Theorem 4.14, where we find that a hypertope is tight if and only if its diagram DD is a forest and, for every non-edge {i,j}\{i,j\} of DD, every ii-element is incident to every jj-element.

We also describe two constructions for building tight hypertopes. The first is combinatorial, adapting the techniques of [5]. Indeed, the construction more generally can be used to build “small” extensions of an existing hypertope, and the properties of this extension are collected in Theorem 5.7. The second construction is algebraic and uses the techniques described in [15] to build tight regular hypertopes.

We start by giving background on incidence systems and hypertopes in § 2. Then we generalize the notion of flatness from polytopes to incidence systems in § 3. In § 4, we establish a lower bound on the size of a hypertope with a given diagram, and we derive several structural results on those hypertopes that meet the lower bound (tight hypertopes). Finally, we describe constructions for flat and tight hypertopes in § 5 and § 6.

Background

Incidence systems

Incidence systems. Incidence systems are a mathematical concept that encodes a great many of the geometrical and combinatorial structures that one may encounter in mathematics. We give here a brief account of the main definitions and terminology of incidence systems. For a more detailed introduction to incidence geometry, we refer to [1].

A triple Γ=(X,∗,t)\Gamma=(X,*,t) is called an incidence system over II if

  • XX is a set whose elements are called the elements of Γ\Gamma,

  • ∗* is a symmetric and reflexive relation (called the incidence relation) on XX, and

  • tt is a map from XX to II, called the type map of Γ\Gamma, such that distinct elements x,y∈Xx,y \in X with x∗yx*y satisfy t(x)≠t(y)t(x) \ne t(y).

Elements of t−1(i)t^{-1}(i) are called the elements of type ii. The rank of Γ\Gamma, denoted as rank⁡(Γ)\operatorname{rank}(\Gamma), is the cardinality of the type set II, which we will assume is finite for the remainder of the paper. A flag in an incidence system Γ\Gamma over II is a set of pairwise incident elements. The type of a flag FF is t(F)t(F), that is the set of types of the elements of FF. The rank of a flag FF is the cardinality of t(F)t(F), while its corank is rank(Γ)−∣t(F)∣(\Gamma) - |t(F)|. A chamber is a flag of type II. An incidence system Γ\Gamma is an incidence geometry if all its maximal flags are chambers. We will denote the set of chambers of Γ\Gamma by Ch⁡(Γ)\operatorname{Ch}(\Gamma).

Let FF be a flag of Γ\Gamma. An element x∈Xx \in X is incident to FF if x∗yx * y for all y∈Fy \in F. The residue of Γ\Gamma with respect to FF, denoted by ΓF\Gamma_F, is the incidence system formed by all the elements of Γ\Gamma incident to FF but not in FF. The rank of the residue ΓF\Gamma_F is equal to rank⁡(Γ)−∣t(F)∣\operatorname{rank}(\Gamma) - |t(F)|, the corank of FF. The type of a residue ΓF\Gamma_F is the set I∖t(F)I \setminus t(F); equivalently, it is the cotype of FF. Thus, a residue of type JJ is the residue of a flag of cotype JJ. Said another way, the type of a residue is the set of types that remains.

The incidence graph of Γ\Gamma is a graph with vertex set XX and where two elements xx and yy of different types are connected by an edge if and only if x∗yx * y. An incidence geometry Γ\Gamma is connected if its incidence graph is connected. It is residually connected if all its residues of rank at least two are connected. An incidence geometry is thin if all its residues of flags of corank one contain exactly two elements.

A thin residually connected incidence geometry is called a hypertope. Unless otherwise stated, we will assume that our hypertopes are finite.

An automorphism of an incidence geometry Γ\Gamma is a bijection ϕ\phi of XX respecting the incidence relation ∗* and such that t(ϕ(x))=t(x)t(\phi(x)) = t(x) for all x∈Xx \in X. The group of all automorphisms of Γ\Gamma is denoted by Aut⁡(Γ)\operatorname{Aut}(\Gamma).

The incidence geometry Γ\Gamma is said to be flag-transitive if the action of Aut⁡(Γ)\operatorname{Aut}(\Gamma) on the set of flags of each type is transitive. It is said to be incidence-transitive if, for each i,j∈Ii,j \in I, the action of Aut⁡(Γ)\operatorname{Aut}(\Gamma) is transitive on pairs (x,y)(x,y) such that t(x)=it(x) = i, t(y)=jt(y) = j, and x∗yx * y. A flag-transitive hypertope is also called regular.

Coset incidence systems

Highly symmetric incidence geometries can be constructed from a group GG together with a set (Gi)i∈I(G_i)_{i \in I} of subgroups of GG as described in [16].

A coset incidence system Γ(G,(Gi)i∈I)\Gamma(G,(G_i)_{i \in I}), denoted also as (G,(Gi)i∈I)(G,(G_i)_{i \in I}), is the incidence system over the type set II where:

  1. The elements of type i∈Ii \in I are left cosets of the form gGigG_i, g∈Gg \in G.

  2. The incidence relation is given by non-empty intersection. More precisely, the element gGigG_i is incident to the element kGjkG_j if and only if gGi∩kGj≠∅gG_i \cap kG_j \ne\varnothing.

We usually refer to the subgroups GiG_i as the (standard) maximal parabolic subgroups. For J⊆IJ \subseteq I, we define (standard) parabolic subgroups as GJ=⋂i∈JGiG_J = \bigcap_{i \in J} G_i, and define G∅=GG_{\varnothing} = G. When J=I∖{i}J = I \setminus\{i\}, we say that GJ=GI∖{i}G_J = G_{I \setminus\{i\}} are the minimal parabolic subgroups, and we will denote them as GiG^i. Finally, the subgroup GI=⋂i∈IGiG_I = \bigcap_{i \in I} G_i is called the Borel subgroup of Γ\Gamma.

Any incidence system that contains a chamber and is incidence-transitive can be constructed as a coset incidence system.

Lemma 2.1. [1 Lemma 1.8.2] Let Γ\Gamma be an incidence system such that G=Aut⁡(Γ)G = \operatorname{Aut}(\Gamma) acts incidence-transitively on Γ\Gamma. Let {xi∣i∈I}\{x_i \mid i \in I\} be a chamber of Γ\Gamma. Then, we have Γ≅(G,(Gi)i∈I)\Gamma\cong(G,(G_i)_{i \in I}) where GiG_i is the stabilizer of xix_i in GG.

Chamber systems

The concept of chamber system was invented by Tits in [17]. A chamber system over II is a pair C=(C,(∼i,i∈I))\mathcal{C} = (C,(\sim_i,i \in I)) consisting of a set CC, whose members are called chambers, and a collection of equivalence relations ∼i\sim_i on CC, indexed by i∈Ii \in I. Two chambers cc and dd are called ii-adjacent if c∼idc \sim_i d. For i∈Ii \in I, each ∼i\sim_i-equivalence class is called an ii-panel. The rank of C\mathcal{C} is ∣I∣|I|. The chamber system C\mathcal{C} is called thin if every ii-panel is of size exactly 2.

For J⊆IJ \subseteq I, we denote by ∼J\sim_J the union of all ∼j\sim_j with j∈Jj \in J. A connected component of the graph of (C,∼J)(\mathcal{C},\sim_J) is called a JJ-cell of C\mathcal{C}. For i∈Ii \in I, the (I∖{i})(I \setminus\{i\})-cells are called ii-elements of C\mathcal{C} (called ii-objects in [1 Definition 3.2.2]). In this article, we will only deal with thin chamber systems. In that case, the chamber system C\mathcal{C} over I={0,1,⋯ ,n−1}I = \{0,1,\cdots,n-1\} can be thought of as an nn-valent graph with a proper nn-coloring of its edges. The vertices of this graph correspond to the chambers of C\mathcal{C} and two distinct chambers are connected by an ii-colored edge if and only if they are ii-adjacent. Whenever C\mathcal{C} is thin, it guarantees that there is exactly one edge of each color at each vertex. In this context, a JJ-cell of C\mathcal{C} is a connected component of the graph obtained by deleting all edges of color in I−JI-J.

A chamber system CC over II is residually connected if, for every subset JJ of II and every system of jj-elements ZjZ_j, one for each j∈Jj \in J, with the property that any two have a non-empty intersection, it follows that ⋂j∈JZj\bigcap_{j\in J} Z_j is an (I∖J)(I \setminus J)-cell.

It is always possible to construct an incidence system from a chamber system. Beware that in general, information can be lost in the process. If CC is a chamber system, the incidence system of CC, denoted by Γ(C)\Gamma(C), is the incidence system over II determined as follows. Its ii-elements, for i∈Ii \in I are the pairs (x,i)(x, i) with xx an ii-element of CC; two elements (x,k)(x, k) and (y,l)(y, l) of Γ(C)\Gamma(C) are incident if and only if x∩y≠∅x \cap y \ne\emptyset in CC, i.e., xx and yy have a chamber in common. (The pairs contain the type ii because it may happen that for some i≠ji \ne j, an ii-element and a jj-element consist of the same chambers, and we need to distinguish the corresponding elements in the incidence system.)

The next theorem is a simplified version of Theorem 3.4.6 in [1]. Please refer to the original textbook for the more precise statement, which requires a bit more background that what is given in this section.

Theorem 2.2. Let G(I)\mathcal{G}(I) be the collection of all residually connected geometries over II and let C(I)\mathcal{C}(I) be the collection of all residually connected chamber systems over II. There is a bijective homomorphism preserving structures between the collections G(I)\mathcal{G}(I) and C(I)\mathcal{C}(I). Thus, each residually connected geometry over II corresponds to a unique residually connected chamber system over II (up to isomorphism), with the same automorphism group, and vice versa.

Therefore, when considering only residually connected incidence geometries and residually connected chamber system, one can freely go back and forth between the two view points.

Diagrams

Francis Buekenhout introduced a diagram associated to incidence geometries [1]. The main idea is that local information, given by rank two residues, often imposes global conditions on the whole geometry. The definition of diagram holds for any incidence geometry. Here, we will give the definition in the context of hypertopes only, for simplicity.

Let H\mathcal{H} be a hypertope, that is, a residually connected and thin incidence geometry. We have seen in § 2.3 that we can also consider H\mathcal{H} as a chamber system. Let II be the type set of H\mathcal{H}. For any i≠j∈Ii \ne j \in I, since H\mathcal{H} is thin, any rank two residue of type {i,j}\{i,j\} of H\mathcal{H} must be the geometry of an nn-gon, for some integer n≥2n \ge2. From the point of view of thin chamber systems, the graph obtained from the chamber graph of H\mathcal{H} by deleting all edges of color not equal to ii or jj must be a cyclic graph of length 2n2n with edges of alternating colors. Of course, the integer nn usually depends on the rank two residue we chose, and not only on the types ii and jj. The integer nn is called the parameter of the residue.

In this paper, a diagram is a labeled graph DD whose vertex set is II and whose edges are labeled by integers strictly greater than two. By convention, a missing edge in DD should be understood as an edge with label 2. We say that H\mathcal{H} belongs to DD, or that H\mathcal{H} has diagram DD, if, for every i≠j∈Ii \ne j \in I, all rank two residues of type {i,j}\{i,j\} of H\mathcal{H} have the same parameter pijp_{ij} (that is, the residues are pijp_{ij}-gons), and that this parameter is exactly the label of the edge {i,j}\{i,j\} of DD.1

Note that a hypertope belonging to a linear diagram is an abstract polytope (see [18]). In fact, it is an equivelar polytope, which is to say that it has a Schläfli symbol. Thus, we may think of hypertopes that belong to a diagram DD as being a generalization of equivelar abstract polytopes.

Diagrams for more general incidence geometries are more detailed, because the rank two residues are much more varied. In this paper, our results that involve diagrams of incidence geometries only need information about the underlying graph, so will not worry about these additional parameters.

The diagram of a hypertope provides an easy lower bound on the number of ii-elements for each ii:

Proposition 2.3. In a hypertope with diagram DD, the number of jj-elements is at least max⁡j≠ipij\max_{j\ne i} p_{ij}.

Proof. In order for the residue of type {i,j}\{i,j\} to be a pijp_{ij}-gon, there must be at least pijp_{ij} ii-elements. □\square

In other words, if we look at node ii in the diagram, then there are at least as many ii-elements as the largest label among all edges incident to ii. Recall that if there is no edge between ii and jj in the diagram, we consider that the same as there being an edge of label 2, so that pij=2p_{ij}=2 in this case.

Note that if H\mathcal{H} is regular, it is incidence-transitive, and thus can be constructed as a coset geometry (G,(Gi)i∈I)(G,(G_i)_{i\in I}) (see Lemma 2.1). The thinness of H\mathcal{H}, together with residual-connectedness, forces the minimal parabolic subgroups Gi=⋂j≠iGjG^i=\bigcap_{j\ne i}G_j to be cyclic groups of order two. Hence Gi=⟨ρi⟩G^i=\langle\rho_i\rangle for some involution ρi∈G=Aut⁡(H)\rho_i\in G=\operatorname{Aut}(\mathcal{H}). Then, when H\mathcal{H} is regular, the integer pijp_{ij} can also be understood as the order of ρiρj\rho_i\rho_j in GG. Therefore, in the context of highly symmetric hypertopes, the Buekenhout diagram of H\mathcal{H} is equivalent to the Coxeter diagram of G=Aut⁡(H)G=\operatorname{Aut}(\mathcal{H}).

For some diagrams, the incidence relation gains some degree of transitivity (not to be confused with incidence-transitivity, which is about the action of Aut⁡(Γ)\operatorname{Aut}(\Gamma) on incident pairs). Though we are mainly interested in hypertopes, the following result applies to any incidence geometry with a diagram.

Lemma 2.4. Let Γ\Gamma be a residually connected incidence geometry with diagram DD. Suppose that xx is an ii-element, yy is a jj-element, and zz is a kk-element such that x∗yx*y and y∗zy*z. If ii and kk are in separate connected components of D∖jD\setminus j, then x∗zx*z.

Proof. Both xx and zz are in the residue Γy\Gamma_y, and the diagram of Γy\Gamma_y is a subgraph of D∖jD\setminus j, and so has ii and kk in separate components. It follows that every ii-element of Γy\Gamma_y is incident to every kk-element by [1 Theorem 2.1.6], and so x∗zx*z.

Direct sums

There is a natural notion of the direct sum of incidence geometries [1 Definition 2.1.8]. Specifically, suppose {Ij}j∈J\{I_j\}_{j\in J} is a collection of pairwise disjoint sets, and consider a collection of incidence geometries Γj=(Xj,∗j,tj)\Gamma_j=(X_j,*_j,t_j) over IjI_j, where the sets XjX_j are pairwise disjoint. Then the direct sum of these is an incidence geometry Γ=(X,∗,t)\Gamma=(X,*,t) over ⋃jIj\bigcup_j I_j, where X=⋃jXjX=\bigcup_j X_j and the type function tt is defined by t∣Xj=tjt|_{X_j}=t_j. Incidence is defined by requiring ∗∣Xj=∗j*|_{X_j}=*_j, and then setting x∗yx*y for all x∈Xi,y∈Xkx\in X_i,y\in X_k for i≠ki\ne k. That is, incidence between two elements in the same XjX_j remains unchanged, and every pair of elements from different sets XiX_i and XkX_k is incident.

For Γ\Gamma as defined above, if the individual summands Γj\Gamma_j have diagram DjD_j, then Γ\Gamma has diagram ⨆jDj\bigsqcup_j D_j. The converse is true for hypertopes:

Proposition 2.5. [1 Corollary 2.1.10] If H\mathcal{H} is a hypertопe with diagram DD, and DD has connected components D1,…,DmD_1,\ldots,D_m, then H\mathcal{H} is isomorphic to the direct sum of H1,…,Hm\mathcal{H}_1,\ldots,\mathcal{H}_m, where Hj\mathcal{H}_j is the hypertopе corresponding to DjD_j (that is, the induced geometry with all elements of the types in DjD_j).

Flatness

In the theory of abstract polytopes, a polytope is said to be (i,j)(i,j)-flat if every ii-face (ii-element) is incident to every jj-face (see [13 Section 4E]). It makes sense to extend this definition to arbitrary incidence systems (though we will primarily be concerned with hypertopes):

Definition 3.1. For types i≠ji\ne j, an incidence system Γ\Gamma is (i,j)(i,j)-flat if every ii-element is incident to every jj-element. If an incidence system is (i,j)(i,j)-flat for some choice of ii and jj, then we also say simply that it is flat.

Equivalently, an incidence system is (i,j)(i,j)-flat if the subgraph of the incidence graph that is induced by all nodes of types ii and jj is a complete bipartite graph.

Proposition 3.2. Suppose that Γ\Gamma is an incidence system that is (i,j)(i,j)-flat. Then the residues of Γ\Gamma that contain types ii and jj are (i,j)(i,j)-flat. Furthermore, if Γ\Gamma is an incidence geometry with diagram DD, then ii and jj are not adjacent in DD.

Proof. The first part is clear: every ii-element is incident to every jj-element, and this incidence persists in the residue. When Γ\Gamma is an incidence geometry, being (i,j)(i,j)-flat means that the residues of type {i,j}\{i,j\} are generalized digons ([1 Definition 2.1.1]). Thus, ii and jj are not adjacent in the diagram of Γ\Gamma.

Next, we prove an analogue of [13 Lemma 4E3] for hypertopes, which is a partial converse to Proposition 3.2.

Proposition 3.3. Let H\mathcal{H} be a residually connected incidence geometry with diagram DD. Consider types ii, jj, and kk such that ii and kk are in separate connected components of D∖jD \setminus j. If the residue of every kk-element of H\mathcal{H} is (i,j)(i,j)-flat, then H\mathcal{H} is (i,j)(i,j)-flat.

Proof. Let xx be an ii-element and yy be a jj-element of H\mathcal{H}. Since H\mathcal{H} is residually connected, there is a sequence

x,z1,y1,z2,y2,…,zm,ym=yx,z_1,y_1,z_2,y_2,\ldots,z_m,y_m=y

such that each yay_a is a jj-element, each zaz_a is a kk-element, and every pair of adjacent elements is incident (see [1 Lemma 1.6.3]). Now, xx is an ii-element that is incident to z1z_1 and y1y_1 is a jj-element that is incident to z1z_1 so that xx and y1y_1 are in the residue Hz1\mathcal{H}_{z_1}. Since the residue at every kk-element is (i,j)(i,j)-flat, it follows that xx is incident to y1y_1. Furthermore, y1y_1 is incident to z2z_2. Then by Lemma 2.4, it follows that xx is incident to z2z_2. Then again by flatness of every kk-residue, we get that xx is incident to y2y_2, and continuing in this way we establish that xx is incident to yy, and so H\mathcal{H} is (i,j)(i,j)-flat. □

Proposition 3.4. Let H\mathcal{H} be an nn-hypertotope with diagram DD. Suppose that for some ii and jj, the number of ii-elements is exactly pijp_{ij}. Then:

  1. H\mathcal{H} is (i,k)(i,k)-flat for each k≠i,jk \ne i,j.

  2. pik=2p_{ik}=2 for each k≠i,jk \ne i,j. In particular, in the diagram D(H)D(\mathcal{H}), ii is either an isolated node or a leaf node connected only to type jj.

Proof. Suppose the number of ii-elements is pijp_{ij} and let xx be a kk-element of H\mathcal{H}. For any flag FF of type I∖{i,j}I \setminus\{i,j\} containing xx, the residue HF\mathcal{H}_F is a pijp_{ij}-gon. Hence, all ii-elements appear in this residue and therefore, we get that xx is incident to all ii-elements of H\mathcal{H}. Therefore, every kk-element is incident to every ii-element, so H\mathcal{H} is (i,k)(i,k)-flat.

The second part follows from the first and Proposition 3.2. □

Corollary 3.5. Let H\mathcal{H} be an nn-hypertotope with diagram DD. The number of ii-elements of H\mathcal{H} is pijp_{ij} if and only if H\mathcal{H} is (i,k)(i,k)-flat for each k≠i,jk \ne i,j.

Proof. Proposition 3.4 shows one direction. For the other, suppose H\mathcal{H} is (i,k)(i,k)-flat for each k≠i,jk \ne i,j, and consider a residue of type {i,j}\{i,j\}. Since in H\mathcal{H} every ii-element is incident to every kk-element for every k∈I∖{i,j}k \in I \setminus\{i,j\}, it follows that every ii-element is in this residue. On the other hand, the residue is a pijp_{ij}-gon, so there are pijp_{ij} exactly ii-elements. □

Tightness in hypertopes

Our next goal will be to establish a lower bound on the size of hypertopes with a given diagram, and to further characterize those hypertopes that meet the lower bound. Let us start by reviewing the situation for polytopes (hypertopes with a linear diagram).

An equivelar polytope P\mathcal{P} of type {p1,…,pn−1}\{p_1,\ldots,p_{n-1}\} has at least 2p1⋯pn−12p_1\cdots p_{n-1} chambers [3 Proposition 3.3], and when P\mathcal{P} has exactly that many chambers, it is called tight. In rank 3 there is an obvious necessary condition that follows from the fact that the number of chambers must be a multiple of 4, and this condition turns out to be sufficient:

Theorem 4.1 ([3 Theorem 5.1]). There is a tight polyhedron (3-polytope) of type {p,q}\{p,q\} if and only if pp and qq are not both odd.

If we are looking for tight regular (flag-transitive) polyhedra, then the situation is more nuanced. The result below follows from [2 Theorem 3.4] and [7 Theorem 5.10].

Theorem 4.2. There is a tight regular polyhedron of type {p,q}\{p,q\} if and only if one of the following is true:

  1. pp and qq are even.

  2. pp is odd and qq is an even divisor of 2p2p or qq is odd and pp is an even divisor of 2q2q.

  3. pp is an odd multiple of 3 and q=4q=4 or qq is an odd multiple of 3 and p=4p=4.

The lower bound for the size of equivelar polytopes follows from a simple chamber-counting argument, so let us try the same idea for hypertopes. To begin, we make the following definition: if T\mathcal{T} is a tree with edge labels, then the weight of T\mathcal{T}, denoted w(T)w(\mathcal{T}), is the product of the edge labels. The weight of a single isolated vertex is defined to be 1. The weight of a forest T=T1⊔⋯⊔Tm\mathcal{T} = T_1 \sqcup\cdots\sqcup T_m is defined to be w(T)=2m−1∏i=1mw(Ti)w(\mathcal{T}) = 2^{m-1} \prod_{i=1}^{m} w(T_i). Thought of another way, given a forest T\mathcal{T} consisting of mm trees, we can turn it into a tree by adding m−1m-1 edges with label 2, and then w(T)w(\mathcal{T}) is the weight of this new tree.

Proposition 4.3. If H\mathcal{H} is a hypertope with connected diagram DD, and TT is a maximum-weight spanning tree of DD, then H\mathcal{H} has at least 2w(T)2w(T) chambers.

Proof. It suffices to prove the bound for all spanning trees of DD. When n=1n=1, the diagram DD is a single node, which has weight 1, and H\mathcal{H} must be the geometry of an edge, with two chambers. When n=2n=2, H\mathcal{H} must be a polygon and the result is also true. Now suppose n≥3n \ge3 and that the result is true for (n−1)(n-1)-hypertopes. Let TT be a spanning tree of DD, and let jj be one of the leaf nodes. Suppose an edge of weight pp connects jj to T∖jT \setminus j. Then we may specify a chamber of H\mathcal{H} by picking an element xx of type jj, and then picking a chamber in the residue of xx. There are at least pp choices of element xx (by Proposition 2.3), and since T∖jT \setminus j is a spanning tree of D∖jD \setminus j, the residue Hx\mathcal{H}_x has at least 2w(T∖j)2w(T \setminus j) chambers, by inductive hypothesis. The result follows.

We will be discussing spanning forests of a diagram DD in this section, by which we will always mean a collection of spanning trees, one per connected component of DD. (For example, we are not considering taking a spanning tree of a connected graph and then just removing one edge.)

Corollary 4.4. If T=T1⊔⋯⊔Tm\mathcal{T} = T_1 \sqcup\cdots\sqcup T_m is a spanning forest of the diagram DD of H\mathcal{H}, then ∣Ch⁡(H)∣≥2w(T)|\operatorname{Ch}(\mathcal{H})| \ge2w(\mathcal{T}).

Proof. Since T\mathcal{T} is a spanning forest with mm trees, H\mathcal{H} has diagram D=D1⊔⋯⊔DmD = D_1 \sqcup\cdots\sqcup D_m, where TiT_i is a spanning tree of DiD_i for each ii. Then, since H\mathcal{H} is residually connected, H\mathcal{H} is the direct sum of the geometries corresponding to each DiD_i (by Proposition 2.5), and the number of chambers is at least the product of the number of chambers from each geometry. So the number of chambers is at least

∏i=1m2w(Ti)=2w(T).\prod_{i=1}^{m} 2w(T_i) = 2w(\mathcal{T}).

Notice that if H\mathcal{H} has a linear diagram, then this is precisely the same bound as what was already known for polytopes.

With this new size bound for hypertopes, we make the following definition:

Definition 4.5. If H\mathcal{H} is a hypertope with diagram DD, we say that H\mathcal{H} is tight if ∣Ch⁡(H)∣=2w(T)|\operatorname{Ch}(\mathcal{H})| = 2w(\mathcal{T}) for a maximum-weight spanning forest T\mathcal{T} of DD.

Proposition 4.6. Suppose H\mathcal{H} has diagram D1⊔⋯⊔DmD_1 \sqcup\cdots\sqcup D_m. Then H\mathcal{H} is the direct sum of hypertopes H1,…,Hm\mathcal{H}_1,\ldots,\mathcal{H}_m, and H\mathcal{H} is tight if and only if each Hi\mathcal{H}_i is tight.

Proof. That H\mathcal{H} is a direct sum follows from Proposition 2.5. Furthermore, since every element of Hi\mathcal{H}_i is incident to every element of Hj\mathcal{H}_j whenever i≠ji \ne j, it follows that

∣Ch⁡(H)∣=∣Ch⁡(H1)∣⋯∣Ch⁡(Hm)∣.|\operatorname{Ch}(\mathcal{H})| = |\operatorname{Ch}(\mathcal{H}_1)| \cdots|\operatorname{Ch}(\mathcal{H}_m)|.

Then H\mathcal{H} is tight if and only if ∣Ch⁡(H)∣=2w(T)=2m∏i=1mw(Ti)|\operatorname{Ch}(\mathcal{H})| = 2w(\mathcal{T}) = 2^m \prod_{i=1}^{m} w(T_i), where T=T1⊔⋯⊔Tm\mathcal{T} = T_1 \sqcup\cdots\sqcup T_m is a max weight spanning forest of H\mathcal{H}. Now, Proposition 4.3 says that

∣Ch⁡(H1)∣⋯∣Ch⁡(Hm)∣≥2m∏i=1mw(Ti),|\operatorname{Ch}(\mathcal{H}_1)| \cdots|\operatorname{Ch}(\mathcal{H}_m)| \ge2^m \prod_{i=1}^{m} w(T_i),

and we get equality only when ∣Ch⁡(Hi)∣=2w(Ti)|\operatorname{Ch}(\mathcal{H}_i)| = 2w(T_i) for each ii, and the result follows.

Proposition 4.7. Suppose that H\mathcal{H} is a tight nn-hypertope with n≥3n \ge3, and let ii be a leaf of a maximum-weight spanning forest TT.

(a) If ii is isolated, then there are two ii-elements, and otherwise the number of ii-elements is equal to the weight of the edge at ii.

(b) The residue of H\mathcal{H} at every ii-element is tight.

(c) For every type k≠ik \ne i not adjacent to ii in TT, the hypertope H\mathcal{H} is (i,k)(i,k)-flat.

Proof. Let nin_i be the number of ii-elements. Let pp be the label of the unique edge at ii, or 2 if there is no label. We can pick a chamber of H\mathcal{H} by first picking an ii-element xx and then picking a chamber in the residue Hx\mathcal{H}_x. By Proposition 4.4, the residue Hx\mathcal{H}_x has at least 2w(T∖i)2w(T \setminus i) chambers, so

∣Ch⁡(H)∣≥ni2w(T∖i).\lvert\operatorname{Ch}(\mathcal{H})\rvert\ge n_i 2^{w(T \setminus i)}.

On the other hand, by assumption we have

∣Ch⁡(H)∣=2w(T)=2w(T∖i)p,\lvert\operatorname{Ch}(\mathcal{H})\rvert= 2^{w(T)} = 2^{w(T \setminus i)}p,

and so ni≤pn_i \le p. Since we also have ni≥pn_i \ge p by Proposition 2.3, it follows that ni=pn_i = p. Then it also follows that ∣Ch⁡(Hx)∣=2w(T∖i)\lvert\operatorname{Ch}(\mathcal{H}_x)\rvert= 2^{w(T \setminus i)}, which means that Hx\mathcal{H}_x is tight. The last part follows from the first part and from Proposition 3.4. □

Now we come to our first big structural result about tight hypertopes.

Theorem 4.8. If H\mathcal{H} is a tight hypertope with connected diagram DD, then DD is a tree.

Proof. We prove the result by induction on nn, the rank of the hypertope. It is trivially true for n=2n = 2. Suppose that n≥3n \ge3 and that the result is true for (n−1)(n - 1)-hypertopes. Let TT be a maximum-weight spanning tree for the diagram DD. Let ii be one of the leaf nodes of TT, and let jj be the neighbor of ii in TT. Then by Proposition 4.7(c), the hypertope H\mathcal{H} is (i,k)(i,k)-flat for each k≠i,jk \ne i,j, and by Proposition 3.2, that implies that pik=2p_{ik} = 2 for k≠i,jk \ne i,j. In other words, the only proper edge in DD that involves ii is the edge between ii and jj. Now, by Proposition 4.7(b), any residue Hx\mathcal{H}_x with xx an ii-element is tight, and so by inductive hypothesis, the diagram D∖iD \setminus i is a tree. Since DD is obtained from D∖iD \setminus i by adding a leaf node that is attached to a tree, no new cycles are created, and so DD is itself a tree. □

Corollary 4.9. All tight 3-hypertopes are polytopes (polyhedra).

Proof. A tight 3-hypertope either has a disconnected diagram, or a tree diagram which must then be linear, and a hypertope with such a diagram is precisely an abstract polytope. □

Now we can extend Theorem 4.7 to show that every residue of a tight hypertope is tight.

Corollary 4.10. If H\mathcal{H} is a tight nn-hypertope with n≥3n \ge3, then every residue of H\mathcal{H} is tight.

Proof. It suffices to prove that the residue at any element xx of type ii is tight. By Proposition 4.6 and Theorem 4.8, we may assume that the diagram DD of H\mathcal{H} is a forest. Let D1⊔⋯⊔DmD_1 \sqcup\cdots\sqcup D_m be the connected components of the diagram D∖iD \setminus i. Since H\mathcal{H} is residually connected, Hx\mathcal{H}_x is a direct sum of hypertopes H1,…,Hm\mathcal{H}_1,\ldots,\mathcal{H}_m with these diagrams. For each rr, we can obtain DrD_r by successively deleting leaves from DD, and so Proposition 4.7(b) proves that each Hr\mathcal{H}_r is tight. It follows that Hx\mathcal{H}_x itself is tight. □

All together, Theorem 4.8, Corollary 4.9, and Corollary 4.10 tell us that tight hypertopes are made by suitably gluing together tight polyhedra such that the diagram is a tree. Thus, restrictions on tight polyhedra yield restrictions on tight hypertopes of all ranks.

Proposition 4.11. If H\mathcal{H} is a tight hypertope with connected diagram DD, then DD is a tree with no adjacent odd labels.

Proof. Two adjacent edges in DD determine a residue that must be a tight polyhedron, and the result follows from Theorem 4.1. □

Proposition 4.12. If H\mathcal{H} is a tight regular hypertope with connected diagram DD, then each edge with an odd label pp is only adjacent to edges with labels that are even divisors of 2p2p, unless pp is a multiple of 33 and is adjacent only to edges with labels 66 and 44.

Proof. This follows from Theorem 4.2 considering that every rank 33 residue of a tight regular hypertope must be a tight regular polyhedron.

Having worked out several necessary conditions for tight hypertopes, we will describe some constructions for them in the next section. First, though, let us work out more of the combinatorial information of tight hypertopes. First, we determine the number of elements of each type.

Proposition 4.13. Suppose H\mathcal{H} is a tight nn-hypertotope with diagram TT. Then the number of ii-elements is

21−deg⁡T(i)∏j:pij>2pij.2^{1-\deg_T(i)}\prod_{j:p_{ij}>2}p_{ij}.

Proof. If TT is disconnected, so that H\mathcal{H} is a direct sum (see Proposition 2.5), then the number of ii-elements of H\mathcal{H} is the same as the number of ii-elements of the hypertope H′\mathcal{H}' that corresponds to the connected component of ii in TT. Thus, we may assume that TT is connected. Then since H\mathcal{H} is tight, TT is a tree, and ∣Ch⁡(H)∣=2w(T)|\operatorname{Ch}(\mathcal{H})|=2w(T). On the other hand, we can specify a chamber by first picking an ii-element xx, and then picking a chamber in the residue of xx. Thus

2w(T)=ni∣Ch⁡(Hx)∣,2w(T)=n_i|\operatorname{Ch}(\mathcal{H}_x)|,

where nin_i is the number of ii-elements. Now, Hx\mathcal{H}_x is tight by Corollary 4.10. The diagram of Hx\mathcal{H}_x consists of d:=deg⁡T(i)d:=\deg_T(i) connected components, trees T1⊔⋯⊔TdT_1\sqcup\cdots\sqcup T_d, so we have

∣Ch⁡(Hx)∣=2w(T∖i)=2d∏i=1dw(Ti).|\operatorname{Ch}(\mathcal{H}_x)|=2w(T\setminus i)=2^d\prod_{i=1}^{d}w(T_i).

Now,

∏i=1dw(Ti)=w(T)∏j:pij>2pij,\prod_{i=1}^{d}w(T_i)=\frac{w(T)}{\prod_{j:p_{ij}>2}p_{ij}},

and so

ni=2w(T)∣Ch⁡(Hx)∣=2w(T)2d∏i=1dw(Ti)=21−d∏j:pij>2pij.n_i=\frac{2w(T)}{|\operatorname{Ch}(\mathcal{H}_x)|} =\frac{2w(T)}{2^d\prod_{i=1}^{d}w(T_i)} =2^{1-d}\prod_{j:p_{ij}>2}p_{ij}.

Now we can relate tightness to flatness in a way that extends [3 Theorem 4.4].

Theorem 4.14. A hypertope H\mathcal{H} is tight if and only if it has a forest diagram TT and, for every non-edge {i,j}\{i,j\} of TT, H\mathcal{H} is (i,j)(i,j)-flat.

Proof. We prove the result by induction on the rank nn. The result is trivial for n=2n=2, since every polygon is tight. Suppose n≥3n\geq3 and that the result is true for (n−1)(n-1)-hypertopes.

First, suppose H\mathcal{H} is tight. Proposition 4.6 and Theorem 4.8 establish that H\mathcal{H} has a forest diagram TT. Now consider types ii and jj not joined by an edge in TT. If ii and jj are in separate connected components of TT, by [1 Theorem 2.1.6], every ii-element is incident to every jj-element, so H\mathcal{H} is (i,j)(i,j)-flat. Otherwise, we may suppose that ii and jj are in the same connected component.

If either ii or jj is a leaf node, then Proposition 4.7(c) says that H\mathcal{H} is (i,j)(i,j)-flat. Otherwise, there is some node kk that is adjacent to jj and not on the unique path between ii and jj. Now, for every kk-element xx, the residue Hx\mathcal{H}_x is a tight hypertope (by Corollary 4.10). Since the diagram of Hx\mathcal{H}_x is T∖kT\setminus k, and there is no edge (i,j)(i,j) in T∖kT\setminus k, it follows that Hx\mathcal{H}_x is (i,j)(i,j)-flat by inductive hypothesis. Note that removing node jj from TT would disconnect ii from kk; otherwise there would be a cycle. Thus Proposition 3.3 implies that H\mathcal{H} itself is (i,j)(i,j)-flat.

Conversely, suppose that H\mathcal{H} has a forest diagram TT and is (i,j)(i,j)-flat whenever (i,j)(i,j) is not an edge of TT. Consider a leaf node ii of TT, connected by an edge of label pp to some type jj. By assumption, H\mathcal{H}

is (i,k)(i,k)-flat for each k∈I∖{i,j}k \in I \setminus\{i,j\}, and so by Corollary 3.5, the number of ii-elements is pp. Now, note that the residue at any given ii-element xx also satisfies our hypotheses, and so is tight by inductive hypothesis. It follows that

∣Ch⁡(H)∣=p∣Ch⁡(Hx)∣=2pw(T∖i)=2w(T),\lvert\operatorname{Ch}(\mathcal{H})\rvert= p\lvert\operatorname{Ch}(\mathcal{H}_x)\rvert= 2p w(\mathcal{T}\setminus i) = 2w(\mathcal{T}),

with the last equality following from the fact that the only edge weight missing from T∖i\mathcal{T}\setminus i compared to T\mathcal{T} is the edge of weight pp at type ii. Thus, H\mathcal{H} is tight. □

Flat extensions of hypertopes

Let H\mathcal{H} be an nn-hypert ope. We say that the (n+1)(n+1)-hypert ope K\mathcal{K} is an extension of H\mathcal{H} if there is a type ii such that all residues of K\mathcal{K} of cotype ii are isomorphic to H\mathcal{H}. Suppose we start with H\mathcal{H} and want to build an extension K\mathcal{K}. How can we accomplish this?

One way that always works is to add a new type ii to the types II of H\mathcal{H}, make exactly two elements of that type, and then make those two elements incident to every other element of H\mathcal{H}. Indeed, this is merely the direct sum of H\mathcal{H} with a 11-hypert ope. It is straightforward to show that this yields a hypert ope. We will call hypertopes obtained this way trivial extensions. Note that, due to thinness, we cannot try to do the same thing with three or more elements (that is, we cannot add three new elements and make them incident to every element of H\mathcal{H}).

What other “small” extensions can we try? Suppose we want K\mathcal{K} to have 44 ii-elements xx such that Kx\mathcal{K}_x is isomorphic to H\mathcal{H}. While we cannot make these elements incident to all other elements of H\mathcal{H}, could we pick a single type jj of H\mathcal{H} and make each ii-element incident to all elements of H\mathcal{H} other than the jj-elements? In other words, we want K\mathcal{K} to be (i,k)(i,k)-flat for k≠i,jk \ne i,j. By Theorem 3.2, this requires that pik=2p_{ik}=2, which is to say that type ii is not connected to type kk in the diagram of K\mathcal{K}. So the diagram of K\mathcal{K} consists of the diagram of H\mathcal{H}, with a single node added and connected to a single type jj of H\mathcal{H} with an edge labeled pi,j≥3p_{i,j} \ge3. We will call an extension of this type a flat extension of H\mathcal{H}. Without loss of generality, we will usually assume that H\mathcal{H} has types {0,…,n−1}\{0,\ldots,n-1\} and that j=n−1j=n-1. (Though let us emphasize that the linear order of types has no bearing here.)

In Flat extensions of abstract polytopes, the first author considered this problem in the context of polytopes [5]. The idea was to take several copies of the flag graph (chamber system) of the nn-polytope H\mathcal{H} and then add a new perfect matching corresponding to the new type. The new polytope had facets (the residues of the new elements) isomorphic to H\mathcal{H}, and it was (n−2,n)(n-2,n)-flat. The same construction works here with very little modification; let us describe it.

Fix an nn-hypert ope H\mathcal{H} with chamber system C\mathcal{C}. For i∈{0,…,n−1}i \in\{0,\ldots,n-1\}, let rir_i be the permutation of the chambers of C\mathcal{C} induced by edges of color ii. If CC is a chamber of H\mathcal{H} and w=⟨r0,…,rn−1⟩w=\langle r_0,\ldots,r_{n-1}\rangle, we will use Cw\mathcal{C}^w to denote the image of C\mathcal{C} under ww. We start with kk copies of C\mathcal{C}, with k≥2k \ge2 an even integer, so that our chambers can be written as (C,j)(C,j) with j∈Zkj \in\mathbb{Z}_k being the layer of the chamber. We want to add edges of color nn such that the resulting chamber system is a hypert ope and such that the diagram is the diagram of H\mathcal{H} but with an extra leaf node of type nn connected to type n−1n-1 with a branch of label kk. Then the induced permutation rnr_n must commute with rir_i for i<n−1i<n-1 since the {i,n}\{i,n\}-residue must be a digon for each i<n−1i<n-1. This implies that once we determine where rnr_n sends a chamber (C,j)(C,j), we know where it sends (Cw,j)(C^w,j) for every w∈⟨r0,…,rn−2⟩w \in\langle r_0,\ldots,r_{n-2}\rangle.

Let us now follow the idea of the construction in [5 Section 3.1]. Suppose that H\mathcal{H} is (n−1)(n-1)-element bipartite, meaning that it is possible to color the chambers of C\mathcal{C} with colors red and blue such that adjacent chambers are different colors if and only if they are (n−1)(n-1)-adjacent. Color each copy of C\mathcal{C} according to such a (single) coloring. Then we will match (C,2m)(C,2m) with (C,2m+1)(C,2m+1) if CC is red, and we will match (C,2m)(C,2m) with (C,2m−1)(C,2m-1) otherwise. We will denote this new graph by C∣k\mathcal{C}\mid k, and the resulting incidence system by H∣k\mathcal{H}\mid k. Clearly C∣k\mathcal{C}\mid k is a chamber system with kk copies of C\mathcal{C} as its nn-elements. We need to show that the result is a hypert ope with the desired diagram.

Proposition 5.1. If k≥4k \ge4 is even and C\mathcal{C} is the chamber system of a hypert ope, then so is C∣k\mathcal{C}\mid k.

Proof. Clearly C∣k\mathcal{C}\mid k is thin, so we need only to show that it is residually connected. To do so, we make use of [1 Lemma 3.4.9(ii)]. Let ZJZ_J (ZK,ZLZ_K,Z_L) be a JJ-cell (KK-cell, LL-cell) such that they have pairwise nonempty intersection; we need to show that ZJ∩ZK∩ZLZ_J \cap Z_K \cap Z_L is a (J∩K∩L)(J \cap K \cap L)-cell. Set Jˉ=J∖{n}\bar{J}=J\setminus\{n\} and similarly define Kˉ\bar{K} and Lˉ\bar{L}. Note that since rnr_n only affects the second coordinate of a chamber (C,a)(C,a), a JJ-cell of C∣k\mathcal{C}|k induces a Jˉ\bar{J}-cell of C\mathcal{C}. Indeed, every JJ-cell ZJZ_J is contained in a set of the form ZJˉ×ZkZ_{\bar{J}} \times Z_k, where ZJˉZ_{\bar{J}} is a Jˉ\bar{J}-cell.

First, suppose n∉J∩K∩Ln\notin J\cap K\cap L, and suppose without loss of generality that n∉Jn\notin J. Then ZJZ_J is a subset of C×{a}C\times\{a\} for some aa, and since ZJZ_J has nonempty intersection with ZKZ_K and ZLZ_L, then these cells also intersect C×{a}C\times\{a\}. It follows that ZJ∩ZK∩ZL=(ZJˉ∩ZKˉ∩ZLˉ)×{a}Z_J\cap Z_K\cap Z_L=(Z_{\bar{J}}\cap Z_{\bar{K}}\cap Z_{\bar{L}})\times\{a\}. Since CC is a hypertope, ZJˉ∩ZKˉ∩ZLˉZ_{\bar{J}}\cap Z_{\bar{K}}\cap Z_{\bar{L}} is a (Jˉ∩Kˉ∩Lˉ)(\bar{J}\cap\bar{K}\cap\bar{L})-cell of CC, which is a (J∩K∩L)(J\cap K\cap L)-cell of C∣k\mathcal{C}|k since n∉J∩K∩Ln\notin J\cap K\cap L.

Next, suppose n∈J∩K∩Ln\in J\cap K\cap L but n−1∉Jn-1\notin J. Then ZJZ_J is a subset of C×{a,a±1}C\times\{a,a\pm1\} (with the choice of a+1a+1 or a−1a-1 depending on the color of the chambers in ZJZ_J). Indeed, since n−1∉Jn-1\notin J, all chambers in ZJZ_J must have the same color. Thus ZKZ_K and ZLZ_L also intersect this same set. Furthermore, from the definition of rnr_n we see that ZJ∩ZK∩ZL=(ZJˉ∩ZKˉ∩ZLˉ)×{a,a±1}Z_J\cap Z_K\cap Z_L=(Z_{\bar{J}}\cap Z_{\bar{K}}\cap Z_{\bar{L}})\times\{a,a\pm1\}. Again, CC is a hypertope, so ZJˉ∩ZKˉ∩ZLˉZ_{\bar{J}}\cap Z_{\bar{K}}\cap Z_{\bar{L}} is a (Jˉ∩Kˉ∩Lˉ)(\bar{J}\cap\bar{K}\cap\bar{L})-cell of CC, and since the matching rnr_n commutes with every other matching rir_i with i<n−1i<n-1, it follows that ZJ∩ZK∩ZLZ_J\cap Z_K\cap Z_L is a ({n}∪(Jˉ∩Kˉ∩Lˉ))(\{n\}\cup(\bar{J}\cap\bar{K}\cap\bar{L}))-cell; that is, a (J∩K∩L)(J\cap K\cap L)-cell.

Finally, suppose that n∈J∩K∩Ln\in J\cap K\cap L and n−1∈Jn-1\in J. If n−1∉Kn-1\notin K or LL, the same argument as above holds symmetrically. Thus, we can assume {n−1,n}⊆J∩K∩L\{n-1,n\}\subseteq J\cap K\cap L. Then ZJZ_J contains chambers of both colors, and it follows that ZJ=ZJˉ×ZkZ_J=Z_{\bar{J}}\times Z_k, and similarly for ZKZ_K and ZLZ_L. Thus ZJ∩ZK∩ZL=(ZJˉ∩ZKˉ∩ZLˉ)×ZkZ_J\cap Z_K\cap Z_L=(Z_{\bar{J}}\cap Z_{\bar{K}}\cap Z_{\bar{L}})\times Z_k, and the argument is similar to before. □

Thus, if H\mathcal{H} is a hypertope, then H∣k\mathcal{H}|k, the incidence system associated to C∣k\mathcal{C}|k, is also a hypertope. We will occasionally blur the line between these two ideas, mostly by using C∣k\mathcal{C}|k to compute things about H∣k\mathcal{H}|k.

Proposition 5.2. If H\mathcal{H} is a hypertope with diagram DD and k≥4k\geq4 is even, then H∣k\mathcal{H}|k has a diagram obtained from DD by adding a new node nn connected to type n−1n-1 with an edge of label kk.

Proof. It is clear from the construction that the nn-elements of C∣k\mathcal{C}|k are isomorphic to CC, so we only need to show that the {i,n}\{i,n\}-cells of C∣k\mathcal{C}|k are digons if i<n−1i<n-1 and kk-gons if i=n−1i=n-1. This is clear from the construction: every {i,n}\{i,n\}-cell has 4 chambers if i<n−1i<n-1 and 2k2k chambers if i=n−1i=n-1. □

Proposition 5.3. If H\mathcal{H} is a hypertope and k≥4k\geq4 is even, then H∣k\mathcal{H}|k is (i,n)(i,n)-flat for every i<n−1i<n-1, and it is tight if and only if H\mathcal{H} is tight.

Proof. The nn-elements of C∣k\mathcal{C}|k are the kk copies of CC, and the ii-elements correspond to the connected components of C∣k\mathcal{C}|k after we remove all edges of label ii. We note that, for every chamber (C,a)(C,a) of C∣k\mathcal{C}|k, each {n−1,n}\{n-1,n\}-cell intersects every copy of CC (that is, every nn-element). Hence, for i<n−1i<n-1, since each ii-element contain {n−1,n}\{n-1,n\}-cells, every ii-element intersects every nn-element, which is to say that C∣k\mathcal{C}|k is (i,n)(i,n)-flat.

For the second part, the necessity follows from Proposition 5.2 and Proposition 4.7(b). Conversely, if CC is tight, then since CC has kk times as many chambers and has a spanning forest for its diagram with one additional edge labeled kk, it follows that C∣k\mathcal{C}|k is tight. □

Next, let us show that the extensions we build this way are as symmetric as we could hope for.

Proposition 5.4. Aut⁡(H∣k)≅Aut⁡(H)×Dk/2\operatorname{Aut}(\mathcal{H}|k)\cong\operatorname{Aut}(\mathcal{H})\times D_{k/2}, where Dk/2D_{k/2} is the dihedral group of order kk.

Proof. We will work with CC and C∣k\mathcal{C}|k directly. Let us first show how to extend an automorphism φ\varphi of CC to an automorphism of C∣k\mathcal{C}|k. Recall that CC is (n−1)(n-1)-element bipartite; let us fix a bicoloring of CC (say, using red and blue) and color each copy of CC in C∣k\mathcal{C}|k by the same bicoloring. For φ∈Aut⁡(C)\varphi\in\operatorname{Aut}(C), we define φ~\widetilde{\varphi} by

(C,j)φ~=(Cφ,j)(C,j)\widetilde{\varphi}=(C\varphi,j)

if φ\varphi is color-preserving, and

(C,j)φ~=(Cφ,−j)(C,j)\widetilde{\varphi}=(C\varphi,-j)

if φ\varphi is color-reversing. It is clear that φ~\widetilde{\varphi} commutes with rir_i for i<ni<n. Next, if φ\varphi is color-preserving, we have

(C,j)rnφ~=(C,j±1)φ~=(Cφ,j±1)=(Cφ,j)rn=((C,j)φ~)rn.(1)(C,j)r_n\widetilde{\varphi}=(C,j\pm1)\widetilde{\varphi}=(C\varphi,j\pm1)=(C\varphi,j)r_n=((C,j)\widetilde{\varphi})r_n. \tag*{(1)}

where the third equality follows since the color of CφC\varphi is the same color as CC, so rnr_n acts on the second coordinate of (C,j)(C,j) and (Cφ,j)(C\varphi,j) in the same way. On the other hand, if φ\varphi is color-reversing, we have

(C,j)nφ~=(C,j±1)φ~=(Cφ,−j∓1)=(Cφ,−j)n=((C,j)φ~)n(C,j)^n\widetilde{\varphi}=(C,j\pm1)\widetilde{\varphi}=(C\varphi,-j\mp1)=(C\varphi,-j)^n=((C,j)\widetilde{\varphi})^n

where the third equality follows since CφC\varphi has the opposite color of CC. So every automorphism of CC induces an automorphism of C∣k\mathcal{C}|_k that fixes layer 0.

Now, from the structure of C∣k\mathcal{C}|_k it is clear that there is an automorphism xx that sends each (C,j)(C,j) to (C,1−j)(C,1-j) and an automorphism yy that sends each (C,j)(C,j) to (C,−1−j)(C,-1-j). These are involutions whose product composition sends (C,j)(C,j) to (C,j+2)(C,j+2), and so they generate a dihedral group of order kk.

The rest of the argument is essentially the same as that of [5 Proposition 3.9]; every element of ⟨x,y⟩\langle x,y\rangle fixes the first coordinate of every pair (C,j)(C,j), and so it is a normal subgroup, and every automorphism of C∣k\mathcal{C}|_k can be written as a product of some φ~\widetilde{\varphi} with an element of ⟨x,y⟩\langle x,y\rangle.

Corollary 5.5. H∣k\mathcal{H}|_k is regular if and only if H\mathcal{H} is.

Proof. The necessity is obvious. Conversely, if H\mathcal{H} is regular, then Proposition 5.4 implies that Aut⁡(H∣k)\operatorname{Aut}(\mathcal{H}|_k) acts transitively on chambers, and so H∣k\mathcal{H}|_k is regular.

Finally, let us show that the construction can be iterated.

Proposition 5.6. Let kk be a positive even integer. Then C∣k\mathcal{C}|_k is nn-element bipartite, and for every j∈{0,…,n−1}j\in\{0,\ldots,n-1\}, if CC is jj-element bipartite then so is C∣k\mathcal{C}|_k.

Proof. Since kk is even and rnr_n changes the layer of a chamber by ±1\pm1, it is clear that every closed path in C∣k\mathcal{C}|_k uses nn-edges an even number of times. Furthermore, if CC is jj-element bipartite, then the induced bipartition on C∣k\mathcal{C}|_k (where every layer is bipartitioned in the same way) is a witness that C∣k\mathcal{C}|_k is jj-element bipartite.

Putting everything together, we have proven the following.

Theorem 5.7. Suppose H\mathcal{H} is a hypertope with diagram DD that is (n−1)(n-1)-element bipartite. Then:

  1. H∣k\mathcal{H}|_k is a hypertope.

  2. The diagram of H∣k\mathcal{H}|_k is obtained from DD by adding a new node nn connected to type n−1n-1 with an edge of label kk.

  3. H∣k\mathcal{H}|_k is tight if and only if H\mathcal{H} is.

  4. H∣k\mathcal{H}|_k is regular if and only if H\mathcal{H} is.

  5. H∣k\mathcal{H}|_k is nn-element bipartite and is jj-element bipartite for each jj such that H\mathcal{H} is jj-element bipartite.

Corollary 5.8. Suppose DD is a tree. Suppose that DD has only even labels, or that it has the form

Diagram with three nodes connected by edges labelled p and q, followed by a dashed oval

Figure 1.

where pp is odd and q≡2(mod4)q\equiv2\pmod{4}. Then there is a tight hypertope with diagram DD.

Proof. Our goal is to find a facet-bipartite, tight polyhedron to start with. If there is a single odd label pp with q≡2(mod4)q\equiv2\pmod{4}, then the tight polyhedron of type {q,p}\{q,p\} described in [3 Theorem 5.1] has a bipartite graph, so its dual is a facet-bipartite tight polyhedron of type {p,q}\{p,q\}. Similarly, if every label of DD is even and we pick pp and qq to be arbitrary adjacent labels, then the tight polyhedron of type {p,q}\{p,q\} described in [3 Theorem 5.1] is also facet-bipartite. Then by Theorem 5.7, we can build a tight hypertope with diagram DD repeatedly adding a new type and a new edge with an even label connected to this new type and an existing type.

Remark 5.9. In [6 Example 4.1], flat extensions of facet-bipartite polytopes were also described in terms of a voltage graph as follows. Start with a facet-bipartite nn-polytope (or more precisely, with its chamber system), and color the chambers according to a bipartition of the facets. Add semi-edges of

label nn to every chamber. Then take a dihedral group ⟨x,y⟩\langle x,y\rangle and give voltage xx to semi-edges attached to chambers of one color and voltage yy to the remaining semi-edges. Then the derived graph is precisely the flat extension described in [5 Section 3.1].

We expect that this same construction should work more generally here with hypertopes, but proving this is beyond the scope of this paper.

Constructing tight hypertopes using twisting

The construction of Section 5 can be applied to hypertopes regardless of any symmetry, and by Theorem 5.7, it can build tight regular (n+1)(n+1)-hypertopes from tight regular nn-hypertopes. In this section, we explore a purely algebraic way to build these regular tight hypertopes.

In order to do so, we will use a construction called twisting. The notion of twisting was originally defined in some very specific cases for polytopes (see [13]). It was extended to general coset geometries in [14], and generalized even further in [15]. As we only need a specific instance of twisting in this paper, we will introduce only a simplified version of the construction here. For more details, we advise to consult [15].

Let H\mathcal{H} be a regular nn-hypertope. We will need some basic facts about the automorphism groups of regular hypertopes; the full details can be found in [9]. Choose a base chamber cc of H\mathcal{H}. As H\mathcal{H} is thin, for each i=0,1,…,n−1i=0,1,\ldots,n-1, there is a unique chamber cic^{i} that is ii-adjacent to cc. Because H\mathcal{H} is regular, there must then exist an automorphism αi\alpha_i sending cc to cic^{i}. Also, this automorphism is unique and must be an involution. Moreover, it turns out that Aut⁡(H)=⟨α0,α1,…,αn−1⟩\operatorname{Aut}(\mathcal{H})=\langle\alpha_0,\alpha_1,\ldots,\alpha_{n-1}\rangle. Actually, the pair (Aut⁡(H),{α0,α1,…,αn−1})(\operatorname{Aut}(\mathcal{H}),\{\alpha_0,\alpha_1,\ldots,\alpha_{n-1}\}) forms a CC-group, but that is not crucial for understanding the construction (for more details on CC-groups and their relation with regular hypertopes, we refer to [9]).

Let Dm=⟨x,y⟩D_m=\langle x,y\rangle be a dihedral group of order 2m2m, which is the automorphism group of a regular mm-gon, and let Aut⁡(H)=⟨α0,α1,…,αn−1⟩\operatorname{Aut}(\mathcal{H})=\langle\alpha_0,\alpha_1,\ldots,\alpha_{n-1}\rangle be the automorphism group of a regular hypertope. Choose any j∈{0,1,…,m−1}j\in\{0,1,\ldots,m-1\} and let φj:Aut⁡(H)→Aut⁡(Dm)\varphi_j:\operatorname{Aut}(\mathcal{H})\to\operatorname{Aut}(D_m) be the map defined as follows:

  • If i≠ji\ne j, we set φj(αi)\varphi_j(\alpha_i) to be the identity.

  • If i=ji=j, we set φj(αj)\varphi_j(\alpha_j) to be the order two automorphism switching xx and yy.

This map φj\varphi_j extends to a group homomorphism, and thus to an action of Aut⁡(H)\operatorname{Aut}(\mathcal{H}) on DmD_m, if and only if all relators in Aut⁡(H)\operatorname{Aut}(\mathcal{H}) with respect to the generating set {α0,α1,…,αm−1}\{\alpha_0,\alpha_1,\ldots,\alpha_{m-1}\} contain an even number of αj\alpha_j’s. Suppose that this is the case and let G=Dm⋊φjAut⁡(H)G=D_m\rtimes_{\varphi_j}\operatorname{Aut}(\mathcal{H}).

Theorem 6.1. The group G=⟨α0,α1,…,αn−1,x⟩G=\langle\alpha_0,\alpha_1,\ldots,\alpha_{n-1},x\rangle is the automorphism group of a regular (n+1)(n+1)-hypertope H′\mathcal{H}'. Moreover, the diagram of H′\mathcal{H}' is obtained from the diagram of H\mathcal{H} by adding a new node connected by an edge labeled 2m2m to the jj-node of the diagram of H\mathcal{H}. Finally, the hypertope H′\mathcal{H}' is tight if H\mathcal{H} is tight.

Proof. Clearly, G=Dm⋊φjAut⁡(H)G=D_m\rtimes_{\varphi_j}\operatorname{Aut}(\mathcal{H}) is generated by {α0,α1,…,αn−1,x}\{\alpha_0,\alpha_1,\ldots,\alpha_{n-1},x\} since yy can be recovered by letting αj\alpha_j act on xx. For ease of notations, set αn=x\alpha_n=x. The first part of the claim is proved by showing that the coset geometry (G,(Gi)i=0,…,n)(G,(G_i)_{i=0,\ldots,n}) is a regular hypertope. This is precisely the content of [15 Theorem 3.16, Corollary 3.17], in the specific case where B=Aut⁡(H)B=\operatorname{Aut}(\mathcal{H}) and A=Dm=⟨x,y⟩A=D_m=\langle x,y\rangle.

For the diagrams, notice that the diagram of H′\mathcal{H}' is characterized by the orders in G=Dm⋊φjAut⁡(H)G=D_m\rtimes_{\varphi_j}\operatorname{Aut}(\mathcal{H}) of the elements αiαj\alpha_i\alpha_j for i≠j∈{0,1,…,n}i\ne j\in\{0,1,\ldots,n\}. It is clear that if both ii and jj are different from nn, so that αi\alpha_i and αj\alpha_j are generators of Aut⁡(H)\operatorname{Aut}(\mathcal{H}), the order of αiαj\alpha_i\alpha_j in GG is the same as it was in Aut⁡(H)\operatorname{Aut}(\mathcal{H}). Thus, the sub-diagram of H′\mathcal{H}' spanned by all nodes except the node nn is the diagram of H\mathcal{H}. Finally, we need to compute the order of αiαn\alpha_i\alpha_n for i=0,1,…,n−1i=0,1,\ldots,n-1. First of all, if i≠ji\ne j, the generator αi\alpha_i acts trivially on DmD_m and it is thus clear that αi\alpha_i and αn\alpha_n commute in GG. If i=ji=j, let us properly identify αn\alpha_n with (αn,e)(\alpha_n,e) and αj\alpha_j with (e,αj)(e,\alpha_j) in G=Dm⋊φjAut⁡(H)G=D_m\rtimes_{\varphi_j}\operatorname{Aut}(\mathcal{H}). Then, we have that (e,αj)(αn,e)(e,αj)(αn,e)=(y,αj)(e,αj)(x,e)=(y,e)(x,e)=(yx,e)(e,\alpha_j)(\alpha_n,e)(e,\alpha_j)(\alpha_n,e)=(y,\alpha_j)(e,\alpha_j)(x,e)=(y,e)(x,e)=(yx,e). Hence, the order of αjαn\alpha_j\alpha_n must be twice the order of yxyx, which is mm.

Suppose now that H\mathcal{H} is tight. This means that its diagram is a tree TT and that the number of chambers of H\mathcal{H} is equal to twice the product of all labels appearing on the edges of the tree. The diagram of H′\mathcal{H}' is also a tree T′T', with an additional edge. The weight w(T′)w(T') is thus equal to $w(T)\cdot 2m where 2m2m is the label of the new edge. The number of chambers of H′\mathcal{H}' is equal to the cardinality of G=Dm⋊φiAut⁡(H)G=D_m\rtimes_{\varphi_i}\operatorname{Aut}(\mathcal{H}) which is equal to 2m⋅2w(T)=4mw(T)=2w(T′)2m\cdot2w(T)=4mw(T)=2w(T'). This shows that H′\mathcal{H}' is tight whenever H\mathcal{H} is tight. □

We now note that, when starting from a regular hypertope H\mathcal{H}, the flat-extension described in § 5 is exactly the construction obtained from Theorem 6.1. Let us denote by T(Dm,H)\mathcal{T}(D_m,\mathcal{H}) the hypertope H′\mathcal{H}' obtained in Theorem 6.1 from Aut⁡(H)\operatorname{Aut}(\mathcal{H}) and DmD_m.

Theorem 6.2. Let H\mathcal{H} be a regular nn-hypertope which is (n−1)(n-1)-element bipartite and with diagram DD. Then T(Dm,H)=H∣2m\mathcal{T}(D_m,\mathcal{H})=\mathcal{H}|2m. Moreover, if the presentation of Aut⁡(H)\operatorname{Aut}(\mathcal{H}) is ⟨α0,…,αn−1∣Rα⟩\langle\alpha_0,\ldots,\alpha_{n-1}\mid R_\alpha\rangle, then Aut⁡(H∣2m)\operatorname{Aut}(\mathcal{H}|2m) has presentation

⟨α0,…,αn−1,x∣Rα,x2=(xαn−1)2m=[x,αi]=[[αn−1,x],αi]=1, for i<n−1⟩\langle\alpha_0,\ldots,\alpha_{n-1},x\mid R_\alpha,x^2=(x\alpha_{n-1})^{2m}=[x,\alpha_i]=[[\alpha_{n-1},x],\alpha_i]=1,\text{ for }i<n-1\rangle

Proof. Let H\mathcal{H} be a regular nn-hypertope H\mathcal{H} which is (n−1)(n-1)-element bipartite and let CC be its chamber system. Let Aut⁡(H)=⟨α0,…,αn−1⟩\operatorname{Aut}(\mathcal{H})=\langle\alpha_0,\ldots,\alpha_{n-1}\rangle, as before. Since CC is (n−1)(n-1)-element bipartite, all closed loops starting in a chamber c∈Cc\in C must cross an even number of (n−1)(n-1)-labeled edges. As closed loops in CC correspond to words on the alphabet {α0,…,αn−1}\{\alpha_0,\ldots,\alpha_{n-1}\} which are sent to the identity in Aut⁡(H)\operatorname{Aut}(\mathcal{H}), this is equivalent to all relators of Aut⁡(H)\operatorname{Aut}(\mathcal{H}) containing an even number of αn−1\alpha_{n-1}. Hence, we see that the condition for φn−1\varphi_{n-1} to extend to a group homomorphism is equivalent to H\mathcal{H} being (n−1)(n-1)-element bipartite.

Let Dm=⟨x,y∣x2=y2=(xy)m=1⟩D_m=\langle x,y\mid x^2=y^2=(xy)^m=1\rangle as before, and let G=Aut⁡(T(Dm,H))G=\operatorname{Aut}(\mathcal{T}(D_m,\mathcal{H})). Since G=Dm⋊Aut⁡(H)G=D_m\rtimes\operatorname{Aut}(\mathcal{H}), and y=αn−1xαn−1y=\alpha_{n-1}x\alpha_{n-1}, we can eliminate yy from a presentation for GG to get the presentation

⟨Sα,x∣Rα,x2=(xαn−1)2m=1,(xαi)2=(αn−1xαn−1αi)2=1, for i<n−1⟩\langle S_\alpha,x\mid R_\alpha,x^2=(x\alpha_{n-1})^{2m}=1,(x\alpha_i)^2=(\alpha_{n-1}x\alpha_{n-1}\alpha_i)^2=1,\text{ for }i<n-1\rangle

where Sα={α0,…,αn−1}S_\alpha=\{\alpha_0,\ldots,\alpha_{n-1}\} and RαR_\alpha is the set of relators of Aut⁡(H)\operatorname{Aut}(\mathcal{H}) with respect to SαS_\alpha. Note that (αn−1xαn−1αi)2=1(\alpha_{n-1}x\alpha_{n-1}\alpha_i)^2=1 can be alternatively written as [[αn−1,x],αi]=1[[\alpha_{n-1},x],\alpha_i]=1 and (xαi)2=1(x\alpha_i)^2=1 as [x,αi]=1[x,\alpha_i]=1.

Now consider the regular hypertope obtained from the construction H∣2m\mathcal{H}|2m, with automorphism group Aut⁡(H∣2m)=Aut⁡(H)⋉Dm\operatorname{Aut}(\mathcal{H}|2m)=\operatorname{Aut}(\mathcal{H})\ltimes D_m. As described in Proposition 5.4, Aut⁡(H∣2m)=⟨Sα,x,y⟩\operatorname{Aut}(\mathcal{H}|2m)=\langle S_\alpha,x,y\rangle, where xx is defined as the automorphism (C,j)↦(C,1−j)(C,j)\mapsto(C,1-j) and yy as the automorphism (C,j)↦(C,−1−j)(C,j)\mapsto(C,-1-j). We will now prove that Aut⁡(H∣2m)\operatorname{Aut}(\mathcal{H}|2m) satisfies all the relations of GG. First, note that all generators and relations of Aut⁡(H)\operatorname{Aut}(\mathcal{H}) are still true in Aut⁡(H∣2m)\operatorname{Aut}(\mathcal{H}|2m). Moreover, one can easily check that y=αn−1xαn−1=φn−1(αn−1)(x)y=\alpha_{n-1}x\alpha_{n-1}=\varphi_{n-1}(\alpha_{n-1})(x), meaning that (xαn−1xαn−1)m=(xαn−1)2m(x\alpha_{n-1}x\alpha_{n-1})^m=(x\alpha_{n-1})^{2m} must fix every chamber. Additionally, for all αi\alpha_i with i<n−1i<n-1, we have that (xαi)2(x\alpha_i)^2 and (yαi)2(y\alpha_i)^2 fixes every chamber. Substituting y=αn−1xαn−1y=\alpha_{n-1}x\alpha_{n-1}, we get that (αn−1xαn−1αi)2(\alpha_{n-1}x\alpha_{n-1}\alpha_i)^2 fixes every chamber. Hence Aut⁡(H∣2m)\operatorname{Aut}(\mathcal{H}|2m) is a quotient of the group GG. However, since both are tight, they must have the same order and hence, are the same group. Therefore, H∣2m\mathcal{H}|2m is the same tight regular hypertope as T(Dm,H)\mathcal{T}(D_m,\mathcal{H}). □

Starting from a coset geometry of kk-gon, with kk even, we can iterate the construction above to create a regular tight hypertope whose diagram is any tree with all even labels.

Corollary 6.3. Let TT be any labeled tree, with all labels equal to even integers. Then there exists a regular tight hypertope whose diagram is TT and whose automorphism group has presentation

⟨αi,i∈V(T) | αi2=1,[αi,αj]=1 for {i,j}∉E(T),(αiαj)pij=1 for {i,j}∈E(T) with even label pij,[[αi,αj],αk]=1 for all distinct i,j,k∈V(T)⟩.\left\langle\alpha_i,i\in V(T)\ \middle|\ \begin{aligned} &\alpha_i^2=1,[\alpha_i,\alpha_j]=1\text{ for }\{i,j\}\notin E(T),\\ &(\alpha_i\alpha_j)^{p_{ij}}=1\text{ for }\{i,j\}\in E(T)\text{ with even label }p_{ij},\\ &[[\alpha_i,\alpha_j],\alpha_k]=1\text{ for all distinct }i,j,k\in V(T) \end{aligned}\right\rangle.

Proof. Let kk be an even number and Dk=⟨α0,α1⟩D_k=\langle\alpha_0,\alpha_1\rangle be a dihedral group of order 2k2k. Then DkD_k is the automorphism group of a regular kk-gon, which is a tight regular polytope, and thus hypertope. Notice that, since kk is even, every defining relation of DkD_k contains an even number of α0\alpha_0 and an even number of α1\alpha_1. We can thus let DkD_k act on any other dihedral group Dm=⟨x,y⟩D_m=\langle x,y\rangle via φ0\varphi_0 or φ1\varphi_1. Doing so, by Theorem 6.1 we form a regular tight hypertope H\mathcal{H} (actually a polytope in this case) whose diagram is a path of length two with labels kk and 2m2m. Since kk and mm can be chosen arbitrarily, the theorem is proven for ∣T∣=3|T|=3. As mentioned in Theorem 6.1, the choice of φi\varphi_i adds a new edge of label 2m2m to the previous diagram. The automorphism group of the hypertope H\mathcal{H} is Dm⋊φiDk=⟨α0,α1,α2=x⟩D_m \rtimes_{\varphi_i} D_k = \langle\alpha_0, \alpha_1, \alpha_2 = x\rangle, with presentation

⟨α0,α1,α2∣α02=(α0α1)k=(α1α2)2m=(α0α2)2=[[α1,α2],α0]=1⟩\langle\alpha_0, \alpha_1, \alpha_2 \mid\alpha_0^2 = (\alpha_0\alpha_1)^k = (\alpha_1\alpha_2)^{2m} = (\alpha_0\alpha_2)^2 = [[\alpha_1,\alpha_2],\alpha_0] = 1\rangle

if φi=φ1\varphi_i = \varphi_1, otherwise

⟨α0,α1,α2∣α02=(α0α1)k=(α0α2)2m=(α1α2)2=[[α0,α2],α1]=1⟩.\langle\alpha_0, \alpha_1, \alpha_2 \mid\alpha_0^2 = (\alpha_0\alpha_1)^k = (\alpha_0\alpha_2)^{2m} = (\alpha_1\alpha_2)^2 = [[\alpha_0,\alpha_2],\alpha_1] = 1\rangle.

The relators for this group all contain an even number of any of the three generators since the relations introduced by semi-direct products are all of the form zwz−1=φi(z)(w)zwz^{-1}=\varphi_i(z)(w). Therefore, we can now use H\mathcal{H} as our starting hypertope, let it act on any dihedral group and form a new regular tight hypertope whose diagram has one more edge with even label. Iterating this process, we can construct tight regular hypertopes whose diagram is any tree whose labels are all even integers. Additionally, by Theorems 6.1 and 6.2, this iteration results in the presentation given in the statement of the corollary. (Note that, if the subdiagram with vertices {i,j,k}\{i,j,k\} is not connected, then in fact the relation [[αi,αj],αk]=1[[\alpha_i,\alpha_j],\alpha_k]=1 is already implied by the other relations.) □\square

Note that if we are given a regular tight hypertope whose diagram has some odd labels, we can still attempt to use this process to reach bigger trees. We will only need to check that φi\varphi_i is a homomorphism. Notice that if the vertex ii of the diagram is the end point of any odd labeled edge, φi\varphi_i will never be a homomorphism. So we can only hope to glue more edges to vertices that only have even labeled edges to start with.

Finally, we show that whenever all labels are equal to 4, there is a unique tight regular hypertope, just as it is the case for polytopes.

Theorem 6.4. There is a unique regular tight hypertope whose diagram is a tree with all labels equal to 4.

Proof. Let TT be a tree whose edges are all labeled by 4. By Corollary 6.3, we know that there exists a tight regular hypertope with diagram TT and that its automorphism group GG has presentation

G:=⟨ρi, i∈V(T) | ρi2=1for all i∈I,[[ρi,ρj],ρk]=1for all i,j,k∈V(T),[ρi,ρj]=1whenever {i,j}∉E(T)⟩.G := \left\langle\rho_i,\ i \in V(T) \ \middle| \ \begin{aligned} \rho_i^2 &= 1 && \text{for all } i \in I,\\ [[\rho_i,\rho_j],\rho_k] &= 1 && \text{for all } i,j,k \in V(T),\\ [\rho_i,\rho_j] &= 1 && \text{whenever } \{i,j\} \notin E(T) \end{aligned} \right\rangle.

Notice that this group is precisely the one constructed in [11] when the underlying graph is a tree.

We now show that the automorphism group of any tight regular hypertope H\mathcal{H} whose diagram is a tree TT all of whose labels are 4 must be isomorphic to GG. It is clear that Aut⁡(H)\operatorname{Aut}(\mathcal{H}) is generated by ∣V∣|V| involutions α1,α2,…,α∣V∣\alpha_1,\alpha_2,\ldots,\alpha_{|V|}. Moreover, whenever αi\alpha_i and αj\alpha_j correspond to non-adjacent vertices of TT, it must be that αi\alpha_i and αj\alpha_j commute. We now show that [[αi,αj],αk]=1[[\alpha_i,\alpha_j],\alpha_k]=1 for any i,j,k∈Ii,j,k \in I. The only non-trivial case to check is when i−j−ki-j-k forms a path in TT. Consider the {i,j,k}\{i,j,k\}-residue of H\mathcal{H}. This residue is a thin, residually connected and regular geometry whose diagram is a path of length three with labels equal to four. It must hence be the unique tight polytope P\mathcal{P} of type {4,4}\{4,4\}, which is commonly denoted by {4,4}(2,0)\{4,4\}_{(2,0)}. (See, for example, the data at [10].) The automorphism group of this polytope is precisely the group GG when TT is the only tree with three vertices (this can be verified in a computer algebra system). In particular, the desired commutator relation holds in P\mathcal{P}. Since H\mathcal{H} is a regular hypertope, this residue can be identified with the coset geometry (⟨αi,αj,αk⟩,⟨αj,αk⟩,⟨αi,αk⟩,⟨αi,αj⟩)(\langle\alpha_i,\alpha_j,\alpha_k\rangle, \langle\alpha_j,\alpha_k\rangle, \langle\alpha_i,\alpha_k\rangle, \langle\alpha_i,\alpha_j\rangle). Therefore, we conclude that [[αi,αj],αk]=1[[\alpha_i,\alpha_j],\alpha_k]=1.

We have shown that the automorphism group of any tight regular hypertope H\mathcal{H} with diagram TT is generated by ∣V∣|V| involutions and that these involutions satisfy all the relations in the presentation of GG. Hence, the group of H\mathcal{H} must be a quotient of GG. But since H\mathcal{H} is tight and regular, ∣Aut⁡(H)∣=2⋅4∣E∣=∣G∣|\operatorname{Aut}(\mathcal{H})|=2\cdot4^{|E|}=|G|. Therefore, we can indeed deduce that Aut⁡(H)≅G\operatorname{Aut}(\mathcal{H})\cong G. □\square

Acknowledgments

The authors thank Maria Elisa Fernandes for helpful discussions about small hypertopes and CC-groups.

The second author is funded by an Action de Recherche Concertée – ARC – from the Communauté Française Wallonie-Bruxelles.

The third author is a Postdoctoral Researcher of the Fonds de la Recherche Scientifique – FNRS.

References

References

  1. [1]Francis Buekenhout and Arjeh M Cohen. Diagram geometry: related to classical groups and buildings, volume 57. Springer Science & Business Media, 2013.DOI
  2. [2]Marston Conder and Gabe Cunningham. Tight orientably-regular polytopes. Ars Math. Contemp., 8(1):68–81, 2015.
  3. [3]Gabe Cunningham. Minimal equivelar polytopes. Ars Math. Contemp., 7(2):299–315, 2014.DOI
  4. [4]Gabe Cunningham. Tight chiral polyhedra. Combinatorica, 38(1):115–142, 2018.arxiv.org/abs/1508.07451
  5. [5]Gabe Cunningham. Flat extensions of abstract polytopes. Art Discrete Appl. Math., 4(3):Paper No. 3.06, 14, 2021.arxiv.org/abs/2001.07677
  6. [6]Gabe Cunningham, Elías Mochán, and Antonio Montero. Cayley extensions of maniplexes and polytopes. Journal of Combinatorial Theory, Series A, 212:106000, 2025.arxiv.org/abs/2305.11843
  7. [7]Gabe Cunningham and Daniel Pellicer. Classification of tight regular polyhedra. J. Algebraic Combin., 43(3):665–691, 2016.DOI
  8. [8]Gabe Cunningham and Daniel Pellicer. Tight chiral polytopes. J. Algebraic Combin., 54(3):837–878, 2021.arxiv.org/abs/2009.04566
  9. [9]Maria Elisa Fernandes, Dimitri Leemans, and Asia Ivić Weiss. Highly symmetric hypertopes. Aequationes Math., 90(5):1045–1067, 2016.arxiv.org/abs/1604.03162
  10. [10]M. I. Hartley. An atlas of small regular abstract polytopes. Periodica Mathematica Hungarica, 53:149–156, 2006. available online at http://www.abstract-polytopes.com/atlas/.DOI
  11. [11]Dimitri Leemans, Pablo Spiga, and Philippe Tranchida. Hypertopes with prescribed diagram symmetries, 2026.arxiv.org/abs/2610.09711
  12. [12]Dimitri Leemans, Klara Stokes, and Philippe Tranchida. Groups represented by incidence geometries. arXiv preprint arXiv:2506.10464, 2025.arxiv.org/abs/2506.10464
  13. [13]Peter McMullen and Egon Schulte. Abstract regular polytopes, volume 92 of Encyclopedia of Mathematics and its Applications. Cambridge University Press, Cambridge, 2002.
  14. [14]Claudio Alexandre Piedade and Philippe Tranchida. From group operations to geometric structures: Amalgamations, hnn-extensions, and twisting in coset geometries. Journal of Algebra, 707:83–144, 2026.arxiv.org/abs/2505.24662
  15. [15]Claudio Alexandre Piedade and Philippe Tranchida. The geometry of wreath and semi-direct products. arXiv preprint arXiv:2605.04646, 2026.arxiv.org/abs/2605.04646
  16. [16]J. Tits. Sur les analogues algébriques des groupes semi-simples complexes. In Colloque d’algèbre supérieure, tenu à Bruxelles du 19 au 22 décembre 1956, Centre Belge de Recherches Mathématiques, pages 261–289. Etablissements Ceuterick, Louvain, 1957.
  17. [17]J. Tits. A local approach to buildings. In The geometric vein, pages 519–547. Springer, New York-Berlin, 1981.DOI

Paper details

vlnc:W_zxfm4jy3
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

Not recorded for this paper

View statement
Looking for related papers…
Contents