betaFeedback

WorksMathematics

CAT(0) Coxeter cell complexes, Bipodality and the Parallel Wall Property

Abstract

Coxeter cell complexes are a simultaneous generalization to cube complexes and the Davis complex of a Coxeter group. We show that 1-skeleta of CAT(0) Coxeter cell complexes isometrically embed into a CAT(0) cube complex. When the resulting cube complex is finite dimensional, we show that the condition of bipodality (satisfied by both CAT(0) cube complexes and Davis complexes of Coxeter groups) implies local finiteness of the cubulation. To this end, we show a Parallel Wall Theorem for bipodal CAT(0) Coxeter cell complexes satisfying a uniform bound on the size of a collection of pairwise intersecting walls. In particular, our methods give a new, purely geometric proof of the Parallel Wall Theorem for Coxeter groups.

Introduction

A Coxeter cell complex is a cell complex built out of polytopes whose symmetry groups are finite Coxeter groups. More precisely, Coxeter cells are Davis complexes of finite Coxeter groups, in the sense of [5]. We say a Coxeter cell complex is CAT(0) if it is built out of finitely many types of Coxeter cells and its induced piecewise Euclidean metric is a CAT(0) metric. Motivating examples of such complexes are finite dimensional cube complexes, whose nn-cells correspond the Davis complex of the Coxeter group (Z/2Z)n(\mathbb{Z}/2\mathbb{Z})^n, and the Davis complex of any Coxeter group, whose cells are formed by the Coxeter cells associated to the group’s finite parabolic subgroups. As in these examples, CAT(0) Coxeter cell complexes come equipped with a natural wallspace structure. For Coxeter groups, this wallspace structure has been thoroughly studied to deduce combinatorial properties of the groups. At the heart of this study lie two key properties:

The BPI Property (Bounded Pairwise Intersections): There is a bound, depending only on the complex, on the number of pairwise intersecting walls.

The PW Property (Parallel Wall Property): There is a constant CC, depending only on the complex, such that for any wall WW and vertex xx at distance ≥C\ge C from WW, there is a wall W′W' separating xx from WW.

These two properties were first shown to hold for Coxeter groups by Brink and Howlett in [1]. In general, the PW Property is difficult to establish. We have shown the PW Property always holds in the 2-dimensional case in [10]. The question is currently open in higher dimensions. Here, we give a criterion allowing us to guarantee the PW Property.

We say that XX is bipodal if it satisfies the following: Suppose that W,VW,V are intersecting walls sharing a common adjacent vertex yy, and that xx is a vertex not separated from yy by WW and VV. If WW is separated from xx by a wall W′W', then for any wall U≠VU \ne V containing W∩VW \cap V, there is a wall U′U' separating xx from UU. It is shown by Dyer and Hohwleg in [6] that the Davis complex of a Coxeter group is bipodal (see also [13] for an elementary proof).

Theorem 1.1. Suppose that XX is a bipodal CAT(0) Coxeter cell complex satisfying the BPI Property. Then XX satisfies the PW Property.

We emphasize that the original proof of the Parallel Wall theorem for Coxeter groups makes extensive use of the algebraic language coming from root systems. As such, the proof of Theorem 1.1 gives a new, geometric proof of the Parallel Wall theorem.

Using the BPI and PW Properties, it can be deduced from [12] and [4] that Coxeter groups act properly on finite dimensional, locally finite CAT(0) cube complexes. In general, Sageev’s construction allows us to isometrically embed the 1-skeleton of a CAT(0) Coxeter cell complex XX into a cube complex C(X)C(X), whose nn-cubes correspond to collections of nn pairwise intersecting walls in XX. In [9 Thm 5.7], Hruska and Wise show that if the complex is itself locally finite, then local finiteness of C(X)C(X) is equivalent to XX having compact-wall separation, an analog of the PW Property for compact sets. We show that for locally finite CAT(0) Coxeter cell complexes, the PW Property implies the compact-wall separation (Corollary 3.5). In particular, we obtain the following:

Corollary 1.2. Suppose that XX is a bipodal locally finite CAT(0) Coxeter cell complex satisfying the BPI Property. Then X(1)X^{(1)} isometrically embeds into a finite dimensional, locally finite CAT(0) cube complex.

Furthermore, metrically proper group actions on XX extend to metrically proper actions on C(X)C(X) [9]. It follows, for example, that if GG is a group acting properly on XX, and GG has a bound on the size of its finite subgroups (e.g. when the action is cocompact or, more generally, when XX is uniformly locally finite), then GG satisfies the Tits alternative [15].

Corollary 1.3. If GG acts properly by automorphisms on a bipodal locally finite CAT(0) Coxeter cell complex satisfying the BPI Property, then GG acts properly on a locally finite, finite dimensional CAT(0) cube complex.

Contrary to much of the theory of CAT(0) cube complexes, we rely on the CAT(0) metric in most of our proofs. Basic results in CAT(0) geometry, as well as background on Coxeter cell complexes, can be found in Section 2. In Section 3, we recall Sageev’s construction and show Corollary 1.2 under the assumption of Theorem 1.1. In Section 4, we introduce the notion of angle between walls, and use this to give a CAT(0)-like inequality for “wall triangles”. We also generalize [1 Lem 1.7], showing that these angles can be used to compute the distance between a vertex and a wall. Finally, we prove Theorem 1.1 in Section 5.

Acknowledgements. The author would like to thank Piotr Przytycki for his helpful discussions throughout this project, and for reviewing preliminary versions of this paper. The author would also like to thank Dani Wise for providing initial feedback on the author’s thesis, which served as the foundation for this work. We acknowledge that AI was not used in any capacity in the process of writing this paper, or proving the results therein.

Background

CAT(0) Geometry

Let (X,d)(X,d) be a geodesic metric space. We denote by [x,y][x,y] a geodesic segment connecting xx and yy. A geodesic triangle in XX consists of a triple of points x,y,z∈Xx,y,z\in X, together with geodesic segments between the three points. We denote by Δ(x,y,z)\Delta(x,y,z) the union of these segments. For every geodesic triangle in XX with vertices x,y,z∈Xx,y,z\in X, there is a unique (up to isometry) Euclidean triangle Δ‾(x,y,z)\overline{\Delta}(x,y,z) in R2\mathbb{R}^{2}, called the comparison triangle, with vertices xˉ,yˉ,zˉ∈R2\bar{x},\bar{y},\bar{z}\in\mathbb{R}^{2} and side lengths dR2(xˉ,yˉ)=d(x,y)d_{\mathbb{R}^{2}}(\bar{x},\bar{y})=d(x,y), dR2(xˉ,zˉ)=d(x,z)d_{\mathbb{R}^{2}}(\bar{x},\bar{z})=d(x,z) and dR2(yˉ,zˉ)=d(y,z)d_{\mathbb{R}^{2}}(\bar{y},\bar{z})=d(y,z). A point p∈[x,y]p\in[x,y] (resp [x,z][x,z] and [y,z][y,z]) is called a comparison point for pˉ∈[xˉ,yˉ]\bar{p}\in[\bar{x},\bar{y}] (resp [xˉ,zˉ][\bar{x},\bar{z}] and [yˉ,zˉ])[\bar{y},\bar{z}]) if dR2(xˉ,pˉ)=d(x,p)d_{\mathbb{R}^{2}}(\bar{x},\bar{p})=d(x,p).

Definition 2.1. A geodesic space XX is CAT(0) if for any geodesic triangle Δ\Delta and comparison points pˉ,qˉ∈Δ‾\bar{p},\bar{q}\in\overline{\Delta} for p,q∈Δp,q\in\Delta, we have d(p,q)≤dR2(pˉ,qˉ)d(p,q)\leq d_{\mathbb{R}^{2}}(\bar{p},\bar{q}). We say that XX is non-positively curved if it is locally CAT(0).

It is a standard exercise to show that CAT(0) spaces are uniquely geodesic and thus contractible. Conversely, the Cartan-Hadamard theorem states that if XX is complete, non-positively curved and simply connected, then XX is CAT(0) [2 Thm II.4.1].

There is also an equivalent definition for CAT(0) spaces in terms of angles. Given x,y,z∈Xx,y,z\in X, the comparison angle ∠‾z(x,y)\overline{\angle}_{z}(x,y) is the Euclidean angle in the comparison triangle Δ‾(x,y,z)\overline{\Delta}(x,y,z) at the vertex zˉ\bar{z}. Given geodesics γ1,γ2\gamma_{1},\gamma_{2} with γ1(0)=γ2(0)=p\gamma_{1}(0)=\gamma_{2}(0)=p, we may then define the Alexandrov angle between γ1\gamma_{1} and γ2\gamma_{2} as

∠(γ1,γ2)=lim⁡ϵ→0sup⁡t,t′≤ϵ∠‾p(γ1(t),γ2(t′))\angle(\gamma_{1},\gamma_{2})=\lim_{\epsilon\to0}\sup_{t,t'\leq\epsilon}\overline{\angle}_{p}\bigl(\gamma_{1}(t),\gamma_{2}(t')\bigr)

We denote by ∠z(x,y)\angle_{z}(x,y) the Alexandrov angle between segments [z,x][z,x] and [z,y][z,y].

Proposition 2.2 ([2 Prop II.1.7]). XX is CAT(0) if and only if ∠z(x,y)≤∠‾z(x,y)\angle_{z}(x,y)\leq\overline{\angle}_{z}(x,y) for all x,y,z∈Xx,y,z\in X. In particular, the sum of Alexandrov angles in a geodesic triangle of a CAT(0) space is bounded by π\pi.

The metric of a CAT(0) space is convex, and due to this CAT(0) spaces behave nicely with respect to their convex subsets.

Proposition 2.3 ([2 Prop II.2.4]). Let XX be a complete CAT(0) space, and C⊂XC \subset X a closed convex subset of XX. Then for x∈Xx \in X, there is a unique point πC(x)∈C\pi_C(x) \in C such that d(x,πC(x))=d(x,C)d(x,\pi_C(x)) = d(x,C). Furthermore, if x∉Cx \notin C then πC(x)\pi_C(x) is the unique point in XX satisfying ∠πC(x)(x,y)≥π/2\angle_{\pi_C(x)}(x,y) \ge\pi/2 for all y∈C∖{πC(x)}y \in C \setminus\{\pi_C(x)\}.

As a consequence, one can show that the function x→d(x,C)x \to d(x,C) is a convex function [2 Cor II.2.5], and therefore the rr-neighborhood of CC, defined by Nr(C)={x∈X:d(x,C)<r}N_r(C) = \{x \in X : d(x,C) < r\}, is convex in XX for any r≥0r \ge0.

Coxeter groups and Coxeter cell complexes

Let Γ\Gamma be a simplicial graph with edge labels taking values in {2,3,…}\{2,3,\ldots\}. Then the Coxeter group with defining graph Γ\Gamma is

W_\Gamma= \left\langle V(\Gamma) \mathrel{\left|} s^2 = 1, (st)^{m_{st}} = 1 \text{ for each } s \mathbin{\text{---}} t \in E(\Gamma) \right.\right\rangle

A Coxeter group WΓW_\Gamma has an associated faithful representation called the Tits representation. We consider a vector space VV with basis {αs}s∈V(Γ)\{\alpha_s\}_{s \in V(\Gamma)}, equipped with the symmetric bilinear form

⟨αs,αt⟩=−cos⁡πmst.\langle\alpha_s,\alpha_t\rangle= -\cos\frac{\pi}{m_{st}}.

Then WΓW_\Gamma acts faithfully on VV via

s⋅v=v−2⟨αs,v⟩αss \cdot v = v - 2\langle\alpha_s,v\rangle\alpha_s

for s∈V(Γ)s \in V(\Gamma) [5 Sec 6.12]. A Coxeter group WΓW_\Gamma is finite if and only if the bilinear form arising from the Tits representation is positive definite [5 Thm 6.12.9]. In this case, the representation gives an action of WΓW_\Gamma by isometries on a Euclidean space VV, where conjugates of elements in V(Γ)V(\Gamma) act by reflections along hyperplanes in VV. These hyperplanes partition VV into a set of cones, on which WΓW_\Gamma acts transitively. The fundamental cone in VV is the cone supported by the hyperplanes fixed by the generators of WΓW_\Gamma. We denote by ee the point at distance 12\frac{1}{2} from each supporting hyperplane in the fundamental cone.

Definition 2.4. The Coxeter cell of type Γ\Gamma, denoted by ZΓZ_\Gamma, is the convex hull of WΓ⋅eW_\Gamma\cdot e.

The cell structure on ZΓZ_\Gamma can be described in terms of the Coxeter subgroups WΓ′≤WΓW_{\Gamma'} \leq W_\Gamma generated by subgraphs Γ′⊂Γ\Gamma' \subset\Gamma. As shown in [5 Lem 7.3.3], every codimension-1 face of a Coxeter cell ZΓZ_\Gamma corresponds to a coset gWΓ′gW_{\Gamma'}, where Γ′⊂Γ\Gamma' \subset\Gamma is a full subgraph with size #V(Γ)−1\#V(\Gamma)-1. The face corresponding to gWΓ′gW_{\Gamma'} is isometric to the convex hull of gWΓ′⋅egW_{\Gamma'} \cdot e. Thus each ZΓ′Z_{\Gamma'} embeds into ZΓZ_\Gamma, allowing us to inductively define the cell structure on ZΓZ_\Gamma. In particular, the vertices of ZΓZ_\Gamma correspond to elements of WΓW_\Gamma, and we have a clear vertex-transitive action of WΓW_\Gamma on ZΓZ_\Gamma.

Definition 2.5. A Coxeter cell complex is a Euclidean polyhedral complex whose cells are isometric to Coxeter cells. A Coxeter cell complex is CAT(0) if its set of isometry classes of cells is finite, and its induced piecewise Euclidean metric is a CAT(0) metric (as defined in [2 Def I.7.4]).

Note that in general, the induced piecewise Euclidean (pseudo)metric on a polyhedral complex may fail to be complete or even geodesic. In fact, it may fail to be a metric at all. However, these pathologies are avoided when the cell complex has a finite set of isometry classes of cells [2 Thm I.7.19]. We denote the set of isometry classes of cells in a Coxeter cell complex XX by Shapes⁡(X)\operatorname{Shapes}(X).

Figure 1. Coxeter cells of type ∙—3∙\bullet\mathrel{\mathop{\text{---}}\limits^{3}} \bullet, 2∙2╱2╲∙—2∙\begin{array}{c}\phantom{2}\bullet\phantom{2}\\[-6pt]\diagup\phantom{2}\diagdown\\[-6pt]\bullet\mathrel{\mathop{\text{---}}\limits_{2}}\bullet\end{array} with labels 22, 22, and 22, and 2∙2╱22╲3∙—4∙\begin{array}{c}\phantom{2}\bullet\phantom{2}\\[-6pt]\diagup^{2}\phantom{2}\diagdown^{3}\\[-6pt]\bullet\mathrel{\mathop{\text{---}}\limits_{4}}\bullet\end{array} (generated using SageMath [14]).

Example 2.6. Let WΓW_{\Gamma} be a Coxeter group. The Davis complex (or Davis–Moussong complex) X=XΓX=X_{\Gamma} has a Coxeter cell of type Γ′\Gamma' for each coset gWΓ′gW_{\Gamma'}, where Γ′⊆Γ\Gamma'\subseteq\Gamma is the defining graph of a finite Coxeter subgroup WΓ′≤WΓW_{\Gamma'}\leq W_{\Gamma}. The faces of Coxeter cells in XX are identified whenever they correspond to the same coset. Note that WΓW_{\Gamma} acts freely and transitively on the vertices of XX.

Example 2.7. If Γ\Gamma is the complete graph on nn vertices with edge labels all equal to 22, then WΓ=(Z/2Z)nW_{\Gamma}=(\mathbb{Z}/2\mathbb{Z})^n and ZΓZ_{\Gamma} is an nn-dimensional cube. Thus cube complexes are also examples of Coxeter cell complexes.

The work of Moussong in [11] provides a combinatorial condition on the links of vertices in XX that guarantees that XX is CAT(0). The combinatorial link of a vertex v∈X(0)v\in X^{(0)} has an nn-simplex for each (n+1)(n+1)-cell incident to vv, where each face of the simplex corresponding to a cell Z⊂XZ\subset X corresponds to a lower dimensional cell of ZZ incident to vv. Vertices of the combinatorial link correspond to edges of XX and these vertices are connected by an edge whenever the corresponding edges in XX bound a 2n2n-gon. The labeled link, denoted by Lk⁡ℓ(v,X)\operatorname{Lk}_{\ell}(v,X) is the combinatorial link of vv, where edges of Lk⁡ℓ(v,X)\operatorname{Lk}_{\ell}(v,X) corresponding to 2n2n-gons in XX have label nn.

If XΓX_{\Gamma} is the Davis complex of the Coxeter group WΓW_{\Gamma}, then Lk⁡ℓ(v,X)\operatorname{Lk}_{\ell}(v,X) has as 1-skeleton Γ\Gamma, and has an nn-simplex for each full subgraph Γ′⊂Γ\Gamma'\subset\Gamma with WΓ′W_{\Gamma'} finite. In [11 Cor 10.2], Moussong shows that the metric link of the identity in XΓX_{\Gamma} is CAT(1). As a consequence of Moussong’s work and [2 Thm II.5.4], we obtain:

Theorem 2.8. Let XX be a Coxeter cell complex with Shapes⁡(X)\operatorname{Shapes}(X) finite. Then XX is non-positively curved if and only if every vertex in XX is labelled-flag:

  1. Lk⁡ℓ(v,X)\operatorname{Lk}_{\ell}(v,X) is a simplicial complex,

  1. every full subgraph Γ⊂Lk⁡ℓ(v,X)\Gamma\subset\operatorname{Lk}_{\ell}(v,X) with WΓW_{\Gamma} finite spans a simplex.

Figure 2. Examples of walls in a CAT(0) Coxeter cell complex

When XX is a cube complex, then a full subgraph Γ⊂Lkℓ(v,X)\Gamma\subset Lk_{\ell}(v, X) has WΓW_{\Gamma} finite if and only if Γ\Gamma is a clique. Simplicial complexes for which every clique spans a simplex are called flag complexes. As such, Theorem 2.8 is a natural generalization of the flag link condition typically given as the defining property of non-positively curved cube complexes [2 Thm II.5.20].

Walls in CAT(0) Coxeter cell complexes

Here, we develop the basic theory of walls in CAT(0) Coxeter cell complexes. As we will show, many of the definitions and properties in this section bear a striking resemblance to those present in both the theories of Coxeter groups and CAT(0) cube complexes. Most of the results in this section are direct generalizations of those in [10], and so their proofs will be omitted. We reiterate that throughout XX will be assumed to be a Coxeter cell complex built out of finitely many types of Coxeter cells. In particular, XX is a complete geodesic metric space.

A midcell in a Coxeter cell Z=ZΓZ = Z_{\Gamma} is the fixed point set of a reflection in WΓW_{\Gamma}. Note that each edge ee in ZZ intersects a unique midcell in ZZ, denoted by M(e,Z)M(e, Z). We say two edges e,e′∈Ze, e' \in Z are equivalent if M(e,Z)=M(e′,Z)M(e, Z) = M(e', Z). We may then extend this equivalence relation to all edges of XX.

Definition 2.9. The wall in XX dual to an edge ee of XX is the abstract complex ⋃e′∈[e]M(e′,Z)/∼\bigcup_{e'\in[e]} M(e', Z)/\sim, where the cell M(e′,F)M(e', F) is naturally identified with the appropriate subcomplex of M(e′,Z)M(e', Z) whenever FF is a face of ZZ.

Remark 2.10. In the case where XX is the Davis complex of a Coxeter group, walls are fixed point sets of reflections. Therefore, they are closed, convex [2 Prop II.6.2] and separate XX into two connected components. Furthermore, the combinatorial distance between vertices in XX (defined by the path metric on the 1-skeleton of XX) is exactly the number of walls separating them. Note that two walls in a Davis complex intersect if and only if the underlying reflections generate a finite dihedral subgroup.

Proposition 2.11 ([10 Prop 3.6]). Let XX be a CAT(0) Coxeter cell complex, and W\mathcal{W} a wall in XX. Then

  1. The map f:W→Xf : \mathcal{W} \to X that sends cells in W\mathcal{W} to their appropriate midcells in XX is an isometric embedding. In particular, W\mathcal{W} can be identified with a closed and convex subset of XX.

  1. Under this identification, N12(W)N_{\frac{1}{2}}(\mathcal{W}) convex, and isometric to W×(−12,12)\mathcal{W} \times\left(-\frac{1}{2},\frac{1}{2}\right),

  1. W⊂X\mathcal{W} \subset X separates XX into two convex components, called (open) halfspaces.

In what follows, the wall W\mathcal{W} will always be identified with its image f(W)⊂Xf(\mathcal{W}) \subset X.

Corollary 2.12 ([10 Cor 3.7]). The combinatorial distance (denoted by d1d_1) between vertices v,w∈X(0)v,w \in X^{(0)} is equal to the number of walls in S(v,w)S(v,w), where S(v,w)S(v,w) is the set of walls separating vv from ww.

The carrier of a wall W\mathcal{W}, denoted by N(W)N(\mathcal{W}) is defined to be the union of all cells in XX intersecting W\mathcal{W}. In contrast to cube complexes, the carrier is in general distinct from N12(W)N_{\frac{1}{2}}(\mathcal{W}), and may fail to be convex both with the CAT(0) metric and the combinatorial metric on X(1)X^{(1)} (Figure 3). However, the inclusion N(W)(1)↪X(1)N(\mathcal{W})^{(1)} \hookrightarrow X^{(1)} remains an isometric embedding [10 Lem 3.10].

Figure 3. N(W)N(\mathcal{W}) fails to be convex in both the CAT(0) metric and the combinatorial metric.

Many of the properties enjoyed by walls are inherited by intersections of walls.

Definition 2.13 (c.f. [9 Def 2.5]). Let XX be a CAT(0) Coxeter cell complex. A codimension-kk wall Wk\mathcal{W}^k is an intersection of kk walls W1,…,Wk\mathcal{W}_1,\ldots,\mathcal{W}_k in XX such that for any cell ZZ intersecting Wk\mathcal{W}^k, Wk∩Z\mathcal{W}^k \cap Z is a codimension-kk subspace of ZZ.

In particular, a codim-1 wall is a wall as previously defined. We will also consider XX to be a codim-0 wall. Note that if W\mathcal{W} is a codim-kk wall, then W\mathcal{W} is still closed and convex in XX and N12(W)N_{\frac{1}{2}}(\mathcal{W}) is isometric to W×Bk\mathcal{W} \times B^k, where BkB^k is the ball of radius 12\frac{1}{2} in Rk\mathbb{R}^k.

Lemma 2.14. Let Wk\mathcal{W}^k be a codim-kk wall, and V\mathcal{V} a wall such that U=Wk∩V\mathcal{U} = \mathcal{W}^k \cap\mathcal{V} is a non-empty, proper subset of Wk\mathcal{W}^k. Then U\mathcal{U} is a codim-(k+1)(k+1) wall, and separates Wk\mathcal{W}^k into two connected components.

Proof. Let V±\mathcal{V}^{\pm} be the two halfspaces of V\mathcal{V}. Let ZZ be a cell whose interior intersects U\mathcal{U}. Then (Wk∩V)∩Z⊊Wk∩Z(\mathcal{W}^k \cap\mathcal{V}) \cap Z \subsetneq\mathcal{W}^k \cap Z, so U\mathcal{U} is codim-(k+1)(k+1) in ZZ and Wk∩V±\mathcal{W}^k \cap\mathcal{V}^{\pm} are non-empty. Since V±\mathcal{V}^{\pm} and Wk\mathcal{W}^k are convex, the sets V±∩Wk\mathcal{V}^{\pm} \cap\mathcal{W}^k are also convex and therefore connected. It follows that V∩Wk\mathcal{V} \cap\mathcal{W}^k separates Wk\mathcal{W}^k into two components.

Cubulating Coxeter cell complexes

We briefly summarize the description of Sageev’s construction by Hruska and Wise in [9], which in our setting associates to each CAT(0) Coxeter cell complex XX a connected, CAT(0) cube complex C(X)C(X):

Definition 3.1 (c.f. [7 Def 2.3, 2.6]). Let W\mathcal{W} be a (possibly infinite) set of walls in XX. A coorientation of walls in W\mathcal{W} is an map σ\sigma that assigns to each wall W∈WW \in\mathcal{W} one of its halfspaces σ(W)\sigma(W). A coorientation is consistent if σ(W1)∩σ(W2)≠∅\sigma(W_1) \cap\sigma(W_2) \ne\varnothing for any pair W1,W2∈WW_1,W_2 \in\mathcal{W}. We say a coorientation satisfies the Descending Chain Condition (DCC) if every sequence of halfspaces σ(W1)⊇σ(W2)⊇⋯\sigma(W_1) \supseteq\sigma(W_2) \supseteq\cdots with Wk∈WW_k \in\mathcal{W} has a minimal halfspace σ(Wn)\sigma(W_n).

The vertex set of C(X)C(X) is the set of all consistent coorientations of walls in XX satisfying the DCC. We put an edge between two coorientations if and only if the two maps differ on a single wall of XX. Finally, vertices in C(X)C(X) span a cube if and only if they span the 1-skeleton of a cube in C(X)(1)C(X)^{(1)}. There is a natural bijection between the hyperplanes of C(X)C(X) and the walls in XX. By Corollary 2.12, we have an isometric embedding X(1)⊂C(X)(1)X^{(1)} \subset C(X)^{(1)}.

Observe that the edges in C(X)C(X) incident to a vertex x∈X(0)⊂C(X)(0)x \in X^{(0)} \subset C(X)^{(0)} are in 1-1 correspondence with the set of walls that are not separated from xx by any other wall in XX. We call this set the set of xx-elementary walls, and denote it by Ex\mathcal{E}_x. If XX is locally finite and satisfies the PW Property, then this set is finite for any x∈X(0)x \in X^{(0)}. However, this does not a priori give us local finiteness for all of C(X)C(X). More precisely, the following is a special case of [9 Thm 5.7] (see also [12 Thm 3]):

Theorem 3.2. Suppose XX is locally finite. Then C(X)C(X) is locally finite if and only if XX has compact-wall separation: there is a constant f(K)f(K) for each compact K⊂XK \subset X such that any wall at distance ≥f(K)\ge f(K) from a compact subset KK is separated from KK by another wall.

Under the assumption that XX is equipped with a geometric group action, it can be deduced from [8 Prop 3.7, Thm B] that the PW Property implies compact-wall separation. In general, we will need the following lemma:

Lemma 3.3. Suppose that XX satisfies the PW Property. Then there is a constant C0C_0 such that for any vertices x,y∈Xx,y \in X and W∈ExW \in\mathcal{E}_x separating xx from yy, there is a vertex zz lying on a combinatorial geodesic from xx to yy separated from xx by WW at combinatorial distance ≤C0\le C_0 from xx.

Proof. The proof is identical to that of [13 Lem 4.3]. We reformulate their proof in the language of CAT(0) Coxeter cell complexes. Let z∈X(1)z \in X^{(1)} be a vertex as desired at minimal combinatorial distance from xx. Note that by minimality of zz, WW is the only wall adjacent to zz separating xx from zz.

Since XX is quasi-isometric to X(1)X^{(1)}, it suffices to uniformly bound the CAT(0) distance d(z,x)d(z,x). We use the following quantities to do so:

  • Let α\alpha be the minimal angle formed by a geodesic [m,m′][m,m'] in the midcell of a cell Z⊂XZ \subset X and a geodesic connecting mm to a vertex of ZZ.

  • Let C1C_1 be the maximal distance between a vertex in a cell ZZ and a point in a midcell of ZZ.

  • Let C2C_2 be the uniform bound for d(x,πW(x))d(x,\pi_{\mathcal{W}}(x)) guaranteed by the PW Property.

Now, let mm be the intersection of the CAT(0) geodesic [z,x][z,x] with W\mathcal{W}. Since W\mathcal{W} is the unique wall adjacent to zz intersecting [z,x][z,x], the segment [z,m][z,m] must be contained in a cell ZZ (see Figure 4). Let ZZ be the unique such cell containing mm in its interior. Then we see that ∠m(x,πW(x))≥α\angle_m(x,\pi_{\mathcal{W}}(x)) \geq\alpha, and so by the CAT(0) inequality,

Figure 4. The geodesic [z,x][z,x] passes through the union of cells in N(W)\mathcal{N}(\mathcal{W}) containing zz.

d(z,x)≤C1+d(m,x)≤C1+C2sin⁡α,d(z,x) \leq C_1 + d(m,x) \leq C_1 + C_2\sin\alpha,

giving us the desired bound.

Proposition 3.4. Suppose XX satisfies the PW Property. Then there are constants (Cn)n∈N(C_n)_{n\in\mathbb{N}} such that any vertex x∈Xx\in X at combinatorial distance ≥Cn\geq C_n from a wall W\mathcal{W} is separated from W\mathcal{W} by at least nn distinct walls.

Proof. We claim that we can take Cn=nmax⁡{C,C0}C_n=n\max\{C,C_0\}, where CC and C0C_0 are the constants from the PW Property and Lemma 3.3 respectively. Indeed, suppose that xx is at distance ≥Cn\geq C_n from a wall W\mathcal{W}. By the PW Property, there is an xx-elementary wall W′\mathcal{W}' separating xx from W\mathcal{W}. By Lemma 3.3, there is a vertex x′x' lying on a combinatorial geodesic from xx to W\mathcal{W} of length d(x,W)d(x,\mathcal{W}) such that W′\mathcal{W}' separates x′x' from xx, and d(x′,W)≥Cn−C0≥Cn−1d(x',\mathcal{W})\geq C_n-C_0\geq C_{n-1}. By induction, x′x' is separated from W\mathcal{W} by at least n−1n-1 walls. But such walls must also separate xx from W\mathcal{W}, and so xx is separated from W\mathcal{W} by at least nn walls.

Corollary 3.5. If XX is locally finite and satisfies the PW Property, then C(X)C(X) is locally finite.

Proof. If XX is locally finite, then the number of walls that can intersect a given compact set is also bounded. Therefore, Proposition 3.4 implies that XX has ball-wall separation (i.e. compact-wall separation, where the value of f(K)f(K) depends only on diam⁡(K)\operatorname{diam}(K)). Thus C(X)C(X) is locally finite by Theorem 3.2.

Angles between walls

In analogy to Coxeter groups, we have a well-defined notion of angle between two intersecting codim-kk walls W1\mathcal{W}_1 and W2\mathcal{W}_2, provided they are contained in a common codim-(k−1)(k-1) wall V\mathcal{V}. Fix first a cell ZZ of V\mathcal{V} whose interior intersects W1∩W2\mathcal{W}_1 \cap\mathcal{W}_2. Then Z∩W1Z \cap\mathcal{W}_1 and Z∩W2Z \cap\mathcal{W}_2 correspond to Euclidean hyperplanes in ZZ intersecting at the center of CC. There are now two choices of normal vectors for each hyperplane which, by Lemma 2.14, are determined by a coorientation σ\sigma of W1\mathcal{W}_1 and W2\mathcal{W}_2 in V\mathcal{V}. The angle between W1\mathcal{W}_1 and W2\mathcal{W}_2 in V\mathcal{V} with respect to σ\sigma is π\pi minus the angle between the associated normal vectors in ZZ, and is denoted by ∠V,σ(W1,W2)\angle_{\mathcal{V},\sigma}(\mathcal{W}_1,\mathcal{W}_2). Since this quantity does not change when restricting to a face of ZZ containing W1∩W2\mathcal{W}_1 \cap\mathcal{W}_2, the angle ∠V,σ(W1,W2)\angle_{\mathcal{V},\sigma}(\mathcal{W}_1,\mathcal{W}_2) does not depend on our choice of ZZ.

Remark 4.1. If W1\mathcal{W}_1, W2\mathcal{W}_2 and W3\mathcal{W}_3 are codim-kk-walls contained in a common codim-(k−1)(k-1) wall V\mathcal{V} and W1∩W2∩W3\mathcal{W}_1 \cap\mathcal{W}_2 \cap\mathcal{W}_3 is a codim-(k+2)(k+2) wall, then the angle between W1∩W2\mathcal{W}_1 \cap\mathcal{W}_2 and W1∩W3\mathcal{W}_1 \cap\mathcal{W}_3 can be written in terms of the angles between W1\mathcal{W}_1, W2\mathcal{W}_2 and W3\mathcal{W}_3. Indeed, let θij=∠V,σ(Wi,Wj)\theta_{ij} = \angle_{\mathcal{V},\sigma}(\mathcal{W}_i,\mathcal{W}_j) and let α=∠W1,σ(W1∩W2,W1∩W3)\alpha= \angle_{\mathcal{W}_1,\sigma}(\mathcal{W}_1 \cap\mathcal{W}_2,\mathcal{W}_1 \cap\mathcal{W}_3). Then by the spherical law of cosines,

cos⁡(α)=cos⁡θ23−cos⁡θ12cos⁡θ13sin⁡θ12sin⁡θ13\cos(\alpha) = \frac{\cos\theta_{23} - \cos\theta_{12}\cos\theta_{13}}{\sin\theta_{12}\sin\theta_{13}}

As such, if XX has finitely many types of Coxeter cells, the set of values taken by angles between codim-kk walls is finite.

Remark 4.2. Suppose that XX is the Davis complex of a Coxeter group WW and that Wr,Wq\mathcal{W}_r,\mathcal{W}_q are intersecting walls fixed by reflections r,q∈Wr,q \in W. Let σ\sigma be the coorientation of the Wr\mathcal{W}_r and Wq\mathcal{W}_q containing the identity in WW. Then cos⁡(∠X,σ(Wr,Wq))=−⟨αr,αq⟩\cos(\angle_{X,\sigma}(\mathcal{W}_r,\mathcal{W}_q)) = -\langle\alpha_r,\alpha_q\rangle, where αr,αq\alpha_r,\alpha_q are the positive roots associated to rr and qq in the Tits representation of WW.

Proposition 4.3. Let W1\mathcal{W}_1, W2\mathcal{W}_2 be intersecting codim-kk walls lying in a codim-(k−1)(k-1) wall V\mathcal{V}, and let γ1=[x,z]\gamma_1 = [x,z], γ2=[z,y]\gamma_2 = [z,y] be geodesics in W1,W2\mathcal{W}_1,\mathcal{W}_2 respectively, such that the union [x,z]∪[z,y][x,z] \cup[z,y] is a geodesic in W1∪W2\mathcal{W}_1 \cup\mathcal{W}_2, with the intrinsic path metric. Then

∠(γ1,γ2)≥∠V,σ(W1,W2),\angle(\gamma_1,\gamma_2) \geq\angle_{\mathcal{V},\sigma}(\mathcal{W}_1,\mathcal{W}_2),

where σ\sigma is the coorientation such that σ(W2)\sigma(\mathcal{W}_2) contains γ1\gamma_1 and σ(W1)\sigma(\mathcal{W}_1) contains γ2\gamma_2.

We first prove Proposition 4.3 for arbitrary hyperplanes in R3\mathbb{R}^{3}. Here, we reuse the notation of ∠R3,σ\angle_{\mathbb{R}^{3},\sigma} to refer to the angle between cooriented hyperplanes in R3\mathbb{R}^{3}.

Lemma 4.4. Let H1,H2∈R3H_1,H_2 \in\mathbb{R}^{3} be distinct intersecting hyperplanes and let σ\sigma be a coorientation of H1H_1 and H2H_2. Let x∈H1∩σ(H2)x \in H_1 \cap\sigma(H_2), y∈H2∩σ(H1)y \in H_2 \cap\sigma(H_1), and let z∈H1∩H2z \in H_1 \cap H_2 minimize d(x,z)+d(z,y)d(x,z) + d(z,y). Then ∠z(x,y)≥∠R3,σ(H1,H2)\angle_z(x,y) \geq\angle_{\mathbb{R}^{3},\sigma}(H_1,H_2).

Proof. Assume without loss of generality that z=(0,0,0)z = (0,0,0), that ∥x∥=∥y∥\lVert x\rVert= \lVert y\rVert, and that H∩H′H \cap H' is the ZZ-axis. Let π(x)\pi(x) and π(y)\pi(y) be the projections of xx and yy onto the XYXY-plane. Note that

Proof of Lemma 4.4

Figure 5. Proof of Lemma 4.4

∠z(π(x),π(y))=∠R3,σ(H1,H2)=:α\angle_z(\pi(x),\pi(y))=\angle_{\mathbb{R}^3,\sigma}(H_1,H_2)=:\alpha. Write x=π(x)+h(0,0,1)x=\pi(x)+h(0,0,1) and y=π(y)+h′(0,0,1)y=\pi(y)+h'(0,0,1). By our assumption, the function

f(t)=∥x−t(0,0,1)∥+∥y−t(0,0,1)∥f(t)=\lVert x-t(0,0,1)\rVert+\lVert y-t(0,0,1)\rVert

is minimized at t=0t=0. Thus we have 0=f′(0)=h∥x∥+h′∥y∥0=f'(0)=\frac{h}{\lVert x\rVert}+\frac{h'}{\lVert y\rVert}, and so h′=−hh'=-h. Now, letting λ=∥π(x)∥=∥π(y)∥\lambda=\lVert\pi(x)\rVert=\lVert\pi(y)\rVert and β=∠z(x,y)\beta=\angle_z(x,y), we see that

∥π(x)−π(y)∥2=2λ2−2λ2cos⁡αand∥x−y∥2=2(λ2+h2)−2(λ2+h2)cos⁡β\lVert\pi(x)-\pi(y)\rVert^2=2\lambda^2-2\lambda^2\cos\alpha \qquad\text{and}\qquad \lVert x-y\rVert^2=2(\lambda^2+h^2)-2(\lambda^2+h^2)\cos\beta

Additionally, the Pythagorean theorem gives us the relation ∥x−y∥2=∥π(x)−π(y)∥2+4h2\lVert x-y\rVert^2=\lVert\pi(x)-\pi(y)\rVert^2+4h^2.

Combining the three equalities above gives us that

cos⁡β=λ2cos⁡α−h2λ2+h2\cos\beta=\frac{\lambda^2\cos\alpha-h^2}{\lambda^2+h^2}

and so cos⁡β−cos⁡α=h2(1−cos⁡α)λ2+h2≥0\cos\beta-\cos\alpha=\frac{h^2(1-\cos\alpha)}{\lambda^2+h^2}\geq0.

Proof of Proposition 4.3. Set Y=W1∩W2Y=W_1\cap W_2, so that z∈Yz\in Y, and assume without loss of generality that x,yx,y are contained in W1∩N12(W2)W_1\cap N_{\frac{1}{2}}(W_2) and W2∩N12(W1)W_2\cap N_{\frac{1}{2}}(W_1) respectively. Note that we have an isometric embedding i:N12(Y)↪Y×R2i:N_{\frac{1}{2}}(Y)\hookrightarrow Y\times\mathbb{R}^2, under which Wi∩N12(Wj)W_i\cap N_{\frac{1}{2}}(W_j) gets mapped into Y×LiY\times L_i for some distinct intersecting lines L1L_1 and L2L_2 in R2\mathbb{R}^2. Since the projection of a geodesic onto a factor is itself geodesic, J:=πY([x,z]∪[z,y])J:=\pi_Y([x,z]\cup[z,y]) is isometric to either a line segment or a point. This gives us a further isometric embedding of J×R2↪R×R2J\times\mathbb{R}^2\hookrightarrow\mathbb{R}\times\mathbb{R}^2, under which i([x,z])i([x,z]) and i([z,y])i([z,y]) map into H1=R×L1H_1=\mathbb{R}\times L_1 and H2=R×L2H_2=\mathbb{R}\times L_2 respectively. Now, H1H_1 and H2H_2 are hyperplanes in R3\mathbb{R}^3, with coorientations determined by σ\sigma, and ∠R3,σ(H1,H2)=∠V,σ(W1,W2)\angle_{\mathbb{R}^3,\sigma}(H_1,H_2)=\angle_{V,\sigma}(W_1,W_2). Therefore, by Lemma 4.4 we have that ∠z(x,y)≥∠R3,σ(H1,H2)=∠V,σ(W1,W2)\angle_z(x,y)\geq\angle_{\mathbb{R}^3,\sigma}(H_1,H_2)=\angle_{V,\sigma}(W_1,W_2).

Corollary 4.5 (CAT(0) inequality for walls). Suppose that W1\mathcal{W}_1, W2\mathcal{W}_2, and W3\mathcal{W}_3 are three pairwise intersecting codim-kk walls contained in a codim-(k−1)(k-1) wall V\mathcal{V}. Suppose further that the walls have empty global intersection, and let σ\sigma be the unique coorientation of walls such that σ(Wi)⊃Wj∩Wℓ\sigma(\mathcal{W}_i) \supset\mathcal{W}_j \cap\mathcal{W}_\ell for i,j,ki,j,k distinct. Then

∠V,σ(W1,W2)+∠V,σ(W1,W3)+∠V,σ(W2,W3)≤π\angle_{\mathcal{V},\sigma}(\mathcal{W}_1,\mathcal{W}_2)+\angle_{\mathcal{V},\sigma}(\mathcal{W}_1,\mathcal{W}_3)+\angle_{\mathcal{V},\sigma}(\mathcal{W}_2,\mathcal{W}_3)\le\pi

Proof. Let Y⊂VY\subset\mathcal{V} be the union W1∪W2∪W3\mathcal{W}_1\cup\mathcal{W}_2\cup\mathcal{W}_3, endowed with the intrinsic path metric. By [2 Thm II.11.1], YY is non-positively curved. Also, YY has fundamental group Z\mathbb{Z}, generated by geodesic triangles with vertices in Wi∩Wj\mathcal{W}_i\cap\mathcal{W}_j. By [3], the action of Z\mathbb{Z} on the universal cover Y~\widetilde{Y} is hyperbolic, and so there is a geodesic line in Y~\widetilde{Y} fixed by Z\mathbb{Z}. Projecting this line onto YY yields a geodesic triangle Δ\Delta with edges γi⊂Wi\gamma_i\subset\mathcal{W}_i. Since Y~→Y\widetilde{Y}\to Y is a local isometry, γi∪γj\gamma_i\cup\gamma_j is geodesic in Wi∪Wj\mathcal{W}_i\cup\mathcal{W}_j for each 1≤i,j≤31\le i,j\le3. By Proposition 4.3, we thus have that ∠V,σ(Wi,Wj)≤∠(γi,γj)\angle_{\mathcal{V},\sigma}(\mathcal{W}_i,\mathcal{W}_j)\le\angle(\gamma_i,\gamma_j). Since V\mathcal{V} is CAT(0), Proposition 2.2 yields

∑i<j∠V,σ(Wi,Wj)≤∑i<j∠(γi,γj)≤π\sum_{i<j}\angle_{\mathcal{V},\sigma}(\mathcal{W}_i,\mathcal{W}_j)\le\sum_{i<j}\angle(\gamma_i,\gamma_j)\le\pi

The angles between walls provide a method to easily calculate the combinatorial distance between a vertex and a given wall, generalizing [1 Lem 1.7]:

Proposition 4.6. Let u,v∈X(0)u,v\in X^{(0)} be adjacent vertices separated by a wall W′\mathcal{W}', and let W\mathcal{W} be a wall intersecting W′\mathcal{W}'. Take σ\sigma to be the coorientation of W\mathcal{W} and W′\mathcal{W}' containing uu. Then

d1(u,W)={d1(v,W)+1if ∠X,σ(W,W′)>π/2d1(v,W)if ∠X,σ(W,W′)=π/2d1(v,W)−1if ∠X,σ(W,W′)<π/2d_1(u,\mathcal{W})= \begin{cases} d_1(v,\mathcal{W})+1 & \text{if }\angle_{X,\sigma}(\mathcal{W},\mathcal{W}')>\pi/2\\ d_1(v,\mathcal{W}) & \text{if }\angle_{X,\sigma}(\mathcal{W},\mathcal{W}')=\pi/2\\ d_1(v,\mathcal{W})-1 & \text{if }\angle_{X,\sigma}(\mathcal{W},\mathcal{W}')<\pi/2 \end{cases}

The idea of the proof of Proposition 4.6 is to show that a vertex of W\mathcal{W} close to the CAT(0) projection πW(u)\pi_{\mathcal{W}}(u) minimizes d1(u,W)d_1(u,\mathcal{W}). More precisely, we choose a particular vertex pp lying in the same closed halfspaces as πW(m)\pi_{\mathcal{W}}(m), where mm is the midpoint of the edge [u,v][u,v]. Note that there are many choices for such pp, specified by the choice of halfspaces of the walls containing πW(m)\pi_{\mathcal{W}}(m). We then use Proposition 4.3 to show that the halfspace of W′\mathcal{W}' containing πW(m)\pi_{\mathcal{W}}(m) is determined by ∠X,σ(W,W′)\angle_{X,\sigma}(\mathcal{W},\mathcal{W}') (Lemma 4.8). Finally, to compare d1(u,p)d_1(u,p) with d1(u,W)d_1(u,\mathcal{W}), we introduce the notion of a reflection and show that each wall separating uu from pp has a reflection that does not separate uu from pp, except possibly for W′\mathcal{W}' in the case where W′\mathcal{W}' is perpendicular to W\mathcal{W}.

Definition 4.7. Suppose that V\mathcal{V} and W\mathcal{W} are intersecting walls in a CAT(0) Coxeter cell complex XX. Let ZΓZ_\Gamma be a cell whose interior intersects V∩W\mathcal{V}\cap\mathcal{W}. Then the intersection of V\mathcal{V} and W\mathcal{W} with ZΓZ_\Gamma are midcells in ZΓZ_\Gamma. Thus W∩ZΓ\mathcal{W}\cap Z_\Gamma is the fixed point set of a (unique) reflection r∈WΓr\in W_\Gamma. The midcell r⋅(V∩ZΓ)r\cdot(\mathcal{V}\cap Z_\Gamma) is the cell of a new wall V‾\overline{\mathcal{V}} in XX called the reflection of V\mathcal{V} along W\mathcal{W}. One may check that the wall obtained through this procedure does not depend on our choice of ZZ.

Lemma 4.8. Let x∈Xx\in X, let W\mathcal{W} be a wall that does not contain xx, and let pp be a vertex of W\mathcal{W} lying in all the same closed halfspaces as πW(x)\pi_{\mathcal{W}}(x). Suppose V\mathcal{V} is a wall intersecting W\mathcal{W} that either

Proof of Lemma 4.8

Figure 6. Proof of Lemma 4.8

  1. separates xx from pp, or

  1. contains xx but not πW(x)\pi_{\mathcal{W}}(x).

Then ∠X,σ(V,W)<π/2\angle_{X,\sigma}(\mathcal{V},\mathcal{W}) < \pi/2, where σ(W)\sigma(\mathcal{W}) is chosen to contain xx and σ(V)\sigma(\mathcal{V}) is chosen to contain pp. Furthermore, the reflection V‾\overline{\mathcal{V}} of V\mathcal{V} along W\mathcal{W} does not satisfy either of the conditions (1) and (2) above.

Proof. We distinguish two cases. If V\mathcal{V} does not contain πW(x)\pi_{\mathcal{W}}(x), then the segment [x,πW(x)][x,\pi_{\mathcal{W}}(x)] intersects V\mathcal{V} at some point x′x' distinct from πW(x)\pi_{\mathcal{W}}(x). By the CAT(0) inequality for angles (Proposition 2.2), ∠z(x′,πW(x))<π/2\angle_z(x',\pi_{\mathcal{W}}(x)) < \pi/2 for any z∈V∩Wz \in\mathcal{V} \cap\mathcal{W}. Picking z∈V∩Wz \in\mathcal{V} \cap\mathcal{W} lying on the geodesic between x′x' and πW(x)\pi_{\mathcal{W}}(x) in V∪W\mathcal{V} \cup\mathcal{W} (with the intrinsic path metric on V∪W\mathcal{V} \cup\mathcal{W}) and applying Proposition 4.3 then gives us that ∠X,σ(V,W)≤∠z(x′,πW(x))<π/2\angle_{X,\sigma}(\mathcal{V},\mathcal{W}) \le\angle_z(x',\pi_{\mathcal{W}}(x)) < \pi/2. Additionally, pp and πW(x)\pi_{\mathcal{W}}(x) lie in the same halfspace of V∩W=V‾∩W\mathcal{V} \cap\mathcal{W} = \overline{\mathcal{V}} \cap\mathcal{W} (as a codim-2 wall in W\mathcal{W}), and therefore they lie in the same halfspace of V‾\overline{\mathcal{V}}. So, if V‾\overline{\mathcal{V}} satisfied (1) or (2), then after choosing σ(V‾)\sigma(\overline{\mathcal{V}}) to be the halfspace containing pp, we would have that ∠X,σ(V‾,W)=π−∠X,σ(V,W)<π/2\angle_{X,\sigma}(\overline{\mathcal{V}},\mathcal{W}) = \pi- \angle_{X,\sigma}(\mathcal{V},\mathcal{W}) < \pi/2, which is a contradiction.

Now, if V\mathcal{V} contains πW(x)\pi_{\mathcal{W}}(x), then πW(x)\pi_{\mathcal{W}}(x) is contained in the interior of some cell ZZ, and pp is the midpoint of an edge of ZZ. We may assume without loss of generality that xx lies in ZZ. Choose y∈σ(V)∩Wy \in\sigma(\mathcal{V}) \cap\mathcal{W} so that the projection of yy onto V∩W\mathcal{V} \cap\mathcal{W} is πW(x)\pi_{\mathcal{W}}(x). Then V\mathcal{V} separates xx from yy, and so the segment [x,y][x,y] intersects V\mathcal{V} at some point x′x'. But then

∠X,σ(V,W)=∠πW(x)(x′,y)<∠πW(x)(x,y)=π/2\angle_{X,\sigma}(\mathcal{V},\mathcal{W}) = \angle_{\pi_{\mathcal{W}}(x)}(x',y) < \angle_{\pi_{\mathcal{W}}(x)}(x,y) = \pi/2

as desired. A similar argument as in the first case then shows that V‾\overline{\mathcal{V}} does not separate xx from yy, and therefore does not separate xx from pp.

Proof of Proposition 4.6. Suppose that p′∈W∩X(1)p' \in\mathcal{W} \cap X^{(1)} minimizes d1(u,W)d_1(u,\mathcal{W}). Let m∈W′m \in\mathcal{W}' be the midpoint of the edge [u,v][u,v], and let p∈W∩X(1)p \in\mathcal{W} \cap X^{(1)} be the vertex of W\mathcal{W} lying in the same closed halfspaces as πW(m)\pi_{\mathcal{W}}(m) and separated from p′p' by the walls containing πW(m)\pi_{\mathcal{W}}(m).

We claim that if ∠X,σ(W,W′)≠π/2\angle_{X,\sigma}(\mathcal{W},\mathcal{W}') \ne\pi/2, then d1(u,p)=d1(u,W)d_1(u,p) = d_1(u,\mathcal{W}), and d1(u,p)=d1(u,W)+1d_1(u,p) = d_1(u,\mathcal{W}) + 1 otherwise. Observe that to compare d1(u,p)d_1(u,p) with d1(u,W)d_1(u,\mathcal{W}), it suffices to look at the walls in the symmetric difference S(p,p′)=S(u,p)△S(u,p′)S(p,p') = S(u,p) \mathbin{\triangle} S(u,p'). Furthermore, the only wall in S(p,p′)S(p,p') that may contain both mm and πW(m)\pi_{\mathcal{W}}(m) is W′\mathcal{W}' in the special case where ∠X,σ(W,W′)=π/2\angle_{X,\sigma}(\mathcal{W},\mathcal{W}') = \pi/2.

Figure 7. The three cases in Proposition 4.6.

Case 1: ∠X,σ(W,W′)>π/2\angle_{X,\sigma}(\mathcal{W},\mathcal{W}') > \pi/2. If V∈S(p,p′)\mathcal{V} \in S(p,p') separates uu from pp, then it either separates mm from pp or equals W′\mathcal{W}'. By Lemma 4.8, the reflection V‾\overline{\mathcal{V}} of V\mathcal{V} along W\mathcal{W} neither separates mm from pp, nor does it contain mm and not πW(m)\pi_{\mathcal{W}}(m). Consequently, V‾\overline{\mathcal{V}} does not separate uu from pp and since V‾∈S(u,p)△S(u,p′)\overline{\mathcal{V}} \in S(u,p) \mathbin{\triangle} S(u,p'), V‾\overline{\mathcal{V}} separates uu from p′p'. It follows that at least half of the walls in S(p,p′)S(p,p') lie in S(u,p)S(u,p), and thus d1(u,p)≤d1(u,p′)=d1(u,W)d_1(u,p) \le d_1(u,p') = d_1(u,\mathcal{W}), giving us the desired claim.

Now, if σ(W′)\sigma(\mathcal{W}') contained pp, then by Lemma 4.8, we would have ∠X,σ(W,W′)<π/2\angle_{X,\sigma}(\mathcal{W},\mathcal{W}') < \pi/2, a contradiction. Therefore, W′\mathcal{W}' separates pp from uu, and so

d1(u,W)=d1(u,p)=d1(v,p)+1=d1(v,W)+1.d_1(u,\mathcal{W}) = d_1(u,p) = d_1(v,p) + 1 = d_1(v,\mathcal{W}) + 1.

Case 2: ∠X,σ(W,W′)=π/2\angle_{X,\sigma}(\mathcal{W},\mathcal{W}') = \pi/2. Similarly to the previous case, any wall in S(p,p′)∖{W′}S(p,p') \setminus\{\mathcal{W}'\} separating uu from pp has a reflection that separates uu from p′p'. Therefore, d1(u,p)=d1(u,p′)+1d_1(u,p) = d_1(u,p') + 1 if W′\mathcal{W}' separates uu from pp and d1(u,p)=d1(u,p′)−1d_1(u,p) = d_1(u,p') - 1 if W′\mathcal{W}' separates uu from p′p'. But then by minimality of p′p', we must have d1(u,p)=d1(u,p′)+1=d1(u,W)+1d_1(u,p) = d_1(u,p') + 1 = d_1(u,\mathcal{W}) + 1. Therefore, W′\mathcal{W}' separates uu from p′p' and

d1(u,W)=d1(u,p′)=d1(u,p)−1=d1(v,p)≥d1(v,W).d_1(u,\mathcal{W}) = d_1(u,p') = d_1(u,p) - 1 = d_1(v,p) \ge d_1(v,\mathcal{W}).

The other inequality follows from interchanging uu and vv, so d1(u,W)=d1(v,W)d_1(u,\mathcal{W}) = d_1(v,\mathcal{W}).

Finally, the remaining case where ∠X,σ(W,W′)<π/2\angle_{X,\sigma}(\mathcal{W},\mathcal{W}') < \pi/2 follows from the first case by interchanging uu and vv as well.

Bipodality and the Parallel Wall theorem

Let XX be a CAT(0) Coxeter cell complex, and let W,W′\mathcal{W}, \mathcal{W}' be distinct walls in XX sharing a common adjacent vertex vv. The intersection of the halfspaces of W,W′\mathcal{W}, \mathcal{W}' containing vv is called a sector of XX bounded by W\mathcal{W} and W′\mathcal{W}'. A sector is said to be sharp-angled if its bounding walls are not perpendicular to one another.

Definition 5.1. Let x∈X(0)x \in X^{(0)} and let W\mathcal{W} and W′\mathcal{W}' be two intersecting walls. Consider V\mathcal{V}, the set of all walls containing the codim-2 wall W∩W′\mathcal{W} \cap\mathcal{W}', and choose σ(V)\sigma(\mathcal{V}) to be the halfspace of V\mathcal{V} containing xx for every V∈V\mathcal{V} \in\mathcal{V}. Then

Dx(W,W′)=⋂V∈Vσ(V)D_x(\mathcal{W},\mathcal{W}') = \bigcap_{\mathcal{V} \in\mathcal{V}} \sigma(\mathcal{V})

is a sector containing xx, called the canonical sector for W,W′\mathcal{W},\mathcal{W}' containing xx.

Remark 5.2. Note that W,W′\mathcal{W},\mathcal{W}' contain V∩V′\mathcal{V}\cap\mathcal{V}' for all V,V′∈V\mathcal{V},\mathcal{V}'\in\mathscr{V}, and that the size of V\mathscr{V} is bounded by the largest number of sides of a polygon in Shapes⁡(X)\operatorname{Shapes}(X). As such, the number of walls that may pair to give the same canonical sector is bounded by a constant depending only on Shapes⁡(X)\operatorname{Shapes}(X).

Definition 5.3 ([6 Def 4.14]). Fix a basepoint xx. A set of walls W\mathcal{W} is said to be bipodal if it satisfies the following: if W\mathcal{W} and W′\mathcal{W}' bound a sector containing xx, and V≠W,W′\mathcal{V}\ne\mathcal{W},\mathcal{W}' is a wall containing W∩W′\mathcal{W}\cap\mathcal{W}', then V∈W\mathcal{V}\in\mathcal{W} implies that W,W′∈W\mathcal{W},\mathcal{W}'\in\mathcal{W}.

Recall that a wall is xx-elementary if it is not separated from xx by another wall of XX. The definition for bipodality of a CAT(0) Coxeter cell complex XX given in the introduction is equivalent to requiring that the set Ex\mathcal{E}_x of xx-elementary walls is bipodal for any basepoint x∈X(0)x\in X^{(0)}. Note that bipodality does not hold for arbitrary CAT(0) Coxeter cell complexes (see Figure 8). The following is a slightly stronger version of Theorem 1.1:

Figure 8. V\mathcal{V} is xx-elementary, but W\mathcal{W} and W′\mathcal{W}' are not.

Theorem 5.4. Let XX be a CAT(0) Coxeter cell complex satisfying the Bounded Pairwise Intersection Property. Then there is a bound CC, depending only on XX, such that for any x∈X(0)x\in X^{(0)} with Ex\mathcal{E}_x bipodal, the walls in Ex\mathcal{E}_x are at distance ≤C\le C from xx.

Proof. Let x∈X(0)x\in X^{(0)} be such that Ex\mathcal{E}_x is bipodal, and let W\mathcal{W} be a wall. Fix a point p∈W∩X(1)p\in\mathcal{W}\cap X^{(1)} minimizing d1(x,W)d_1(x,\mathcal{W}), and assume that each wall in S(x,p)\mathcal{S}(x,p) intersects W\mathcal{W}. Then for each wall in V∈S(x,p)\mathcal{V}\in\mathcal{S}(x,p), we may consider the canonical sector containing xx for the pair of walls V\mathcal{V} and W\mathcal{W}. Since V\mathcal{V} is not perpendicular to W\mathcal{W} by Proposition 4.6, the sectors we obtain are sharp-angled, and their bounding walls are distinct from W\mathcal{W}. By Remark 5.2, the number of distinct sectors obtained this way increases as the size of S(x,p)\mathcal{S}(x,p) increases. Enumerate these sectors as D1,…,DrD_1,\ldots,D_r. Note that one of the walls bounding DiD_i, say Vi\mathcal{V}_i, separates xx from pp.

Let NN be the bound on the number of pairwise intersecting walls in XX, and choose C≥0C\ge0 so that r>Nr>N whenever #S(x,p)>C\#\mathcal{S}(x,p)>C. Then there is a pair of disjoint walls Vi\mathcal{V}_i and Vj\mathcal{V}_j such that Vi\mathcal{V}_i separates xx from Vj\mathcal{V}_j. But then Vj∉Ex\mathcal{V}_j\notin\mathcal{E}_x, and since W\mathcal{W} contains the intersection of the two walls bounding DjD_j, we obtain under the assumption of bipodality that W∉Ex\mathcal{W}\notin\mathcal{E}_x. We therefore conclude that W∉Ex\mathcal{W}\notin\mathcal{E}_x for any wall W\mathcal{W} with d1(x,W)>Cd_1(x,\mathcal{W})>C.

References

  1. [1][BH93] Brigitte Brink and Robert B. Howlett, A finiteness property and an automatic structure for Coxeter groups, Math. Ann. 296 (1993), no. 1, 179–190.
  2. [2][BH99] Martin R. Bridson and André Haefliger, Metric spaces of non-positive curvature, Grundlehren der mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences], vol. 319, Springer-Verlag, Berlin, 1999.DOI
  3. [3][Bri99] Martin R. Bridson, On the semisimplicity of polyhedral isometries, Proc. Amer. Math. Soc. 127 (1999), no. 7, 2143–2146.DOI
  4. [4][Cap06] Pierre-Emmanuel Caprace, Conjugacy of 2-spherical subgroups of Coxeter groups and parallel walls, Algebr. Geom. Topol. 6 (2006), 1987–2029.
  5. [5][Dav08] Michael W. Davis, The geometry and topology of Coxeter groups, London Mathematical Society Monographs Series, vol. 32, Princeton University Press, Princeton, NJ, 2008.
  6. [6][DH16] Matthew Dyer and Christophe Hohlweg, Small roots, low elements, and the weak order in Coxeter groups, Adv. Math. 301 (2016), 739–784.arxiv.org/abs/1505.02058
  7. [7][Gur05] Dan P. Guralnik, Coarse decompositions for boundaries of CAT(0) groups, PhD thesis, Haifa, 2005. Available at arXiv: 0611006.arxiv.org/abs/math/0611006
  8. [8][Gur06] ———, Local finiteness of cubulations and CAT(0) groups (2006), available at arXiv:0610950v2.
  9. [9][HW14] G. C. Hruska and Daniel T. Wise, Finiteness properties of cubulated groups, Compos. Math. 150 (2014), no. 3, 453–506.DOI
  10. [10][KT25] Carl Kristof-Tessier, The parallel wall theorem for CAT(0) even 2-complexes (2025), available at arXiv:2510.07448.
  11. [11][Mou88] Gabor Moussong, Hyperbolic Coxeter groups, ProQuest LLC, Ann Arbor, MI, 1988. Thesis (Ph.D.)–The Ohio State University.
  12. [12][NR03] G. A. Niblo and L. D. Reeves, Coxeter groups act on CAT(0) cube complexes, J. Group Theory 6 (2003), no. 3, 399–413.DOI
  13. [13][OP25] Damian Osajda and Piotr Przytycki, Coxeter groups are biautomatic, Invent. Math. 242 (2025), no. 3, 627–637.arxiv.org/abs/2206.07804
  14. [14][Sag23] Sage Developers, Sagemath, the Sage Mathematics Software System (Version 10.2), 2023. https://www.sagemath.org.
  15. [15][SW05] Micah Sageev and Daniel T. Wise, The Tits alternative for CAT(0) cubical complexes, Bull. London Math. Soc. 37 (2005), no. 5, 706–710.DOI

Paper details

vlnc:W_e2e3qx4h
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

None listed by the authors

View statement
Looking for related papers…
Contents