In mathematics, a Coxeter group is an abstract group defined by a presentation consisting of a finite set of generators S = \{s_1, \dots, s_n\}, each of order 2 (i.e., s_i^2 = 1), together with relations (s_i s_j)^{m_{ij}} = 1 for i \neq j, where the m_{ij} are integers greater than or equal to 2 (or infinity, indicating no relation beyond the generators' orders), encoded in a symmetric Coxeter matrix M = (m_{ij}) with m_{ii} = 1.[1][2] These relations arise combinatorially from a Coxeter diagram, a graph where vertices represent generators and edges are labeled by m_{ij} \geq 3 (unlabeled edges imply m_{ij} = 3).[3]Named after the Canadian geometer H.S.M. Coxeter (1907–2003), who systematized their study in the 1930s through work on reflection groups and regular polytopes, Coxeter groups provide a uniform algebraic framework for groups generated by reflections in Euclidean, spherical, or hyperbolic spaces.[2] Every finite Coxeter group is isomorphic to a finite reflection group acting faithfully on Euclidean space, and their classification—into irreducible types A_n (n \geq 1), B_n (n \geq 2), D_n (n \geq 4), E_6, E_7, E_8, F_4, G_2, H_3, H_4, and I_2(m) (m \geq 3)—corresponds to the connected Coxeter-Dynkin diagrams and determines the symmetry groups of regular polytopes and tessellations.[3] Infinite families include affine Coxeter groups, which tile Euclidean space, while hyperbolic ones generate discrete subgroups acting on hyperbolic space.[2]Coxeter groups are foundational across mathematics, serving as Weyl groups in the classification of semisimple Lie algebras and simple algebraic groups, where root systems and positive systems align with their reflection representations.[2] They appear in combinatorics through enumerative properties like the length function and Bruhat order, in topology via CAT(0) spaces and buildings (developed further by Jacques Tits in the 1960s), and in algebraic geometry for resolving surface singularities and studying hypergeometric functions.[3][2] Their geometric realizations as isometry groups underscore applications in symmetry analysis, from classical polyhedral geometry to modern areas like cluster algebras and representation theory.[1]
Definition and Formalism
Coxeter Presentation
A Coxeter group is defined algebraically as a group W generated by a finite set S = \{s_1, s_2, \dots, s_n\} of involutions, satisfying the relations s_i^2 = 1 for all i and (s_i s_j)^{m_{ij}} = 1 for i \neq j, where each m_{ij} is either an integer greater than or equal to 2 or infinity (indicating no relation beyond the free product).[4] This presentation captures the essential structure through these braid-like relations, which dictate the orders of products of distinct generators.[4]The pair (W, S) forms a Coxeter system, where S serves as a distinguished generating set consisting of involutions, and the relations are precisely those implied by the m_{ij}. Any group admitting such a presentation is a Coxeter group, with S playing the role of a set of "reflections" in the abstract sense, though the geometric interpretation is not part of this algebraic definition.[5]This framework was introduced by H.S.M. Coxeter in 1931 as an abstraction of geometric reflection groups, allowing the study of their combinatorial and algebraic properties independently of embeddings in vector spaces.[6][5]To illustrate, consider the symmetric group S_n, which arises as a Coxeter group with generators corresponding to adjacent transpositions s_i = (i \ i+1), each of order 2. The relations follow from permutation properties: (s_i s_j)^2 = 1 if |i - j| > 1 (commuting transpositions), and (s_i s_{i+1})^3 = 1 if adjacent (the braid relation for order 3), generalizing to longer braids that encode all permutations via reduced decompositions.[4] This example demonstrates how the Coxeter presentation extends the familiar structure of permutation groups to broader classes via parameterized relations.[4]
Coxeter Matrix and Diagram
The Coxeter matrix of a Coxeter system (W, S), where S is a finite set of generators, is the symmetric |S| \times |S| matrix M = (m_{st}) with entries in \{1, 2, \dots, \infty\} such that m_{ss} = 1 for all s \in S and m_{st} = m_{ts} \geq 2 (or \infty) for distinct s, t \in S.[7][8] This matrix encodes the defining relations of W, specifically (st)^{m_{st}} = e for s \neq t when m_{st} is finite, providing a combinatorial specification of the group's structure beyond the abstract presentation.[7]Associated with the Coxeter matrix is the Schläfli matrix (also known as the bilinear form matrix or cosine matrix) B = (c_{st}), a symmetric matrix with c_{ss} = 1 and c_{st} = -\cos(\pi / m_{st}) for s \neq t (where \cos(\pi / \infty) = -1).[7][8] This matrix links the combinatorial data to geometric interpretations, such as the dihedral angles between reflection hyperplanes in faithful representations, though its primary role here is as an algebraic tool derived from M.[8]The Coxeter-Dynkin diagram (or Coxeter diagram) is an undirected graph with vertex set S, where vertices s and t are connected by an edge if m_{st} \geq 3; no edge appears if m_{st} = 2, while edges are labeled by m_{st} if m_{st} > 3 (with conventional shortcuts like double bonds for m_{st} = 4 and triple bonds for m_{st} = 6).[7][8] If the diagram is disconnected, the corresponding Coxeter group is the direct product of the groups associated to its connected components.[7]For example, the symmetric group S_3 (isomorphic to the dihedral group of order 6) has Coxeter matrixM = \begin{pmatrix} 1 & 3 \\ 3 & 1 \end{pmatrix}with respect to two adjacent transposition generators, and its Coxeter-Dynkin diagram consists of two vertices joined by an edge labeled 3.[7][8] The corresponding Schläfli matrix isB = \begin{pmatrix} 1 & -1/2 \\ -1/2 & 1 \end{pmatrix},since \cos(\pi/3) = 1/2.[8]
Basic Examples
Dihedral Groups
The dihedral group D_n serves as the prototypical example of a non-trivial Coxeter group of rank 2, defined by a Coxeter matrix where the off-diagonal entry m_{12} = n for n \geq 3. It is generated by two reflections s and t satisfying the relation (st)^n = 1, with the product st corresponding to a rotation by $2\pi/n.[9] This structure arises as the specialization of the general Coxeter presentation to two generators.[10]The group presentation is \langle s, t \mid s^2 = t^2 = (st)^n = 1 \rangle, and for finite n \geq 3, D_n has order $2n.[11] Geometrically, D_n realizes the symmetries of a regular n-gon in the Euclidean plane, comprising n rotations and n reflections across axes of symmetry that pass either through opposite vertices or through midpoints of opposite sides.[12] These reflections generate the full group, providing an intuitive reflection representation in two dimensions.In the infinite case, where m_{12} = \infty, the relation (st)^n = 1 is omitted, yielding the infinite dihedral group, which is isomorphic to \mathbb{Z} \rtimes \mathbb{Z}_2.[10] This group acts on the real line as the group of isometries generated by translations (corresponding to powers of st) and reflections ( s and t), extending the finite dihedral action to an unbounded setting.[11]
Classical Polyhedral Groups
The classical polyhedral groups are the rank-3 finite Coxeter groups that realize the full reflection symmetries of the Platonic solids in three-dimensional Euclidean space, serving as fundamental examples that connect geometric intuition to the abstract structure of Coxeter presentations. These irreducible groups, denoted by the types A_3, B_3, and H_3, correspond to the tetrahedral, octahedral (or cubic), and icosahedral (or dodecahedral) symmetries, respectively, and highlight the role of reflection generators in producing finite groups via spherical Coxeter diagrams. Unlike reducible cases that decompose into direct products of lower-rank groups, these classical types are indecomposable and form the exceptional components in the broader classification of irreducible finite Coxeter groups.[2]The tetrahedral group of type A_3 has order 24 and acts as the full symmetry group of the regular tetrahedron, including both rotations and reflections. Its Coxeter diagram is a linear chain of three vertices connected by single edges, implying m_{12} = m_{23} = 3 for adjacent generators s_1, s_2, s_3, while non-adjacent generators commute with m_{13} = 2. The presentation is \langle s_1, s_2, s_3 \mid s_i^2 = 1, (s_1 s_2)^3 = (s_2 s_3)^3 = 1, (s_1 s_3)^2 = 1 \rangle, where the reflections correspond to the planes bisecting opposite edges of the tetrahedron.[2][13]The octahedral group (also called the cubic group) of type B_3 has order 48 and captures the symmetries of the regular octahedron or its dual, the cube. The Coxeter diagram consists of a linear chain of three vertices, with the bond between s_1 and s_2 labeled 3 (single edge) and the bond between s_2 and s_3 labeled 4 (double edge), reflecting the dihedral angles in the octahedral arrangement. Verification via the presentation yields \langle s_1, s_2, s_3 \mid s_i^2 = 1, (s_1 s_2)^3 = (s_2 s_3)^4 = 1, (s_1 s_3)^2 = 1 \rangle, where the relation (s_2 s_3)^4 = 1 enforces the quadruple order for the pair associated with the double bond.[2][13]The icosahedral group of type H_3 has order 120 and represents the symmetries of the regular icosahedron or its dual, the dodecahedron. Its Coxeter diagram is a linear chain of three vertices, with bonds labeled 3 (single edge between s_1 and s_2) and 5 (triple edge between s_2 and s_3), corresponding to the specific dihedral angles of these solids. The presentation is \langle s_1, s_2, s_3 \mid s_i^2 = 1, (s_1 s_2)^3 = (s_2 s_3)^5 = 1, (s_1 s_3)^2 = 1 \rangle, with the reflections acting on the planes through the center and vertices of the icosahedron.[2][13]Each of these groups contains rank-2 dihedral subgroups generated by pairs of adjacent reflections, mirroring the 2D symmetries embedded in the 3D polyhedral structure. Their irreducibility ensures that no nontrivial direct product decomposition into lower-rank Coxeter groups is possible, distinguishing them from composite spherical types.[2]
Connection to Reflection Groups
Reflection Representations
A Coxeter group W generated by a set of involutions S = \{s_1, \dots, s_n\} admits a faithful linear representation, known as the Tits representation, on the vector space V = \mathbb{R}^n, where the dimension equals the number of generators (the rank). This representation is defined with respect to a symmetric bilinear form B on V induced by the Coxeter matrix, given explicitly by B(\alpha_i, \alpha_j) = -\cos(\pi / m_{ij}) for i \neq j and B(\alpha_i, \alpha_i) = 1 (or often normalized to 2 in root system contexts), where \{\alpha_1, \dots, \alpha_n\} is a basis of simple roots for V. Each generator s_i acts as a reflection across the hyperplane H_i = \{ v \in V \mid B(v, \alpha_i) = 0 \} orthogonal to \alpha_i with respect to B, fixing H_i pointwise and sending \alpha_i to -\alpha_i. The group W embeds faithfully as a subgroup of the orthogonal group O(V, B) preserving B, providing the core connection between the abstract Coxeter system and reflection geometry.[14]When W is finite, B is positive definite, so V carries a Euclidean structure, and W acts as a finite reflection group—a discrete subgroup of the orthogonal group O(n) with respect to this metric—generated by reflections across hyperplanes through the origin. For infinite Coxeter groups, the signature of B varies: positive semi-definite for affine types (realizing affine Weyl groups acting on Euclidean space \mathbb{R}^{n-1}), and indefinite for hyperbolic types (acting on hyperbolic space). The explicit formula for the action of a reflection s_i on a vector v \in V iss_i(v) = v - 2 \frac{B(v, \alpha_i)}{B(\alpha_i, \alpha_i)} \alpha_i.If the simple roots are normalized so that B(\alpha_i, \alpha_i) = 1, this simplifies to s_i(v) = v - 2 B(v, \alpha_i) \alpha_i. The angles between simple roots, determined by the Coxeter numbers m_{ij}, ensure the braid relations are satisfied in this representation. This construction is orthogonal with respect to B, as each s_i has determinant -1 and preserves the form.[14]The Tits representation, introduced by Jacques Tits, canonically embeds any abstract Coxeter group into \mathrm{GL}(V) via these linear reflections, with trivial kernel, generalizing beyond finite Euclidean cases to all Coxeter systems. It relies on the bilinear form B to define the reflections, ensuring the defining relations hold precisely.[14]Under this representation, W acts by permuting the walls of a fundamental chamber, defined as the simplicial cone C = \{ v \in V \mid B(v, \alpha_i) > 0 \ \forall i \}. The orbit of C under W tiles the Tits cone \{ v \in V \mid B(v, v) > 0 \} (the entire space V when B is positive definite), with adjacent chambers separated by reflection hyperplanes, forming a Coxeter complex whose chambers are congruent via the group action. This tiling realizes the group's combinatorial structure geometrically, with C as the fundamental domain.[14]The geometric interpretation of Coxeter groups as reflection groups traces its roots to Ludwig Schläfli's 1853 classification of regular polytopes, where he described their symmetries in terms of reflections across facets, laying early groundwork for higher-dimensional kaleidoscopic constructions. This idea was further developed by H.S.M. Coxeter in the 1930s through his studies of kaleidoscopes, where he explored systems of mirrors generating finite or infinite reflection groups, providing the inspiration for the modern abstract formalism.
Abstraction and Generalization
The Coxeter presentation provides an abstract algebraic framework that generalizes the structure of reflection groups beyond their geometric realizations. Specifically, a Coxeter group is defined as any group W generated by a set S of involutions (elements s \in S satisfying s^2 = 1) subject to relations of the form (st)^{m_{st}} = 1 for s, t \in S with s \neq t, where each m_{st} \geq 2 is an integer or \infty (indicating no relation beyond the involutions). This presentation captures the essential combinatorial relations among reflections without requiring a faithful linear representation in Euclidean space or a geometric action by isometries. As a result, Coxeter groups encompass not only classical reflection groups but also structures that may lack a natural geometric embedding, allowing the theory to extend to purely algebraic and combinatorial contexts.[14]A prominent example of this abstraction is the class of right-angled Coxeter groups, which arise when all m_{st} are either 2 (commutativity: st = ts) or \infty (no further relation). These groups are determined by a simplicial graph \Gamma with vertex set S, where vertices correspond to generators and edges indicate commuting pairs. The presentation is then W_\Gamma = \langle S \mid s^2 = 1 \ \forall s \in S, \ (st)^2 = 1 \ \text{if } \{s,t\} \in E(\Gamma) \rangle. Right-angled Coxeter groups often fail to act faithfully as reflection groups on Euclidean space but retain the Coxeter relations combinatorially, highlighting the independence of the abstract definition from geometry. Removing the involution relations s^2 = 1 from this presentation yields the corresponding right-angled Artin group, which generalizes further to braid-like structures while preserving the underlying diagram.[15]Some Coxeter groups realize reflection actions only virtually, meaning they act as reflection groups on orbifolds or quotients of geometric spaces rather than directly on manifolds. For instance, certain infinite Coxeter groups generated by reflections in hyperbolic space may have finite-index subgroups that act properly discontinuously on orbifold fundamental domains, where the orbifold structure accounts for singular loci corresponding to non-free orbits under the group action. This virtual realization bridges the abstract Coxeter structure to geometric contexts via covering spaces or orbifold quotients, without requiring a direct isometric reflection representation.The Coxeter diagram—a graph with vertices for generators and edges labeled by m_{st} > 2 (unlabeled for m_{st} = 3, absent for m_{st} = 2, and sometimes marked \infty)—provides a combinatorial classification of all finite and infinite Coxeter groups up to isomorphism. This diagrammatic approach abstracts the relations into a visual and algebraic tool, enabling the study of Coxeter groups through graph-theoretic properties independent of any embedding. The universal Coxeter group of rank n, corresponding to the diagram with no edges (all m_{st} = \infty), is the free product \mathbb{Z}/2\mathbb{Z} * \cdots * \mathbb{Z}/2\mathbb{Z} (n factors), serving as the freest example where only the involution relations hold.
Finite Coxeter Groups
Classification of Irreducible Types
A Coxeter group is called irreducible if it cannot be expressed as a direct product of two nontrivial Coxeter groups, which corresponds to its Coxeter diagram being connected.[16][17]The finite irreducible Coxeter groups are completely classified up to isomorphism by their connected Coxeter-Dynkin diagrams. There are four infinite families—A_n for n ≥ 1, B_n (isomorphic to C_n) for n ≥ 2, D_n for n ≥ 4—and exceptional types including six of higher rank: E_6, E_7, E_8, F_4, G_2, H_3, H_4, plus the rank-2 dihedral family I_2(m) for m ≥ 5 (with I_2(3) = A_2, I_2(4) = B_2, I_2(6) = G_2).[18][16] The groups of types A_n, B_n, D_n, E_6, E_7, E_8, F_4, and G_2 are crystallographic (Weyl groups), while H_3, H_4, and I_2(m) (m ≥ 5, m ≠ 6) are non-crystallographic.[18] Every finite Coxeter group is isomorphic to a direct product of these irreducible ones, with the product corresponding to the disjoint union of their diagrams.[17]The Coxeter-Dynkin diagram encodes the relations among the generators: nodes represent simple reflections, unlabelled edges indicate m_{ij} = 3, double edges indicate m_{ij} = 4, and explicit labels are used for m_{ij} ≥ 5 (triple edges sometimes denote m=6). For type A_n, the diagram is a linear chain of n nodes connected by single edges, representing the symmetries of an n-simplex.[16] Type B_n (or C_n, with the distinguishing edge at the opposite end) consists of a linear chain of n nodes where the end edge is a double bond (m=4) and the rest are single.[18] For D_n, it is a chain of n-2 nodes with single edges, forked at one end by two additional nodes each connected by a single edge to the penultimate node.[16]The exceptional diagrams are as follows: E_6 is a chain of five nodes with a single node branching from the third; E_7 extends this to a chain of six nodes with the branch from the third; E_8 has a chain of seven nodes with the branch from the third.[18]F_4 is a linear chain of four nodes with bonds single-double-single (m=3,4,3). G_2 has two nodes connected by a triple bond (m=6). H_3 is a linear chain of three nodes with a single edge (m=3) followed by a labeled edge (m=5), and H_4 is a chain of four nodes with two single edges (m=3,3) followed by a labeled edge (m=5).[18][16]The following table summarizes the ranks, orders, and geometric realizations for these types (orders for infinite families are given by formulas; exceptional ones are fixed).[16]
A Weyl group arises in the context of semisimple Lie algebras over the complex numbers, where it serves as the group of symmetries for the associated root system. Specifically, for a root system \Phi in a finite-dimensional Euclidean space V, the Weyl group W is the finite Coxeter group generated by the reflections s_\alpha across the hyperplanes \alpha^\perp for \alpha \in \Phi, acting faithfully and orthogonally on V while permuting the roots in \Phi. This group normalizes the fundamental Weyl chamber, which is the closure of a connected component of V minus the union of all reflection hyperplanes. Hermann Weyl introduced this structure in 1925 as part of his foundational work on the representation theory of semisimple Lie groups, identifying W as the symmetry group preserving the root system.[19][14]A root system \Phi is a finite set of nonzero vectors in V that spans V, is closed under inversion (if \alpha \in \Phi then -\alpha \in \Phi), and satisfies the property that for any \alpha, \beta \in \Phi, the reflection of \beta across \alpha^\perp lies in \Phi. The reflections are defined bys_\alpha(v) = v - 2 \frac{(\alpha, v)}{(\alpha, \alpha)} \alphafor v \in V, where (\cdot, \cdot) denotes the inner product on V. A subset \Delta = \{\alpha_1, \dots, \alpha_r\} of \Phi consisting of linearly independent roots is called a set of simple roots if every root in \Phi can be uniquely expressed as an integer linear combination of elements from \Delta with all coefficients nonnegative or all nonpositive. The positive roots \Phi^+ are those with nonnegative coefficients in this basis, and the reflections s_i = s_{\alpha_i} generate W as a Coxeter group with relations determined by the angles between simple roots.[20][14]The action of W on \Phi is transitive on the roots of a given length, ensuring that W permutes the roots while preserving the inner product structure. The irreducible finite Weyl groups are precisely those of types A_n, B_n, C_n, D_n, E_6, E_7, E_8, F_4, and G_2, forming the crystallographic subset of irreducible Coxeter groups; their Coxeter diagrams coincide with the corresponding Dynkin diagrams, with multiple edges indicating roots of different lengths.[20][14]A concrete example is the root system of type A_2, associated to the Lie algebra \mathfrak{[sl](/page/SL)}_3(\mathbb{C}). Here, \Phi consists of six roots in the plane V = \{(x,y,z) \in \mathbb{R}^3 \mid x+y+z=0\}: \pm(e_1 - e_2), \pm(e_2 - e_3), \pm(e_3 - e_1), where e_i are the standard basis vectors and the inner product is the induced Euclidean one. The simple roots can be taken as \alpha_1 = e_1 - e_2 and \alpha_2 = e_2 - e_3, with reflections generating W \cong S_3, the symmetric group on three letters, which acts by permuting the roots and has order $6. This illustrates how W$ faithfully realizes the symmetries of the equilateral triangle formed by the positive roots in the plane.[20]
Key Properties
A finite Coxeter group W generated by a set S of reflections can be characterized as finite if and only if the associated bilinear form on the reflection representation is positive definite.[16] Equivalently, the eigenvalues of the Schläfli matrix L, defined by L_{ii} = 1 and L_{ij} = -\cos(\pi / m_{ij}) for i \neq j where m_{ij} are the entries of the Coxeter matrix, are all positive.[21] This condition aligns with the group's Coxeter diagram being of spherical type, corresponding to the finite irreducible types in the classification (A_n, B_n, etc.).[16]The length function \ell: W \to \mathbb{N} assigns to each element w \in W the minimal number of generators from S needed in a reduced expression for w.[9] The inversion set \mathrm{Inv}(w) consists of the positive roots that w maps to negative roots in the reflection representation, and its cardinality equals \ell(w).[9]The Coxeter complex of a finite Coxeter group W, which is the order complex of the poset of proper parabolic subgroups ordered by inclusion, is shellable.[22] Shellability implies that the complex is a pure simplicial complex with a linear ordering of maximal faces such that each face intersects previous ones in a pure subcomplex, ensuring connectivity of links and acyclicity in certain homology degrees.[22] This property further establishes that the Stanley-Reisner ring of the complex is Cohen-Macaulay, reflecting deep algebraic structure tied to the group's combinatorial geometry.[22]In the representation theory of finite Coxeter groups over \mathbb{C}, the Kazhdan-Lusztig two-sided cells induce a partition of the irreducible representations into families.[23] Two-sided cells partition W based on equivalence relations from the Hecke algebra structure, capturing modular representations and linking to affine Hecke algebras.[23] Character values of these representations are given by formulas involving Kazhdan-Lusztig polynomials, which are defined recursively via the R-polynomials and satisfy positivity properties essential for understanding the group's character table.[23]The subgroup W^+ of even-length elements in W, defined by \{w \in W \mid \ell(w) \equiv 0 \pmod{2}\}, is the kernel of the signhomomorphism \varepsilon: W \to \{\pm 1\} given by \varepsilon(w) = (-1)^{\ell(w)}, and has index 2 in W.[7] For simply-laced Coxeter groups (those with Coxeter diagrams having no multiple edges, i.e., types A, D, E), this even subgroup aligns closely with the rotation subgroup in the geometric realization as reflection groups.[7]
Infinite Coxeter Groups
Affine Coxeter Groups
Affine Coxeter groups are a class of infinite Coxeter groups that extend finite irreducible Weyl groups by incorporating a lattice of translations, resulting in groups of rank n+1 acting faithfully and discretely on Euclidean n-space via affine reflections.[24] These groups are crystallographic reflection groups, meaning their reflections generate a discretesubgroup of the affine transformation group preserving a lattice.[24] The finite Weyl group appears as a finite-index quotient of the affine Coxeter group by the translation subgroup.[25]The presentation of an affine Coxeter group mirrors that of its finite counterpart but includes an additional generator s_0, with relations (s_i s_j)^{m_{ij}} = 1 for i,j \in \{0,1,\dots,n\}, where the exponents m_{ij} are determined by the affine Dynkin diagram and yield an infinite group overall.[24] Affine Dynkin diagrams are constructed by adjoining one vertex to the finite Dynkin diagram in a manner that extends the root system; for instance, the diagram for \tilde{A}_n (n ≥ 1) forms a cycle of n+1 nodes connected by single edges, representing equal angles between adjacent reflections.[24]Geometrically, affine Coxeter groups act on \mathbb{R}^n by reflections across a set of affine hyperplanes arranged such that the fundamental domain is a bounded alcove—a simplex whose reflections tile the entire space without overlaps or gaps, covering \mathbb{R}^n periodically.[24] This action is properly discontinuous and free outside the hyperplanes, ensuring the group is infinite while the stabilizer of any point is finite.[24]The irreducible affine Coxeter groups are classified by their correspondence to the finite irreducible types, yielding the ten families: \tilde{A}_n (n ≥ 1), \tilde{B}_n (n ≥ 2), \tilde{C}_n (n ≥ 2), \tilde{D}_n (n ≥ 4), \tilde{E}_6, \tilde{E}_7, \tilde{E}_8, \tilde{F}_4, and \tilde{G}_2.[24] Each has infiniteorder but is virtually abelian, with the abelian translation subgroup having finite index equal to the order of the finite Weyl group.[25] For example, the affine group \tilde{A}_1 is the infinite dihedral group generated by two reflections s and t with no relation between them beyond s^2 = t^2 = 1, acting on the real line \mathbb{R} by reflections at the integers, which generates translations by multiples of 2 and tiles the line with fundamental intervals of length 1.[26]
Hyperbolic Coxeter Groups
Hyperbolic Coxeter groups are a class of infinite Coxeter groups that arise as discrete reflection groups acting on hyperbolic space \mathbb{H}^n, generated by reflections in the sides of a convex Coxeter polyhedron P serving as a fundamental domain of finite volume. These polyhedra have dihedral angles of the form \pi/m_{ij} where m_{ij} \geq 2 or \infty, and the representation is faithful if P is non-degenerate. The geometry requires that the sum of dihedral angles incident to each codimension-2 face is less than (n-1)\pi, ensuring the action occurs in negatively curved space rather than Euclidean or spherical.[27][28]The associated Coxeter diagrams are indefinite, meaning the quadratic form defined by the Gram matrix G with entries G_{ii} = 1 and G_{ij} = -\cos(\pi/m_{ij}) has signature (n,1), confirming the hyperbolic nature. For instance, a rank-3 diagram with labels m_{12}=3, m_{13}=\infty, m_{23}=3 corresponds to the right-angled ideal triangle group, where two vertices lie at infinity and the finite vertex has angle \pi/3. This diagram yields a non-compact fundamental domain with cusps.[29][30]Classification of hyperbolic Coxeter groups poses significant challenges, with no complete enumeration known even in low dimensions due to the infinite possibilities for diagrams. In \mathbb{H}^2, they coincide with hyperbolic triangle groups satisfying $1/p + 1/q + 1/r < 1, but higher-dimensional cases rely on partial results like Andreev's theorem for simplicial polytopes in \mathbb{H}^3. Vinberg's algorithm provides a practical method to verify hyperbolicity: construct the Gram matrix and check its eigenvalues for n positive and one negative, ensuring the polyhedron is non-empty and of finite volume; the process iteratively resolves simplicial chambers until termination.[31][27][30]Representative examples include the [3,\infty,3] group in \mathbb{H}^2, which tessellates the hyperbolic plane by ideal triangles and achieves the minimal growth rate among non-cocompact finite-volume Coxeter groups in this space. Another key example is the [2,3,\infty] Coxeter group, whose index-2 rotation subgroup is isomorphic to the modular group \mathrm{PSL}(2,\mathbb{Z}), acting on the hyperbolic plane with a fundamental domain being a hyperbolic triangle with angles \pi/2, \pi/3, and $0. This group highlights connections to number theory.[29][32]Hyperbolic Coxeter groups exhibit exponential growth, with the growth rate \tau_W > 2 given by the spectral radius of the Cartan matrix, distinguishing them from affine groups where \tau_W = 2. Fundamental domains may be compact, yielding cocompact actions, or non-compact with cusps, leading to finite-volume but non-compact quotients; the [3,\infty,3] example illustrates the latter with parabolic subgroups at infinity.[29][32]
Structural Features
Length Function and Coxeter Elements
In a Coxeter system (W, S), the length function l: W \to \mathbb{N}_0 assigns to each element w \in W the minimal number of simple reflections from S needed to express w as a product; such a minimal-length expression is called reduced.[2] This function satisfies the key property that for any w \in W and s \in S, l(ws) = l(w) \pm 1, reflecting whether right multiplication by s lengthens or shortens a reduced expression for w.[2] Moreover, the deletion property holds: if a word in the generators of length k > l(w) equals w, then there exist indices $1 \leq i < j \leq k such that deleting the i-th and j-th letters yields another word of length k-2 also equaling w, allowing iterative reduction to a minimal expression.[33]A Coxeter element c in a Coxeter group W with generating set S = \{s_1, \dots, s_n\} is the product of all elements of S in some order, such as c = s_1 s_2 \cdots s_n. In finite Coxeter groups, all such products, regardless of order, are conjugate in W, forming a single conjugacy class.[34] In the case of a finite irreducible Coxeter group, every Coxeter element has order equal to the Coxeter number h of the group, which is the smallest positive integer such that c^h = 1.[2] Acting linearly on the associated reflection representation, a Coxeter element c has eigenvalues of the form \exp(2\pi i m / h), where the m are the exponents of the group (rank many distinct positive integers between 1 and h-1).[34][35]For a concrete illustration, consider the irreducible Coxeter group of type A_2, which is isomorphic to the symmetric group S_3 generated by adjacent transpositions s_1 = (1\,2) and s_2 = (2\,3). Here, the length l(w) of a permutation w equals the number of inversions in w.[33] The Coxeter number is h=3, and a Coxeter element such as c = s_1 s_2 = (1\,2\,3) acts as a $120^\circ rotation in the plane, with eigenvalues \exp(2\pi i / 3) and \exp(-2\pi i / 3).[34]
Partial Orders
Coxeter groups admit rich partial order structures that capture combinatorial and geometric aspects of their elements. Two fundamental posets are the weak order and the Bruhat order, both graded by the length function \ell: W \to \mathbb{N}, where the rank of an element w \in W is \ell(w). These orders arise naturally from the generating set S and the relations among reduced words, facilitating the study of reduced decompositions and subexpression properties.[7]The right weak order on W is the partial order generated by the covering relations w \prec ws for s \in S whenever \ell(ws) = \ell(w) + 1. Equivalently, u \leq v if v = u y for some y with \ell(v) = \ell(u) + \ell(y), or some reduced word for v has a reduced word for u as an initial subword.[36] This order refines the coset structure, with right-weak order on right cosets W / W_J for parabolic subgroups W_J. The left weak order is defined analogously by left multiplication. For finite Coxeter groups, the weak order is a lattice, with meets and joins computable via certain projections.[7]The Bruhat order, often called the strong Bruhat order to distinguish it from the weak order, is a finer partial order defined by u \leq v if some reduced word for v contains a reduced word for u as a (not necessarily contiguous) subsequence. This is equivalent to the containment of inversion sets: u \leq v if and only if N(u) \subseteq N(v), where N(w) = \{\alpha \in \Phi^+ \mid \ell(s_\alpha w) < \ell(w)\} is the set of positive roots inverted by w, with \Phi^+ the positive roots in the reflection representation (for finite groups). The Bruhat order covers the weak order, meaning every weak order relation is a Bruhat relation, but the converse holds only for covering relations involving simple reflections.[37]A variant known as the strong Bruhat order in the context of Hecke algebras refers to the Bruhat order restricted or extended to double cosets W_J \backslash W / W_K for parabolic subgroups W_J, W_K, which indexes the basis elements and structure constants in Iwahori-Hecke algebras via parabolic induction and decomposition.[7]For finite Coxeter groups, both the weak and Bruhat orders are ranked posets with the identity as minimal element and the longest element w_0 as maximal. They are Eulerian posets, satisfying \sum_{u \leq v} (-1)^{\ell(v) - \ell(u)} = 0 for all intervals [u, v] with u < v, which implies their rank-generating polynomials are Eulerian and aids in enumerative combinatorics. Additionally, the weak order is shellable, meaning its order complex admits a shelling (a linear extension where each facet shares an initial segment with the previous), ensuring Cohen-Macaulay topology and connectivity properties.[7]A concrete example occurs in the symmetric group S_n, the Coxeter group of type A_{n-1}. Here, the Bruhat order on permutations corresponds to componentwise inequality on inversion tables: for permutations \sigma, \tau \in S_n with inversion tables (a_1, \dots, a_n) and (b_1, \dots, b_n) where a_i counts elements j > i with \sigma(j) < \sigma(i), we have \sigma \leq \tau if and only if a_i \leq b_i for all i. The strong Bruhat order in this setting aligns with comparisons of permutation matrix entries via row and column dominance, reflecting the poset's role in Schubert calculus.[7]
Homological and Topological Aspects
Homology Computations
The homology of Coxeter groups can be computed algebraically using resolutions derived from combinatorial posets and geometric models associated to their presentations. For right-angled Coxeter groups, the Salvetti complex of the corresponding right-angled Artin group provides a finite CW-complex homotopy equivalent to the complement of the associated hyperplane arrangement in \mathbb{R}^n. The group W acts on this complex, and the homology H_*(W; \mathbb{Z}) is the homology of the quotient space, which can be calculated via the spectral sequence arising from the universal cover.[38]For general Coxeter groups, the Davis resolution furnishes an explicit free resolution of the trivial \mathbb{Z}W-module \mathbb{Z}, constructed from the poset of cosets of spherical parabolic subgroups. The chain complex is the tensor product \mathbb{Z}W \otimes_{\mathbb{Z}} C_*(\mathcal{P}), where C_*(\mathcal{P}) is the cellular chain complex of the geometric realization of the poset \mathcal{P}; its Hom complex computes the group cohomology, while Tor groups yield the homology H_*(W; \mathbb{Z}). This resolution is particularly efficient, with length equal to the number of spherical subsets of the generating set.[39]The Solomon-Tits theorem underpins these computations for finite Coxeter groups W of rank n. It asserts that the rational homology of the Coxeter complex \Sigma(W,S) vanishes in intermediate degrees: \tilde{H}_i(\Sigma(W,S); \mathbb{Q}) = 0 for $0 < i < n-1, with \tilde{H}_{n-1}(\Sigma(W,S); \mathbb{Q}) affording the trivial W-representation. Consequently, the top homology carries a permutation module structure induced by the W-action on the set of maximal simplices (chambers), simplifying the syzygy computations in the Davis resolution and revealing the structure of H_*(W; \mathbb{Z}) as torsion in low degrees with free parts concentrated near the ends.[40]As a representative example, consider the symmetric group S_n, a Coxeter group of type A_{n-1}. Its homology H_*(S_n; \mathbb{Z}) is generated by classes corresponding to cycles in the bar resolution, stabilized by transfers from subgroups; the Betti numbers \dim H_i(S_n; \mathbb{Q}) vanish for i > 0.For infinite Coxeter groups, particularly affine ones, the Davis resolution reveals virtual duality properties. Affine Coxeter groups of rank n are virtual duality groups of dimension n, meaning a finite-index torsion-free subgroup F \leq W satisfies H_i(F; \mathbb{Z}) = 0 for $0 < i < n and H_n(F; \mathbb{Z}) \cong \mathbb{Z}. This arises from the contractibility of the Davis complex and the fact that its link at the vertex corresponding to the essential spherical subset is a homology (n-1)-sphere. Moreover, the Davis manifold—a compact aspherical manifold homotopy equivalent to BF constructed by resolving singularities of the Davis complex—satisfies Poincaré duality H_i(M; \mathbb{Z}) \cong H^{n-i}(M; \mathbb{Z}).[41]
Davis Complexes and CAT(0) Geometry
The Davis complex \Sigma associated to a Coxeter system (W, S) is a piecewise Euclidean cell complex that serves as a geometric realization on which W acts properly and cocompactly by isometries.[42] Its vertices are the elements of W, and the 1-skeleton is the Cayley graph of W with respect to the generating set S. Higher-dimensional cells correspond to cosets w W_T for spherical parabolic subgroups W_T (where T \subseteq S generates a finite subgroup), with each such cell of dimension |T| and realized as a Coxeter polytope in the associated Euclidean space.[42] The fundamental chamber is a Coxeter polytope K in the Tits representation, and \Sigma is constructed by gluing translates w K along their faces, ensuring that mirrors (codimension-1 faces) are identified via the reflections in S.[42]The group W acts freely on the vertices of [\Sigma](/page/Sigma) and simply transitively, with the action extending to a proper, cocompact isometricaction on the entire complex.[42] This action preserves the cell structure, where each generator s \in S acts as a reflection across the corresponding hyperplanes. Equipped with the geodesic metric induced by the Euclidean structure on each cell, [\Sigma](/page/Sigma) is a complete CAT(0) space for any Coxeter group, as established by Moussong's criterion: the link of every vertex is a spherical Coxeter complex, which is a flag complex ensuring non-positive curvature.[42] This CAT(0) property implies that [\Sigma](/page/Sigma) is contractible and simply connected, providing a model for the classifying space \underline{E}W.[42]The visual boundary \partial \Sigma of the Davis complex is the sphere at infinity, compactified by adding endpoints of geodesic rays, and is homeomorphic to the nerve L(W, S) of the Coxeter system—a simplicial complex whose simplices are the spherical subsets of S.[42] Coxeter elements, products of all generators in S in some order, act on \Sigma as rotations in the Tits cone and induce dynamics on \partial \Sigma, often as pseudo-rotations preserving the spherical structure and fixed sets corresponding to parabolic subgroups.[42]For the infinite dihedral group D_\infty with presentation \langle s, t \mid (st)^2 = 1 \rangle, the Davis complex \Sigma is isometric to the Euclidean plane \mathbb{E}^2 tessellated by copies of a fundamental domain consisting of two adjacent half-planes, or equivalently, a tree-like strip where the action generates translations along lines.[42]An alternative construction, the Moussong complex, provides a simplicial model for the Davis complex, particularly useful for hyperbolic Coxeter groups where the Euclidean structure may not suffice.[43] It is built by simplicially subdividing the cells of \Sigma, yielding a piecewisehyperbolic or Euclidean metric that remains CAT(0), and is especially effective for computing topological invariants in word-hyperbolic cases.[43]
Applications
Symmetry of Polytopes and Tessellations
Finite Coxeter groups act as the full symmetry groups, including reflections and rotations, of regular polytopes embedded in Euclidean space of dimension equal to the rank of the group.[14] These groups generate the isometries that map the polytope to itself while preserving its regular structure, where all faces, edges, and vertices are congruent and symmetrically arranged. Specific irreducible finite Coxeter groups of types A_n, B_n, F_4, H_3, H_4, G_2, and I_2(m) act as the full symmetry groups of regular polytopes in dimensions 2 through 4, as classified by their Coxeter diagrams.[44] For instance, the exceptional Coxeter group of type H_4, of rank 4, is the symmetry group of the regular 120-cell, a four-dimensional polytope composed of 120 regular dodecahedral cells meeting five at each vertex, with the group having order 14400.In three dimensions, the classical examples include the tetrahedral group of type A_3, symmetry group of the regular tetrahedron; the octahedral group of type B_3, for the cube and octahedron; and the icosahedral group of type H_3, for the dodecahedron and icosahedron. These finite groups ensure the polytopes are bounded and compact, contrasting with infinite cases in other geometries. The enumeration of such finite regular polytopes is limited, with five in three dimensions (the Platonic solids) and six in four dimensions (including the 24-cell alongside simplices, hypercubes, and their duals).[44]Affine Coxeter groups, which are infinite but virtually abelian, arise as the symmetry groups of regular tessellations, or honeycombs, that tile Euclidean space \mathbb{R}^n without gaps or overlaps.[14] These groups extend the finite cases by including translations alongside reflections, generating periodic structures. There are nine irreducible affine Coxeter types, \tilde{A}_n, \tilde{B}_n, \tilde{C}_n, \tilde{D}_n, \tilde{E}_6, \tilde{E}_7, \tilde{E}_8, \tilde{F}_4, and \tilde{G}_2, each corresponding to a unique regular honeycomb in dimension equal to its rank, starting from dimension 2 for planar tilings up to dimension 8. For example, the affine Coxeter group \tilde{A}_2 of rank 3 acts as the symmetry group of the equilateral triangular tiling of the Euclidean plane, where reflections across the lines of the tiling generate the full group.[45] In three dimensions and higher, these include the cubic honeycomb (type \tilde{C}_3) and prismatic honeycombs, providing Euclidean analogs to the finite polytopes.Hyperbolic Coxeter groups are infinite discrete subgroups of the isometry group of hyperbolic space \mathbb{H}^n, acting as symmetry groups of regular tessellations that fill \mathbb{H}^n with regular polytopes meeting in a highly symmetric but unbounded manner.[43] These groups lead to indefinite quadratic forms (of signature (n-1,1)) that yield hyperbolic geometry. A prominent example is the Coxeter group with diagram [3,7,3], of rank 4, which is the full symmetry group of the regular {3,7,3} honeycomb in \mathbb{H}^3; this tessellation consists of regular {3,7} apeirohedra (infinite-sided polyhedra with heptagonal faces and three at each vertex) meeting three at each edge. Such hyperbolic tessellations exist when the Coxeter diagram's angles sum to less than the Euclidean case, allowing infinite extent without closure.The connection between these geometric realizations and the algebraic structure of Coxeter groups is captured through Schläfli symbols, which encode the regularity of the polytopes or tessellations and directly correspond to the labels on Coxeter-Dynkin diagrams. For a regular polytope or tiling denoted by the Schläfli symbol {p, q}, p indicates the number of sides of each face, and q the number meeting at each vertex; this corresponds to the Coxeter diagram [p, q], where the branches are labeled by the dihedral angles \pi/p and \pi/q. In higher dimensions, extended symbols like {p, q, r} link to linear diagrams [p, q, r], omitting implicit 3's for tetrahedral vertex figures. This notation unifies the finite, affine, and hyperbolic cases: finite when the symbol yields positive definite form (spherical geometry), affine for semi-definite (Euclidean), and hyperbolic for indefinite with signature (n,1).[44]
Buildings and Combinatorial Geometry
A building associated to a Coxeter system (W, S) is a simplicial complex \Sigma that decomposes as a disjoint union of simplices, where each apartment—a maximal subcomplex—is isomorphic to the Coxeter complex of (W, S), and the group W acts transitively on the chambers, which are the maximal simplices of \Sigma.[46] This structure, introduced by Jacques Tits, provides a geometric framework for understanding groups containing W as a quotient of the Weyl group in a BN-pair.[47]For finite W, the resulting buildings are spherical. The thin spherical building coincides with the Coxeter complex itself, serving as the basic model. In contrast, thick spherical buildings emerge from groups with BN-pairs, such as \mathrm{GL}_n(K) over a field K, where the building encodes the geometry of parabolic subgroups and their flag varieties.[47] These thick versions generalize the thin case by allowing multiple chambers per panel, reflecting the richness of the ambient group structure.[46]A concrete example arises for the irreducible Coxeter type A_n, where the spherical building for the BN-pair in \mathrm{SL}_{n+1}(\mathbb{F}_q) over a finite field \mathbb{F}_q is the flag complex consisting of partial flags of subspaces in an (n+1)-dimensional vector space; vertices correspond to proper nontrivial subspaces, and simplices to chains of inclusions.[46]Affine buildings correspond to infinite affine Coxeter systems, yielding Euclidean structures that are infinite in extent. These are linked to affine Grassmannians in the theory of reductive groups over local fields, as developed in the Bruhat-Tits construction, where apartments are Euclidean Coxeter complexes tiled by the affine Weyl group.[48]Spherical buildings of rank at least 2 are simply connected and homotopy equivalent to a wedge of spheres in the top dimension equal to the rank minus one.[46] This homotopy type facilitates homological computations, as captured by the Solomon-Tits theorem, which identifies the top homology group of the building with the Steinberg representation of the associated finite group of Lie type.[49]