OctoInformBeta

OpenAI preprint, September 2026 · §3. Dihedral intervals and transported reflection orders

Lemma 3.1: For the plane Π\Pi of two positive roots, DΠD_\Pi is a Coxeter system with the ambient reflections and Bruhat graph, copied onto each coset

Reflections with roots in a plane generate a dihedral group.

Hover over a blue or green term in the text, or a node in the graph, to find it in the other. Select it to read its declaration both in Lean and in words.

The Setting

Let (W,S)(W,S) be a Coxeter system of finite rank, with simple reflections sis_i indexed by a finite set II, Coxeter matrix (mij)(m_{ij}), length ℓ\ell and set of reflections TT. The Bruhat graph of WW is the directed graph on WW with an edge x→yx \to y whenever y=txy = tx for a reflection tt and ℓ(x)<ℓ(y)\ell(x) < \ell(y); the edge is labelled by t=yx−1t = yx^{-1}✦, and the Bruhat order ≤\le is the reflexive transitive closure of the edges.✦

The group WW acts on the real vector space ℝI\mathbb{R}^I with a basis of simple roots, the Tits space, and each reflection of WW has a root in it, all made precise below; the reflections whose roots lie in a plane Π\Pi of ℝI\mathbb{R}^I generate a reflection subgroup DΠD_\Pi of WW. In 𝔖4\mathfrak{S}_4, the plane spanned by the roots of (12)(12) and (34)(34) holds no other positive root, and DΠ=⟨(12),(34)⟩D_\Pi = \langle (12),(34)\rangle is a Klein four-group; the plane of (12)(12) and (23)(23) also holds the root of (13)(13), and DΠD_\Pi is the symmetric group on {1,2,3}\{1,2,3\}. The question is how the length and the Bruhat graph of WW restrict to such a subgroup and to its cosets DΠaD_\Pi a, and how the plane is recovered from a product of two reflections or from a path of two edges.

Roots in the Tits Space

The Tits space is ℝI\mathbb{R}^I, and the simple root αi\alpha_i is the coordinate vector of i∈Ii \in I.✦ The Tits form is the bilinear form ⟨⋅,⋅⟩\langle\cdot,\cdot\rangle on ℝI\mathbb{R}^I whose values on the simple roots are the Coxeter Gram entries ⟨αi,αj⟩=−2cos⁡(π/mij)(i,j∈I),\langle \alpha_i, \alpha_j \rangle = -2\cos(\pi/m_{ij}) \qquad (i, j \in I),(1)✦ an entry mij=∞m_{ij} = \infty giving −2-2. The geometric simple reflection σi\sigma_i is the linear map v↦v−⟨αi,v⟩αiv \mapsto v - \langle \alpha_i, v \rangle\,\alpha_i of ℝI\mathbb{R}^I✦, and the Tits action of WW is the homomorphism from WW to GL(ℝI)\mathrm{GL}(\mathbb{R}^I) that sends sis_i to σi\sigma_i.✦

A root of (W,S)(W,S) is a vector of ℝI\mathbb{R}^I of the form w(αi)w(\alpha_i) with w∈Ww \in W and i∈Ii \in I✦; a positive root is a root with all coordinates nonnegative, and Φ+\Phi^+ is the set of positive roots.✦ A vector aa of ℝI\mathbb{R}^I represents a reflection tt of WW when a=w(αi)a = w(\alpha_i) and t=wsiw−1t = ws_iw^{-1} for some w∈Ww \in W and i∈Ii \in I.✦ Every reflection tt has a positive root αt\alpha_t representing it, fixed once and for all✦, and every positive root aa represents a reflection sas_a.✦ The root of an edge x→yx \to y of the Bruhat graph is the positive root αt\alpha_t of its label t=yx−1t = yx^{-1}.✦

The height of a vector vv of ℝI\mathbb{R}^I is ht(v)=∑i∈Ivi\mathrm{ht}(v) = \sum_{i \in I} v_i.✦ A geometric functional is a linear functional ff on ℝI\mathbb{R}^I for which f(a)/ht(a)f(a)/\mathrm{ht}(a) takes distinct values on distinct positive roots aa; one such ff is fixed✦, and the geometric slope of a positive root aa is f(a)/ht(a)f(a)/\mathrm{ht}(a).✦

Plane Subgroups and Plane Bases

For positive roots aa and bb, the root plane Π(a,b)=ℝa+ℝb\Pi(a,b) = \mathbb{R}a + \mathbb{R}b is the subspace of ℝI\mathbb{R}^I they span.✦ For a subspace Π\Pi of ℝI\mathbb{R}^I, the plane subgroup DΠD_\Pi is the subgroup of WW generated by the reflections represented by a vector of Π\Pi.✦

An element ww of WW inverts a positive root aa when every coordinate of w−1(a)w^{-1}(a) is at most 00✦, and the inversion set N(w)N(w) is the set of positive roots that ww inverts.✦ The relative inversion count of ww for a subspace Π\Pi is the number of positive roots in Π\Pi that ww inverts: ℓΠ(w)=#(N(w)∩Π).\ell_\Pi(w) = \#\bigl(N(w) \cap \Pi\bigr).(2)✦

A plane basis of a subspace Π\Pi of ℝI\mathbb{R}^I is an ordered pair (α,β)(\alpha,\beta) of distinct positive roots with Π(α,β)=Π\Pi(\alpha,\beta) = \Pi such that, among the positive roots lying in Π\Pi, α\alpha has the least geometric slope and β\beta the greatest.✦

For distinct elements ss and tt of a group GG with s2=t2=1s^2 = t^2 = 1, the dihedral Coxeter matrix of the pair is the Coxeter matrix on two indices whose off-diagonal entry is the order mm of stst, with m=∞m = \infty when stst has infinite order.✦ For distinct elements ss and tt of a group GG with s2=t2=1s^2 = t^2 = 1 and s,t≠1s, t \ne 1, the dihedral Coxeter system of the pair ss, tt is the subgroup ⟨s,t⟩\langle s,t \rangle of GG taken as a Coxeter system with simple reflections ss and tt and the dihedral Coxeter matrix of the pair as its Coxeter matrix.✦ For a plane basis (α,β)(\alpha,\beta) of Π\Pi, the plane Coxeter system is the dihedral Coxeter system of the pair sαs_\alpha, sβs_\beta on DΠD_\Pi✦, and its Coxeter matrix is the plane dihedral matrix.✦

The Claim

Theorem 1. Let (W,S)(W,S) be a Coxeter system of finite rank.

  1. If aa and bb are distinct positive roots, then Π(a,b) has a plane basis.\Pi(a,b) \text{ has a plane basis.}(3)✦

  2. Let Π\Pi be a subspace of ℝI\mathbb{R}^I with a plane basis (α,β)(\alpha,\beta), and let D=DΠD = D_\Pi carry the plane Coxeter system, with set of reflections TDT_D, length ℓD\ell_D, Bruhat order ≤D\le_D and Bruhat graph edges g→Dhg \to_D h. Then TD=T∩D,T_D = T \cap D,(4)✦ for every w∈Dw \in D ℓΠ(w)=ℓD(w),\ell_\Pi(w) = \ell_D(w),(5)✦ and for all g,h∈Dg, h \in D g→h⇔g→Dh.g \to h \iff g \to_D h.(6)✦

  3. Let Π\Pi be a subspace of ℝI\mathbb{R}^I. For every a∈Wa \in W there is a0∈Wa_0 \in W with a∈DΠa0andℓ(a0)≤ℓ(za0) for all z∈DΠ.a \in D_\Pi a_0 \quad\text{and}\quad \ell(a_0) \le \ell(za_0) \ \text{ for all } z \in D_\Pi.(7)✦ If a0a_0 has least length in its coset DΠa0D_\Pi a_0, then for all g,h∈DΠg, h \in D_\Pi ga0→ha0⇔g→h.ga_0 \to ha_0 \iff g \to h.(8)✦ If moreover Π\Pi has a plane basis (α,β)(\alpha,\beta) and DD is as in (b), then for g,h∈Dg, h \in D with g≤Dhg \le_D h ga0≤ha0,ga_0 \le ha_0,(9)✦ and for every w∈Dw \in D ℓΠ(wa0)=ℓD(w).\ell_\Pi(wa_0) = \ell_D(w).(10)✦

  4. Let EE be a vector space over a field KK and, for u∈Eu \in E and a linear functional φ\varphi on EE, let ru,φ(x)=x−φ(x)ur_{u,\varphi}(x) = x - \varphi(x)\,u. If φ\varphi and ψ\psi are linearly independent linear functionals on EE and u,v∈Eu, v \in E, then im⁡(ru,φrv,ψ−1)=Ku+Kv.\operatorname{im}\bigl(r_{u,\varphi}\,r_{v,\psi} - 1\bigr) = Ku + Kv.(11)✦ If κ\kappa is a nondegenerate bilinear form on EE, ru=ru,κ(u,⋅)r_u = r_{u,\kappa(u,\cdot)}, and u,v,u′,v′∈Eu, v, u', v' \in E satisfy κ(u,u)=κ(v,v)=κ(u′,u′)=κ(v′,v′)=2\kappa(u,u) = \kappa(v,v) = \kappa(u',u') = \kappa(v',v') = 2, with u,vu, v linearly independent, u′,v′u', v' linearly independent and rurv=ru′rv′r_u r_v = r_{u'} r_{v'}, then Ku+Kv=Ku′+Kv′.Ku + Kv = Ku' + Kv'.(12)✦

  5. Let Π\Pi, (α,β)(\alpha,\beta) and DD be as in (b), let a0∈Wa_0 \in W, let g1a0→g2a0→g3a0g_1a_0 \to g_2a_0 \to g_3a_0 be a path of two edges in the Bruhat graph of WW with g1,g2,g3∈Dg_1, g_2, g_3 \in D, and let γ1\gamma_1 and γ2\gamma_2 be the roots of its two edges. Then Π(γ1,γ2)=Π,\Pi(\gamma_1,\gamma_2) = \Pi,(13)✦ and every z∈Wz \in W with edges g1a0→z→g3a0g_1a_0 \to z \to g_3a_0 satisfies z∈Da0.z \in Da_0.(14)✦

In the paper, the result reads: for a plane Π\Pi spanned by two independent roots, D=DΠD = D_\Pi is dihedral with its canonical generators and its reflections are exactly the ambient reflections with roots in Π\Pi; each coset DaDa has a unique element a0a_0 of minimal length, w↦wa0w \mapsto wa_0 identifies the labelled directed Bruhat graph of DD with the ambient graph on DaDa, x≤Dyx \le_D y implies x≤yx \le y, and ℓD(x)\ell_D(x) is the number of positive roots γ\gamma of DD with x−1γ<0x^{-1}\gamma < 0; two factorizations of an element into two distinct reflections have root pairs spanning the same plane, and all directed two-edge paths with fixed endpoints have labels in one maximal dihedral subgroup.

What Is Checked

The claim is stated over the Tits space ℝI\mathbb{R}^I with its simple roots, the Tits form (1) and the Tits action built from the geometric simple reflections; over the roots and positive roots of (W,S)(W,S), the reflections they represent, the Bruhat graph and Bruhat order of WW and the roots of its edges; and over the root planes, the plane subgroups DΠD_\Pi, the inversion sets and relative inversion counts (2), the height and the geometric slope, plane bases, and the plane Coxeter system with its plane dihedral matrix, an instance of the dihedral Coxeter system of a pair. The statements (11) and (12) are over an arbitrary vector space over a field.

The library also proves that the label yx−1yx^{-1} of an edge x→yx \to y of the Bruhat graph is a reflection✦, that ⟨αi,αi⟩=2\langle \alpha_i, \alpha_i \rangle = 2 for every i∈Ii \in I✦, that the geometric simple reflections satisfy the Coxeter relations (σiσj)mij=1(\sigma_i\sigma_j)^{m_{ij}} = 1 for finite mijm_{ij}✦, and that a geometric functional exists.✦ For distinct elements ss and tt of a group GG with s2=t2=1s^2 = t^2 = 1, the homomorphism from the Coxeter group of their dihedral Coxeter matrix to GG that sends the two generators to ss and tt has image ⟨s,t⟩\langle s,t \rangle.✦ For a plane basis (α,β)(\alpha,\beta) of Π\Pi, the reflections sαs_\alpha and sβs_\beta are distinct✦ and generate DΠD_\Pi.✦