OctoInformBeta

OpenAI preprint, September 2026 · §3. Comparison with an arbitrary aspherical presentation

Lemma 3.3: Signed counts survive evaluation after any group homomorphism, from every starting vertex

Injective relator boundaries determine translated counts.

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

The comparison concerns the signed relator counts of a finite product of conjugates, recorded at vertices of a Cayley graph. Let Γ\Gamma be a group, let SS be a generator alphabet with assigned values v:S→Γv:S\to\Gamma, and let F(S)F(S) be the free group on SS. Word evaluation w↦w‾w\mapsto\bar w is the homomorphism F(S)→ΓF(S)\to\Gamma extending vv.✦ Distinct letters of SS remain distinct even when their values agree.

Let RR index relator words rρ∈F(S)r_\rho\in F(S). With Λ=ℤ[Γ]\Lambda=\mathbb Z[\Gamma] acting on the left, the based edge module C1C_1 and the relator module C2C_2 are the free left Λ\Lambda-modules in (1); for a vertex g∈Γg\in\Gamma, their integral basis elements are written e(g,s)=gese_{(g,s)}=ge_s and e(g,ρ)=geρe_{(g,\rho)}=ge_\rho. C1=⨁s∈SΛes,C2=⨁ρ∈RΛeρ.C_1=\bigoplus_{s\in S}\Lambda e_s, \qquad C_2=\bigoplus_{\rho\in R}\Lambda e_\rho.(1) An edge labelled ss runs from gg to gv(s)gv(s). The based path chain [w]g∈C1[w]_g\in C_1 is the signed integral edge chain obtained by traversing ww from gg, with reverse traversals counted negatively; write [w]=[w]1[w]=[w]_1.✦ The relator boundary map ∂2:C2→C1\partial_2:C_2\to C_1 is the integral linear map specified by (2). ∂2(geρ)=[rρ]g.\partial_2(ge_\rho)=[r_\rho]_g.(2)✦

Normal Factors and Labels

A normal closure factor f=(u,ρ,σ)∈F(S)×R×{1,−1}f=(u,\rho,\sigma)\in F(S)\times R\times\{1,-1\} is a triple consisting of a conjugating word, a relator index and a sign.✦ For a normal closure factor f=(u,ρ,σ)∈F(S)×R×{1,−1}f=(u,\rho,\sigma)\in F(S)\times R\times\{1,-1\} and a vertex g∈Γg\in\Gamma, the conjugate relator word W(f)∈F(S)W(f)\in F(S) and the translated signed relator count Ng(f)∈C2N_g(f)\in C_2 are defined by (3). W(f)=urρσu−1,Ng(f)=σ(gu‾)eρ.W(f)=u r_\rho^\sigma u^{-1}, \qquad N_g(f)=\sigma(g\bar u)e_\rho.(3)

For a group KK, an edge labelling a:Γ×S→Ka:\Gamma\times S\to K is a function assigning an element of KK to each based edge. The semidirect label homomorphism ℓa:F(S)→KΓ⋊Γ\ell_a:F(S)\to K^\Gamma\rtimes\Gamma extends s↦(a(⋅,s),v(s))s\mapsto(a(\mathord\cdot,s),v(s)), with the shift action on functions.✦ The path holonomy Av,a(w,g)∈KA_{v,a}(w,g)\in K is the value at gg of the function component of ℓa(w)\ell_a(w), equivalently the product of edge labels in traversal order, taking inverses on reverse traversals.✦

Let H=SU(2)H=\mathrm{SU}(2) be the group of unit quaternions, consisting of the quaternions of norm one.✦ For an index set II, a relator factor (h,i,σ)∈H×I×{1,−1}(h,i,\sigma)\in H\times I\times\{1,-1\} is a triple consisting of an element of HH, an index in II and a sign.✦ The signed relator count sc⁡(L)∈ℤ(I)\operatorname{sc}(L)\in\mathbb Z^{(I)} of a finite list L=((hj,ij,σj))jL=((h_j,i_j,\sigma_j))_j is defined by (4), where eie_i denotes the integral basis vector at ii. sc⁡(L)=∑jσjeij.\operatorname{sc}(L)=\sum_j\sigma_j e_{i_j}.(4)✦ For labels a:Γ×S→Ha:\Gamma\times S\to H, the evaluated normal factor Egv,a(f)∈H×(Γ×R)×{1,−1}E_g^{v,a}(f)\in H\times(\Gamma\times R)\times\{1,-1\} is the relator factor in (5). Egv,a(u,ρ,σ)=(Av,a(u,g),(gu‾,ρ),σ).E_g^{v,a}(u,\rho,\sigma) =\bigl(A_{v,a}(u,g),(g\bar u,\rho),\sigma\bigr).(5)✦

The Claim

Theorem 1. Suppose every relator rηr_\eta evaluates to the identity in Γ\Gamma and the relator boundary map ∂2\partial_2 is injective. Let ρ∈R\rho\in R and z∈F(S)z\in F(S) satisfy [z]=[rρ][z]=[r_\rho]. For a finite list of normal closure factors l=(fj)jl=(f_j)_j satisfying z=∏jW(fj)z=\prod_j W(f_j), the identity (6) holds. ∑jN1(fj)=eρ.\sum_j N_1(f_j)=e_\rho.(6)✦ For any finite list l=(fj)jl=(f_j)_j satisfying (6), let q:Γ→Dq:\Gamma\to D be a group homomorphism and let a:D×S→Ha:D\times S\to H be an edge labelling. At every starting vertex g∈Dg\in D, the identity (7) holds, with e(g,ρ)e_{(g,\rho)} the integral basis vector in ℤ(D×R)\mathbb Z^{(D\times R)}. sc⁡((Egq∘v,a(fj))j)=e(g,ρ).\operatorname{sc}\bigl((E_g^{q\circ v,a}(f_j))_j\bigr)=e_{(g,\rho)}.(7)✦

In the paper, the result reads: for the aligned presentation, with ρz\rho_z denoting the final relator index corresponding to zz, and fixed finite normal closure expressions z=∏juz,jrρz,jσz,juz,j−1z=\prod_j u_{z,j}r_{\rho_{z,j}}^{\sigma_{z,j}}u_{z,j}^{-1}, one has ∑jσz,ju‾z,jeρz,j=eρz\sum_j\sigma_{z,j}\bar u_{z,j}e_{\rho_{z,j}}=e_{\rho_z} in the free module on all final relator indices.

What Is Checked

The claims are stated over arbitrary groups and alphabets: (6) requires closed relators with injective boundary map and a specified normal closure expression with the prescribed based chain, while (7) requires the count identity and allows arbitrary homomorphisms and HH-valued edge labels.