OctoInform Beta

OpenAI preprint, September 2026 · §3. Deformation inside a division algebra · §3

Lemma 3.3 [An invertible witness for a nonzero formula]: A formula defined and nonzero at rational d×dd\times d matrices is defined and invertible at d×dd\times d matrices over the Hahn series ℂ((ℤ×ℤ))\mathbb{C}(\!(\mathbb{Z}\times\mathbb{Z})\!)

A nonzero formula takes an invertible value over Hahn series.

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

Take a rational formula ff in nn variables and rational d×dd\times d matrices W1,…,WnW_1,\dots,W_n, d>0d>0, at which ff has a defined value a≠0a\neq0. The value aa may be singular, and then the tuple W=(W1,…,Wn)W=(W_1,\dots,W_n) does not hit ff, that is, ff has no invertible value at WW. The result of this page, Lemma 3.3 of the paper, passes to d×dd\times d matrices over a field of Hahn series (formal series with well-ordered support) with complex coefficients and exponents in ℤ×ℤ\mathbb{Z}\times\mathbb{Z}, and finds there a point at which ff is defined with an invertible value. The point is built from WW: each WiW_i is lifted to a series Wi*W_i^* whose coefficients are multiples of the Weyl monomials, certain matrices defined below, and the series is read as a d×dd\times d matrix of Hahn series.

Formulas and Their Values

A rational formula ff in nn variables is a finite tree built from the variables x1,…,xnx_1,\dots,x_n and rational constants by addition f+gf+g, ordered multiplication f⋅gf\cdot g and inversion f−1f^{-1}.✦

Let RR be a ring receiving ℚ\mathbb{Q} through a ring homomorphism c:ℚ→Rc\colon\mathbb{Q}\to R, and let X=(X1,…,Xn)∈RnX=(X_1,\dots,X_n)\in R^n. The evaluation at XX is the least relation f(X)↓af(X)\downarrow a, read “ff has value aa at XX”, between formulas ff and elements aa of RR that obeys the following rules, where qq is a rational constant and a′a' an element of RR xi(X)↓Xi,q(X)↓c(q),f(X)↓ag(X)↓a′(f+g)(X)↓a+a′(f⋅g)(X)↓aa′,f(X)↓aaa′=a′a=1f−1(X)↓a′.x_i(X)\downarrow X_i,\qquad q(X)\downarrow c(q),\qquad \frac{f(X)\downarrow a\quad g(X)\downarrow a'}{(f+g)(X)\downarrow a+a'\quad (f\cdot g)(X)\downarrow aa'},\qquad \frac{f(X)\downarrow a\quad aa'=a'a=1}{f^{-1}(X)\downarrow a'}.(1)✦ In a matrix ring Matd(R)\mathrm{Mat}_d(R) the constant qq is the scalar matrix qIdqI_d. A ring homomorphism ψ:R→R′\psi\colon R\to R' makes R′R' receive ℚ\mathbb{Q} through ψ∘c\psi\circ c, and ψ(X)\psi(X) denotes (ψ(X1),…,ψ(Xn))(\psi(X_1),\dots,\psi(X_n)).

Hahn Series and Weyl Monomials

For a ring AA and an ordered abelian group GG, A((G))A(\!(G)\!) is the ring of Hahn series z=∑g∈Gzgtgz=\sum_{g\in G}z_gt^g with coefficients zg∈Az_g\in A and well-ordered support. The exponent group Γ\Gamma is ℤ×ℤ\mathbb{Z}\times\mathbb{Z} under addition, ordered lexicographically: first coordinates are compared first.✦ For a finite set ι\iota, the splitting σ:Matι(A)((G))→Matι(A((G)))\sigma\colon\mathrm{Mat}_\iota(A)(\!(G)\!)\to\mathrm{Mat}_\iota(A(\!(G)\!)) is the ring homomorphism that sends a series zz with matrix coefficients to the matrix of its entry series zij=∑g(zg)ijtgz_{ij}=\sum_g(z_g)_{ij}t^g.✦

Fix a field KK of characteristic zero, an integer d>0d>0 and an additive character χ\chi of ℤ/dℤ\mathbb{Z}/d\mathbb{Z} with values in KK; the d×dd\times d matrices over KK have rows and columns indexed by ℤ/dℤ\mathbb{Z}/d\mathbb{Z}. Let SS be the cyclic shift, the matrix with ones at the positions (i+1,i)(i+1,i), and let Da=diag(χ(ai))iD^a=\mathrm{diag}(\chi(ai))_i be the clock matrix. For residues a,ba,b, the Weyl monomial DaSbD^aS^b is the product of DaD^a and SbS^b, the matrix with entries (DaSb)ij={χ(ai)if i=j+b,0otherwise,(D^aS^b)_{ij}=\begin{cases}\chi(ai)&\text{if }i=j+b,\\0&\text{otherwise},\end{cases}(2) For g=(a,b)∈Γg=(a,b)\in\Gamma, the monomial unit ugu_g is the invertible matrix DaSbD^aS^b with aa and bb reduced modulo dd.✦

The ambient ring 𝒜\mathcal{A} is the ring Matd(K)((Γ))\mathrm{Mat}_d(K)(\!(\Gamma)\!) of Hahn series with exponents in Γ\Gamma whose coefficients are d×dd\times d matrices over KK.✦ The Fourier coefficients of a d×dd\times d matrix WW over KK are Wˆ(a,b)=1d∑j∈ℤ/dℤχ(−aj)Wj,j−b,\hat W(a,b)=\frac1d\sum_{j\in\mathbb{Z}/d\mathbb{Z}}\chi(-aj)\,W_{j,\,j-b},(3)✦ and the lift of WW is the element of 𝒜\mathcal{A} W*=∑a,b∈ℤ/dℤWˆ(a,b)DaSbt(a,b),W^*=\sum_{a,b\in\mathbb{Z}/d\mathbb{Z}}\hat W(a,b)\,D^aS^b\,t^{(a,b)},(4)✦ with aa and bb in the exponent taken as their representatives in {0,…,d−1}\{0,\dots,d-1\}.

Let AA be an algebra over KK and (eg)g∈G(e_g)_{g\in G} a family of units of AA with e0=1e_0=1 and each product egehe_ge_h a scalar multiple of eg+he_{g+h}. The coefficient line of gg is the subspace KegKe_g of AA spanned by ege_g✦, and the quantum coefficient ring of the family is the subring of A((G))A(\!(G)\!) of the series whose coefficient at every gg lies in KegKe_g.✦ The division subring 𝒟χ\mathcal{D}_\chi is the quantum coefficient ring of the monomial units in 𝒜\mathcal{A}: the series whose coefficient at every (a,b)(a,b) is a scalar multiple of DaSbD^aS^b.✦ It contains the lift W*W^* of every d×dd\times d matrix WW over KK.✦ The representation πχ:𝒟χ→Matd(K((Γ)))\pi_\chi\colon\mathcal{D}_\chi\to\mathrm{Mat}_d(K(\!(\Gamma)\!)) is the restriction of the splitting σ\sigma to 𝒟χ\mathcal{D}_\chi.✦

The Claim

Theorem 1. (i) Let ff be a rational formula in nn variables, d>0d>0, and W1,…,WnW_1,\dots,W_n rational d×dd\times d matrices at which ff has a value a≠0a\neq0. Then there are d×dd\times d matrices X1,…,XnX_1,\dots,X_n and vv over the field ℂ((Γ))\mathbb{C}(\!(\Gamma)\!) with f(X1,…,Xn)↓vandv∈GLd(ℂ((Γ))).f(X_1,\dots,X_n)\downarrow v\quad\text{and}\quad v\in\mathrm{GL}_d\bigl(\mathbb{C}(\!(\Gamma)\!)\bigr).(5)✦ (ii) Let ff and dd be as in (i), KK a field of characteristic zero, χ\chi a primitive additive character of ℤ/dℤ\mathbb{Z}/d\mathbb{Z} with values in KK (that is, an injective one), and W1,…,WnW_1,\dots,W_n matrices in Matd(K)\mathrm{Mat}_d(K) at which ff has a value a≠0a\neq0. Then there is a matrix vv in Matd(K((Γ)))\mathrm{Mat}_d(K(\!(\Gamma)\!)) with f(πχ(W1*),…,πχ(Wn*))↓vandv∈GLd(K((Γ))).f\bigl(\pi_\chi(W_1^*),\dots,\pi_\chi(W_n^*)\bigr)\downarrow v\quad\text{and}\quad v\in\mathrm{GL}_d\bigl(K(\!(\Gamma)\!)\bigr).(6)✦

In the paper, the result reads: every admissible nonzero rational formula over ℚ\mathbb{Q} has a defined evaluation at matrices over an extension field of ℚ\mathbb{Q} at which its value is invertible. Here admissible and nonzero refer to the evaluations of the formula at rational d×dd\times d matrices with d>0d>0.

What Is Checked

The claim is stated over rational formulas and their evaluation (1), at d×dd\times d matrices over ℚ\mathbb{Q}, KK, ℂ((Γ))\mathbb{C}(\!(\Gamma)\!) and K((Γ))K(\!(\Gamma)\!) with scalar constants; its second part is stated over the lifts (4), the division subring 𝒟χ\mathcal{D}_\chi and the representation πχ\pi_\chi.

The library proves that χ(−ab)D−aS−b\chi(-ab)D^{-a}S^{-b} is a left✦ and a right inverse of DaSbD^aS^b✦. The monomial units satisfy u(0,0)=Idu_{(0,0)}=I_d✦, and each product uguhu_gu_h is a scalar multiple of ug+hu_{g+h}.✦

Evaluation follows ring homomorphisms. Let ψ:R→R′\psi\colon R\to R' be a ring homomorphism and X∈RnX\in R^n. If f(X)↓af(X)\downarrow a in RR, then f(ψ(X))↓ψ(a)in R′.f(\psi(X))\downarrow\psi(a)\quad\text{in }R'.(7)✦ If ψ\psi reflects units, so that r∈Rr\in R is a unit whenever ψ(r)\psi(r) is, and f(ψ(X))↓af(\psi(X))\downarrow a in R′R', then there is v∈Rv\in R with f(X)↓vandψ(v)=a.f(X)\downarrow v\quad\text{and}\quad\psi(v)=a.(8)✦ If RR is a division ring, R′R' is a nontrivial ring and f(ψ(X))↓af(\psi(X))\downarrow a with a≠0a\neq0, then a∈(R′)×.a\in(R')^{\times}.(9)✦

For a ring homomorphism φ\varphi from a commutative ring Λ\Lambda to a field LL, the denominators of φ\varphi are the elements pp of Λ\Lambda with φ(p)≠0\varphi(p)\neq0, a multiplicative subset of Λ\Lambda.✦ The regular ring Λφ\Lambda_\varphi of φ\varphi is the localization of Λ\Lambda at the denominators of φ\varphi.✦ The evaluation φˆ:Λφ→L\hat\varphi\colon\Lambda_\varphi\to L is the ring homomorphism p/s↦φ(p)φ(s)−1p/s\mapsto\varphi(p)\varphi(s)^{-1} that extends φ\varphi.✦ This evaluation reflects units, for elements λ\lambda of Λφ\Lambda_\varphi and for square matrices YY over Λφ\Lambda_\varphi indexed by a finite set ι\iota, with φˆ\hat\varphi applied entrywise: φˆ(λ)∈L×⇒λ∈Λφ×,\hat\varphi(\lambda)\in L^\times\ \Longrightarrow\ \lambda\in\Lambda_\varphi^\times,(10)✦ φˆ(Y)∈GLι(L)⇒Y∈GLι(Λφ).\hat\varphi(Y)\in\mathrm{GL}_\iota(L)\ \Longrightarrow\ Y\in\mathrm{GL}_\iota(\Lambda_\varphi).(11)✦