OctoInformBeta

OpenAI preprint, September 2026 · §4. The adjugate scheme in characteristic two

§4 (setting): spectral pencils and adjugate ideals: Cofactor ideals and permanent coefficients for affine pencils, with the gradient identity in characteristic two

Spectral pencils and their selector gradients.

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

A diagonal direction in matrix space gives a spectral pencil with two scalar parameters. The cofactor ideal and the critical equations of a selected permanent coefficient describe loci attached to this pencil.

Let RR be a commutative ring, let II be a finite index set, and let θ=(θi)i∈I\theta=(\theta_i)_{i\in I} be a family in RR. Write D=diag⁡(θi)D=\operatorname{diag}(\theta_i). For the affine constructions over a field KK, take a natural number kk; the matrix size is n=k+1n=k+1 and I={0,…,k}I=\{0,\ldots,k\}.

The binary linear form ℓλ∈K[s,t]\ell_\lambda\in K[s,t] associated with λ∈K\lambda\in K is given by (1). ℓλ(s,t)=s+λt.\ell_\lambda(s,t)=s+\lambda t.(1)✦ The diagonal binary pencil Sθ∈Mat⁡I(K[s,t])S_\theta\in\operatorname{Mat}_I(K[s,t]) is the diagonal matrix whose entry indexed by i∈Ii\in I is the binary linear form ℓθi\ell_{\theta_i}, as in (2). Sθ(s,t)=diag⁡(ℓθi(s,t))i∈I.S_\theta(s,t)=\operatorname{diag}\bigl(\ell_{\theta_i}(s,t)\bigr)_{i\in I}.(2) For A∈Mat⁡I(K)A\in\operatorname{Mat}_I(K), the bivariate spectral pencil Mθ,A∈Mat⁡I(K[s,t])M_{\theta,A}\in\operatorname{Mat}_I(K[s,t]) is the matrix in (3). Mθ,A(s,t)=Sθ(s,t)+A.M_{\theta,A}(s,t)=S_\theta(s,t)+A.(3)✦

Coefficient Selection

Let U=(uij)i,j∈IU=(u_{ij})_{i,j\in I} be the generic matrix, with independent variables 𝒖=(uij)\mathbf u=(u_{ij}). The generic affine pencil Pθ(U)∈Mat⁡I(R[t][s][𝒖])P_\theta(U)\in\operatorname{Mat}_I(R[t][s][\mathbf u]) is defined by (4). Pθ(U)=sI+tD+U.P_\theta(U)=sI+tD+U.(4)✦ For x∈Mat⁡I(R)x\in\operatorname{Mat}_I(R), the specialized affine pencil Pθ(x)∈Mat⁡I(R[t][s])P_\theta(x)\in\operatorname{Mat}_I(R[t][s]) is defined by (5). Pθ(x)=sI+tD+x.P_\theta(x)=sI+tD+x.(5)✦

For natural numbers aa and bb, the bivariate coefficient map ℰa,b:R[t][s][𝒖]→R[𝒖]\mathcal E_{a,b}:R[t][s][\mathbf u]\to R[\mathbf u] is the additive map that extracts the coefficient of satbs^a t^b from every coefficient polynomial, as specified by (6), where α\alpha ranges over exponent vectors of the variables 𝒖\mathbf u. ℰa,b(∑αfα(s,t)𝒖α)=∑α[tb]([sa]fα(s,t))𝒖α.\mathcal E_{a,b}\!\left(\sum_\alpha f_\alpha(s,t)\mathbf u^\alpha\right) =\sum_\alpha [t^b]\bigl([s^a]f_\alpha(s,t)\bigr)\mathbf u^\alpha.(6) For a natural number qq, the affine selector Gθ,q∈R[𝒖]G_{\theta,q}\in R[\mathbf u] is the polynomial defined in (7), where per\operatorname{per} denotes the permanent. Gθ,q=ℰq,q(per⁡Pθ(U)).G_{\theta,q}=\mathcal E_{q,q}\bigl(\operatorname{per}P_\theta(U)\bigr).(7)✦

The diagonal bidegree associated with qq is the exponent vector (q,q)(q,q) on the two variables.✦ For a matrix BB over a commutative ring and weights θ\theta in that ring, its diagonal permanent coefficient Hθ,q(B)H_{\theta,q}(B) is the scalar defined by (8). Hθ,q(B)=[sqtq]per⁡(B+sI+tD).H_{\theta,q}(B)=[s^q t^q]\operatorname{per}(B+sI+tD).(8)✦ In particular, Hθ,q(U)H_{\theta,q}(U) is a polynomial in R[𝒖]R[\mathbf u].

The Claim

Definition 1 (Spectral pencils and their ideals). Let KK be a field, let A∈Mat⁡n(K)A\in\operatorname{Mat}_n(K), and let θ∈Kn\theta\in K^n, where n=k+1n=k+1. The diagonal pencil Bθ,A(t)∈Mat⁡n(K[t])B_{\theta,A}(t)\in\operatorname{Mat}_n(K[t]) is defined by (9). Bθ,A(t)=tD+A.B_{\theta,A}(t)=tD+A.(9)✦ For a matrix B∈Mat⁡I(R)B\in\operatorname{Mat}_I(R) over a commutative ring, its spectral pencil LB(s)∈Mat⁡I(R[s])L_B(s)\in\operatorname{Mat}_I(R[s]) is defined by (10). LB(s)=sI+B.L_B(s)=sI+B.(10)✦ The cofactor ideal 𝒞(B)\mathcal C(B) is the ideal of R[s]R[s] generated by all entries of the adjugate of LB(s)L_B(s), as in (11). 𝒞(B)=⟨adj(LB(s))ij:i,j∈I⟩⊆R[s].\mathcal C(B)= \left\langle\operatorname{adj}(L_B(s))_{ij}:i,j\in I\right\rangle \subseteq R[s].(11)✦ For B=Bθ,A(t)B=B_{\theta,A}(t), this is an ideal in K[t][s]K[t][s].

The spectral curve equation fθ,A∈K[s,t]f_{\theta,A}\in K[s,t] is the determinant specified in (12). fθ,A(s,t)=det⁡Mθ,A(s,t).f_{\theta,A}(s,t)=\det M_{\theta,A}(s,t).(12)✦

The selector gradient ideal Jθ,qJ_{\theta,q} is the ideal of K[𝒖]K[\mathbf u] generated by every entrywise partial derivative of Hθ,q(U)H_{\theta,q}(U), as in (13). Jθ,q=⟨∂Hθ,q(U)∂uij:i,j∈I⟩⊆K[𝒖].J_{\theta,q}= \left\langle \frac{\partial H_{\theta,q}(U)}{\partial u_{ij}}:i,j\in I \right\rangle \subseteq K[\mathbf u].(13)✦

Theorem 1 (The selector and its gradient). Let RR be a commutative ring, let II be finite, let θ∈RI\theta\in R^I, and let qq be a natural number. For the generic matrix UU, the diagonal coefficient and affine selector satisfy (14). Hθ,q(U)=Gθ,q.H_{\theta,q}(U)=G_{\theta,q}.(14)✦ Suppose that RR has characteristic two. For every x∈Mat⁡I(R)x\in\operatorname{Mat}_I(R) and a,b∈Ia,b\in I, the gradient identity is (15). ∂Gθ,q∂uab|U=x=[tq]([sq]adj⁡(Pθ(x))ba).\left.\frac{\partial G_{\theta,q}}{\partial u_{ab}}\right|_{U=x} =[t^q]\bigl([s^q]\operatorname{adj}(P_\theta(x))_{ba}\bigr).(15)✦

In the paper, the result reads: over K=𝔽¯2K=\overline{\mathbb F}_2, with matrix size n≥2n\ge2 (the paper’s kk), distinct θi\theta_i, q=⌊(n−1)/3⌋q=\lfloor(n-1)/3\rfloor, r=n−2qr=n-2q, c=r−1c=r-1 and T=(q+22)T=\binom{q+2}{2}, for an n×nn\times n matrix XX over KK and homogeneous coordinates s,t,us,t,u, set M=sI+tD+uXM=sI+tD+uX, CX=V(det⁡M)⊆ℙ2C_X=V(\det M)\subseteq\mathbb P^2 and let Z⊆CXZ\subseteq C_X be the subscheme cut out by all adjugate entries, while the gradient locus of the selector, denoted GDG_D in the paper, includes all entrywise partials and does not include the equation GD=0G_D=0.

What Is Checked

The field constructions in (9), (12) and (13) are stated in matrix size n=k+1n=k+1 over KK. The constructions in (10) and (11), and the coefficient identity (14), are stated over a commutative ring with a finite matrix index set; (15) requires characteristic two.