OctoInformBeta

OpenAI preprint, September 2026 · §3. Quadratic signature inequalities · §3

The quadratic signature property: The slot signature property of ff: every face Hessian of every iterated derivative of ff along pairs has at most one positive direction

A signature condition on the face Hessians of slot totals.

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

OpenAI’s preprint Approximate counting of common bases of two matroids (2026), the paper behind the library, is about approximately counting the common bases of two matroids M1M_1 and M2M_2 on {1,…,n}\{1,\dots,n\}. The analysis of the counting algorithm, from its third section to its sixth, runs over a finite set Γ\Gamma, a list of disjoint pairs {x1,y1},…,{xn,yn}\{x_1,y_1\},\dots,\{x_n,y_n\} of elements of Γ\Gamma and a positive function ff on the subsets of Γ\Gamma. In that analysis the matroids enter only through one hypothesis on ff, the slot signature property: a condition, of the kind that defines Lorentzian polynomials, on the signature of quadratic forms read off from the derivatives of ff along pairs.

Derivatives Along Pairs

The multiaffine polynomial FF of a real function ff on the subsets of Γ\Gamma, in the variables ziz_i, i∈Γi\in\Gamma, is the polynomial (1); ff is its coefficient function. F(z)=∑S⊆Γf(S)∏i∈SziF(z)=\sum_{S\subseteq\Gamma} f(S)\prod_{i\in S} z_i(1)

For a vector a∈ℝΓa\in\mathbb{R}^{\Gamma}, the derivative of ff along aa is the function ∂af\partial_a f on the subsets of Γ\Gamma given by (∂af)(A)=∑i∈Γ\Aaif(A∪{i}),(\partial_a f)(A)=\sum_{i\in\Gamma\setminus A} a_i\, f(A\cup\{i\}),(2)✦ the coefficient function of the derivative ∑i∈Γai∂F/∂zi\sum_{i\in\Gamma} a_i\,\partial F/\partial z_i of (1) in the direction aa. For a list a1,…,aka_1,\dots,a_k of vectors in ℝΓ\mathbb{R}^{\Gamma}, the iterated derivative of ff along the list is the function ∂a1,…,akf=∂a1∂a2⋯∂akf,\partial_{a_1,\dots,a_k} f=\partial_{a_1}\partial_{a_2}\cdots\partial_{a_k} f,(3)✦ and ff itself for the empty list.

The indicator vector 𝟏E∈ℝΓ\mathbf{1}_E\in\mathbb{R}^{\Gamma} of a set E⊆ΓE\subseteq\Gamma has entry 11 at the elements of EE and 00 elsewhere.✦ The direction of a pair p=(x,y)p=(x,y) of elements of Γ\Gamma is the indicator vector 𝟏p=𝟏{x,y}\mathbf{1}_p=\mathbf{1}_{\{x,y\}} of the set {x,y}\{x,y\}.✦

The slot total of ff along a finite list P=(p1,…,pk)P=(p_1,\dots,p_k) of pairs of elements of Γ\Gamma is the iterated derivative TPf=∂𝟏p1,…,𝟏pkfT_P f=\partial_{\mathbf{1}_{p_1},\dots,\mathbf{1}_{p_k}} f of ff along their directions.✦ When the pairs are disjoint and avoid a set AA, its value (TPf)(A)(T_P f)(A) is the sum of ff over the sets obtained from AA by adding one element of each pair.

Face Hessians and One Positive Direction

The face Hessian of a real function gg on the subsets of Γ\Gamma at a set A⊆ΓA\subseteq\Gamma is the matrix HA(g)H_A(g) with rows and columns indexed by Γ\A\Gamma\setminus A and entries HA(g)ij={g(A∪{i,j}),i≠j,0,i=j.H_A(g)_{ij}=\begin{cases} g(A\cup\{i,j\}), & i\neq j,\\ 0, & i=j.\end{cases}(4)✦ It is the Hessian at the origin, in the variables outside AA, of the polynomial with coefficient function gg differentiated once in each variable of AA.

A bilinear form BB on a real vector space WW has at most one positive direction when, for all e,v∈We,v\in W, B(e,e)>0andB(v,e)=0⇒B(v,v)≤0.B(e,e)>0\ \text{and}\ B(v,e)=0\quad\Longrightarrow\quad B(v,v)\le 0.(5)✦ A real matrix HH with rows and columns indexed by a finite set II has at most one positive direction when the form (u,v)↦u⊤Hv(u,v)\mapsto u^{\top}Hv on ℝI\mathbb{R}^{I} does.

The Definition

Definition 1. Let Γ\Gamma be a finite set. A function ff from the subsets of Γ\Gamma to ℝ\mathbb{R} has the slot signature property if the face Hessians of its slot totals satisfy HA(TPf)has at most one positive directionH_A(T_P f)\ \text{has at most one positive direction}(6)✦ for every finite list PP of pairs of elements of Γ\Gamma and every set A⊆ΓA\subseteq\Gamma.

In the paper, the definition reads: a polynomial has the quadratic signature property if every homogeneous quadratic obtained from it by partial differentiation and nonnegative linear substitutions has a Hessian with at most one positive eigenvalue.

What Is Checked

The definition (6) is stated over a real function ff on the subsets of a finite set Γ\Gamma, through its slot totals, the iterated derivatives (3) built from (2) along the directions of finite lists of pairs, their face Hessians (4) at every subset of Γ\Gamma, and the condition (5) of at most one positive direction.