OctoInformBeta

OpenAI preprint, September 2026 · §5. Lagrangian counts and ordered triple orbits · §5

§5 (setting): Lagrangians, the matrix chart and the kernel J: Self-orthogonal subspaces of V⊕V*V\oplus V^{*}, graphs of symmetric matrices A,BA,B meeting in dimension D−rank⁡(A−B)D-\operatorname{rank}(A-B), and the kernel 𝟏{dim⁡(L∩M)=r}/πD,r\mathbf{1}\{\dim(L\cap M)=r\}/\pi_{D,r}

Lagrangian subspaces, their matrix chart and an intersection kernel.

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 kernel JJ in the sense of graph limits, a weighted graph on a finite vertex set, is built here from symplectic linear algebra. Its vertices are the Lagrangian subspaces of a symplectic space of dimension 2D2D, for a natural number DD, over a finite field KK, such as 𝔽q\mathbb{F}_q, taken with the uniform measure, and the weight of a pair of Lagrangians depends only on the dimension of their intersection.

Symmetric matrices give a chart on this vertex set: a symmetric matrix XX names the Lagrangian ΓX\Gamma_X, its graph. Whether two such graphs meet in a given dimension becomes a question about the difference of the two matrices.

The Symplectic Space and Symmetric Matrices

Let KK be a field, VV a vector space over KK and V*V^{*} its dual, and write the elements of V⊕V*V\oplus V^{*} as pairs (x,ξ)(x,\xi) with x∈Vx\in V and ξ∈V*\xi\in V^{*}. The canonical symplectic form ωV\omega_V on V⊕V*V\oplus V^{*} is the bilinear form given by ωV((x,ξ),(y,η))=η(x)−ξ(y).\omega_V\bigl((x,\xi),(y,\eta)\bigr)=\eta(x)-\xi(y).(1)✦ The vertical subspace of V⊕V*V\oplus V^{*} is the kernel 0⊕V*0\oplus V^{*} of the projection onto VV, the set of pairs (0,ξ)(0,\xi) with ξ∈V*\xi\in V^{*}.✦

For a natural number DD, the space SDS_D of symmetric matrices of size DD over KK is the subspace of the D×DD\times D matrices over KK consisting of those XX with X𝖳=XX^{\mathsf T}=X.✦

The Claim

Definition 1. Let ω\omega be a bilinear form on a vector space EE over KK, and let the orthogonal of a subspace LL of EE be L⟂ω={y∈E:ω(x,y)=0for allx∈L}L^{\perp_\omega}=\{y\in E:\ \omega(x,y)=0\ \text{for all}\ x\in L\}. A Lagrangian subspace of EE for ω\omega is a subspace LL of EE equal to its orthogonal, and these form the set Lag(E,ω)={L≤E:L⟂ω=L}.\mathrm{Lag}(E,\omega)=\{L\le E:\ L^{\perp_\omega}=L\}.(2)✦ A Lagrangian of VV is a Lagrangian subspace of V⊕V*V\oplus V^{*} for the canonical form (1), and these form the set Lag(V)=Lag(V⊕V*,ωV).\mathrm{Lag}(V)=\mathrm{Lag}\bigl(V\oplus V^{*},\omega_V\bigr).(3)✦ Given a subspace UU of VV and a bilinear form BB on UU, the graph of BB over UU is the subspace ΓU(B)={(x,ξ)∈V⊕V*:x∈Uandξ(u)=B(x,u)for allu∈U}\Gamma_U(B)=\{(x,\xi)\in V\oplus V^{*}:\ x\in U\ \text{and}\ \xi(u)=B(x,u)\ \text{for all}\ u\in U\}(4)✦ of V⊕V*V\oplus V^{*}. The graph of a symmetric matrix X∈SDX\in S_D is the subspace ΓX={(x,ξ)∈KD⊕(KD)*:ξ(y)=x𝖳Xyfor ally∈KD}\Gamma_X=\{(x,\xi)\in K^D\oplus (K^D)^{*}:\ \xi(y)=x^{\mathsf T}Xy\ \text{for all}\ y\in K^D\}(5)✦ of KD⊕(KD)*K^D\oplus (K^D)^{*}. For a natural number rr, the stratum Lagr(V)\mathrm{Lag}_r(V) is the set of Lagrangians of VV whose intersection with the vertical subspace has dimension rr.✦ When KK is finite, the stratum mass πD,r\pi_{D,r} is the proportion πD,r=|Lagr(KD)||Lag(KD)|,\pi_{D,r}=\frac{|\mathrm{Lag}_r(K^D)|}{|\mathrm{Lag}(K^D)|},(6)✦ and the kernel JJ is the real function on pairs of Lagrangians L,M∈Lag(KD)L,M\in\mathrm{Lag}(K^D) given by J(L,M)={πD,r−1ifdim⁡(L∩M)=r,0otherwise.J(L,M)=\begin{cases}\pi_{D,r}^{-1} & \text{if}\ \dim(L\cap M)=r,\\ 0 & \text{otherwise.}\end{cases}(7)✦

Theorem 1. Let KK be a field, DD a natural number and A,B∈SDA,B\in S_D two symmetric matrices over KK. Then the graphs of AA and BB satisfy dim⁡(ΓA∩ΓB)+rank⁡(A−B)=D.\dim(\Gamma_A\cap\Gamma_B)+\operatorname{rank}(A-B)=D.(8)✦

In the paper, the result reads: in a symplectic space of dimension 2N2N, NN being the paper’s name for the DD above, a Lagrangian is a self-orthogonal subspace; ΓX={(x,Xx)}\Gamma_X=\{(x,Xx)\} is the graph chart of a symmetric matrix XX; and J(L,M)=𝟏{dim⁡(L∩M)=r}/πD,rJ(L,M)=\mathbf{1}\{\dim(L\cap M)=r\}/\pi_{D,r}.

What Is Checked

The definitions and the rank identity (8) are stated over a field KK and a vector space VV over KK with its dual, the canonical form (1) and the vertical subspace 0⊕V*0\oplus V^{*}, and, where matrices enter, over V=KDV=K^D with the space SDS_D of symmetric matrices, the strata Lagr(KD)\mathrm{Lag}_r(K^D) and the set Lag(KD)\mathrm{Lag}(K^D) of Lagrangians of KDK^D on which JJ lives.

Beyond these definitions, the library proves that the graph ΓX\Gamma_X of every symmetric matrix X∈SDX\in S_D is a Lagrangian of KDK^D.