OctoInformBeta

OpenAI preprint, September 2026 · §7. The singular contribution · §7

Proposition 7.1 [Global profile cost]: For thirteen random Lagrangians of a 2D2D-dimensional symplectic space over a field of odd order qq, at most κD11q−𝒞\kappa_D^{11}q^{-\mathcal{C}} if all cj≤Dc_j\le D

The profile cost bounds the probability of a point profile.

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

Let DD be a natural number. Attach to each of 13 points a Lagrangian subspace of a symplectic space of dimension 2D2D over a finite field with an odd number qq of elements, chosen independently and uniformly at random. The points are one vertex class of the bipartite graph HH of the counterexample to Sidorenko’s conjecture, whose other class is 22 triples of points, the faces, each point joined to the faces containing it. The question is how likely a prescribed pattern of intersection dimensions along the faces and along the pairs of points inside them is, as a power of qq.

Lagrangian Subspaces and Point Profiles

A symplectic space (E,ω)(E,\omega) over a field KK is a finite-dimensional vector space EE over KK with a nondegenerate alternating bilinear form ω\omega on it. A Lagrangian subspace of EE is a subspace LL of EE equal to its orthogonal: L=L⟂={y∈E:ω(x,y)=0for allx∈L}.L=L^{\perp}=\{y\in E:\ \omega(x,y)=0\ \text{for all}\ x\in L\}.(1)✦ The set of Lagrangian subspaces of EE is written ℒ\mathcal{L}.

The 22 faces are the following three-element subsets of the points {0,…,12}\{0,\dots,12\}, each face jj listed with a label βj∈{0,1}3\beta_j\in\{0,1\}^3: jpointsβjjpointsβj0{0,1,3}00011{3,10,11}1011{0,1,9}01012{4,8,9}1012{0,2,3}00113{4,8,11}0113{0,2,9}11014{5,6,7}0004{1,3,10}00115{5,6,11}1105{1,7,10}00016{5,7,10}0106{1,7,12}10017{5,10,11}1117{1,9,12}11018{6,7,12}0018{2,3,4}10119{6,8,11}0109{2,4,9}10020{6,8,12}01110{3,4,11}11121{8,9,12}111\begin{array}{rllrll} j & \text{points} & \beta_j & \qquad j & \text{points} & \beta_j\\ 0 & \{0,1,3\} & 000 & \qquad 11 & \{3,10,11\} & 101\\ 1 & \{0,1,9\} & 010 & \qquad 12 & \{4,8,9\} & 101\\ 2 & \{0,2,3\} & 001 & \qquad 13 & \{4,8,11\} & 011\\ 3 & \{0,2,9\} & 110 & \qquad 14 & \{5,6,7\} & 000\\ 4 & \{1,3,10\} & 001 & \qquad 15 & \{5,6,11\} & 110\\ 5 & \{1,7,10\} & 000 & \qquad 16 & \{5,7,10\} & 010\\ 6 & \{1,7,12\} & 100 & \qquad 17 & \{5,10,11\} & 111\\ 7 & \{1,9,12\} & 110 & \qquad 18 & \{6,7,12\} & 001\\ 8 & \{2,3,4\} & 101 & \qquad 19 & \{6,8,11\} & 010\\ 9 & \{2,4,9\} & 100 & \qquad 20 & \{6,8,12\} & 011\\ 10 & \{3,4,11\} & 111 & \qquad 21 & \{8,9,12\} & 111 \end{array}(2)✦ The 33 pairs are the sides of the faces: each pair ee is a set of two points and lies in exactly two faces. A profile is a pair (c,h)(c,h) of functions with natural number values, cc on the faces and hh on the pairs.

The point profile of a 13-tuple L=(L0,…,L12)L=(L_0,\dots,L_{12}) in ℒ13\mathcal{L}^{13} is the profile (c,h)(c,h) given by the dimensions over KK cj=dim⁡⋂i∈jLi,he=dim⁡⋂i∈eLi.c_j=\dim\bigcap_{i\in j}L_i,\qquad h_e=\dim\bigcap_{i\in e}L_i.(3)✦

The probability P(c,h)P(c,h) of a profile (c,h)(c,h) is the proportion of 13-tuples in ℒ13\mathcal{L}^{13} with point profile (c,h)(c,h): P(c,h)=|{L∈ℒ13:Lhas point profile(c,h)}||ℒ|13.P(c,h)=\frac{\bigl|\{L\in\mathcal{L}^{13}:\ L\ \text{has point profile}\ (c,h)\}\bigr|} {|\mathcal{L}|^{13}}.(4)✦

The Profile Cost

The separation count σe\sigma_e of a pair ee is the Hamming distance between the labels of the two faces containing ee, a number from 00 to 33.✦

The face cost CjC_j of a face jj, for a profile (c,h)(c,h) in dimension DD, is the natural number Cj=(D+1)cj+∑e((he−cj)++12),C_j=(D+1)\,c_j+\sum_{e}\binom{(h_e-c_j)_+ +1}{2},(5)✦ the sum running over the three sides ee of jj, where x+=max⁡(x,0)x_+=\max(x,0).

The profile cost of a profile (c,h)(c,h) in dimension DD is the real number 𝒞D(c,h)=12∑jCj−16∑e(3−σe)b(he),\mathcal{C}_D(c,h)=\frac12\sum_{j}C_j-\frac16\sum_{e}(3-\sigma_e)\,b(h_e),(6)✦ the sums running over the 22 faces and the 33 pairs, where b(n)=n(n+1)/2b(n)=n(n+1)/2.

Let ΛD=(D+1)2D\Lambda_D=(D+1)\,2^D be the Lagrangian constant✦ and MD=ΛD44DM_D=\Lambda_D^4\,4^D the profile mass constant of a natural number DD.✦ The full profile constant κD\kappa_D of DD is the real number κD=ΛD2DMD=(D+1)528D,\kappa_D=\Lambda_D\,2^D M_D=(D+1)^5\,2^{8D},(7)✦

The Claim

Theorem 1 (Global profile cost). Let KK be a finite field of odd characteristic with qq elements, and let (E,ω)(E,\omega) be a symplectic space of dimension 2D2D over KK. Let (c,h)(c,h) be a profile with cj≤Dc_j\le D for every face jj. Then the number of 13-tuples of Lagrangian subspaces of EE with point profile (c,h)(c,h) satisfies |{L∈ℒ13:Lhas point profile(c,h)}|≤κD11|ℒ|13q−𝒞D(c,h),\bigl|\{L\in\mathcal{L}^{13}:\ L\ \text{has point profile}\ (c,h)\}\bigr| \le\kappa_D^{11}\,|\mathcal{L}|^{13}\,q^{-\mathcal{C}_D(c,h)},(8)✦ and the probability of the profile satisfies P(c,h)≤κD11q−𝒞D(c,h).P(c,h)\le\kappa_D^{11}\,q^{-\mathcal{C}_D(c,h)}.(9)✦

In the paper, the result reads: for every prescribed profile (c,h)(c,h), its probability is OD(q−𝒞)O_D(q^{-\mathcal{C}}) with 𝒞=12(∑jCj−∑e(1−ae)b(he))\mathcal{C}=\frac12\bigl(\sum_j C_j-\sum_e(1-a_e)\,b(h_e)\bigr), where ae=σe/3a_e=\sigma_e/3.

What Is Checked

The claim is stated over a symplectic space of dimension 2D2D over a finite field of odd characteristic, its Lagrangian subspaces, the point profile of a 13-tuple of them along the 22 faces and the 33 pairs, the profile cost 𝒞D(c,h)\mathcal{C}_D(c,h) built from the face costs and the separation counts, and the constant κD\kappa_D.