OctoInformBeta

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

Proposition 7.7 [Vanishing singular contribution]: For D=2r≥D0D=2r\ge D_0, the part of the density of an active subgraph of HH from non-transverse placements tends to 00 as q→∞q\to\infty

In large dimension the singular contribution tends to zero.

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

The graph HH is the incidence graph of 13 points and 22 faces, each face a triple of points: its vertices are the points and the faces, and each face is joined to its three points. The points of a face jj fill its slots 0,1,20,1,2 in increasing order, v0(j)<v1(j)<v2(j)v_0(j)<v_1(j)<v_2(j), and a corner (j,i)(j,i) is a face jj with one of its slots ii. Two points of a common face form a point pair; there are 33 point pairs, and each lies in exactly two faces.

Put a Lagrangian subspace LiL_i of a symplectic space over a finite field at each point ii, and a Lagrangian subspace YjY_j, the center of jj, at each face jj. An activity pattern TT marks at each face jj a set Tj⊆{0,1,2}T_j\subseteq\{0,1,2\} of active slots, and the edges ijij of HH from a face jj to the points ii in its active slots form the activity set BB. With the edge ijij weighted by J(Li,Yj)J(L_i,Y_j), for a kernel JJ on pairs of Lagrangian subspaces, the product of the weights over BB, averaged over all placements, is the homomorphism density t(B,J)t(B,J). The singular contribution is the part of this average that comes from placements in which the subspaces at the two points of some point pair meet nontrivially. The question is whether it vanishes as the field grows, with one threshold on the dimension for all activity patterns, and whether the same holds in a canonical symplectic space and in a parallel model built from symmetric matrices.

Lagrangian Subspaces and the Singular Tail

Let ω\omega be a symplectic form on a finite-dimensional vector space EE over a finite field KK. A Lagrangian subspace LL of EE is a subspace of EE equal to its own orthogonal L⟂ω={y∈E:ω(x,y)=0 for all x∈L}L^{\perp_\omega}=\{y\in E:\ \omega(x,y)=0\text{ for all }x\in L\}, and Λ\Lambda denotes the set of Lagrangian subspaces of EE.✦

Fix a Lagrangian subspace AA and a natural number rr. The center stratum Σr(A)\Sigma_r(A) is the set of Lagrangian subspaces LL with dim⁡(L∩A)=r\dim(L\cap A)=r.✦ For a triple 𝑳=(L0,L1,L2)\mathbf L=(L_0,L_1,L_2) of Lagrangian subspaces and a set T⊆{0,1,2}T\subseteq\{0,1,2\} of active slots, the active center set 𝒞r(𝑳;T)\mathcal C_r(\mathbf L;T) is the set of Lagrangian subspaces YY with dim⁡(Li∩Y)=r\dim(L_i\cap Y)=r for every i∈Ti\in T.✦ The kernel JJ on pairs of Lagrangian subspaces is the function J(L,Y)=|Λ||Σr(A)|𝟏[dim⁡(L∩Y)=r](L,Y∈Λ),J(L,Y)=\frac{|\Lambda|}{|\Sigma_r(A)|}\,\mathbf 1[\dim(L\cap Y)=r]\qquad(L,Y\in\Lambda),(1) and the active face density of 𝑳\mathbf L with active slots TT, centered at AA at rank rr, is the real number Φr(A;𝑳;T)=|𝒞r(𝑳;T)||Λ|(|Λ||Σr(A)|)|T|=1|Λ|∑Y∈Λ∏i∈TJ(Li,Y).\Phi_r(A;\mathbf L;T)=\frac{|\mathcal C_r(\mathbf L;T)|}{|\Lambda|} \Bigl(\frac{|\Lambda|}{|\Sigma_r(A)|}\Bigr)^{|T|} =\frac1{|\Lambda|}\sum_{Y\in\Lambda}\ \prod_{i\in T}J(L_i,Y).(2)✦

The actual singular tail Ψr(A;T)\Psi_r(A;T) of an activity pattern TT is the real number obtained by averaging the product of the face densities (2) over all placements 𝑳=(L0,…,L12)\mathbf L=(L_0,\dots,L_{12}) of Lagrangian subspaces at the 13 points, counting only the placements in which some point pair is not transverse, that is, in which the subspaces at its two points meet nontrivially: Ψr(A;T)=1|Λ|13∑𝑳∈Λ13:dim⁡(Li∩Lk)>0for some point pair {i,k}∏jΦr(A;(Lv0(j),Lv1(j),Lv2(j));Tj),\Psi_r(A;T)=\frac1{|\Lambda|^{13}} \sum_{\substack{\mathbf L\in\Lambda^{13}:\ \dim(L_i\cap L_k)>0\\ \text{for some point pair }\{i,k\}}} \ \prod_{j}\Phi_r\bigl(A;(L_{v_0(j)},L_{v_1(j)},L_{v_2(j)});T_j\bigr),(3)✦ the product running over the 22 faces.

The Majorant

A dimension profile in dimension DD is a pair (c,h)(c,h) of a map cc giving each face jj a number cj∈{0,…,D}c_j\in\{0,\dots,D\} and a map hh giving each point pair ee a number he∈{0,…,D}h_e\in\{0,\dots,D\}.✦ Slot 00 of a face jj holds the point pair {v0(j),v1(j)}\{v_0(j),v_1(j)\}, slot 11 the pair {v0(j),v2(j)}\{v_0(j),v_2(j)\} and slot 22 the pair {v1(j),v2(j)}\{v_1(j),v_2(j)\}✦; with ee the pair in slot ii, the profile sizes of jj are sj,i=he−cjs_{j,i}=h_e-c_j, the difference truncated at 00. A geometric profile in dimension DD is a dimension profile (c,h)(c,h) with natural values such that every face jj has cj≤hec_j\le h_e for its three point pairs ee, and sj,0+sj,1+sj,2≤D−cjs_{j,0}+s_{j,1}+s_{j,2}\le D-c_j.✦

Each face carries a three-bit label, the labels being given by a fixed table with one entry for each face, and the separation count 3ae3a_e of a point pair ee is the number of slots at which the labels of its two faces differ. With b(n)=n(n+1)/2b(n)=n(n+1)/2 the triangular function on real numbers, the profile cost of (c,h)(c,h) is P(c,h)=12∑j((D+1)cj+∑i=02(sj,i+12))−12∑e(1−ae)b(he),P(c,h)=\frac12\sum_j\Bigl((D+1)c_j+\sum_{i=0}^{2}\binom{s_{j,i}+1}{2}\Bigr) -\frac12\sum_e(1-a_e)\,b(h_e),(4)✦ The face natural cost of a face jj is the jj-th summand of the first sum in (4).✦

A tail tuple a=(c,u,v,p,t)a=(c,u,v,p,t) consists of natural numbers c,u,vc,u,v and triples p=(p0,p1,p2)p=(p_0,p_1,p_2) and t=(t0,t1,t2)t=(t_0,t_1,t_2) of natural numbers.✦ The tail sizes si=pi+tis_i=p_i+t_i, the tail heights hi=c+sih_i=c+s_i, the maximal tail height h‾=max⁡ihi\bar h=\max_ih_i and the tail offset δ=2u−c\delta=2u-c of aa are defined by these formulas. The planted gain of real numbers c,d,δ,vc,d,\delta,v and real triples p,tp,t is π(c,d,δ,v,p,t)=(d+c2+c+1)c+δ2−2(c+δ2)2+∑ib(pi+ti)−δ∑ipi−∑ipi2−v(d−3δ2−∑ipi+v)−∑ib(ti)+b(v),\begin{split} \pi(c,d,\delta,v,p,t)={}&\Bigl(\frac{d+c}2+c+1\Bigr)\frac{c+\delta}2 -2\Bigl(\frac{c+\delta}2\Bigr)^2+\sum_ib(p_i+t_i)-\delta\sum_ip_i-\sum_ip_i^2\\ &-v\Bigl(\frac{d-3\delta}2-\sum_ip_i+v\Bigr)-\sum_ib(t_i)+b(v), \end{split}(5)✦ the sums running over i∈{0,1,2}i\in\{0,1,2\}. The tail gain of aa in dimension DD with a bit β\beta is gD(1,a)=π(c,D−c,δ,v,p,t)g_D(1,a)=\pi(c,D-c,\delta,v,p,t) for β=1\beta=1 and gD(0,a)=b(⌊h‾/2⌋)g_D(0,a)=b(\lfloor\bar h/2\rfloor) for β=0\beta=0.✦

Fix natural numbers D,r,cD,r,c and sizes s=(s0,s1,s2)s=(s_0,s_1,s_2), and put d=D−cd=D-c. A full tail choice is a pair (u,θ)(u,\theta) of a common parameter, an integer u∈{0,…,c}u\in\{0,\dots,c\} with u≤ru\le r and r−u≤dr-u\le d, and a base parameter, a quadruple θ=(θ0,θ1,θ2,θ3)∈{0,…,d}4\theta=(\theta_0,\theta_1,\theta_2,\theta_3)\in\{0,\dots,d\}^4 feasible for ρ=r−u\rho=r-u: ρ≤θ3\rho\le\theta_3, θi+θk≤ρ\theta_i+\theta_k\le\rho for distinct i,k∈{0,1,2}i,k\in\{0,1,2\}, θ0+θ1+θ2+θ3≤3ρ\theta_0+\theta_1+\theta_2+\theta_3\le3\rho, and θi≤si\theta_i\le s_i, si+2ρ≤d+2θis_i+2\rho\le d+2\theta_i and 2ρ≤θ3+θi2\rho\le\theta_3+\theta_i for each i∈{0,1,2}i\in\{0,1,2\}.✦ A face tail choice with bit 11 is a full tail choice, and with bit 00 it is the single trivial choice.✦ The face tail tuple of a face tail choice is the planted tail tuple (c,u,d−θ3,(θ0,θ1,θ2),(si−θi)i)(c,u,d-\theta_3,(\theta_0,\theta_1,\theta_2),(s_i-\theta_i)_i) when the bit is 11 and the choice is (u,θ)(u,\theta), and the lower tail tuple (c,0,0,s,(0,0,0))(c,0,0,s,(0,0,0)) when the bit is 00.✦

Now let (c,h)(c,h) be a profile in dimension DD, rr a natural number and fullj\mathrm{full}_j a bit at each face jj. A global tail choice xx picks at each face jj a face tail choice xjx_j for DD, rr, cjc_j, the sizes (sj,0,sj,1,sj,2)(s_{j,0},s_{j,1},s_{j,2}) and the bit fullj\mathrm{full}_j.✦ The global tail tuple of a global tail choice xx at a face jj is the face tail tuple aj(x)a_j(x) of xjx_j✦, and the global tail gain of xx is G(x)=∑jgD(fullj,aj(x))G(x)=\sum_jg_D(\mathrm{full}_j,a_j(x)).✦ The face tail constant at a face with bit 00 is the lower tail constant max⁡{1,(r+1)4DCD2}\max\{1,(r+1)\,4^DC_D^2\}, where Cn=(n+1)2nC_n=(n+1)2^n is the Lagrangian constant; at a face with bit 11 and xj=(u,θ)x_j=(u,\theta) it is the full tail constant Fj(x)=ODCD32cj+uCD−u23(D−cj)2θ0+θ1+θ2+θ3+3(r−u)+∑i(sj,i−θi),F_j(x)=O_D\,C_D^3\,2^{c_j+u}\,C_{D-u}\,2^{3(D-c_j)}\, 2^{\theta_0+\theta_1+\theta_2+\theta_3+3(r-u)+\sum_i(s_{j,i}-\theta_i)},(6) The first factor ODO_D is the orbit mass constant, the product of CDC_D and the pair orbit constant 8⋅8DCD38\cdot8^DC_D^3, so that OD=8(D+1)427DO_D=8(D+1)^42^{7D}✦; the last factor is the reduced planted constant.✦

The global tail constant Γ(x)\Gamma(x) of a global tail choice xx is the product over the faces jj of the face tail constants of xx at jj.✦

The full profile constant is QD=2DCD⋅CD44D=(D+1)528DQ_D=2^DC_D\cdot C_D^44^D=(D+1)^52^{8D}, its last factor CD44DC_D^44^D being the profile mass constant; the residual direct constant is RD,h=2∑eheCD13R_{D,h}=2^{\sum_eh_e}C_D^{13}✦ and the residual nondirect constant is RD′=67⋅1313⋅2132CD13R'_D=67\cdot13^{13}\cdot2^{132}C_D^{13}✦. For a global tail choice xx on (c,h)(c,h) and a real number qq, the profile term majorant is m(x;q)={QD11qG(x)−P(c,h)if G(x)<P(c,h),RD,hq−∑eb(he)/3+RD′q1641−Dotherwise.m(x;q)=\begin{cases} Q_D^{11}\,q^{G(x)-P(c,h)}&\text{if }G(x)<P(c,h),\\[2pt] R_{D,h}\,q^{-\sum_eb(h_e)/3}+R'_D\,q^{1641-D}&\text{otherwise}. \end{cases}(7)✦ The profile tail majorant is ℳ(c,h;q)=∑xΓ(x)m(x;q)\mathcal M(c,h;q)=\sum_x\Gamma(x)\,m(x;q), the sum running over the global tail choices✦, and the integrated tail majorant MD,r(full;q)M_{D,r}(\mathrm{full};q) is the real function of qq that is the sum of the profile tail majorants ℳ(c,h;q)\mathcal M(c,h;q) over all dimension profiles (c,h)(c,h) in dimension DD with he≠0h_e\ne0 for some point pair ee.

The singular tail threshold D0D_0 is a fixed integer, at least 16421642, with the following property: if D=2r≥D0D=2r\ge D_0 is even, (c,h)(c,h) is a geometric profile in dimension DD and xx is a global tail choice for some choice of bits, then either G(x)<P(c,h)G(x)<P(c,h), or cj=0c_j=0 at every face, he∈{0,2}h_e\in\{0,2\} at every point pair and G(x)=23∑eb(he)G(x)=\frac23\sum_eb(h_e).✦

The Canonical Space and the Matrix Model

For a vector space VV over KK with dual space V*V^*, the canonical symplectic form ωV\omega_V on V⊕V*V\oplus V^* takes (x,α)(x,\alpha) and (y,β)(y,\beta) to β(x)−α(y)\beta(x)-\alpha(y).✦ The vertical Lagrangian is the vertical subspace 0⊕V*0\oplus V^*, the kernel of the projection onto VV, as a Lagrangian subspace for ωV\omega_V.✦ A prime qq is odd when q≠2q\ne2.✦ For an odd prime qq, let 𝔽q\mathbb F_q be the field with qq elements; the canonical singular tail Ψrcan(T;q)\Psi^{\mathrm{can}}_r(T;q) is the actual singular tail Ψr(0⊕V*;T)\Psi_r(0\oplus V^*;T) computed for ωV\omega_V with V=𝔽q2rV=\mathbb F_q^{2r}.

For a natural number DD, the space SDS_D of symmetric matrices over KK is the subspace of D×DD\times D matrices XX with X𝖳=XX^{\mathsf T}=X.✦ The radical quotient form of a symmetric bilinear form γ\gamma on a finite-dimensional vector space WW over KK, with radical RR, is the form γ‾(x+R,y+R)=γ(x,y)\bar\gamma(x+R,y+R)=\gamma(x,y) on W/RW/R.✦ The discriminant sign of γ\gamma is χ(det⁡M)\chi(\det M), where χ\chi is the quadratic character of KK and MM is the Gram matrix of γ‾\bar\gamma in a basis; it is 11 when γ\gamma has rank zero.✦ The sign of a matrix Z∈SDZ\in S_D is the discriminant sign of the form (x,y)↦x𝖳Zy(x,y)\mapsto x^{\mathsf T}Zy on KDK^D.✦

For a natural number rr and a sign ξ\xi, the signed rank layer 𝒮r,ξ\mathcal S_{r,\xi} is the set of matrices in SDS_D of rank rr and sign ξ\xi✦, and its layer mass is the proportion |𝒮r,ξ|/|SD||\mathcal S_{r,\xi}|/|S_D| of SDS_D it fills✦. The rank kernel κr,ξ\kappa_{r,\xi} is a real function on SDS_D whose value at ZZ depends only on whether ZZ lies in 𝒮r,ξ\mathcal S_{r,\xi}. An assignment XX of a matrix Xi∈SDX_i\in S_D to each point ii is fully transverse when Xi−XkX_i-X_k is invertible for every point pair {i,k}\{i,k\} with i<ki<k.✦

For a set TT of slots, signs (ξ0,ξ1,ξ2)(\xi_0,\xi_1,\xi_2) and a triple (X0,X1,X2)(X_0,X_1,X_2) in S2rS_{2r}, the face local average is the uniform mean of ∏k∈Tκr,ξk(Xk−Y)\prod_{k\in T}\kappa_{r,\xi_k}(X_k-Y) over Y∈S2rY\in S_{2r}, the kernels taken in S2rS_{2r}.✦ Given an activity pattern TT and a sign ξj,i\xi_{j,i} at each corner, the matrix singular tail Ψrmat(T,ξ;q)\Psi^{\mathrm{mat}}_r(T,\xi;q) is the uniform mean, over all assignments XX in S2rS_{2r}, of the configuration product, which is 00 when XX is fully transverse and otherwise the product over the faces jj of the face local averages at (Xv0(j),Xv1(j),Xv2(j))(X_{v_0(j)},X_{v_1(j)},X_{v_2(j)}); over K=𝔽qK=\mathbb F_q it reads Ψrmat(T,ξ;q)=1|S2r|13∑X∈S2r13X not fully transverse∏j1|S2r|∑Y∈S2r∏i∈Tjκr,ξj,i(Xvi(j)−Y).\Psi^{\mathrm{mat}}_r(T,\xi;q)=\frac1{|S_{2r}|^{13}} \sum_{\substack{X\in S_{2r}^{13}\\ X\text{ not fully transverse}}} \ \prod_j\frac1{|S_{2r}|}\sum_{Y\in S_{2r}}\ \prod_{i\in T_j}\kappa_{r,\xi_{j,i}}(X_{v_i(j)}-Y).(8)✦

The Claim

Theorem 1 (Vanishing singular contribution). Let KK be a finite field of odd characteristic, EE a symplectic space of dimension 2D2D over KK, and rr a natural number with D=2r≥D0D=2r\ge D_0 and |K|≥4CD|K|\ge4C_D. For every Lagrangian subspace AA of EE and every activity pattern TT, Ψr(A;T)≤MD,r(fullT;|K|),\Psi_r(A;T)\le M_{D,r}(\mathrm{full}_T;|K|),(9)✦ where fullT\mathrm{full}_T gives the bit 11 exactly to the faces jj with Tj={0,1,2}T_j=\{0,1,2\}. For all natural numbers D≥1642D\ge1642 and rr and every choice full\mathrm{full} of a bit at each face, as the real variable qq tends to ∞\infty, MD,r(full;q)→0.M_{D,r}(\mathrm{full};q)\to0.(10)✦ For every natural number rr with 2r≥D02r\ge D_0 and every activity pattern TT, as q→∞q\to\infty through the odd primes, Ψrcan(T;q)→0,\Psi^{\mathrm{can}}_r(T;q)\to0,(11)✦ and, for every choice of signs ξj,i∈{1,−1}\xi_{j,i}\in\{1,-1\} at the corners, Ψrmat(T,ξ;q)→0.\Psi^{\mathrm{mat}}_r(T,\xi;q)\to0.(12)✦

In the paper, the result reads: for every sufficiently large even D=2rD=2r and every activity set BB, 𝔼𝑳,𝒀[𝟏{dim⁡(Li∩Lk)>0 for some point pair {i,k}}∏ij∈BJ(Li,Yj)]=o(1)\mathbb E_{\mathbf L,\mathbf Y}\bigl[\mathbf 1\{\dim(L_i\cap L_k)>0\text{ for some point pair }\{i,k\}\}\prod_{ij\in B}J(L_i,Y_j)\bigr]=o(1) as q→∞q\to\infty, with one threshold for all activity sets. Here 𝒀=(Yj)\mathbf Y=(Y_j) is the placement of the centers at the faces, and 𝔼𝑳,𝒀\mathbb E_{\mathbf L,\mathbf Y} is the average over all placements 𝑳\mathbf L and 𝒀\mathbf Y.

What Is Checked

The claims (9) to (12) are stated over a symplectic space of dimension 2D2D over a finite field of odd characteristic, its Lagrangian subspaces with their center strata and active center sets, the active face density Φr\Phi_r and the actual singular tail Ψr\Psi_r, the integrated tail majorant MD,rM_{D,r} and the threshold D0D_0, the canonical form ωV\omega_V with its vertical Lagrangian, and the matrix singular tail Ψrmat\Psi^{\mathrm{mat}}_r with its rank kernels, with limits taken through the odd primes.

About these constructions the library proves the global tail dichotomy: an integer D0≥1642D_0\ge1642 with the property that defines the singular tail threshold exists.✦