OctoInformBeta

OpenAI preprint, September 2026 · Section 3. Finite reflection-positive probability laws

Lemma 3.6: Conditional energy contracts for real tests, with selected pair marginals given by square differences and an exceptional uniform component

Large-prime tuple laws have uniform labels.

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 field label at one vertex can be tested against the labels at every other vertex. For the tuple models, the vertices are permutation words, and the selected pairs lie on a diagonal cycle of these words.

Write [m]={0,…,m−1}[m]=\{0,\ldots,m-1\}. A permutation word uu of height mm on a block set CC is a function from CC to the symmetric group Sym⁡([m])\operatorname{Sym}([m]).✦ The fixed parameters and the block set BB are given by (1): h=24,T=80,B=[2T+1],r=h!2,δ=110h2.h=24,\qquad T=80,\qquad B=[2T+1],\qquad r=\frac{h!}{2},\qquad \delta=\frac{1}{10h^2}.(1) The permutation tuple vertices form the set 𝒱=Sym⁡([h])B\mathcal V=\operatorname{Sym}([h])^B.✦ Let cc be the cyclic shift of [h][h].✦ For each nonnegative integer kk , the cyclic tuple vertex ck∈𝒱c_k\in\mathcal V is the constant word specified in (2): ck(b)=ck(b∈B,k≥0).c_k(b)=c^k\qquad (b\in B,\ k\geq0).(2) The height of the permutation words is h=24h=24. Blocks are truncated at T=80T=80 . The block count is 2T+1=1612T+1=161. Lists use base r=h!/2=24!/2r=h!/2=24!/2 . The interaction parameter is δ=1/(10h2)\delta=1/(10h^2).

For an active block set S⊆BS\subseteq B, an inactive-block address aa is a function from B\SB\setminus S to Sym⁡([h])\operatorname{Sym}([h]).✦ The address space ASA_S, list index set ISI_S, and addressed list entry space LSL_S are the finite sets in (3): AS=Sym⁡([h])B\S,IS=[r|S|],LS=AS×IS.A_S=\operatorname{Sym}([h])^{B\setminus S},\qquad I_S=[r^{|S|}],\qquad L_S=A_S\times I_S.(3) The inactive-block projection πS:𝒱→AS\pi_S:\mathcal V\to A_S is restriction to B\SB\setminus S.✦ For a prime pp , given list values Y:LS→𝔽pY:L_S\to\mathbb F_p and a labeling j:𝒱→ISj:\mathcal V\to I_S , the sampled tuple configuration zS,Y,jz_{S,Y,j} is the function from the tuple vertex set to the field defined in (4): zS,Y,j(v)=Y(πS(v),j(v)).z_{S,Y,j}(v)=Y\bigl(\pi_S(v),j(v)\bigr).(4) The empty-block list index is 0∈I⌀0\in I_\varnothing; the leading list identification is the bijection ι:𝒱→L⌀\iota:\mathcal V\to L_\varnothing, ι(v)=(π⌀(v),0)\iota(v)=(\pi_\varnothing(v),0).

Threshold shift adjacency of words u,w:C→Gu,w:C\to G , for a finite block set CC , a group GG , an element a∈Ga\in G , and a natural threshold tt , is the condition |{b∈C:w(b)=u(b)a}|≥t|\{b\in C:w(b)=u(b)a\}|\geq t .✦ For each SS, write ES⊆LS×LSE_S\subseteq L_S\times L_S for the list constraint edges, the ordered pairs prescribing square-difference constraints on the list values.

The Mixture Laws

An ordered position pair is a pair (i,j)∈[h]2(i,j)\in[h]^2 with i<ji<j.✦ The position transposition vb,ij∈𝒱v^{b,ij}\in\mathcal V of a word vv at block bb agrees with vv outside bb and has value v(b)(ij)v(b)(i\ j) at bb.✦ The transposition interaction HSH_S is the real matrix on 𝒱\mathcal V defined by the counting sum in (5): HS(v,w)=∑b∈S∑0≤i<j<hδj−i𝟏{w=vb,ij}.H_S(v,w)=\sum_{b\in S}\ \sum_{0\leq i<j<h} \delta^{j-i}\,\mathbf 1_{\{w=v^{b,ij}\}}.(5)✦ For finite sets I,JI,J, a real matrix AA on II, and labelings j,k:I→Jj,k:I\to J, the label coincidence pairing, index energy, and index Boltzmann weight are respectively the quantities in (6): ⟨j,k⟩A=∑u∈I∑v∈IA(u,v)𝟏{j(u)=k(v)},EA(j)=12⟨j,j⟩A,wA(j)=exp⁡(EA(j)).\begin{aligned} \langle j,k\rangle_A &=\sum_{u\in I}\sum_{v\in I}A(u,v)\mathbf 1_{\{j(u)=k(v)\}},\\ E_A(j)&=\tfrac12\langle j,j\rangle_A, &w_A(j)&=\exp(E_A(j)). \end{aligned}(6) The tuple Boltzmann weight wSw_S on labelings j:𝒱→ISj:\mathcal V\to I_S is the function given by wS(j)=wHS(j)w_S(j)=w_{H_S}(j).

The nonzero square indicator χp\chi_p on 𝔽p\mathbb F_p , for a prime pp , takes value one on nonzero squares and zero elsewhere.✦ For an active block set S⊆BS\subseteq B, the square-constraint indicator VSV_S on list values Y:LS→𝔽pY:L_S\to\mathbb F_p is the function in (7), and the tuple component average Cp,SC_{p,S} is the real linear functional there on observables G:𝔽p𝒱→ℝG:\mathbb F_p^{\mathcal V}\to\mathbb R: VS(Y)=∏(u,w)∈ESχp(Y(w)−Y(u)),Cp,S(G)=𝔼Y∈𝔽pLS𝔼j∈IS𝒱VS(Y)wS(j)G(zS,Y,j).\begin{aligned} V_S(Y)&=\prod_{(u,w)\in E_S}\chi_p\bigl(Y(w)-Y(u)\bigr),\\ C_{p,S}(G)&=\mathbb E_{Y\in\mathbb F_p^{L_S}} \mathbb E_{j\in I_S^{\mathcal V}} V_S(Y)w_S(j)G(z_{S,Y,j}). \end{aligned}(7) All probability averages are uniform; sums over vertices or block subsets are counting sums.

The truncated block family for a natural number nn is 𝒟n={S⊆B:|S|≤n}\mathcal D_n=\{S\subseteq B:|S|\leq n\} , and the terminal block family is 𝒟term={S⊆B:|S|=T+1}\mathcal D_{\mathrm{term}}=\{S\subseteq B:|S|=T+1\} . The block subset mixture weight is p−|S|/16p^{-|S|/16}.✦ For a family 𝒟\mathcal D of block subsets, the restricted mixture functional Mp,𝒟M_{p,\mathcal D} and unnormalized tuple mixture MpM_p are defined in (8): Mp,𝒟(G)=∑S∈𝒟p−|S|/16Cp,S(G),Mp=Mp,𝒟T+1.M_{p,\mathcal D}(G)=\sum_{S\in\mathcal D}p^{-|S|/16}C_{p,S}(G), \qquad M_p=M_{p,\mathcal D_{T+1}}.(8) The normalized restricted mixture ρp,𝒟\rho_{p,\mathcal D} on real observables of 𝔽p𝒱\mathbb F_p^{\mathcal V} is the functional Mp,𝒟/Mp,𝒟(1)M_{p,\mathcal D}/M_{p,\mathcal D}(1). The full tuple law μp\mu_p, normalized good tuple law μp∘\mu_p^\circ, and exceptional subprobability εp\varepsilon_p are the functionals in (9); the exceptional mass is ηp=εp(1)\eta_p=\varepsilon_p(1): μp=MpMp(1),μp∘=ρp,𝒟T,εp=Mp,𝒟termMp(1).\mu_p=\frac{M_p}{M_p(1)},\qquad \mu_p^\circ=\rho_{p,\mathcal D_T},\qquad \varepsilon_p=\frac{M_{p,\mathcal D_{\mathrm{term}}}}{M_p(1)}.(9)

Let Ωp⊆𝔽p2\Omega_p\subseteq\mathbb F_p^2 , for odd pp , be the set of pairs (x,y)(x,y) for which y−xy-x is a nonzero square. For an odd prime pp and a real test F:𝔽p2→ℝF:\mathbb F_p^2\to\mathbb R, the nonzero square pair average Qp(F)Q_p(F) is the quantity in (10): Qp(F)=𝔼q∈𝔽p×𝔼x∈𝔽pF(x,x+q2).Q_p(F)=\mathbb E_{q\in\mathbb F_p^\times} \mathbb E_{x\in\mathbb F_p}F(x,x+q^2).(10)✦

Conditional Energy and Thresholds

The point mass dρ(z)d_\rho(z) of a real linear functional ρ\rho on real functions on a finite space XX , at a point zz in that space, is the quantity given by dρ(z)=ρ(𝟏{z})d_\rho(z)=\rho(\mathbf 1_{\{z\}}) .✦ For v∈𝒱v\in\mathcal V , the vertex decomposition is the natural bijection from {v}⊔(𝒱\{v})\{v\}\sqcup(\mathcal V\setminus\{v\}) onto 𝒱\mathcal V , sending each vertex to itself.✦ Write Yv=𝔽p𝒱\{v}Y_v=\mathbb F_p^{\mathcal V\setminus\{v\}} ; for a field element xx and a configuration yy in this space, let zx,yz_{x,y} be the configuration with value xx at vv and restriction yy elsewhere. For a real linear functional ρ\rho on real functions on 𝔽p𝒱\mathbb F_p^{\mathcal V}, the uniform relative joint density Wρ,vW_{\rho,v} of the separated coordinates is the kernel on 𝔽p×Yv\mathbb F_p\times Y_v in (11): Wρ,v(x,y)=p|𝒱|dρ(zx,y)(x∈𝔽p,y∈Yv).W_{\rho,v}(x,y)=p^{|\mathcal V|}d_\rho(z_{x,y}) \qquad (x\in\mathbb F_p,\ y\in Y_v).(11) The conditional energy ℰ(W,f)\mathcal E(W,f) of a real kernel W:X×Y→ℝW:X\times Y\to\mathbb R on finite sets X,YX,Y and a real function f:X→ℝf:X\to\mathbb R is the quantity in (12), with a zero denominator contributing zero: ℰ(W,f)=𝔼y∈Y(𝔼x∈Xf(x)W(x,y))2𝔼x∈XW(x,y),ℰρ,v(f)=ℰ(Wρ,v,f).\mathcal E(W,f)=\mathbb E_{y\in Y} \frac{\bigl(\mathbb E_{x\in X}f(x)W(x,y)\bigr)^2} {\mathbb E_{x\in X}W(x,y)}, \qquad \mathcal E_{\rho,v}(f)=\mathcal E(W_{\rho,v},f).(12)

The pullback ϕ*E\phi^*E of an ordered edge set EE on a vertex set VV along a bijection ϕ:U→V\phi:U\to V is the ordered edge set on UU given by ϕ*E={(u,w):(ϕ(u),ϕ(w))∈E}\phi^*E=\{(u,w):(\phi(u),\phi(w))\in E\} .✦ For a graph on a disjoint union I⊔JI\sqcup J, a crossing edge is a pair (i,j)∈I×J(i,j)\in I\times J for which either orientation of the joining edge is present.✦ The leading constraint graph is Elead=ι*E⌀E_{\mathrm{lead}}=\iota^*E_\varnothing, with edge count e=|E⌀|e=|E_\varnothing|. For an ordered graph EE on a finite vertex set and a vertex vv, let mE(v)m_E(v) be the number of crossing edges for the decomposition into {v}\{v\} and its complement. The graph conditional energy constant KE(v)K_E(v) for an ordered graph EE on a finite vertex set and a vertex vv is the quantity in (13): KE(v)=(2(1/2)mE(v)+4((1/2)mE(v))2)((32mE(v))1/4)2.K_E(v)=\left(\frac{2}{(1/2)^{m_E(v)}}+ \frac{4}{((1/2)^{m_E(v)})^2}\right) \left((32m_E(v))^{1/4}\right)^2.(13) The leading mass lower bound a0a_0, mass positivity threshold PmP_m, and Boltzmann weight upper bound w*w_* are the constants in (14): a0=2−(e+1),Pm=max⁡{3,(2e+1e)2},w*=exp⁡(|𝒱|(T+1)(h2)δ2).\begin{aligned} a_0&=2^{-(e+1)},& P_m&=\max\{3,(2^{e+1}e)^2\},\\ w_*&=\exp\!\left(\frac{|\mathcal V|(T+1)\binom h2\delta}{2}\right). \end{aligned}(14) Let 𝒫(B)\mathcal P(B) be the family of all subsets of BB. The conditional error numerator A0A_0, decay constant C0C_0, and decay threshold PcP_c are defined in (15): A0=∑v∈𝒱KElead(v)+|𝒫(B)|w*,C0=A0a0,Pc=max⁡{Pm,3,C032}.A_0=\sum_{v\in\mathcal V}K_{E_{\mathrm{lead}}}(v) +|\mathcal P(B)|w_*,\qquad C_0=\frac{A_0}{a_0},\qquad P_c=\max\{P_m,3,C_0^{32}\}.(15)

The Claim

Theorem 1 (Lemma 3.6). Let pp be a prime with p≥Pmp\geq P_m. For every vertex v∈𝒱v\in\mathcal V and real function f:𝔽p→ℝf:\mathbb F_p\to\mathbb R, the one-label marginal identities are (16) and (17): μp(z↦f(zv))=𝔼x∈𝔽pf(x).\mu_p\bigl(z\mapsto f(z_v)\bigr)=\mathbb E_{x\in\mathbb F_p}f(x).(16)✦ μp∘(z↦f(zv))=𝔼x∈𝔽pf(x).\mu_p^\circ\bigl(z\mapsto f(z_v)\bigr)=\mathbb E_{x\in\mathbb F_p}f(x).(17)✦ If also p≥Pcp\geq P_c, then every vertex vv and every real ff with 𝔼x∈𝔽pf(x)=0\mathbb E_{x\in\mathbb F_p}f(x)=0 satisfy (18) and (19): ℰμp,v(f)≤p−1/32𝔼x∈𝔽pf(x)2.\mathcal E_{\mu_p,v}(f)\leq p^{-1/32}\mathbb E_{x\in\mathbb F_p}f(x)^2.(18)✦ ℰμp∘,v(f)≤p−1/32𝔼x∈𝔽pf(x)2.\mathcal E_{\mu_p^\circ,v}(f)\leq p^{-1/32}\mathbb E_{x\in\mathbb F_p}f(x)^2.(19)✦ For every natural number kk and real test F:𝔽p2→ℝF:\mathbb F_p^2\to\mathbb R, the selected-cycle pair identities are (20) and (21): μp∘(z↦F(zck,zck+1))=Qp(F).\mu_p^\circ\bigl(z\mapsto F(z_{c_k},z_{c_{k+1}})\bigr)=Q_p(F).(20)✦ μp(z↦F(zck,zck+1))=(1−ηp)Qp(F)+ηp𝔼x∈𝔽p𝔼y∈𝔽pF(x,y).\mu_p\bigl(z\mapsto F(z_{c_k},z_{c_{k+1}})\bigr) =(1-\eta_p)Q_p(F)+\eta_p\mathbb E_{x\in\mathbb F_p} \mathbb E_{y\in\mathbb F_p}F(x,y).(21)✦

In the paper, the result reads: for the fixed constructed measures there is one absolute threshold PP such that, for every prime p≥Pp\geq P, both μp\mu_p and μp∘\mu_p^\circ have uniform one-label marginals, and, uniformly in the vertex vv and the choice ρ∈{μp,μp∘}\rho\in\{\mu_p,\mu_p^\circ\}, every complex function f:𝔽p→ℂf:\mathbb F_p\to\mathbb C with 𝔼xf(x)=0\mathbb E_xf(x)=0 satisfies ∥𝔼ρ[f(zv)∣(zw)w≠v]∥L2(λρ,v)≤p−1/64∥f∥L2(𝔽p)\|\mathbb E_\rho[f(z_v)\mid(z_w)_{w\ne v}]\|_{L^2(\lambda_{\rho,v})} \leq p^{-1/64}\|f\|_{L^2(\mathbb F_p)}, where λρ,v\lambda_{\rho,v} is the other-label marginal, while every selected edge pair has law Unif⁡(Ωp)\operatorname{Unif}(\Omega_p) under μp∘\mu_p^\circ and law (1−ηp)Unif⁡(Ωp)+ηpUnif⁡(𝔽p2)(1-\eta_p)\operatorname{Unif}(\Omega_p)+\eta_p\operatorname{Unif}(\mathbb F_p^2) under μp\mu_p.

What Is Checked

The assertions concern the particular finite tuple mixture laws in (9), with real test functions, the mass threshold in (14), and the stronger conditional threshold in (15).