OctoInformBeta

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

Lemma 3.5: An additional marked block dominates the unmarked second moment, with paired positivity above the reflection threshold

Marking controls the reflection cut form.

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 reflection cut compares a real observable on one half of a tuple configuration with its value on the reflected half. The configurations are sampled from lists indexed by permutations on the blocks that have not been marked.

For a height hh, a permutation word vv on a block set B0B_0 is a function from B0B_0 to the permutations of [h]={0,…,h−1}[h]=\{0,\ldots,h-1\}.✦ The tuple vertex set 𝒱\mathcal V is the set of permutation words on the block set BB in (1); the height, block cutoff, list base and half vertex count are fixed by h=24,T=80,m=2T+1=161,B={0,…,m−1},r=h!2,M=(h!)m2.h=24,\qquad T=80,\qquad m=2T+1=161,\qquad B=\{0,\ldots,m-1\},\qquad r=\frac{h!}{2},\qquad M=\frac{(h!)^m}{2}.(1)

For a block bb and distinct symbols α,β∈[h]\alpha,\beta\in[h], the positive cut predicate says that α\alpha precedes β\beta in block bb.✦ The positive cut half H⊆𝒱H\subseteq\mathcal V is the set of words in which α\alpha precedes β\beta in block bb, and the block transposition reflection ρ\rho composes the permutation at bb on the left with (αβ)(\alpha\ \beta): H={v∈𝒱:vb−1(α)<vb−1(β)},(ρv)c={(αβ)vb,c=b,vc,c≠b.H=\{v\in\mathcal V:v_b^{-1}(\alpha)<v_b^{-1}(\beta)\},\qquad (\rho v)_c= \begin{cases} (\alpha\ \beta)v_b,&c=b,\\ v_c,&c\ne b. \end{cases}(2) The reflection half decomposition ι:H⊔H→𝒱\iota:H\sqcup H\to\mathcal V sends a word in the first copy to itself and a word in the second copy to its reflection in (2).✦

For an active block set S⊆BS\subseteq B, an inactive-block address aa is a permutation word on B\SB\setminus S.✦ An addressed list index jj is an integer in JSJ_S, and an addressed list entry ℓ\ell is a pair in LSL_S, with these sets and the inactive-block projection defined by (3): AS=Sym⁡([h])B\S,JS={0,…,r|S|−1},LS=AS×JS,πS(v)=v|B\S.A_S=\operatorname{Sym}([h])^{B\setminus S},\qquad J_S=\{0,\ldots,r^{|S|}-1\},\qquad L_S=A_S\times J_S,\qquad \pi_S(v)=v|_{B\setminus S}.(3)

The cyclic shift γ\gamma on [h][h] is the permutation that sends each position to the next and the last to zero.✦ For a finite block set B0B_0, a group GG, a shift c∈Gc\in G and a nonnegative integer qq, threshold shift adjacency of u,v∈GB0u,v\in G^{B_0} is the condition (4): |{a∈B0:v(a)=u(a)c}|≥q.\bigl|\{a\in B_0:v(a)=u(a)c\}\bigr|\ge q.(4)✦ The list constraint edges ES⊆LS×LSE_S\subseteq L_S\times L_S are the directed edges specifying the square-difference constraints between list values.

Let pp be prime. For a field KK and an element t∈Kt\in K, the nonzero square indicator σK(t)\sigma_K(t) is given by (5); for a finite directed edge set EE and a labeling zz of its vertices by 𝔽p\mathbb F_p, the graph square-constraint indicator VE(z)V_E(z) is given by the same display, and the tuple square-constraint indicator is VS=VESV_S=V_{E_S} on list values Y:LS→𝔽pY:L_S\to\mathbb F_p: σK(t)=𝟏{t≠0,t is a square in K},VE(z)=∏(u,v)∈Eσ𝔽p(z(v)−z(u)),VS(Y)=VES(Y).\sigma_K(t)=\mathbf 1_{\{t\ne0,\ t\text{ is a square in }K\}},\qquad V_E(z)=\prod_{(u,v)\in E}\sigma_{\mathbb F_p}(z(v)-z(u)),\qquad V_S(Y)=V_{E_S}(Y).(5)

Weights and Half Moments

An ordered position pair is a pair (a,d)(a,d) in [h]2[h]^2 with a<da<d.✦ For a block c∈Bc\in B and a permutation word v∈𝒱v\in\mathcal V, the position transposition τc,a,dv\tau_{c,a,d}v composes vcv_c on the right with (ad)(a\ d) and fixes the other blocks.✦ The transposition interaction KSK_S on SS, with interaction parameter δ\delta, is the counting sum in (6): δ=110h2,KS(v,w)=∑c∈S∑0≤a<d<hδd−a𝟏{w=τc,a,dv}.\delta=\frac1{10h^2},\qquad K_S(v,w)=\sum_{c\in S}\ \sum_{0\le a<d<h} \delta^{d-a}\mathbf 1_{\{w=\tau_{c,a,d}v\}}.(6)

For a finite index set II, label set JJ, real matrix KK on II and labelings j,k:I→Jj,k:I\to J, the label coincidence pairing 𝒞K(j,k)\mathcal C_K(j,k), index energy ℰK(j)\mathcal E_K(j) and index Boltzmann weight 𝒲K(j)\mathcal W_K(j) are the real quantities in (7): 𝒞K(j,k)=∑x∈I∑y∈IK(x,y)𝟏{j(x)=k(y)},ℰK(j)=12𝒞K(j,j),𝒲K(j)=exp⁡(ℰK(j)).\begin{aligned} \mathcal C_K(j,k)&=\sum_{x\in I}\sum_{y\in I}K(x,y)\mathbf1_{\{j(x)=k(y)\}},\\ \mathcal E_K(j)&=\tfrac12\mathcal C_K(j,j),\qquad \mathcal W_K(j)=\exp(\mathcal E_K(j)). \end{aligned}(7) The tuple Boltzmann weight WSW_S on labelings j:𝒱→JSj:\mathcal V\to J_S is the function given by WS(j)=𝒲KS(j)W_S(j)=\mathcal W_{K_S}(j).

For list values Y:LS→𝔽pY:L_S\to\mathbb F_p and a labeling j:𝒱→JSj:\mathcal V\to J_S, the sampled tuple configuration zS,Y,jz_{S,Y,j} selects the list value at the vertex’s inactive-block address and index.✦ The component functional νp,S\nu_{p,S} on real observables GG is the weighted uniform average in (8): zS,Y,j(v)=Y(πS(v),j(v)),νp,S[G]=𝔼Y∈𝔽pLS𝔼j∈JS𝒱VS(Y)WS(j)G(zS,Y,j).\begin{aligned} z_{S,Y,j}(v)&=Y(\pi_S(v),j(v)),\\ \nu_{p,S}[G]&= \mathbb E_{Y\in\mathbb F_p^{L_S}}\mathbb E_{j\in J_S^{\mathcal V}} V_S(Y)W_S(j)G(z_{S,Y,j}). \end{aligned}(8) All averages here and below use uniform probability on the indicated finite sets. The tuple reflection quadratic form Qp,SbQ^b_{p,S} on real functions F:𝔽pH→ℝF:\mathbb F_p^H\to\mathbb R is defined by (9): Qp,Sb(F)=νp,S[z↦F(z|H)F((z∘ρ)|H)].Q^b_{p,S}(F)=\nu_{p,S}\bigl[z\mapsto F(z|_H)F((z\circ\rho)|_H)\bigr].(9)✦ The symbols α,β\alpha,\beta specifying the cut are suppressed in this notation.

Suppose b∉Sb\notin S. A positive half address a∈ASa\in A_S is an inactive-block address in which α\alpha precedes β\beta in block bb, and a positive half list entry ℓ∈LS\ell\in L_S is a pair consisting of a positive half address and an index in JSJ_S. Their sets and the inactive-block half projection are (10): AS+={a∈AS:ab−1(α)<ab−1(β)},LS+=AS+×JS,πS+=πS|H:H→AS+.A_S^+=\{a\in A_S:a_b^{-1}(\alpha)<a_b^{-1}(\beta)\},\qquad L_S^+=A_S^+\times J_S,\qquad \pi_S^+=\pi_S|_H:H\to A_S^+.(10) The list reflection half decomposition ηS:LS+⊔LS+→LS\eta_S:L_S^+\sqcup L_S^+\to L_S leaves the list index fixed, sending the address to itself on the first copy and to its block transposition reflection on the second.✦

For a directed graph EE on a finite set YY and a bijection e:X→Ye:X\to Y, the pullback e*Ee^*E is the directed graph on XX consisting of the pairs (x,y)(x,y) for which (e(x),e(y))∈E(e(x),e(y))\in E.✦ For a directed graph EE on two copies of a finite set XX, the positive half internal edge set E+E^+ is the set of pairs in X×XX\times X whose corresponding pair in the first copy belongs to EE.✦ The unmarked half graph ẼS\widetilde E_S and its positive half internal edges ES+E_S^+ are specified in (11), where ++ denotes the first copy: ẼS=ηS*ES,ES+={(x,y)∈LS+×LS+:(+x,+y)∈ẼS}.\widetilde E_S=\eta_S^*E_S,\qquad E_S^+=\{(x,y)\in L_S^+\times L_S^+:(+x,+y)\in\widetilde E_S\}.(11) The unmarked quadratic moment Dp,Sb(F)D^b_{p,S}(F) of the real half-label function FF is the real second moment in (12): Dp,Sb(F)=𝔼x∈𝔽pLS+𝔼j∈JSHVES+(x)[F(v↦x(πS+(v),j(v)))]2.D^b_{p,S}(F)= \mathbb E_{x\in\mathbb F_p^{L_S^+}}\mathbb E_{j\in J_S^H} V_{E_S^+}(x) \left[F\bigl(v\mapsto x(\pi_S^+(v),j(v))\bigr)\right]^2.(12)

The interaction lower-bound parameter κ\kappa, Boltzmann weight bound WmaxW_{\max}, and marked and unmarked reflection constants Cm,CuC_{\mathrm m},C_{\mathrm u} are fixed by (13): κ=910δh−1,Wmax=exp⁡(|𝒱|(T+1)(h2)δ2),Cm=(exp⁡(κ)−1)M(rT+1)M,Cu=(32|𝒱|2)1/4Wmax⁡2.\begin{aligned} \kappa&=\tfrac9{10}\delta^{h-1},\qquad W_{\max}=\exp\left(\frac{|\mathcal V|(T+1)\binom h2\delta}{2}\right),\\ C_{\mathrm m}&=\frac{(\exp(\kappa)-1)^M}{(r^{T+1})^M},\qquad C_{\mathrm u}=(32|\mathcal V|^2)^{1/4}W_{\max}^2. \end{aligned}(13) The reflection positivity threshold PrefP_{\mathrm{ref}} is the real number in (14): Pref=max⁡{3,(CurMCm)16}.P_{\mathrm{ref}}=\max\left\{3, \left(\frac{C_{\mathrm u}r^M}{C_{\mathrm m}}\right)^{16}\right\}.(14)

The Claim

Theorem 1. Let pp be prime, let S⊆BS\subseteq B have |S|≤T|S|\le T, and let b∈B\Sb\in B\setminus S. Fix distinct α,β∈[h]\alpha,\beta\in[h], write S′=S∪{b}S'=S\cup\{b\}, and let F:𝔽pH→ℝF:\mathbb F_p^H\to\mathbb R. Then Cmr−MDp,Sb(F)≤Qp,S′b(F).C_{\mathrm m}r^{-M}D^b_{p,S}(F)\le Q^b_{p,S'}(F).(15)✦ If p≥Prefp\ge P_{\mathrm{ref}}, then also 0≤Qp,Sb(F)+p−1/16Qp,S′b(F).0\le Q^b_{p,S}(F)+p^{-1/16}Q^b_{p,S'}(F).(16)✦

In the paper, the result reads: for b∉Sb\notin S, |S|≤T|S|\le T, S′=S∪{b}S'=S\cup\{b\} and every real half-label function FF, one has Qp,S′b(F)≥Cmr−MDp,Sb(F)Q^b_{p,S'}(F)\ge C_{\mathrm m}r^{-M}D^b_{p,S}(F), and consequently νp,S+p−1/16νp,S′\nu_{p,S}+p^{-1/16}\nu_{p,S'} has a positive semidefinite matrix at this cut whenever p≥Pref=max⁡{3,(CurM/Cm)16}p\ge P_{\mathrm{ref}}=\max\{3,(C_{\mathrm u}r^M/C_{\mathrm m})^{16}\}, with T,M,Cm,CuT,M,C_{\mathrm m},C_{\mathrm u} denoting the paper’s t0,n,c*,C*t_0,n,c_*,C_*.

What Is Checked

The claims (15) and (16) are stated for the component functionals and unmarked quadratic moments over prime fields, with real half-label functions, at a cut in an inactive block and with at most TT active blocks; the prime threshold is required for (16). The word and list reflection half decompositions are bijections between their respective finite sets.