OctoInformBeta

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

Proposition 3.1: A normalized good part has uniform square pair marginals, and both laws have squared real conditional decay

Reflection-positive tuple laws above fixed thresholds.

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 configurations are field-valued labels on permutation words, with a distinguished cyclic sequence of labels. The constraints concern square gaps along this sequence and the dependence of one coordinate on all the others.

Fix h=24h=24, T=80T=80, r=h!/2r=h!/2 and the block set B={0,…,2T}B=\{0,\ldots,2T\}. A permutation word vv on BB is a function assigning a permutation of {0,…,h−1}\{0,\ldots,h-1\} to each block of BB.✦ The vertex set VV and the cyclic vertices vkv_k are given by (1), where σ(i)=i+1(mod⁡h)\sigma(i)=i+1\pmod h is the cyclic shift. V=𝔖hB,vk(b)=σk(b∈B,k≥0).V=\mathfrak S_h^B,\qquad v_k(b)=\sigma^k\quad(b\in B,\ k\ge0).(1) There are |B|=2T+1=161|B|=2T+1=161 blocks.

Let pp be prime. Strict squares in 𝔽p\mathbb F_p are its nonzero squares; their set Qp*Q_p^\ast and the corresponding ordered pair space Ωp\Omega_p are defined in (2). Qp*={q2:q∈𝔽p×},Ωp={(x,y)∈𝔽p2:y−x∈Qp*}.Q_p^\ast=\{q^2:q\in\mathbb F_p^\times\},\qquad \Omega_p=\{(x,y)\in\mathbb F_p^2:y-x\in Q_p^\ast\}.(2) Every 𝔼\mathbb E below is a uniform probability average, whereas ∑\sum denotes a counting sum. A quotient with denominator zero is assigned the value zero. For a real function gg on 𝔽p2\mathbb F_p^2, the square pair average 𝒜p(g)\mathcal A_p(g) is the uniform average defined by (3). 𝒜p(g)=𝔼q∈𝔽p×𝔼x∈𝔽pg(x,x+q2).\mathcal A_p(g)=\mathbb E_{q\in\mathbb F_p^\times} \mathbb E_{x\in\mathbb F_p}g(x,x+q^2).(3)✦

Component Averages

For S⊆BS\subseteq B, an inactive-block address uu on B\SB\setminus S is a function assigning a permutation of {0,…,h−1}\{0,\ldots,h-1\} to each block in B\SB\setminus S.✦ The address space 𝒰S\mathcal U_S, list index set ISI_S and list entry set ℒS\mathcal L_S are defined by (4); the projection πS\pi_S restricts a vertex to its inactive blocks. 𝒰S=𝔖hB\S,IS={0,…,r|S|−1},ℒS=𝒰S×IS,πS(v)=v|B\S.\mathcal U_S=\mathfrak S_h^{B\setminus S},\qquad I_S=\{0,\ldots,r^{|S|}-1\},\qquad \mathcal L_S=\mathcal U_S\times I_S,\qquad \pi_S(v)=v|_{B\setminus S}.(4) For a finite block set DD, a group KK, a shift c∈Kc\in K and a nonnegative integer tt, threshold shift adjacency ∼c,t\sim_{c,t} is the relation on group-valued addresses D→KD\to K defined by (5). u∼c,tw⇔|{b∈D:w(b)=u(b)c}|≥t.u\sim_{c,t}w\quad\Longleftrightarrow\quad |\{b\in D:w(b)=u(b)c\}|\ge t.(5) Write ℰS⊆ℒS×ℒS\mathcal E_S\subseteq\mathcal L_S\times\mathcal L_S for the list constraint edges. A field-valued list Y:ℒS→𝔽pY:\mathcal L_S\to\mathbb F_p has square-constraint indicator χS(Y)\chi_S(Y) as in (6); the nonzero square indicator 𝟏Qp*(x)\mathbf1_{Q_p^\ast}(x) is 11 precisely when xx is a strict square and 00 otherwise. χS(Y)=∏(a,a′)∈ℰS𝟏Qp*(Y(a′)−Y(a)).\chi_S(Y)=\prod_{(a,a')\in\mathcal E_S} \mathbf1_{Q_p^\ast}\bigl(Y(a')-Y(a)\bigr).(6) A list indexing λ:V→IS\lambda:V\to I_S selects the output configuration ΦS(Y,λ)\Phi_S(Y,\lambda) in (7). ΦS(Y,λ)v=Y(πS(v),λv).\Phi_S(Y,\lambda)_v=Y\bigl(\pi_S(v),\lambda_v\bigr).(7)✦

An ordered position pair is a pair (i,j)(i,j) with 0≤i<j<h0\le i<j<h.✦ For a vertex v∈Vv\in V, a block b∈Bb\in B and positions 0≤i<j<h0\le i<j<h, the position transposition vb,ij∈Vv^{b,ij}\in V is the word obtained by replacing vbv_b with vb∘(ij)v_b\circ(i\ j) and leaving all other blocks fixed.✦ With δ=1/(10h2)\delta=1/(10h^2), the transposition interaction HSH_S is the matrix in (8). HS(v,w)=∑b∈S∑0≤i<j<hδj−i𝟏{w=vb,ij}.H_S(v,w)=\sum_{b\in S}\sum_{0\le i<j<h} \delta^{j-i}\mathbf1_{\{w=v^{b,ij}\}}.(8) For a real matrix AA on VV and indexings λ,λ′:V→J\lambda,\lambda':V\to J into a finite set JJ, the label coincidence pairing DAD_A, index energy EAE_A and Boltzmann weight wAw_A are defined in (9). The tuple weight WSW_S uses A=HSA=H_S and J=ISJ=I_S. DA(λ,λ′)=∑v∈V∑w∈VA(v,w)𝟏{λv=λw′},EA(λ)=12DA(λ,λ),wA(λ)=exp⁡(EA(λ)),WS(λ)=wHS(λ).\begin{aligned} D_A(\lambda,\lambda') &=\sum_{v\in V}\sum_{w\in V}A(v,w) \mathbf1_{\{\lambda_v=\lambda'_w\}},\\ E_A(\lambda)&=\tfrac12D_A(\lambda,\lambda),\qquad w_A(\lambda)=\exp(E_A(\lambda)),\qquad W_S(\lambda)=w_{H_S}(\lambda). \end{aligned}(9) The tuple component average Cp,SC_{p,S} is the weighted average of a real observable G:𝔽pV→ℝG:\mathbb F_p^V\to\mathbb R defined by (10). Cp,S(G)=𝔼Y∈𝔽pℒS𝔼λ∈ISVχS(Y)WS(λ)G(ΦS(Y,λ)).C_{p,S}(G)=\mathbb E_{Y\in\mathbb F_p^{\mathcal L_S}} \mathbb E_{\lambda\in I_S^V} \chi_S(Y)W_S(\lambda)G\bigl(\Phi_S(Y,\lambda)\bigr).(10)✦

The truncated family ℱm\mathcal F_m consists of block subsets of size at most mm; the terminal family 𝒯\mathcal T consists of those of size T+1T+1. For a family 𝒟\mathcal D of block subsets, the restricted mixture ℳp,𝒟\mathcal M_{p,\mathcal D} and its normalized average Lp,𝒟L_{p,\mathcal D} use the block weights ap(S)a_p(S) in (11). ℱm={S⊆B:|S|≤m},𝒯={S⊆B:|S|=T+1},ap(S)=p−|S|/16,ℳp,𝒟(G)=∑S∈𝒟ap(S)Cp,S(G),Lp,𝒟(G)=ℳp,𝒟(G)ℳp,𝒟(1).\begin{aligned} \mathcal F_m&=\{S\subseteq B:|S|\le m\},\qquad \mathcal T=\{S\subseteq B:|S|=T+1\},\qquad a_p(S)=p^{-|S|/16},\\ \mathcal M_{p,\mathcal D}(G)&=\sum_{S\in\mathcal D}a_p(S)C_{p,S}(G),\qquad L_{p,\mathcal D}(G)= \frac{\mathcal M_{p,\mathcal D}(G)}{\mathcal M_{p,\mathcal D}(1)}. \end{aligned}(11) The full mixture is ℳp=ℳp,ℱT+1\mathcal M_p=\mathcal M_{p,\mathcal F_{T+1}}, with normalizer Zp=ℳp(1)Z_p=\mathcal M_p(1).✦ The full tuple average μp\mu_p and exceptional average ϵp\epsilon_p use this normalizer, while the normalized good average μp∘\mu_p^\circ uses its restricted normalizer, as in (12); ηp\eta_p denotes the exceptional mass. μp(G)=ℳp(G)Zp,ϵp(G)=ℳp,𝒯(G)Zp,μp∘=Lp,ℱT,ηp=ϵp(1).\mu_p(G)=\frac{\mathcal M_p(G)}{Z_p},\qquad \epsilon_p(G)=\frac{\mathcal M_{p,\mathcal T}(G)}{Z_p},\qquad \mu_p^\circ=L_{p,\mathcal F_T},\qquad \eta_p=\epsilon_p(1).(12)

Cuts and Conditional Energy

For b∈Bb\in B and distinct positions α,β\alpha,\beta, the positive cut half Cb;αβ⊆VC_{b;\alpha\beta}\subseteq V consists of words in which α\alpha precedes β\beta in block bb.✦ The reflection θb;αβ:V→V\theta_{b;\alpha\beta}:V\to V, for b∈Bb\in B and distinct positions α,β\alpha,\beta, transposes the values α\alpha and β\beta in block bb, as specified by (13). Cb;αβ={v∈V:vb−1(α)<vb−1(β)},(θb;αβv)d={(αβ)∘vb,d=b,vd,d≠b.\begin{aligned} C_{b;\alpha\beta} &=\{v\in V:v_b^{-1}(\alpha)<v_b^{-1}(\beta)\},\\ (\theta_{b;\alpha\beta}v)_d &=\begin{cases}(\alpha\ \beta)\circ v_b,&d=b,\\v_d,&d\ne b. \end{cases} \end{aligned}(13) For q∈𝔽p×q\in\mathbb F_p^\times and a∈𝔽pa\in\mathbb F_p, the simultaneous square affine map Aq,aA_{q,a} on configurations is defined in (14). (Aq,az)v=q2zv+a.(A_{q,a}z)_v=q^2z_v+a.(14)

For a real linear functional ρ\rho on functions on a finite space Ξ\Xi and a bijection e:Ξ→X×Ye:\Xi\to X\times Y, the point mass dρd_\rho and the joint density Wρ,eW_{\rho,e} relative to uniform probability are defined in (15). dρ(ξ)=ρ(𝟏{ξ}),Wρ,e(x,y)=|Ξ|dρ(e−1(x,y)).d_\rho(\xi)=\rho(\mathbf1_{\{\xi\}}),\qquad W_{\rho,e}(x,y)=|\Xi|\,d_\rho\bigl(e^{-1}(x,y)\bigr).(15) The coordinate decomposition V={v}⊔(V\{v})V=\{v\}\sqcup(V\setminus\{v\}), for v∈Vv\in V, gives the splitting ev(z)=(zv,(zw)w≠v)e_v(z)=(z_v,(z_w)_{w\ne v}); write Wρ,v=Wρ,evW_{\rho,v}=W_{\rho,e_v}. For a real function f:X→ℝf:X\to\mathbb R and a real kernel WW on X×YX\times Y, the conditional energy ℰ(W;f)\mathscr E(W;f) is the average over YY defined by (16). ℰ(W;f)=𝔼y∈Y(𝔼x∈Xf(x)W(x,y))2𝔼x∈XW(x,y).\mathscr 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)}.(16)✦

Fixed Thresholds

The pullback e*Ee^*E of a directed graph EE by a bijection ee is the directed graph whose edges are the pairs (x,y)(x,y) for which (e(x),e(y))∈E(e(x),e(y))\in E.✦ The crossing edges of a directed graph on I⊔JI\sqcup J are the pairs (i,j)∈I×J(i,j)\in I\times J joined in either direction.✦ For a vertex vv of a finite directed graph EE, let mE(v)m_E(v) be the number of crossing edges for the split into {v}\{v\} and its complement. The graph conditional energy constant Γ(E,v)\Gamma(E,v) is defined by (17). Γ(E,v)=(2(1/2)mE(v)+4((1/2)mE(v))2)((32mE(v))1/4)2.\Gamma(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.(17)✦ The leading list identification ι:V→ℒ⌀\iota:V\to\mathcal L_\varnothing is the map ι(v)=(π⌀(v),0)\iota(v)=(\pi_\varnothing(v),0), where 00 is the unique index in I⌀I_\varnothing. The leading constraint graph E0E_0, its edge count e0e_0, the leading mass scale ℓ\ell and the mass threshold MM are specified in (18). E0=ι*ℰ⌀,e0=|ℰ⌀|,ℓ=2−(e0+1),M=max⁡{3,(2e0+1e0)2}.\begin{aligned} E_0&=\iota^*\mathcal E_\varnothing,\qquad e_0=|\mathcal E_\varnothing|,\qquad \ell=2^{-(e_0+1)},\\ M&=\max\{3,(2^{e_0+1}e_0)^2\}. \end{aligned}(18) Put n=(h!)|B|=|V|n=(h!)^{|B|}=|V| and N=n/2N=n/2.✦ The fixed Boltzmann bound UU, interaction scale κ\kappa, and marked and unmarked reflection constants KmK_{\mathrm{m}} and KuK_{\mathrm{u}} are given in (19). U=exp⁡(n(T+1)(h2)δ2),κ=910δh−1,Km=(exp⁡κ−1)N(rT+1)N,Ku=(32n2)1/4U2.\begin{aligned} U&=\exp\left(\frac{n(T+1)\binom h2\delta}{2}\right),\qquad \kappa=\frac9{10}\delta^{h-1},\\ K_{\mathrm{m}}&=\frac{(\exp\kappa-1)^N}{(r^{T+1})^N},\qquad K_{\mathrm{u}}=(32n^2)^{1/4}U^2. \end{aligned}(19) The conditional error numerator BcB_{\mathrm{c}} and conditional decay constant AcA_{\mathrm{c}} are defined in (20). Bc=∑v∈VΓ(E0,v)+2|B|U,Ac=Bc/ℓ.B_{\mathrm{c}}=\sum_{v\in V}\Gamma(E_0,v)+2^{|B|}U,\qquad A_{\mathrm{c}}=B_{\mathrm{c}}/\ell.(20) The reflection threshold RR and conditional decay threshold CC are (21). R=max⁡{3,(KurNKm)16},C=max⁡{M,3,Ac32}.R=\max\left\{3,\left(\frac{K_{\mathrm{u}}r^N}{K_{\mathrm{m}}}\right)^{16}\right\}, \qquad C=\max\{M,3,A_{\mathrm{c}}^{32}\}.(21)

The Claim

Theorem 1. Let pp be prime. If p≥Mp\ge M, the identities (22) and (23) hold: μp=(1−ηp)μp∘+ϵp.\mu_p=(1-\eta_p)\mu_p^\circ+\epsilon_p.(22)✦ μp(1)=1.\mu_p(1)=1.(23)✦ If p≥Rp\ge R, then for b∈Bb\in B, distinct positions α,β\alpha,\beta and every real F:𝔽pCb;αβ→ℝF:\mathbb F_p^{C_{b;\alpha\beta}}\to\mathbb R, reflection positivity takes the form (24): 0≤μp(z↦F((zv)v∈Cb;αβ)F((zθb;αβv)v∈Cb;αβ)).0\le\mu_p\left(z\mapsto F\bigl((z_v)_{v\in C_{b;\alpha\beta}}\bigr) F\bigl((z_{\theta_{b;\alpha\beta}v})_{v\in C_{b;\alpha\beta}}\bigr)\right).(24)✦ For q∈𝔽p×q\in\mathbb F_p^\times, a∈𝔽pa\in\mathbb F_p and every real observable GG, the affine identities (25) and (26) hold, the latter for every S⊆BS\subseteq B: μp(G∘Aq,a)=μp(G).\mu_p(G\circ A_{q,a})=\mu_p(G).(25)✦ Cp,S(G∘Aq,a)=Cp,S(G).C_{p,S}(G\circ A_{q,a})=C_{p,S}(G).(26)✦ If p≥Mp\ge M, then every k≥0k\ge0 and real g:𝔽p2→ℝg:\mathbb F_p^2\to\mathbb R satisfy the selected-pair identity (27): μp∘(z↦g(zvk,zvk+1))=𝒜p(g).\mu_p^\circ\bigl(z\mapsto g(z_{v_k},z_{v_{k+1}})\bigr)=\mathcal A_p(g).(27)✦ Let S⊆BS\subseteq B have |S|≤T|S|\le T, let Y:ℒS→𝔽pY:\mathcal L_S\to\mathbb F_p satisfy χS(Y)≠0\chi_S(Y)\ne0, and let λ:V→IS\lambda:V\to I_S. The output z=ΦS(Y,λ)z=\Phi_S(Y,\lambda) satisfies (28) for every k≥0k\ge0: zvk+1−zvk∈Qp*.z_{v_{k+1}}-z_{v_k}\in Q_p^\ast.(28)✦ If p≥Cp\ge C, then every v∈Vv\in V and real function f:𝔽p→ℝf:\mathbb F_p\to\mathbb R of uniform mean zero satisfy the squared estimates (29) and (30): ℰ(Wμp,v;f)≤p−1/32𝔼x∈𝔽pf(x)2.\mathscr E(W_{\mu_p,v};f)\le p^{-1/32}\mathbb E_{x\in\mathbb F_p}f(x)^2.(29)✦ ℰ(Wμp∘,v;f)≤p−1/32𝔼x∈𝔽pf(x)2.\mathscr E(W_{\mu_p^\circ,v};f)\le p^{-1/32}\mathbb E_{x\in\mathbb F_p}f(x)^2.(30)✦ There is an integer P≥64P\ge64 such that every prime p≥Pp\ge P satisfies (31): max⁡{M,C}≤p,R≤p,ηp≤p−4.\max\{M,C\}\le p,\qquad R\le p,\qquad \eta_p\le p^{-4}.(31)✦

In the paper, the result reads: there is an absolute integer PP such that every prime p≥Pp\ge P admits a probability law μp\mu_p on 𝔽pV\mathbb F_p^V and a nonnegative submeasure ϵp≤μp\epsilon_p\le\mu_p of mass ηp≤p−4\eta_p\le p^{-4}, with μp∘=(μp−ϵp)/(1−ηp)\mu_p^\circ=(\mu_p-\epsilon_p)/(1-\eta_p), such that μp\mu_p is reflection positive, both μp\mu_p and ϵp\epsilon_p are invariant under simultaneous zv↦szv+az_v\mapsto sz_v+a for s∈Qp*s\in Q_p^\ast and a∈𝔽pa\in\mathbb F_p, every one-label marginal of μp\mu_p and μp∘\mu_p^\circ is uniform, under μp∘\mu_p^\circ all gaps zvj+1−zvjz_{v_{j+1}}-z_{v_j} for 0≤j<h0\le j<h are strict squares and each ordered pair (zvj,zvj+1)(z_{v_j},z_{v_{j+1}}) is uniform on Ωp\Omega_p, and for ρp∈{μp,μp∘}\rho_p\in\{\mu_p,\mu_p^\circ\}, every v∈Vv\in V and every complex function f:𝔽p→ℂf:\mathbb F_p\to\mathbb C of uniform mean zero, ∥𝔼ρp[f(zv)∣(zw)w≠v]∥2≤p−1/64∥f∥2\|\mathbb E_{\rho_p}[f(z_v)\mid(z_w)_{w\ne v}]\|_2 \le p^{-1/64}\|f\|_2, where the left norm uses the other-label marginal and the right norm uses uniform probability on 𝔽p\mathbb F_p.

What Is Checked

The statements are over weighted permutation-word configurations in 𝔽pV\mathbb F_p^V, with each assertion using its specified prime threshold. The component averages Cp,SC_{p,S} and the full mixture ℳp\mathcal M_p are linear in their real observable GG.