OctoInformBeta

OpenAI preprint, September 2026 · Section 4. Signed multilinear forms and conductor decay

Proposition 4.1: Product tuple laws at sufficiently large primes have nonnegative diagonal integrals

Signed Hölder bounds from reflection positivity.

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 multilinear averages here place arbitrary real functions at the vertices of a jointly sampled configuration. Their diagonal integrals provide the factors in a signed Hölder bound.

Let VV be a finite vertex set, XX a state space, and LL a real linear functional on real functions on XVX^V. For a family a=(av)v∈Va=(a_v)_{v\in V} of real functions on XX and a real function gg on XX, the multilinear integral 𝒵L(a)\mathcal Z_L(a) and the diagonal integral ZL(g)Z_L(g) are the real numbers given by (1). The diagonal root DL(g)D_L(g) of a real function gg on XX is the real power of its diagonal integral specified in (1), with d=|V|d=|V|. 𝒵L(a)=L(z↦∏v∈Vav(zv)),ZL(g)=𝒵L((g)v∈V),DL(g)=ZL(g)1/d.\mathcal Z_L(a)=L\!\left(z\mapsto\prod_{v\in V}a_v(z_v)\right), \qquad Z_L(g)=\mathcal Z_L((g)_{v\in V}), \qquad D_L(g)=Z_L(g)^{1/d}.(1)✦

For a finite block set BB and a natural number qq, a permutation word is a function w:B→𝔖qw:B\to\mathfrak S_q, where 𝔖q\mathfrak S_q is the permutation group of [q]={0,…,q−1}[q]=\{0,\ldots,q-1\}.✦ Write 𝒲(B,q)=𝔖qB\mathcal W(B,q)=\mathfrak S_q^B. For a word w∈𝒲(B,q)w\in\mathcal W(B,q), b∈Bb\in B and α,β∈[q]\alpha,\beta\in[q], the block transposition reflection Rb;α,βwR_{b;\alpha,\beta}w is the word that agrees with ww off bb and has block (αβ)∘w(b)(\alpha\,\beta)\circ w(b) at bb.✦ A word w∈𝒲(B,q)w\in\mathcal W(B,q) satisfies the positive cut condition at b∈Bb\in B for α,β∈[q]\alpha,\beta\in[q] when w(b)−1(α)<w(b)−1(β)w(b)^{-1}(\alpha)<w(b)^{-1}(\beta)✦; the positive cut half Cb;α,βC_{b;\alpha,\beta} is the subset of 𝒲(B,q)\mathcal W(B,q) given by (2). Cb;α,β={w∈𝒲(B,q):w(b)−1(α)<w(b)−1(β)}.C_{b;\alpha,\beta} =\{w\in\mathcal W(B,q):w(b)^{-1}(\alpha)<w(b)^{-1}(\beta)\}.(2)✦

For a real linear functional LL on functions on X𝒲(B,q)X^{\mathcal W(B,q)}, the word-cut form QLb;α,βQ_L^{b;\alpha,\beta} is the bilinear form on real functions on XCb;α,βX^{C_{b;\alpha,\beta}} defined in (3). QLb;α,β(F,G)=L(z↦F((zw)w∈Cb;α,β)G((zRb;α,βw)w∈Cb;α,β)).Q_L^{b;\alpha,\beta}(F,G) =L\!\left(z\mapsto F((z_w)_{w\in C_{b;\alpha,\beta}}) G((z_{R_{b;\alpha,\beta}w})_{w\in C_{b;\alpha,\beta}})\right).(3) A real linear functional LL on functions on X𝒲(B,q)X^{\mathcal W(B,q)} has cut symmetry when QLb;α,β(F,G)=QLb;α,β(G,F)Q_L^{b;\alpha,\beta}(F,G)=Q_L^{b;\alpha,\beta}(G,F) for every cut and all real functions F,GF,G on XCb;α,βX^{C_{b;\alpha,\beta}}. Cut positivity of a real linear functional LL on functions on X𝒲(B,q)X^{\mathcal W(B,q)} means that QLb;α,β(F,F)≥0Q_L^{b;\alpha,\beta}(F,F)\geq0 for every cut with α≠β\alpha\ne\beta and every real function FF on XCb;α,βX^{C_{b;\alpha,\beta}}.

Tuple Laws

The tuple parameters are the height h=24h=24, block cutoff T=80T=80, block count m=2T+1=161m=2T+1=161, and list base r=h!/2r=h!/2. Put ℬ={0,…,m−1}\mathcal B=\{0,\ldots,m-1\} and 𝒱=𝒲(ℬ,h)\mathcal V=\mathcal W(\mathcal B,h); these are the permutation tuple vertices, whose number is d=(h!)md=(h!)^m.✦

For an active block subset S⊆ℬS\subseteq\mathcal B, an inactive-block address u∈ASu\in A_S is a function assigning a permutation in 𝔖h\mathfrak S_h to every block outside SS.✦ The addressed list indices ISI_S and addressed list entries ESE_S are the finite sets in (4). For an active block subset S⊆ℬS\subseteq\mathcal B and a tuple vertex v∈𝒱v\in\mathcal V, the inactive-block projection πS(v)\pi_S(v) is the address in ASA_S obtained by restricting vv to the blocks outside SS. AS=𝔖hℬ\S,IS={0,…,r|S|−1},ES=AS×IS,πS(v)=v|ℬ\S.A_S=\mathfrak S_h^{\mathcal B\setminus S},\qquad I_S=\{0,\ldots,r^{|S|}-1\},\qquad E_S=A_S\times I_S,\qquad \pi_S(v)=v|_{\mathcal B\setminus S}.(4)✦

The cyclic shift c∈𝔖hc\in\mathfrak S_h sends each position to the next, with the last returning to zero.✦ For a finite block set CC, a group GG, an element c∈Gc\in G, and a natural threshold KK, threshold shift adjacency of u,v∈GCu,v\in G^C is the condition (5). |{b∈C:v(b)=u(b)c}|≥K.\bigl|\{b\in C:v(b)=u(b)c\}\bigr|\geq K.(5)✦

Let ℰS⊆ES×ES\mathscr E_S\subseteq E_S\times E_S be the list constraint edges, the ordered pairs of addressed list entries specifying the square-difference constraints between list values. The nonzero square indicator χF\chi_F on a field FF takes value one at nonzero squares and zero elsewhere.✦ For a prime pp, the square-constraint indicator Φp,S\Phi_{p,S} of list values Y:ES→𝑭pY:E_S\to\mathbf F_p is the product in (6). Φp,S(Y)=∏(e,f)∈ℰSχ𝑭p(Y(f)−Y(e)).\Phi_{p,S}(Y) =\prod_{(e,f)\in\mathscr E_S}\chi_{\mathbf F_p}(Y(f)-Y(e)).(6)✦

An ordered position pair is a pair (s,t)∈[q]2(s,t)\in[q]^2 with s<ts<t.✦ For a word w∈𝒲(B,q)w\in\mathcal W(B,q), a block b∈Bb\in B and an ordered position pair (s,t)∈[q]2(s,t)\in[q]^2, the position transposition τb;s,tw\tau_{b;s,t}w is the word that agrees with ww off bb and has block w(b)∘(st)w(b)\circ(s\,t) at bb.✦ For S⊆BS\subseteq B and δ∈ℝ\delta\in\mathbb R, the transposition interaction HS,δH_{S,\delta} on words is the real-valued function of two words given by the counting sum in (7). HS,δ(v,w)=∑b∈S∑0≤s<t<qδt−s𝟏{w=τb;s,tv}.H_{S,\delta}(v,w) =\sum_{b\in S}\sum_{0\leq s<t<q} \delta^{t-s}\mathbf1_{\{w=\tau_{b;s,t}v\}}.(7)✦

The label coincidence pairing 𝒞A(ℓ,k)\mathcal C_A(\ell,k) of labelings ℓ,k:I→Λ\ell,k:I\to\Lambda, for a real matrix AA on a finite set II and a label set Λ\Lambda, is the real number given by the counting sum in (8). 𝒞A(ℓ,k)=∑x∈I∑y∈IAx,y𝟏{ℓ(x)=k(y)}.\mathcal C_A(\ell,k) =\sum_{x\in I}\sum_{y\in I}A_{x,y}\mathbf1_{\{\ell(x)=k(y)\}}.(8)✦ For a real matrix AA on a finite set II and a labeling ℓ:I→Λ\ell:I\to\Lambda, the index energy ℰA(ℓ)\mathcal E_A(\ell) and index Boltzmann weight ℬA(ℓ)\mathcal B_A(\ell) are the real numbers given by (9). ℰA(ℓ)=12𝒞A(ℓ,ℓ),ℬA(ℓ)=exp⁡(ℰA(ℓ)).\mathcal E_A(\ell)=\tfrac12\mathcal C_A(\ell,\ell),\qquad \mathcal B_A(\ell)=\exp(\mathcal E_A(\ell)).(9)

For tuple vertices take δ=1/(10h2)\delta=1/(10h^2).✦ The tuple Boltzmann weight WS(j)W_S(j) of a labeling j:𝒱→ISj:\mathcal V\to I_S is the real number given by (10).✦ For a natural number pp, list values Y:ES→ℤ/pℤY:E_S\to\mathbb Z/p\mathbb Z, and a labeling j:𝒱→ISj:\mathcal V\to I_S, the sampled tuple configuration zS,Y,j:𝒱→ℤ/pℤz_{S,Y,j}:\mathcal V\to\mathbb Z/p\mathbb Z assigns to each vertex vv the list value at (πS(v),j(v))(\pi_S(v),j(v)), as in (10). WS(j)=ℬHS,δ(j),zS,Y,j(v)=Y(πS(v),j(v)).W_S(j)=\mathcal B_{H_{S,\delta}}(j),\qquad z_{S,Y,j}(v)=Y(\pi_S(v),j(v)).(10)✦ The tuple component average Lp,S(F)L_{p,S}(F) of a real configuration observable FF is the uniform probability average in (11). Lp,S(F)=𝔼Y∈𝑭pES𝔼j∈IS𝒱Φp,S(Y)WS(j)F(zS,Y,j).L_{p,S}(F) =\mathbb E_{Y\in\mathbf F_p^{E_S}} \mathbb E_{j\in I_S^{\mathcal V}} \Phi_{p,S}(Y)W_S(j)F(z_{S,Y,j}).(11)✦

The block subset mixture weight θp(S)\theta_p(S) of S⊆ℬS\subseteq\mathcal B at a natural number pp is the real number p−|S|/16p^{-|S|/16}.✦ For a prime pp, the unnormalized tuple mixture L̃p\widetilde L_p and the full tuple law LpL_p are real linear functionals on real functions on 𝑭p𝒱\mathbf F_p^{\mathcal V} given by (12); the sum is a counting sum over block subsets. L̃p(F)=∑S⊆ℬ|S|≤T+1θp(S)Lp,S(F),Lp(F)=L̃p(F)L̃p(1).\widetilde L_p(F) =\sum_{\substack{S\subseteq\mathcal B\\|S|\leq T+1}} \theta_p(S)L_{p,S}(F),\qquad L_p(F)=\frac{\widetilde L_p(F)}{\widetilde L_p(1)}.(12)

For a real linear functional KK on functions on a set Ω\Omega, the point mass ρK(ω)\rho_K(\omega) of KK at ω∈Ω\omega\in\Omega is the real number ρK(ω)=K(𝟏{ω})\rho_K(\omega)=K(\mathbf1_{\{\omega\}}).✦ The tensor product law ⨂j∈JKj\bigotimes_{j\in J}K_j of real linear functionals KjK_j on functions on XjVX_j^V, where J,VJ,V and the state spaces XjX_j are finite, is the real linear functional on functions on configurations with state space ∏j∈JXj\prod_{j\in J}X_j given by the counting sum in (13), with the coordinate configurations assembled vertex by vertex. (⨂j∈JKj)(F)=∑(zj)∈∏j∈JXjV(∏j∈JρKj(zj))F(v↦(zj(v))j∈J).\left(\bigotimes_{j\in J}K_j\right)(F) =\sum_{(z_j)\in\prod_{j\in J}X_j^V} \left(\prod_{j\in J}\rho_{K_j}(z_j)\right) F\bigl(v\mapsto(z_j(v))_{j\in J}\bigr).(13)✦ For a finite family of primes (pj)j∈J(p_j)_{j\in J}, put XJ=∏j∈J𝑭pjX_J=\prod_{j\in J}\mathbf F_{p_j}. The product tuple law LJ=⨂j∈JLpjL_J=\bigotimes_{j\in J}L_{p_j} is a real linear functional on product configurations; write 𝒵J=𝒵LJ\mathcal Z_J=\mathcal Z_{L_J} and ZJ=ZLJZ_J=Z_{L_J}.✦

The Prime Threshold

Put M=d/2M=d/2.✦ The interaction lower bound κ\kappa and Boltzmann weight upper bound WmaxW_{\max} are the constants in (14). κ=910δh−1,Wmax=exp⁡(d(T+1)(h2)δ2).\kappa=\frac9{10}\delta^{h-1},\qquad W_{\max}=\exp\!\left(\frac{d(T+1)\binom h2\delta}{2}\right).(14) The marked reflection constant CmarkedC_{\mathrm{marked}} and unmarked reflection constant CunmarkedC_{\mathrm{unmarked}} are defined in (15). Cmarked=(exp⁡(κ)−1)M(rT+1)M,Cunmarked=(32d2)1/4Wmax⁡2.C_{\mathrm{marked}} =\frac{(\exp(\kappa)-1)^M}{(r^{T+1})^M},\qquad C_{\mathrm{unmarked}}=(32d^2)^{1/4}W_{\max}^2.(15) The reflection positivity threshold is the real number PrefP_{\mathrm{ref}} in (16). Pref=max⁡{3,(CunmarkedrMCmarked)16}.P_{\mathrm{ref}} =\max\!\left\{3, \left(\frac{C_{\mathrm{unmarked}}r^M}{C_{\mathrm{marked}}}\right)^{16} \right\}.(16)✦

The Claim

Theorem 1. Let BB be a nonempty finite block set, let q≥2q\geq2, and let XX be a state space. Suppose that LL is a real linear functional on functions on X𝒲(B,q)X^{\mathcal W(B,q)} with cut symmetry and cut positivity. For arbitrary real functions aw:X→ℝa_w:X\to\mathbb R, the word signed Hölder bound is (17). |𝒵L((aw)w∈𝒲(B,q))|≤∏w∈𝒲(B,q)ZL(aw)1/|𝒲(B,q)|.\left|\mathcal Z_L((a_w)_{w\in\mathcal W(B,q)})\right| \leq\prod_{w\in\mathcal W(B,q)} Z_L(a_w)^{1/|\mathcal W(B,q)|}.(17)✦ Let JJ be a finite coordinate set and (pj)j∈J(p_j)_{j\in J} a family of primes with pj≥Prefp_j\geq P_{\mathrm{ref}} for every jj. For every real function g:XJ→ℝg:X_J\to\mathbb R, the product diagonal integral satisfies (18). 0≤ZJ(g).0\leq Z_J(g).(18)✦ For every family of real functions gv:XJ→ℝg_v:X_J\to\mathbb R, the product signed Hölder bound is (19). |𝒵J((gv)v∈𝒱)|≤∏v∈𝒱ZJ(gv)1/d.\left|\mathcal Z_J((g_v)_{v\in\mathcal V})\right| \leq\prod_{v\in\mathcal V}Z_J(g_v)^{1/d}.(19)✦

In the paper, with PP a lower threshold for the primes and X𝒫X_{\mathcal P}, Z𝒫Z_{\mathcal P} and 𝒵𝒫\mathcal Z_{\mathcal P} denoting the preceding objects with coordinates indexed by the primes themselves, the result reads: for every finite set 𝒫\mathcal P of primes at least PP and arbitrary real functions g,gv:X𝒫→ℝg,g_v:X_{\mathcal P}\to\mathbb R, Z𝒫(g)≥0Z_{\mathcal P}(g)\geq0 and |𝒵𝒫((gv))|≤∏v∈𝒱Z𝒫(gv)1/d|\mathcal Z_{\mathcal P}((g_v))|\leq\prod_{v\in\mathcal V}Z_{\mathcal P}(g_v)^{1/d}, including zero diagonal integrals.

What Is Checked

The word statement is over real linear functionals on permutation word configurations with a nonempty finite block set and at least two positions, satisfying cut symmetry and cut positivity; the product statements are over tensor products of full tuple laws for a finite family of prime fields of orders at least PrefP_{\mathrm{ref}}.