OctoInformBeta

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

Lemma 4.2: Product tuple laws give mean domination and square-affine invariance, with exact conductor supports for projected lifts

Coordinate averaging contracts translation-invariant diagonal-root seminorms.

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 real function on a product of prime fields can be tested by placing it at every vertex of a random configuration and taking the resulting product average. The quantity of interest is the root of this diagonal average, together with its behavior when some field coordinates are averaged out.

Let VV be a finite nonempty vertex set of cardinality dd, let XX be a state space, and let LL be a real linear functional on real functions on XVX^V. The multilinear integral ΛL\Lambda_L of a family a=(av)v∈Va=(a_v)_{v\in V} of real functions on XX, its diagonal integral DLD_L, and its diagonal root RLR_L are defined by (1). ΛL(a)=L(z↦∏v∈Vav(zv)),DL(g)=ΛL((g)v∈V),RL(g)=DL(g)1/d.\begin{aligned} \Lambda_L(a)&=L\left(z\mapsto\prod_{v\in V}a_v(z_v)\right),\\ D_L(g)&=\Lambda_L((g)_{v\in V}), & R_L(g)&=D_L(g)^{1/d}. \end{aligned}(1) The positivity and multilinear Hölder hypotheses on LL are the conditions in (2), for every real gg and every family aa. DL(g)≥0,|ΛL(a)|≤∏v∈VRL(av).D_L(g)\geq 0, \qquad |\Lambda_L(a)|\leq\prod_{v\in V}R_L(a_v).(2)

The Tuple Laws

A permutation word vv on a block set DD is a function assigning a permutation of the positions to each block of DD.✦ The tuple vertices 𝒱\mathcal V are the permutation words on the block set ℬ\mathcal B with the parameters in (3); MM is half the vertex count and rr is the list base. h=24,T=80,m=2T+1=161,ℬ={0,…,m−1},𝒱=Sym⁡({0,…,h−1})ℬ,r=h!/2,M=(h!)m/2,δ=110h2.\begin{gathered} h=24,\qquad T=80,\qquad m=2T+1=161,\qquad \mathcal B=\{0,\ldots,m-1\},\\ \mathcal V=\operatorname{Sym}(\{0,\ldots,h-1\})^{\mathcal B}, \qquad r=h!/2,\qquad M=(h!)^m/2,\qquad \delta=\frac{1}{10h^2}. \end{gathered}(3) For S⊆ℬS\subseteq\mathcal B, an inactive-block address u∈𝒜Su\in\mathcal A_S is a permutation word on ℬ\S\mathcal B\setminus S.✦ An addressed list entry e∈ℰSe\in\mathcal E_S, for S⊆ℬS\subseteq\mathcal B, is a pair of an inactive-block address and an index in ISI_S, as in (4).✦ For S⊆ℬS\subseteq\mathcal B, the inactive-block projection πS:𝒱→𝒜S\pi_S:\mathcal V\to\mathcal A_S is the map that restricts a tuple vertex to ℬ\S\mathcal B\setminus S. 𝒜S=Sym⁡({0,…,h−1})ℬ\S,IS={0,…,r|S|−1},ℰS=𝒜S×IS,πS(v)=v|ℬ\S.\mathcal A_S=\operatorname{Sym}(\{0,\ldots,h-1\})^{\mathcal B\setminus S}, \qquad I_S=\{0,\ldots,r^{|S|}-1\},\qquad \mathcal E_S=\mathcal A_S\times I_S,\qquad \pi_S(v)=v|_{\mathcal B\setminus S}.(4) The cyclic shift cc rotates the hh positions, returning the last position to zero.✦ For a finite block set DD, a group GG, an element c∈Gc\in G, and a threshold tt, two addresses u,v∈GDu,v\in G^D are threshold shift adjacent when at least tt blocks satisfy vb=ubcv_b=u_bc.✦ Write 𝒢S⊆ℰS×ℰS\mathcal G_S\subseteq \mathcal E_S\times\mathcal E_S for the list constraint edges specifying the square-difference constraints.

An ordered position pair is a pair (i,k)(i,k) with 0≤i<k<h0\leq i<k<h.✦ For a tuple vertex v∈𝒱v\in\mathcal V, a block b∈ℬb\in\mathcal B, and an ordered position pair (i,k)(i,k), the position transposition vb;i,k∈𝒱v^{b;i,k}\in\mathcal V is the tuple vertex obtained by replacing vbv_b by vb∘(ik)v_b\circ(i\ k) and leaving the other block permutations fixed.✦ For S⊆ℬS\subseteq\mathcal B, the transposition interaction HSH_S is the real matrix on 𝒱\mathcal V given by the counting sum in (5). HS(v,w)=∑b∈S∑0≤i<k<hδk−i𝟏{w=vb;i,k}.H_S(v,w)=\sum_{b\in S}\sum_{0\leq i<k<h} \delta^{k-i}\mathbf 1_{\{w=v^{b;i,k}\}}.(5)✦ For a real matrix HH on a finite set II, a label set KK, and labelings λ,μ:I→K\lambda,\mu:I\to K, the label coincidence pairing CHC_H, index energy ℋH\mathcal H_H, and index Boltzmann weight WHW_H are defined by (6); both sums are counting sums. CH(λ,μ)=∑v∈I∑w∈IH(v,w)𝟏{λ(v)=μ(w)},ℋH(λ)=12CH(λ,λ),WH(λ)=exp⁡(ℋH(λ)).C_H(\lambda,\mu)=\sum_{v\in I}\sum_{w\in I} H(v,w)\mathbf 1_{\{\lambda(v)=\mu(w)\}},\qquad \mathcal H_H(\lambda)=\tfrac12C_H(\lambda,\lambda),\qquad W_H(\lambda)=\exp(\mathcal H_H(\lambda)).(6)

For a field FF, the nonzero square indicator sFs_F is one on the nonzero squares of FF and zero elsewhere; write sp=s𝔽ps_p=s_{\mathbb F_p} when pp is prime.✦ Given list values Y:ℰS→𝔽pY:\mathcal E_S\to\mathbb F_p and a labeling λ:𝒱→IS\lambda:\mathcal V\to I_S, the square-constraint indicator Vp,SV_{p,S} and sampled tuple configuration ZS,Y,λZ_{S,Y,\lambda} are defined in (7), together with the tuple Boltzmann weight WSW_S. Vp,S(Y)=∏(e,e′)∈𝒢Ssp(Y(e′)−Y(e)),ZS,Y,λ(v)=Y(πS(v),λ(v)),WS(λ)=WHS(λ).\begin{aligned} V_{p,S}(Y)&=\prod_{(e,e')\in\mathcal G_S}s_p(Y(e')-Y(e)),\\ Z_{S,Y,\lambda}(v)&=Y(\pi_S(v),\lambda(v)), & W_S(\lambda)&=W_{H_S}(\lambda). \end{aligned}(7) For a real observable FF on 𝔽p𝒱\mathbb F_p^{\mathcal V}, the tuple component Cp,S(F)C_{p,S}(F) averages over uniform list values and uniform labelings, as in (8). The unnormalized tuple mixture ℳp\mathcal M_p weights each component by p−|S|/16p^{-|S|/16}; the full tuple law νp\nu_p divides this mixture by its total mass, as in (8). Cp,S(F)=𝔼Y:ℰS→𝔽p𝔼λ:𝒱→ISVp,S(Y)WS(λ)F(ZS,Y,λ),ℳp(F)=∑S⊆ℬ|S|≤T+1p−|S|/16Cp,S(F),νp(F)=ℳp(1)−1ℳp(F).\begin{aligned} C_{p,S}(F)&=\mathbb E_{Y:\mathcal E_S\to\mathbb F_p} \mathbb E_{\lambda:\mathcal V\to I_S} V_{p,S}(Y)W_S(\lambda)F(Z_{S,Y,\lambda}),\\ \mathcal M_p(F)&=\sum_{\substack{S\subseteq\mathcal B\\|S|\leq T+1}} p^{-|S|/16}C_{p,S}(F), &\nu_p(F)&=\mathcal M_p(1)^{-1}\mathcal M_p(F). \end{aligned}(8) Here the outer sum is a counting sum. The reciprocal is taken to be zero when its argument is zero.

For a real linear functional KK on a finite configuration space, its point mass at ww is ρK(w)=K(𝟏{w})\rho_K(w)=K(\mathbf 1_{\{w\}}).✦ The tensor product of functionals KjK_j on XjVX_j^V is the counting sum in (9), where the XjX_j and the coordinate set JJ are finite. For a family of primes p=(pj)j∈Jp=(p_j)_{j\in J}, the product residue space is Xp=∏j∈J𝔽pjX_p=\prod_{j\in J}\mathbb F_{p_j}✦, and the product tuple law is νp=⨂j∈Jνpj\nu_p=\bigotimes_{j\in J}\nu_{p_j}.✦ (⨂j∈JKj)(F)=∑z∈∏j∈JXjV(∏j∈JρKj(zj))F((zj(v))v∈V,j∈J).\left(\bigotimes_{j\in J}K_j\right)(F) =\sum_{z\in\prod_{j\in J}X_j^V} \left(\prod_{j\in J}\rho_{K_j}(z_j)\right) F\bigl((z_j(v))_{v\in V,\,j\in J}\bigr).(9)✦

Let e=|𝒢⌀|e=|\mathcal G_{\varnothing}| be the leading graph edge count.✦ The Boltzmann weight constant WmaxW_{\max} and interaction lower bound κ\kappa are given in (10). The same display defines the marked and unmarked reflection constants CmarkC_{\mathrm{mark}} and CunmarkC_{\mathrm{unmark}}, and the mass and reflection thresholds PmassP_{\mathrm{mass}} and PrefP_{\mathrm{ref}}. Wmax=exp⁡(|𝒱|(T+1)(h2)δ2),κ=910δh−1,Cmark=(exp⁡(κ)−1)M(rT+1)M,Cunmark=(32|𝒱|2)1/4Wmax⁡2,Pmass=max⁡{3,(2e+1e)2},Pref=max⁡{3,(CunmarkrMCmark)16}.\begin{aligned} W_{\max}&=\exp\left(\frac{|\mathcal V|(T+1)\binom h2\delta}{2}\right), &\kappa&=\frac9{10}\delta^{h-1},\\ C_{\mathrm{mark}}&=\frac{(\exp(\kappa)-1)^M}{(r^{T+1})^M}, &C_{\mathrm{unmark}}&=(32|\mathcal V|^2)^{1/4}W_{\max}^2,\\ P_{\mathrm{mass}}&=\max\{3,(2^{e+1}e)^2\}, &P_{\mathrm{ref}}&=\max\left\{3, \left(\frac{C_{\mathrm{unmark}}r^M}{C_{\mathrm{mark}}}\right)^{16}\right\}. \end{aligned}(10)

Coordinate Averaging and Residue Lifts

For a finite product X=∏j∈JXjX=\prod_{j\in J}X_j, a subset B⊆JB\subseteq J, and points x,z∈Xx,z\in X, the coordinate replacement x[B←z]∈Xx[B\leftarrow z]\in X is the point with coordinate zjz_j for j∈Bj\in B and xjx_j otherwise.✦ The coordinate averaging projection EBE_B on a finite product X=∏j∈JXjX=\prod_{j\in J}X_j, for B⊆JB\subseteq J, is the operator on real functions on XX given in (11), with uniform probability average over z∈Xz\in X. (EBf)(x)=𝔼z∈Xf(x[B←z]).(E_Bf)(x)=\mathbb E_{z\in X}f(x[B\leftarrow z]).(11)✦

Keep the prime family pp fixed. For a prime qq, the centered residue delta kqk_q is the real function on 𝔽q\mathbb F_q given in (12), with argument t∈𝔽qt\in\mathbb F_q; for U⊆JU\subseteq J, the centered residue product kUk_U is the real function on XpX_p given there, with argument z∈Xpz\in X_p. The residue lift component FL0,UF_{L_0,U} associated with a real sequence f:ℕ→ℝf:\mathbb N\to\mathbb R, a nonnegative integer length L0L_0, and a support U⊆JU\subseteq J is the real function on XpX_p given by the normalized counting sum in (12). kq(t)=q𝟏{t=0}−1,kU(z)=∏j∈Ukpj(zj),FL0,U(x)=L0−1∑n=0L0−1f(n)kU(x−(nmod⁡pj)j∈J).\begin{aligned} k_q(t)&=q\mathbf 1_{\{t=0\}}-1, &k_U(z)&=\prod_{j\in U}k_{p_j}(z_j),\\ F_{L_0,U}(x)&=L_0^{-1}\sum_{n=0}^{L_0-1} f(n)k_U\bigl(x-(n\bmod p_j)_{j\in J}\bigr). \end{aligned}(12) The interval used here is {0,…,L0−1}\{0,\ldots,L_0-1\}; a sequence on the public interval {1,…,L0}\{1,\ldots,L_0\} enters through f(n)=a(n+1)f(n)=a(n+1). At L0=0L_0=0 the reciprocal in (12) is zero.

The conductor of a support U⊆JU\subseteq J is qU=∏j∈Upjq_U=\prod_{j\in U}p_j. For a nonnegative integer cutoff QQ and B⊆JB\subseteq J, the conductor-bounded support family 𝒮Q\mathscr S_Q and remaining family ℛB,Q\mathscr R_{B,Q} are families of subsets of JJ given in (13); for a real sequence f:ℕ→ℝf:\mathbb N\to\mathbb R and a nonnegative integer length L0L_0, the conductor-truncated residue lift FL0,QF_{L_0,Q} is the real function on XpX_p given there. Each term keeps the complete exact support UU. 𝒮Q={U⊆J:qU≤Q},ℛB,Q={U∈𝒮Q:U∩B=⌀},FL0,Q(x)=∑U∈𝒮QFL0,U(x).\mathscr S_Q=\{U\subseteq J:q_U\leq Q\},\qquad \mathscr R_{B,Q}=\{U\in\mathscr S_Q:U\cap B=\varnothing\},\qquad F_{L_0,Q}(x)=\sum_{U\in\mathscr S_Q}F_{L_0,U}(x).(13)

The Claim

Theorem 1. Let VV be finite and nonempty, and let LL satisfy the positivity and multilinear Hölder hypotheses (2). If |V||V| is even, the map in (14) is a seminorm on the real functions on XX. ∥g∥L:=RL(g).\|g\|_L:=R_L(g).(14)✦

Let JJ be finite and pjp_j prime for every j∈Jj\in J. If pj≥max⁡{Pmass,Pref}p_j\geq\max\{P_{\mathrm{mass}},P_{\mathrm{ref}}\} for every jj, mean domination holds as in (15), for every real g:Xp→ℝg:X_p\to\mathbb R. |𝔼x∈Xpg(x)|≤Rνp(g).\left|\mathbb E_{x\in X_p}g(x)\right|\leq R_{\nu_p}(g).(15)✦

For any finite prime family, choose q,a∈Xpq,a\in X_p with qj≠0q_j\ne0 for every jj, and put A(x)j=qj2xj+ajA(x)_j=q_j^2x_j+a_j. The product tuple law has the square-affine invariance (16) for every real observable FF on Xp𝒱X_p^{\mathcal V}. νp(z↦F((A(zv))v∈𝒱))=νp(F).\nu_p\bigl(z\mapsto F((A(z_v))_{v\in\mathcal V})\bigr)=\nu_p(F).(16)✦

Suppose X=∏j∈JXjX=\prod_{j\in J}X_j with finite additive groups XjX_j, and suppose LL is invariant under simultaneous translations zv↦zv+az_v\mapsto z_v+a for every a∈Xa\in X. Under (2), every real gg and every B⊆JB\subseteq J satisfy the projection contraction (17). RL(EBg)≤RL(g).R_L(E_Bg)\leq R_L(g).(17)✦

For the residue lifts, let L0,QL_0,Q be nonnegative integers, f:ℕ→ℝf:\mathbb N\to\mathbb R, and B⊆JB\subseteq J. The surviving-support identity is (18). EBFL0,Q(x)=∑U∈ℛB,QFL0,U(x).E_BF_{L_0,Q}(x)=\sum_{U\in\mathscr R_{B,Q}}F_{L_0,U}(x).(18)✦ If pj≥Prefp_j\geq P_{\mathrm{ref}} for every jj, these lifts also satisfy the diagonal-integral inequality (19). Dνp(∑U∈ℛB,QFL0,U)≤Dνp(FL0,Q).D_{\nu_p}\left(\sum_{U\in\mathscr R_{B,Q}}F_{L_0,U}\right) \leq D_{\nu_p}(F_{L_0,Q}).(19)✦

In the paper, the result reads: for the finite prime sets 𝒫\mathcal P and real functions gg on ∏p∈𝒫𝔽p\prod_{p\in\mathcal P}\mathbb F_p of this section, g↦∥g∥𝒫g\mapsto\|g\|_{\mathcal P} is a seminorm with |𝔼g|≤∥g∥𝒫|\mathbb E g|\leq\|g\|_{\mathcal P}, square-affine maps A(x)p=rpxp+apA(x)_p=r_px_p+a_p with ap∈𝔽pa_p\in\mathbb F_p and nonzero square rpr_p satisfy ∥g∘A∥𝒫=∥g∥𝒫\|g\circ A\|_{\mathcal P}=\|g\|_{\mathcal P}, and averaging the coordinates in any B⊆𝒫B\subseteq\mathcal P gives ∥EBg∥𝒫\B=∥EBg∥𝒫≤∥g∥𝒫\|E_Bg\|_{\mathcal P\setminus B}=\|E_Bg\|_{\mathcal P}\leq \|g\|_{\mathcal P}.

What Is Checked

The abstract statements concern real configuration functionals on finite nonempty vertex sets under the positivity and Hölder hypotheses, with even vertex count for the seminorm and simultaneous translation invariance for projection contraction. The product statements use finite prime fields with the stated thresholds, and the residue lifts keep the fixed ambient prime family.

The tuple mixture functional and the tensor product functional are real linear. For a finite nonempty vertex set VV, a state space XX, and a real linear functional LL on real functions on XVX^V satisfying the positivity and Hölder hypotheses, the diagonal-root triangle inequality (20) holds for real functions f,gf,g on XX. RL(f+g)≤RL(f)+RL(g).R_L(f+g)\leq R_L(f)+R_L(g).(20)✦