OctoInformBeta

OpenAI preprint, September 2026 · Section 7. Comparison along square-step progressions

Proposition 7.2: Common child bounds control diagonal and mixed tuple forms with ceiling factors and a decaying error

Frozen lifts transfer to shorter progressions.

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

For an integer aa, a positive integer DD and an integer index nn, a square-step progression has the form a+D2na+D^2n. Proposition 7.2 compares a residue slice of a square-difference-free set with sets on the shorter interval associated with this step. The ambient prime family remains fixed when the length changes.

For a natural number NN, write IN={0,…,N−1}I_N=\{0,\ldots,N-1\}. A finite set A⊆ℕA\subseteq\mathbb N is square-difference-free if b≠a+t2b\ne a+t^2 for a,b∈Aa,b\in A and every positive integer tt.✦ A set on the public interval {1,…,N}\{1,\ldots,N\} is represented on INI_N by translation down by one. Its indicator is 1A1_A.

Let JJ be a finite coordinate set, let p=(pj)j∈Jp=(p_j)_{j\in J} be distinct primes, and put Xp=∏j∈J𝔽pjX_p=\prod_{j\in J}\mathbb F_{p_j}. The conductor qUq_U of a coordinate subset U⊆JU\subseteq J is the positive integer defined by qU=∏j∈Upjq_U=\prod_{j\in U}p_j. For a positive integer modulus qq, a positive integer length LL, a natural number cutoff QQ and a real function ff on ℕ\mathbb N, the centered residue delta RqR_q on ℤ/qℤ\mathbb Z/q\mathbb Z, centered residue product GUG_U on XpX_p, residue lift component ℓp,L,Uf\ell_{p,L,U}f on XpX_p, and conductor-truncated lift ℒp,L,Qf\mathcal L_{p,L,Q}f on XpX_p are the real functions in (1); the last sum retains every complete support of conductor at most QQ. Rq(t)=q1{0}(t)−1,GU(x)=∏j∈URpj(xj),(ℓp,L,Uf)(x)=1L∑n∈ILf(n)GU(x−(nmod⁡pj)j∈J),𝒰p(Q)={U⊆J:qU≤Q},ℒp,L,Qf=∑U∈𝒰p(Q)ℓp,L,Uf.\begin{aligned} R_q(t)&=q\,1_{\{0\}}(t)-1, &G_U(x)&=\prod_{j\in U}R_{p_j}(x_j),\\ (\ell_{p,L,U}f)(x) &=\frac1L\sum_{n\in I_L}f(n)G_U\bigl(x-(n\bmod p_j)_{j\in J}\bigr),\\ \mathcal U_p(Q)&=\{U\subseteq J:q_U\le Q\}, &\mathcal L_{p,L,Q}f&=\sum_{U\in\mathcal U_p(Q)}\ell_{p,L,U}f. \end{aligned}(1) Here and below ∑\sum denotes a counting sum, whereas 𝔼\mathbb E denotes a uniform probability average on the indicated finite space. The product residue space XpX_p is the Cartesian product of the residue fields 𝔽pj\mathbb F_{p_j} over j∈Jj\in J. The conductor-bounded support family 𝒰p(Q)\mathcal U_p(Q) consists of the coordinate subsets U⊆JU\subseteq J whose conductor qUq_U is at most QQ.

For a positive integer MM, a residue s∈ℤ/Mℤs\in\mathbb Z/M\mathbb Z, a coordinate subset B⊆JB\subseteq J and z∈Xpz\in X_p, the normalized residue slice fsf_s of AA is a real function on ℕ\mathbb N and the coordinate replacement θB,z\theta_{B,z} is a map from XpX_p to XpX_p, given by (2). The function gs,zN,Qg_{s,z}^{N,Q} is the frozen lift on XpX_p. fs(n)=M1A(n)1{nmod⁡M=s},(θB,zx)j={zjj∈B,xjj∉B,gs,zN,Q(x)=(ℒp,N,Qfs)(θB,zx).\begin{aligned} f_s(n)&=M1_A(n)1_{\{n\bmod M=s\}},\\ (\theta_{B,z}x)_j&= \begin{cases}z_j&j\in B,\\x_j&j\notin B,\end{cases} &g_{s,z}^{N,Q}(x)&=(\mathcal L_{p,N,Q}f_s)(\theta_{B,z}x). \end{aligned}(2)

The Tuple Laws

For a positive integer hh and a block set CC, a permutation word vv of height hh on CC is a function assigning a permutation of {0,…,h−1}\{0,\ldots,h-1\} to each block.✦ Set h=24h=24, T=80T=80 and m=2T+1=161m=2T+1=161, and let 𝔖h\mathfrak S_h denote the symmetric group on {0,…,h−1}\{0,\ldots,h-1\}; the tuple vertex set is 𝒱=(𝔖h)m\mathcal V=(\mathfrak S_h)^m; write d=|𝒱|d=|\mathcal V|, r=h!/2r=h!/2 and δ=1/(10h2)\delta=1/(10h^2). For an active block set S⊆{0,…,m−1}S\subseteq\{0,\ldots,m-1\}, an inactive-block address is a permutation word on the complementary blocks.✦ Write 𝒜S\mathcal A_S for these addresses, πSv=v|Sc\pi_Sv=v|_{S^c} for the inactive-block projection, and ℐS={0,…,r|S|−1}\mathcal I_S=\{0,\ldots,r^{|S|}-1\} for the addressed list indices. An addressed list entry belongs to ℰS=𝒜S×ℐS\mathcal E_S=\mathcal A_S\times\mathcal I_S✦; the empty-block list index is 00.✦

The ordered position pairs are (a,b)(a,b) with 0≤a<b<h0\le a<b<h.✦ For a word v∈𝒱v\in\mathcal V and a block c∈{0,…,m−1}c\in\{0,\ldots,m-1\}, the position transposition vc;a,bv^{c;a,b} replaces vcv_c by vc(ab)v_c(a\ b).✦ For a real interaction HH on 𝒱×𝒱\mathcal V\times\mathcal V and labelings j,k:𝒱→ℐSj,k:\mathcal V\to\mathcal I_S, the transposition interaction HSH_S, label coincidence pairing CHC_H, index energy EHE_H and index Boltzmann weight WHW_H are given by (3); the tuple Boltzmann weight is WS=WHSW_S=W_{H_S}.✦ HS(v,w)=∑c∈S∑0≤a<b<hδb−a1{w=vc;a,b},CH(j,k)=∑v,w∈𝒱H(v,w)1{j(v)=k(w)},EH(j)=12CH(j,j),WH(j)=exp⁡(EH(j)).\begin{aligned} H_S(v,w)&=\sum_{c\in S}\sum_{0\le a<b<h} \delta^{b-a}1_{\{w=v^{c;a,b}\}},\\ C_H(j,k)&=\sum_{v,w\in\mathcal V}H(v,w)1_{\{j(v)=k(w)\}}, &E_H(j)&=\tfrac12C_H(j,j), &W_H(j)&=\exp(E_H(j)). \end{aligned}(3) The cyclic shift chc_h rotates the hh positions.✦ Threshold shift adjacency of two words u,vu,v at threshold tt is the condition |{b:vb=ubch}|≥t|\{b:v_b=u_bc_h\}|\ge t.✦

Let ES⊆ℰS×ℰSE_S\subseteq\mathcal E_S\times\mathcal E_S be the list constraint edges specifying the square-difference constraints. For a prime pp, a list assignment Y∈𝔽pℰSY\in\mathbb F_p^{\mathcal E_S} and labeling j∈ℐS𝒱j\in\mathcal I_S^{\mathcal V}, the sampled tuple configuration is xS,Y,j(v)=Y(πSv,j(v))x_{S,Y,j}(v)=Y(\pi_Sv,j(v)).✦ The strict square indicator χp\chi_p on 𝔽p\mathbb F_p is the function equal to one on nonzero squares and zero elsewhere.✦ The square-constraint indicator VSV_S on list assignments and the tuple component functional 𝒞p,S\mathcal C_{p,S} on real observables FF on 𝔽p𝒱\mathbb F_p^{\mathcal V} are defined in (4). VS(Y)=∏(a,b)∈ESχp(Yb−Ya),𝒞p,S(F)=𝔼Y∈𝔽pℰS𝔼j∈ℐS𝒱VS(Y)WS(j)F(xS,Y,j).\begin{aligned} V_S(Y)&=\prod_{(a,b)\in E_S}\chi_p(Y_b-Y_a),\\ \mathcal C_{p,S}(F) &=\mathbb E_{Y\in\mathbb F_p^{\mathcal E_S}} \mathbb E_{j\in\mathcal I_S^{\mathcal V}} V_S(Y)W_S(j)F(x_{S,Y,j}). \end{aligned}(4)

The restricted tuple mixture Ξp,𝒟\Xi_{p,\mathcal D} for a family 𝒟\mathcal D of block subsets is the weighted sum of the component functionals in (5). Its weights are p−|S|/16p^{-|S|/16}.✦ The unnormalized tuple mixture Ξp\Xi_p uses |S|≤T+1|S|\le T+1✦, and the terminal family 𝒯\mathcal T uses |S|=T+1|S|=T+1.✦ The full tuple law μp\mu_p, exceptional tuple sublaw εp\varepsilon_p, and exceptional mass ηp\eta_p are the normalized functionals in that display. Ξp,𝒟=∑S∈𝒟p−|S|/16𝒞p,S,Ξp=Ξp,{S:|S|≤T+1},𝒯={S:|S|=T+1},μp=Ξp(1)−1Ξp,εp=Ξp(1)−1Ξp,𝒯,ηp=εp(1).\begin{aligned} \Xi_{p,\mathcal D}&=\sum_{S\in\mathcal D}p^{-|S|/16}\mathcal C_{p,S}, &\Xi_p&=\Xi_{p,\{S:|S|\le T+1\}}, &\mathcal T&=\{S:|S|=T+1\},\\ \mu_p&=\Xi_p(1)^{-1}\Xi_p, &\varepsilon_p&=\Xi_p(1)^{-1}\Xi_{p,\mathcal T}, &\eta_p&=\varepsilon_p(1). \end{aligned}(5)

For real functionals LjL_j on finite configuration spaces, the tensor product ⨂jLj\bigotimes_jL_j is the functional given by the counting sum in (6), weighted by the point masses of the LjL_j.✦ The product tuple law PpP_p uses the full law in every coordinate.✦ The mixed tensor law Pp,BP_{p,B} replaces the coordinate law by its exceptional sublaw precisely on BB; its exceptional mass product is ηB\eta_B. (⨂jLj)(F)=∑(zj)j(∏jLj(1{zj}))F((zj(v))v,j),Pp=⨂j∈Jμpj,Pp,B=⨂j∈J{εpjj∈B,μpjj∉B,ηB=∏j∈Bηpj.\begin{aligned} \left(\bigotimes_jL_j\right)(F) &=\sum_{(z_j)_j}\left(\prod_jL_j(1_{\{z_j\}})\right) F\bigl((z_j(v))_{v,j}\bigr),\\ P_p&=\bigotimes_{j\in J}\mu_{p_j}, &P_{p,B}&=\bigotimes_{j\in J} \begin{cases}\varepsilon_{p_j}&j\in B,\\\mu_{p_j}&j\notin B, \end{cases} &\eta_B&=\prod_{j\in B}\eta_{p_j}. \end{aligned}(6) The point mass of LjL_j at zjz_j here is Lj(1{zj})L_j(1_{\{z_j\}}).✦

For a real functional LL on configurations in Xp𝒱X_p^{\mathcal V} and real functions Fv:Xp→ℝF_v:X_p\to\mathbb R indexed by v∈𝒱v\in\mathcal V, the multilinear integral ℐL((Fv)v)\mathcal I_L((F_v)_v) is the real number given by (7). The diagonal integral ZpZ_p uses the same function at all vertices✦, and the set diagonal functional YpY_p applies it to an indicator lift, as in (7).✦ ℐL((Fv)v)=L(x↦∏v∈𝒱Fv(xv)),Zp(g)=ℐPp((g)v),Yp(L,Q,A)=Zp(ℒp,L,Q1A).\begin{aligned} \mathcal I_L((F_v)_v)&=L\left(x\mapsto\prod_{v\in\mathcal V}F_v(x_v)\right),\\ Z_p(g)&=\mathcal I_{P_p}((g)_v), &Y_p(L,Q,A)&=Z_p(\mathcal L_{p,L,Q}1_A). \end{aligned}(7)

Thresholds and Scales

The Boltzmann weight bound WmaxW_{\max}, interaction lower bound κ\kappa, half vertex count b0b_0, and marked and unmarked reflection constants are the numbers in (8). Wmax=exp⁡(d(T+1)(h2)δ/2),κ=910δh−1,b0=(h!)m/2,Cmarked=(exp⁡(κ)−1)b0(rT+1)b0,Cunmarked=(32d2)1/4Wmax⁡2.\begin{aligned} W_{\max}&=\exp\left(d(T+1)\binom h2\delta/2\right), &\kappa&=\tfrac9{10}\delta^{h-1}, &b_0&=(h!)^m/2,\\ C_{\mathrm{marked}}&=\frac{(\exp(\kappa)-1)^{b_0}}{(r^{T+1})^{b_0}}, &C_{\mathrm{unmarked}}&=(32d^2)^{1/4}W_{\max}^2. \end{aligned}(8)

The leading list identification sends vv to (π⌀v,0)(\pi_\varnothing v,0).✦ The leading constraint graph Γ\Gamma is the pullback of E⌀E_\varnothing along this bijection: a pair is an edge when its image is an edge.✦ Write e=|E⌀|e=|E_\varnothing|. For a finite directed graph EE and vertex vv, let nE(v)n_E(v) count the other vertices joined to vv in either orientation. The graph conditional energy constant cE(v)c_E(v), numerical leading mass lower bound a0a_0, and conditional error numerator AcondA_{\mathrm{cond}} are defined in (9); CcondC_{\mathrm{cond}} is the conditional decay constant. cE(v)=(2(1/2)nE(v)+4((1/2)nE(v))2)((32nE(v))1/4)2,a0=2−(e+1),Acond=∑v∈𝒱cΓ(v)+2mWmax,Ccond=Acond/a0.\begin{aligned} c_E(v)&=\left(\frac{2}{(1/2)^{n_E(v)}}+ \frac{4}{((1/2)^{n_E(v)})^2}\right) \left((32n_E(v))^{1/4}\right)^2,\\ a_0&=2^{-(e+1)}, &A_{\mathrm{cond}}&=\sum_{v\in\mathcal V}c_\Gamma(v)+2^mW_{\max}, &C_{\mathrm{cond}}&=A_{\mathrm{cond}}/a_0. \end{aligned}(9) The mass positivity threshold PmassP_{\mathrm{mass}}, conditional decay threshold PcondP_{\mathrm{cond}}, and reflection positivity threshold PrefP_{\mathrm{ref}} are defined by (10). Pmass=max⁡{3,(2e+1e)2},Pcond=max⁡{Pmass,3,Ccond32},Pref=max⁡{3,(Cunmarkedrb0/Cmarked)16}.\begin{aligned} P_{\mathrm{mass}}&=\max\{3,(2^{e+1}e)^2\},\\ P_{\mathrm{cond}}&=\max\{P_{\mathrm{mass}},3,C_{\mathrm{cond}}^{32}\},\\ P_{\mathrm{ref}}&=\max\{3,(C_{\mathrm{unmarked}}r^{b_0}/C_{\mathrm{marked}})^{16}\}. \end{aligned}(10)

The lift size conditions for positive integers L,QL,Q are 2Q≤L2Q\le L and Q5≤LQ^5\le L.✦ For a real exponent aa and nonnegative integers k,Mk,M, the power cutoff Qa(N)Q_a(N), source parameters, and progression transfer factor Tk,MT_{k,M} are defined in (11); abbreviate τ=τd\tau=\tau_d, σ=σd\sigma=\sigma_d and QN=Qβ(N)Q_N=Q_\beta(N). Qa(N)=⌊Na⌋,β=1100000,γ=164,τk=βγ1000(k+1),σk=γτk4,Tk,M(N)=(1+M2N2τk−1)k.\begin{aligned} Q_a(N)&=\lfloor N^a\rfloor, &\beta&=\frac1{100000}, &\gamma&=\frac1{64},\\ \tau_k&=\frac{\beta\gamma}{1000(k+1)}, &\sigma_k&=\frac{\gamma\tau_k}{4}, &T_{k,M}(N)&=(1+M^2N^{2\tau_k-1})^k. \end{aligned}(11) The constant β\beta is 1/1000001/100000. The parameter γ\gamma is 1/641/64. For a nonnegative integer kk, the parameter τk\tau_k is βγ/(1000(k+1))\beta\gamma/(1000(k+1)). For a nonnegative integer kk, the parameter σk\sigma_k is one quarter of γτk\gamma\tau_k. For B⊆JB\subseteq J put D=MqBD=Mq_B. The ceiling child length N′N' and the length factor λL\lambda_L for a proposed child length LL are (12). N′=⌈ND2⌉,λL=D2LN,λ=λN′.N'=\left\lceil\frac{N}{D^2}\right\rceil, \qquad \lambda_L=\frac{D^2L}{N}, \qquad \lambda=\lambda_{N'}.(12)

The Claim

Theorem 1 (Transfer along square-step progressions). For every ϵ>0\epsilon>0 there is Cϵ>0C_\epsilon>0 with the following uniform fiber estimate (13). Suppose every pjp_j is at least PcondP_{\mathrm{cond}} and PrefP_{\mathrm{ref}}, and MM is positive and coprime to every pjp_j. Let N,Q,H,L,QcN,Q,H,L,Q_c be natural numbers such that (N,Q)(N,Q) and (L,Qc)(L,Q_c) satisfy the lift size conditions, Q,Qc≤NQ,Q_c\le N, H>0H>0, HqB≤QHq_B\le Q, H≤QcH\le Q_c and N≤D2LN\le D^2L. Suppose A⊆INA\subseteq I_N is square-difference-free and K≥0K\ge0 bounds Yp(L,Qc,A′)Y_p(L,Q_c,A') for every square-difference-free A′⊆ILA'\subseteq I_L. For every s,zs,z, Zp(gs,zN,Q)≤λLdK+dH−1/64(CϵNϵ)d(λLd+qBMd).Z_p(g_{s,z}^{N,Q})\le \lambda_L^dK+ dH^{-1/64}(C_\epsilon N^\epsilon)^d (\lambda_L^d+q_BM^d).(13)✦

Fix M≥2M\ge2. There are C1>0C_1>0 and N1N_1 such that the asymptotic fiber estimate (14) holds for every N≥N1N\ge N_1, uniformly over the prime family and all fiber choices. Suppose the primes satisfy the same threshold and coprimality conditions, qB≤Nτq_B\le N^\tau, and A⊆INA\subseteq I_N is square-difference-free. If K≥0K\ge0 bounds Yp(N′,QN′,A′)Y_p(N',Q_{N'},A') for every square-difference-free A′⊆IN′A'\subseteq I_{N'}, then for every s,zs,z, Zp(gs,zN,QN)≤Td,M(N)K+C1N−σ.Z_p(g_{s,z}^{N,Q_N})\le T_{d,M}(N)K+C_1N^{-\sigma}.(14)✦

For the mixed tuple estimate (15), suppose every pjp_j is at least PmassP_{\mathrm{mass}} and PrefP_{\mathrm{ref}}. Let Fv:Xp→ℝF_v:X_p\to\mathbb R be real functions and K≥0K\ge0. If Zp(Fv∘θB,z)≤KZ_p(F_v\circ\theta_{B,z})\le K for every v∈𝒱v\in\mathcal V and every z∈Xpz\in X_p, then |ℐPp,B((Fv)v)|≤ηBK.\left|\mathcal I_{P_{p,B}}((F_v)_v)\right|\le\eta_BK.(15)✦

For each fixed M≥2M\ge2 there are C2>0C_2>0 and N2N_2 such that the asymptotic mixed estimate (16) holds for N≥N2N\ge N_2. Suppose MM is coprime to every pjp_j, the primes are at least max⁡{Pmass,Pcond}\max\{P_{\mathrm{mass}},P_{\mathrm{cond}}\} and PrefP_{\mathrm{ref}}, qB≤Nτq_B\le N^\tau, and A⊆INA\subseteq I_N is square-difference-free. If K≥0K\ge0 bounds Yp(N′,QN′,A′)Y_p(N',Q_{N'},A') for every square-difference-free A′⊆IN′A'\subseteq I_{N'}, then arbitrary residues (sv)v∈𝒱(s_v)_{v\in\mathcal V} satisfy |ℐPp,B((ℒp,N,QNfsv)v)|≤ηB(Td,M(N)K+C2N−σ).\left|\mathcal I_{P_{p,B}} ((\mathcal L_{p,N,Q_N}f_{s_v})_v)\right| \le\eta_B\bigl(T_{d,M}(N)K+C_2N^{-\sigma}\bigr).(16)✦

For positive integers N,DN,D, the ceiling factor satisfies (17), with N′N' as in (12). 1≤D2N′N<1+D2N.1\le\frac{D^2N'}N<1+\frac{D^2}N.(17)✦ The child length satisfies (18) for integers N,D≥2N,D\ge2. N′<N.N'<N.(18)✦ For every fixed pair of nonnegative integers k,Mk,M, the progression transfer factor has the limit (19) along integer lengths. limN→∞Tk,M(N)=1.\lim_{N\to\infty}T_{k,M}(N)=1.(19)✦

In the paper, the result reads: for its fixed progression parameter M0M_0 and decay exponent σ0\sigma_0, there are Ctr<∞C_{\mathrm{tr}}<\infty and NtrN_{\mathrm{tr}} such that, for N≥NtrN\ge N_{\mathrm{tr}}, square-difference-free A⊆{1,…,N}A\subseteq\{1,\ldots,N\} and B⊆𝒫NB\subseteq\mathcal P_N with qB≤Nτq_B\le N^\tau, where 𝒫N\mathcal P_N is the paper’s ambient prime family, putting 𝒮=𝒫N\B\mathcal S=\mathcal P_N\setminus B, D=M0qBD=M_0q_B, N′=⌈N/D2⌉N'=\lceil N/D^2\rceil and λ=D2N′/N\lambda=D^2N'/N, every residue and fiber specification s,zBs,z_B gives g=(ℒ𝒫N,N,Qfs)(zB,⋅)g=(\mathcal L_{\mathcal P_N,N,Q}f_s)(z_B,\cdot) satisfying 1≤N′<N1\le N'<N and Z𝒮(g)≤λdmax⁡A′⊆{1,…,N′},A′ square-difference-freeY(N′,A′)+CtrN−σ0Z_{\mathcal S}(g)\le\lambda^d \max_{A'\subseteq\{1,\ldots,N'\},\ A'\text{ square-difference-free}} Y(N',A')+C_{\mathrm{tr}}N^{-\sigma_0}, with QQ the paper’s conductor cutoff and YY its set functional, the same upper bound holds for the absolute mixed form of the corresponding gvg_v under arbitrary choices sv,zB,vs_v,z_{B,v} in every slot, all constants are uniform in N,A,BN,A,B and the fiber specifications, and 1≤λ<1+D2/N1\le\lambda<1+D^2/N with λd=1+Od(D2/N)=1+Od,M0(N−1+2τ)\lambda^d=1+O_d(D^2/N)=1+O_{d,M_0}(N^{-1+2\tau}).

What Is Checked

The statements concern real conductor-truncated lifts of square-difference-free subsets of INI_N, product tuple laws and their exceptional-coordinate replacements, with the same ambient prime family at parent and child lengths. The tuple component, mixture and tensor product functionals preserve sums of real observables and multiplication by real scalars.