OctoInformBeta

OpenAI preprint, September 2026 · Section 6. A signed kernel for actual square differences

Proposition 6.1: Signed rational models and separated interval lifts with truncation parameter HH and an explicit H−1/3H^{-1/3} error

Ordered square kernels have controlled pairings.

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 bilinear form couples two ordered sets of integers through rational frequencies, selected by the primes dividing their denominators. The square restriction concerns all differences between the two supports. Write [N]={1,…,N}[N]=\{1,\ldots,N\}; the natural range {0,…,N−1}\{0,\ldots,N-1\} used for the interval lifts is identified with [N][N] by n=t+1n=t+1. A finite set AA of natural numbers is square-difference-free when it contains no pair a,ba,b with b=a+m2b=a+m^2 for a positive integer mm.✦

The Signed Rational Model

Let II be finite, let (pi)i∈I(p_i)_{i\in I} be a family of primes, and fix R⊆IR\subseteq I and residues ci∈ℤ/piℤc_i\in\mathbb Z/p_i\mathbb Z. The local quadratic modulus λ(p)\lambda(p), root square modulus DD, and conductor d(U)d(U) of a set U⊆IU\subseteq I are the integers in (1). λ(p)={8,p=2,p,p≠2,D=∏i∈Rλ(pi),d(U)=∏i∈Upi.\lambda(p)=\begin{cases}8,&p=2,\\p,&p\ne2,\end{cases} \qquad D=\prod_{i\in R}\lambda(p_i),\qquad d(U)=\prod_{i\in U}p_i.(1) The root residue weight ρ\rho is a function on the natural numbers, and the centered prime sieve weight ψp\psi_p is a real function on ℤ/pℤ\mathbb Z/p\mathbb Z, defined by (2). Here and below 𝟏E\mathbf1_E denotes the indicator of a condition or set EE. ρ(n)=∏i∈Rpi𝟏nmod⁡pi=ci,ψp(x)=p𝟏x=0−1p−1.\rho(n)=\prod_{i\in R}p_i\mathbf1_{n\bmod p_i=c_i}, \qquad \psi_p(x)=\frac{p\mathbf1_{x=0}-1}{p-1}.(2) For T⊆IT\subseteq I, the restricted sieve weight ψT\psi_T, signed divisor weight wHw_H, and complete sieved root weight w∞w_\infty are complex functions on the natural numbers defined by (3). ψT(n)=∏i∈T(𝟏i∉Rψpi(nmod⁡pi)),wH(n)=ρ(n)∑T⊆Id(T)≤H(−1)|T|ψT(n),w∞(n)=ρ(n)∏i∉R(1−ψpi(nmod⁡pi)).\begin{aligned} \psi_T(n)&=\prod_{i\in T}\bigl(\mathbf1_{i\notin R}\psi_{p_i}(n\bmod p_i)\bigr),\\ w_H(n)&=\rho(n)\sum_{\substack{T\subseteq I\\d(T)\le H}}(-1)^{|T|}\psi_T(n),\\ w_\infty(n)&=\rho(n)\prod_{i\notin R}(1-\psi_{p_i}(n\bmod p_i)). \end{aligned}(3)

The exponential phase is e(t)=exp⁡(2πit)e(t)=\exp(2\pi\mathrm i t).✦ For a complex sequence aa, its periodic mean of length TT is av⁡T(a)\operatorname{av}_T(a) in (4); for a rational frequency ξ\xi, the complete rational coefficient γ(ξ)\gamma(\xi) is the complex number given by the specified mean, where q(ξ)q(\xi) denotes the reduced positive denominator of ξ\xi. Reciprocals at zero are assigned value zero. av⁡T(a)=1T∑n=0T−1a(n),γ(ξ)=av⁡q(ξ)∏i∈Ipi(n↦w∞(n+1)e(ξ(n+1)2)).\begin{aligned} \operatorname{av}_T(a)&=\frac1T\sum_{n=0}^{T-1}a(n),\\ \gamma(\xi)&=\operatorname{av}_{q(\xi)\prod_{i\in I}p_i} \bigl(n\mapsto w_\infty(n+1)e(\xi(n+1)^2)\bigr). \end{aligned}(4) For a denominator bound LL, (5) defines the denominator prime indices I(q)I(q) as a subset of II, the prime power modulus P(q,p)P(q,p) as an integer, the truncated rational coefficient kH(ξ)k_H(\xi) as a complex number, and the bounded-denominator frequency set ℬL\mathcal B_L as a set of rational frequencies, where qq is a positive denominator, pp is prime, and vp(q)v_p(q) is the exponent of pp in qq. I(q)={i∈I:pi∣q},P(q,p)=pmax⁡(1,vp(q)),kH(ξ)=γ(ξ)𝟏d(I(q(ξ))\R)≤H,ℬL={ξ∈ℚ:0≤ξ<1,q(ξ)≤L}.\begin{gathered} I(q)=\{i\in I:p_i\mid q\},\qquad P(q,p)=p^{\max(1,v_p(q))},\\ k_H(\xi)=\gamma(\xi)\mathbf1_{d(I(q(\xi))\setminus R)\le H},\qquad \mathcal B_L=\{\xi\in\mathbb Q:0\le\xi<1,\ q(\xi)\le L\}. \end{gathered}(5)

For N≥1N\ge1 and a natural root length MM, the normalized linear exponential sum 𝒟N\mathcal D_N, weighted square transform 𝒮HM,N\mathcal S_H^{M,N}, and rational model transform ℳHL,N\mathcal M_H^{L,N} are complex functions of a real variable, and the quadratic exponential weight sNs_N is a complex function of a real variable and a natural number, all defined by (6). The sums over mm and nn here are counting sums. 𝒟N(y)=1N∑n=1Ne(ny),sN(α,m)=2(m+1)Ne(α(m+1)2),𝒮HM,N(α)=∑m=0M−1sN(α,m)wH(m+1),ℳHL,N(α)=∑ξ∈ℬLkH(ξ)𝒟N(α−ξ).\begin{aligned} \mathcal D_N(y)&=\frac1N\sum_{n=1}^Ne(ny),& s_N(\alpha,m)&=\frac{2(m+1)}N e(\alpha(m+1)^2),\\ \mathcal S_H^{M,N}(\alpha)&=\sum_{m=0}^{M-1}s_N(\alpha,m)w_H(m+1),& \mathcal M_H^{L,N}(\alpha)&=\sum_{\xi\in\mathcal B_L} k_H(\xi)\mathcal D_N(\alpha-\xi). \end{aligned}(6) For F:ℤ/qℤ→ℂF:\mathbb Z/q\mathbb Z\to\mathbb C, the normalized interval Fourier transform FˆN\widehat F_N is the complex function of a real frequency defined by (7), with x̃\widetilde x the representative of xx in {0,…,q−1}\{0,\ldots,q-1\}. The absolute constant CC is fixed there as well. FˆN(α)=1N∑x∈ℤ/qℤF(x)e(−x̃α),C=264+3(4+8π)+156+16.\widehat F_N(\alpha)=\frac1N\sum_{x\in\mathbb Z/q\mathbb Z} F(x)e(-\widetilde x\alpha),\qquad C=2^{64}+3(4+8\pi)+156+16.(7)

Interval Residue Lifts

For a finite coordinate set JJ and prime family (pj)j∈J(p_j)_{j\in J}, the product residue space XpX_p is ∏j∈Jℤ/pjℤ\prod_{j\in J}\mathbb Z/p_j\mathbb Z.✦ Let a,L,Qa,L,Q be natural numbers and let ff be a real function on the natural numbers. For a prime pp and a coordinate set U⊆JU\subseteq J, the centered residue delta gpg_p is a real function on ℤ/pℤ\mathbb Z/p\mathbb Z and the centered residue product gUg_U is a real function on XpX_p, given in (8). A conductor-bounded support U⊆JU\subseteq J is a whole coordinate set belonging to 𝒰Q\mathcal U_Q in (8). gp(x)=p𝟏x=0−1,gU(x)=∏j∈Ugpj(xj),𝒰Q={U⊆J:∏j∈Upj≤Q},[n]p=(nmod⁡pj)j∈J,Pa,L,Uf(x)=1L∑t=0L−1f(a+t)gU(x−[a+t]p),ℒa,L,Qf(x)=∑U∈𝒰QPa,L,Uf(x).\begin{gathered} g_p(x)=p\mathbf1_{x=0}-1,\qquad g_U(x)=\prod_{j\in U}g_{p_j}(x_j),\qquad \mathcal U_Q=\left\{U\subseteq J:\prod_{j\in U}p_j\le Q\right\},\\ [n]_p=(n\bmod p_j)_{j\in J},\qquad P_{a,L,U}f(x)=\frac1L\sum_{t=0}^{L-1}f(a+t)g_U(x-[a+t]_p),\\ \mathcal L_{a,L,Q}f(x)=\sum_{U\in\mathcal U_Q}P_{a,L,U}f(x). \end{gathered}(8) For a real function ff on the natural numbers, the interval lift component Pa,L,UfP_{a,L,U}f and interval residue lift ℒa,L,Qf\mathcal L_{a,L,Q}f are the real functions on XpX_p defined by (8). At L=0L=0 both are zero. For B⊆JB\subseteq J and z∈Xpz\in X_p, coordinate replacement ΦB,z:Xp→Xp\Phi_{B,z}:X_p\to X_p is the map whose jjth coordinate equals zjz_j for j∈Bj\in B and equals xjx_j otherwise.✦

Let (pis)i∈S(p_i^{\mathrm{s}})_{i\in S} be a second finite prime family. Its combined small modulus D0D_0 and residue projections πi\pi_i are given in (9). For a prime pp, a strict local square x∈ℤ/λ(p)ℤx\in\mathbb Z/\lambda(p)\mathbb Z is a residue equal to 11 modulo 88 at p=2p=2, and a nonzero square in 𝔽p\mathbb F_p at an odd prime.✦ A strict pair s,t∈ℤ/D0ℤs,t\in\mathbb Z/D_0\mathbb Z is a pair for which every πi(t−s)\pi_i(t-s) is a strict local square.✦ For a residue ss modulo D0D_0, the normalized residue slice fsf_s of a finite set A⊆ℕA\subseteq\mathbb N is the real function on the natural numbers defined by (9). D0=∏i∈Sλ(pis),πi:ℤ/D0ℤ→ℤ/λ(pis)ℤ,πi(x)=xmod⁡λ(pis),fs(n)=D0𝟏A(n)𝟏nmod⁡D0=s.\begin{gathered} D_0=\prod_{i\in S}\lambda(p_i^{\mathrm{s}}),\qquad \pi_i:\mathbb Z/D_0\mathbb Z\longrightarrow \mathbb Z/\lambda(p_i^{\mathrm{s}})\mathbb Z,\quad \pi_i(x)=x\bmod\lambda(p_i^{\mathrm{s}}),\\ f_s(n)=D_0\mathbf1_A(n)\mathbf1_{n\bmod D_0=s}. \end{gathered}(9) The combined prime family pextp^{\mathrm{ext}} is the family on the disjoint union S⊔JS\sqcup J equal to psp^{\mathrm{s}} on SS and to pp on JJ.✦

The Good Tuple Law

Fix height h=24h=24, truncation level T=80T=80, block count b*=2T+1=161b_*=2T+1=161, list base r=h!/2r=h!/2, and interaction parameter η=1/(10h2)\eta=1/(10h^2). A permutation word on a block set EE assigns each block a permutation of {0,…,h−1}\{0,\ldots,h-1\}; the tuple vertex set VV consists of these words for E={0,…,b*−1}E=\{0,\ldots,b_*-1\}. An ordered position pair is a pair (u,v)(u,v) with 0≤u<v<h0\le u<v<h.✦ For a vertex x∈Vx\in V, a block b∈Eb\in E, and an ordered position pair (u,v)(u,v), the position transposition τb,u,vx\tau_{b,u,v}x is the vertex obtained by replacing xbx_b by xb(uv)x_b(u\ v) and leaving the other blocks unchanged.✦ More generally, for a finite block set EE, a group GG, and κ∈G\kappa\in G, threshold shift adjacency of u,v:E→Gu,v:E\to G at threshold tt means |{b∈E:vb=ubκ}|≥t|\{b\in E:v_b=u_b\kappa\}|\ge t.✦

For the tuple block set E={0,…,b*−1}E=\{0,\ldots,b_*-1\}, write Sym⁡(h)\operatorname{Sym}(h) for the permutation group of the hh positions. For S0⊆ES_0\subseteq E, an inactive-block address aa is a permutation word in 𝒜S0=Sym⁡(h)E\S0\mathcal A_{S_0}=\operatorname{Sym}(h)^{E\setminus S_0}, an addressed list index ii is an integer in ℐS0={0,…,r|S0|−1}\mathcal I_{S_0}=\{0,\ldots,r^{|S_0|}-1\}, and an addressed list entry (a,i)(a,i) is a pair in 𝒜S0×ℐS0\mathcal A_{S_0}\times\mathcal I_{S_0}. The inactive-block projection pr⁡S0(x)\operatorname{pr}_{S_0}(x) is the restriction of a vertex xx to E\S0E\setminus S_0.✦ For a prime pp, list values YY in 𝔽p\mathbb F_p on the addressed entries, and a labeling ℓ:V→ℐS0\ell:V\to\mathcal I_{S_0}, the sampled tuple configuration ZS0(Y,ℓ)Z_{S_0}(Y,\ell) is the function from VV to 𝔽p\mathbb F_p defined by (10). Let ℰS0\mathcal E_{S_0} denote the prescribed finite set of ordered list constraint edges. The nonzero square indicator on a field FF is σF(x)=𝟏x≠0,x a square\sigma_F(x)=\mathbf1_{x\ne0,\ x\text{ a square}}, written σp\sigma_p for F=𝔽pF=\mathbb F_p; the square-constraint indicator VS0(Y)V_{S_0}(Y) is defined by the other formula in (10). ZS0(Y,ℓ)(v)=Y(pr⁡S0(v),ℓ(v)),VS0(Y)=∏(u,v)∈ℰS0σp(Y(v)−Y(u)).Z_{S_0}(Y,\ell)(v)=Y(\operatorname{pr}_{S_0}(v),\ell(v)),\qquad V_{S_0}(Y)=\prod_{(u,v)\in\mathcal E_{S_0}}\sigma_p(Y(v)-Y(u)).(10)

For a real matrix AA on a finite set KK and labelings ℓ,k\ell,k on KK with values in a common label set, the label coincidence pairing BA(ℓ,k)B_A(\ell,k), index energy EA(ℓ)E_A(\ell), and index Boltzmann weight WA(ℓ)W_A(\ell) are those in (11). The transposition interaction AS0A_{S_0} is the real matrix on VV, and the tuple Boltzmann weight WS0W_{S_0} is the real function on labelings of VV in ℐS0\mathcal I_{S_0}, defined by (11). BA(ℓ,k)=∑u,v∈KA(u,v)𝟏ℓ(u)=k(v),EA(ℓ)=12BA(ℓ,ℓ),WA(ℓ)=exp⁡(EA(ℓ)),AS0(x,y)=∑b∈S0∑0≤u<v<hηv−u𝟏y=τb,u,vx,WS0(ℓ)=WAS0(ℓ).\begin{aligned} B_A(\ell,k)&=\sum_{u,v\in K}A(u,v)\mathbf1_{\ell(u)=k(v)},& E_A(\ell)&=\tfrac12 B_A(\ell,\ell),\quad W_A(\ell)=\exp(E_A(\ell)),\\ A_{S_0}(x,y)&=\sum_{b\in S_0}\sum_{0\le u<v<h} \eta^{v-u}\mathbf1_{y=\tau_{b,u,v}x},& W_{S_0}(\ell)&=W_{A_{S_0}}(\ell). \end{aligned}(11) For prime pp, the tuple component functional Cp,S0C_{p,S_0} is the real linear functional on real observables GG of 𝔽pV\mathbb F_p^V whose value, the component average, is given by (12). Both expectations there are uniform probability averages over the indicated finite spaces. For a finite family 𝒟\mathscr D of block subsets, the restricted mixture Mp,𝒟M_{p,\mathscr D} has block subset mixture weights p−|S0|/16p^{-|S_0|/16}, and its normalization μp,𝒟\mu_{p,\mathscr D} is the normalized restricted mixture in the same display. The truncated block family ℱn\mathscr F_n and good tuple law μp∘\mu_p^\circ are defined there as well. Cp,S0(G)=𝔼Y:𝒜S0×ℐS0→𝔽p𝔼ℓ:V→ℐS0VS0(Y)WS0(ℓ)G(ZS0(Y,ℓ)),Mp,𝒟(G)=∑S0∈𝒟p−|S0|/16Cp,S0(G),μp,𝒟(G)=Mp,𝒟(G)Mp,𝒟(1),ℱn={S0⊆E:|S0|≤n},μp∘=μp,ℱT.\begin{aligned} C_{p,S_0}(G)&=\mathbb E_{Y:\mathcal A_{S_0}\times\mathcal I_{S_0}\to\mathbb F_p} \mathbb E_{\ell:V\to\mathcal I_{S_0}} V_{S_0}(Y)W_{S_0}(\ell)G(Z_{S_0}(Y,\ell)),\\ M_{p,\mathscr D}(G)&=\sum_{S_0\in\mathscr D}p^{-|S_0|/16}C_{p,S_0}(G),\qquad \mu_{p,\mathscr D}(G)=\frac{M_{p,\mathscr D}(G)}{M_{p,\mathscr D}(1)},\\ \mathscr F_n&=\{S_0\subseteq E:|S_0|\le n\},\qquad \mu_p^\circ=\mu_{p,\mathscr F_T}. \end{aligned}(12) The leading graph edge count e*=|ℰ⌀|e_*=|\mathcal E_\varnothing| gives the mass positivity threshold P*=max⁡{3,(2e*+1e*)2}P_* =\max\{3,(2^{e_*+1}e_*)^2\}. The cyclic shift κ\kappa rotates the hh positions by one, and for a natural number kk, the cyclic tuple vertex vk∈Vv_k\in V has value κk\kappa^k in every block.

A real linear functional LL on functions on a finite space Ω\Omega has point mass dL(z)=L(𝟏{z})d_L(z)=L(\mathbf1_{\{z\}}) at z∈Ωz\in\Omega.✦ For finite coordinate spaces XjX_j and functionals LjL_j on observables of XjVX_j^V, their tensor product is the functional in (13), a counting sum over coordinate configurations. The law used for the interval pair is ν=⨂j∈Jμpj∘\nu=\bigotimes_{j\in J}\mu_{p_j}^\circ. (⨂j∈JLj)(G)=∑z∈∏j∈JXjV(∏j∈JdLj(zj))G(v↦(zj(v))j∈J).\left(\bigotimes_{j\in J}L_j\right)(G) =\sum_{z\in\prod_{j\in J}X_j^V} \left(\prod_{j\in J}d_{L_j}(z_j)\right) G\bigl(v\mapsto(z_j(v))_{j\in J}\bigr).(13)✦

The Claim

Theorem 1. Let (pi)i∈I(p_i)_{i\in I} be an injective finite prime family. Suppose N≥1N\ge1, H≥1H\ge1, and the root data satisfy the scale and prime-cover conditions (14): D≤N1/1000,H≤N1/1000,8≤N1/16,q(ξ)∣∏i∈IP(q(ξ),pi)for every ξ∈ℚ with q(ξ)≤N1/8,pi=2⇒i∈R,ci=1 if pi=2,ci≠0 otherwise,i∈R.\begin{gathered} D\le N^{1/1000},\qquad H\le N^{1/1000},\qquad 8\le N^{1/16},\\ q(\xi)\mid\prod_{i\in I}P(q(\xi),p_i) \quad\text{for every }\xi\in\mathbb Q\text{ with }q(\xi)\le N^{1/8},\\ p_i=2\ \Longrightarrow\ i\in R,\qquad c_i=1\text{ if }p_i=2,\quad c_i\ne0\text{ otherwise},\quad i\in R. \end{gathered}(14) Put LH=⌈DH⌉L_H=\lceil DH\rceil and MN=⌊N⌋M_N=\lfloor\sqrt N\rfloor. For every real α\alpha, |𝒮HMN,N(α)−ℳHLH,N(α)|≤CH−1/3.\left|\mathcal S_H^{M_N,N}(\alpha)-\mathcal M_H^{L_H,N}(\alpha)\right| \le C H^{-1/3}.(15)✦

Let q>Nq>N and F,G:ℤ/qℤ→ℂF,G:\mathbb Z/q\mathbb Z\to\mathbb C. The ordered square-shift conditions with root length MM are (16): ∑x|F(x)|2≤N,∑y|G(y)|2≤N,F(x)G(y)≠0⇒x̃<ỹ≤N,F(x)G(y)=0whenever y−x=(m+1)2 in ℤ/qℤ,0≤m<M.\begin{gathered} \sum_x|F(x)|^2\le N,\qquad \sum_y|G(y)|^2\le N,\\ F(x)G(y)\ne0\ \Longrightarrow\ \widetilde x<\widetilde y\le N,\\ F(x)G(y)=0\quad\text{whenever }y-x=(m+1)^2\text{ in }\mathbb Z/q\mathbb Z, \quad 0\le m<M. \end{gathered}(16) Under these conditions with M=MNM=M_N, and the preceding kernel hypotheses, |∑ξ∈ℬLHkH(ξ)FˆN(−ξ)GˆN(ξ)|≤CH−1/3.\left|\sum_{\xi\in\mathcal B_{L_H}} k_H(\xi)\widehat F_N(-\xi)\widehat G_N(\xi)\right| \le C H^{-1/3}.(17)✦

Independently, let N≥1N\ge1, q>Nq>N, M≥0M\ge0, and let KK be finite. Fix frequencies θb∈ℝ\theta_b\in\mathbb R, coefficients ab∈ℂa_b\in\mathbb C, a weight w:ℕ→ℂw:\mathbb N\to\mathbb C, and δ≥0\delta\ge0. Let F,G:ℤ/qℤ→ℂF,G:\mathbb Z/q\mathbb Z\to\mathbb C satisfy (16) with this MM, and suppose the uniform approximation (18) holds: |∑m=0M−1sN(α,m)w(m+1)−∑b∈Kab𝒟N(α−θb)|≤δ(α∈ℝ).\left|\sum_{m=0}^{M-1}s_N(\alpha,m)w(m+1) -\sum_{b\in K}a_b\mathcal D_N(\alpha-\theta_b)\right|\le\delta \qquad(\alpha\in\mathbb R).(18) Then |∑b∈KabFˆN(−θb)GˆN(θb)|≤δ.\left|\sum_{b\in K}a_b\widehat F_N(-\theta_b)\widehat G_N(\theta_b)\right| \le\delta.(19)✦

For the interval statement, suppose the combined prime family pextp^{\mathrm{ext}} is injective, pj≥64p_j\ge64 and pj≥P*p_j\ge P_* for every j∈Jj\in J, and every prime at most NN occurs in pextp^{\mathrm{ext}}. Let a,b,L,Q,H,Na,b,L,Q,H,N be natural numbers with N,H≥1N,H\ge1, and fix B⊆JB\subseteq J. Write dB=∏j∈Bpjd_B=\prod_{j\in B}p_j and assume the interval scale conditions (20): a+L≤b,HdB≤Q,D0dB≤N1/1000,H≤N1/1000,8≤N1/16.a+L\le b,\qquad H d_B\le Q,\qquad D_0d_B\le N^{1/1000},\qquad H\le N^{1/1000},\qquad 8\le N^{1/16}.(20) Let A⊆{0,…,N−1}A\subseteq\{0,\ldots,N-1\} be square-difference-free, let s,ts,t be a strict pair modulo D0D_0, and choose z,w∈Xpz,w\in X_p such that wj−zjw_j-z_j is a nonzero square for each j∈Bj\in B. Put U=ℒa,L,QfsU=\mathcal L_{a,L,Q}f_s and W=ℒb,L,QftW=\mathcal L_{b,L,Q}f_t. Suppose the real second-moment bounds 𝔼x∈XpU(x)2≤ℛ2\mathbb E_{x\in X_p}U(x)^2\le\mathcal R^2 and 𝔼x∈XpW(x)2≤ℛ2\mathbb E_{x\in X_p}W(x)^2\le\mathcal R^2 hold for some ℛ∈ℝ\mathcal R\in\mathbb R, where these expectations use uniform probability on XpX_p. For every natural kk, |ν(X↦U(ΦB,z(X(vk)))W(ΦB,w(X(vk+1))))|≤dBH−1/3((NL)2D0C+ℛ2).\left|\nu\!\left(X\mapsto U(\Phi_{B,z}(X(v_k)))W(\Phi_{B,w}(X(v_{k+1})))\right)\right| \le d_B H^{-1/3}\left(\left(\frac NL\right)^2D_0C+\mathcal R^2\right).(21)✦

In the paper, the result reads: for an exponent β\beta, N≥1N\ge1, H=Nβ/2H=N^{\beta/2} and D=8d0≤NβD=8d_0\le N^\beta with d0d_0 odd and squarefree, complex functions F,J:[N]→ℂF,J:[N]\to\mathbb C bounded in modulus by one, supported respectively in classes a,ba,b modulo DD with b−ab-a a strict square class, and satisfying x<yx<y and y−xy-x not a positive integer square for every x∈supp⁡Fx\in\operatorname{supp}F and y∈supp⁡Jy\in\operatorname{supp}J, obey (22) with the paper’s coefficient 𝔤(ξ)\mathfrak g(\xi), reduced denominator q(ξ)q(\xi), and Fourier transforms Fˆ(ξ)=N−1∑n=1NF(n)e(−nξ)\widehat F(\xi)=N^{-1}\sum_{n=1}^NF(n)e(-n\xi) and likewise Jˆ\widehat J, D|∑ξ∈ℚ∩[0,1)q(ξ)≤H,q(ξ) squarefreegcd⁡(q(ξ),D)=1𝔤(ξ)Fˆ(−ξ)Jˆ(ξ)|≪H−1/3,D\left|\sum_{\substack{\xi\in\mathbb Q\cap[0,1)\\q(\xi)\le H,\ q(\xi)\text{ squarefree}\\ \gcd(q(\xi),D)=1}} \mathfrak g(\xi)\widehat F(-\xi)\widehat J(\xi)\right| \ll H^{-1/3},(22) with an absolute implied constant independent of all the data.

What Is Checked

The estimates (15) and (17) are stated over a fixed finite prime family with the scale and prime-cover hypotheses in (14); (19) is stated over an arbitrary finite frequency family and the ordered square-shift conditions (16). The interval estimate (21) is stated over real interval lifts with complete conductor-bounded supports, the combined prime family, and the tensor product of normalized good tuple laws.

Each local quadratic modulus in the small-prime family divides its combined modulus, as in (23).✦ λ(pis)∣D0(i∈S).\lambda(p_i^{\mathrm{s}})\mid D_0\qquad(i\in S).(23)✦