OctoInformBeta

OpenAI preprint, September 2026 · Section 3. Small Fourier signals

Lemma 3.1: Quantile jitter and passage families for finite angular packets, with smooth convergence of their lifts

Positive Fourier laws for smooth passages.

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

An angular Fourier packet fm,Af_{m,A} on the three-torus is a complex function obtained by multiplying an amplitude AA on the two slow coordinates by the Fourier character of an integer angular mode mm. A passage changes the angular coordinate used to expand such a packet. The small Fourier signal is prescribed by a smooth function of the normalized old mode, while a positive law supplies the jitter in the new temporal speed.

Write 𝕋=ℝ/ℤ\mathbb T=\mathbb R/\mathbb Z and e(t)=exp⁡(2πit)e(t)=\exp(2\pi i t). We identify the coordinate torus with the product 𝕋×𝕋2\mathbb T\times\mathbb T^2, writing a point as (θ,u,a)(\theta,u,a).✦ The covering projection π:ℝ3→𝕋3\pi:\mathbb R^3\to\mathbb T^3 reduces each coordinate modulo one.✦ Circle Haar measure dzdz, slow-coordinate measure dudadu\,da, and three-torus measure μ=dθduda\mu=d\theta\,du\,da are normalized. A torus map is smooth when it has smooth local lifts through π\pi.✦ For m∈ℤ\{0}m\in\mathbb Z\setminus\{0\} and an amplitude A:𝕋2→ℂA:\mathbb T^2\to\mathbb C, the angular packet fm,Af_{m,A} is given by (1); for a circle-valued speed φ\varphi on 𝕋\mathbb T, the twist iterates there use every k∈ℤk\in\mathbb Z. fm,A(θ,u,a)=A(u,a)e(mθ),Tφk(θ,u,a)=(θ+kφ(u),u,a).\begin{aligned} f_{m,A}(\theta,u,a)&=A(u,a)e(m\theta),\\ T_\varphi^k(\theta,u,a)&=(\theta+k\varphi(u),u,a). \end{aligned}(1)

For an index set ℐ\mathcal I, smooth passage packet data F=(mi,Ai,A0,i,o,l,u,R,V)i∈ℐF=(m_i,A_i,A_{0,i},o,l,u,R,V)_{i\in\mathcal I} consist of nonzero integer modes, smooth amplitudes on 𝕋2\mathbb T^2, and smooth real-coordinate lifts A0,i(a,x)A_{0,i}(a,x), with o≤l≤u≤o+1o\le l\le u\le o+1 and R>0R>0. The lift agrees with Ai(xmod⁡1,a)A_i(x\bmod1,a) for o<x≤o+1o<x\le o+1 and is supported in (l,u](l,u] for every spare angle aa; its profiles are twice continuously differentiable, and the lifts and their first xx derivatives are jointly continuous. The real speed VV is smooth and one-periodic, with V′>0V'>0 on [l−2R,u+2R][l-2R,u+2R]. An increasing good arc (δ,1−δ)(\delta,1-\delta) for the passage packet data FF is an arc for which o=0o=0, δ<l≤u<1−δ\delta<l\le u<1-\delta, and V′>0V'>0 on (δ,1−δ)(\delta,1-\delta).✦

Spare coordinate norm invariance of the packet data FF is the condition that both |A0,i(a,x)||A_{0,i}(a,x)| and |Ai(u,a)||A_i(u,a)| are independent of aa.✦ The lifted amplitude support CiC_i of packet ii is the closed support of A0,i(0,⋅)A_{0,i}(0,\cdot), given by Ci=supp⁡clA0,i(0,⋅)C_i=\operatorname{supp}_{\mathrm{cl}}A_{0,i}(0,\cdot).✦ A reflection of FF is a map τ:ℐ→ℐ\tau:\mathcal I\to\mathcal I satisfying (2). mτi=−mi,Aτi=Ai¯,A0,τi=A0,i¯.m_{\tau i}=-m_i,\qquad A_{\tau i}=\overline{A_i},\qquad A_{0,\tau i}=\overline{A_{0,i}}.(2)✦

The Positive Law

For α>0\alpha>0, the squared sinc density sαs_\alpha and the averaged sinc density qαq_\alpha are real functions on ℝ\mathbb R, with averaging radius aαa_\alpha, specified in (3); sinc is the sine of its argument divided by that argument, with value one at zero. The function DdD_d there is the real part of the Fourier transform of a complex frequency amplitude d:ℝ→ℂd:\mathbb R\to\mathbb C, with dˆ(R)=∫ℝd(v)e(−Rv)dv\widehat d(R)=\int_{\mathbb R}d(v)e(-Rv)\,dv.✦ aα=(2α)−1,sα(R)=αsinc⁡(παR)2,qα(R)=12aα∫−aαaαsα(R−t)dt,Dd(R)=Re⁡dˆ(R).\begin{aligned} a_\alpha&=(2\alpha)^{-1},& s_\alpha(R)&=\alpha\operatorname{sinc}(\pi\alpha R)^2,\\ q_\alpha(R)&=\frac1{2a_\alpha} \int_{-a_\alpha}^{a_\alpha}s_\alpha(R-t)\,dt,& D_d(R)&=\operatorname{Re}\widehat d(R). \end{aligned}(3)

The primorial sieve modulus associated with an integer L≥2L\ge2 is PL=∏p≤L,pprimepP_L=\prod_{p\le L,\ p\ \mathrm{prime}}p.✦ For a positive modulus PP, a positive signal strength γ\gamma, and h∈ℤ/Pℤh\in\mathbb Z/P\mathbb Z, the Ramanujan sum cP(h)c_P(h) and its real part rP(h)r_P(h) define the joint signal law qhq_h and the signal jitter density ρ\rho in (4). The symbol h̃\widetilde h denotes the representative of hh in {0,…,P−1}\{0,\ldots,P-1\}, and B,c>0B,c>0. cP(h)=∑v∈(ℤ/Pℤ)×e(−vh/P),rP(h)=Re⁡cP(h),qh(R)=qα(R)+γrP(h)Dd(R)P,ρ(y)=cB∑h∈ℤ/Pℤqh(B(cy−h̃/P)).\begin{aligned} c_P(h)&=\sum_{v\in(\mathbb Z/P\mathbb Z)^\times}e(-vh/P),& r_P(h)&=\operatorname{Re}c_P(h),\\ q_h(R)&=\frac{q_\alpha(R)+\gamma r_P(h)D_d(R)}{P},& \rho(y)&=cB\sum_{h\in\mathbb Z/P\mathbb Z} q_h\bigl(B(cy-\widetilde h/P)\bigr). \end{aligned}(4)

For a positive probability density ρ\rho on ℝ\mathbb R, its distribution function is Fρ(y)=∫−∞yρ(t)dtF_\rho(y)=\int_{-\infty}^y\rho(t)\,dt, and its quantile QρQ_\rho is the inverse of FρF_\rho on (0,1)(0,1). Quantile cutoffs of width δ\delta are smooth compactly supported real functions ξ,ψ\xi,\psi whose closed supports lie in (0,1)(0,1) and (δ,1−δ)(\delta,1-\delta), respectively, with ξ=1\xi=1 on [δ,1−δ][\delta,1-\delta], ψ=1\psi=1 on [2δ,1−2δ][2\delta,1-2\delta], and 0≤ψ≤10\le\psi\le1.✦ The periodic quantile jitter J0J_0, its circle descent JJ, and the neutral circle cutoff Ψ\Psi are defined by (5). 𝒫f(x)=∑k∈ℤf(x+k),J0=𝒫(ξQρ),J(xmod⁡1)=J0(x),Ψ(xmod⁡1)=𝒫ψ(x).\begin{aligned} \mathcal P f(x)&=\sum_{k\in\mathbb Z}f(x+k),& J_0&=\mathcal P(\xi Q_\rho),\\ J(x\bmod1)&=J_0(x),& \Psi(x\bmod1)&=\mathcal P\psi(x). \end{aligned}(5)

Passages and Packet Densities

For a circle map HH, the angular shear SH(θ,u,a)=(θ+H(u),u,a)S_H(\theta,u,a)=(\theta+H(u),u,a) and cyclic coordinate permutation C(θ,u,a)=(u,a,θ)C(\theta,u,a)=(u,a,\theta) give the passage chart 𝒞H=C∘S−H\mathcal C_H=C\circ S_{-H}. For circle maps H,σH,\sigma, the product torus passage TH,σT_{H,\sigma} is the bijection in (6); the coordinate torus passage is the same map under the identification with 𝕋×𝕋2\mathbb T\times\mathbb T^2. TH,σ(θ,u,a)=(θ+H(u+σ(a))−H(u),u+σ(a),a),TH,σ−1(θ,u,a)=(θ−H(u)+H(u−σ(a)),u−σ(a),a).\begin{aligned} T_{H,\sigma}(\theta,u,a) &=\bigl(\theta+H(u+\sigma(a))-H(u),u+\sigma(a),a\bigr),\\ T_{H,\sigma}^{-1}(\theta,u,a) &=\bigl(\theta-H(u)+H(u-\sigma(a)),u-\sigma(a),a\bigr). \end{aligned}(6)✦

For real functions v,J0v,J_0, time r∈ℝr\in\mathbb R, and parameters p=(p1,p2)∈ℝ2p=(p_1,p_2)\in\mathbb R^2, the two-parameter passage lift Φrv,J0(p;⋅):ℝ3→ℝ3\Phi_r^{v,J_0}(p;\cdot):\mathbb R^3\to\mathbb R^3 is defined by (7). s=rp1(1+p1J0(x2)),Φrv,J0(p;x)=(x0+r(1+p1J0(x2))∫01(v(x1+ts)+p2)dt,x1+s,x2).\begin{aligned} s&=rp_1\bigl(1+p_1J_0(x_2)\bigr),\\ \Phi_r^{v,J_0}(p;x) &=\left(x_0+r\bigl(1+p_1J_0(x_2)\bigr) \int_0^1\bigl(v(x_1+ts)+p_2\bigr)\,dt,\ x_1+s,\ x_2\right). \end{aligned}(7)✦

A smooth quantile passage family for F,α,γ,d,P,BF,\alpha,\gamma,d,P,B consists of a positive integer cc, a width 0<δ<1/80<\delta<1/8 with quantile cutoffs, and real numbers Δn\Delta_n with |Δn|≤n−1|\Delta_n|\le n^{-1} for n>0n>0 and Δn→0\Delta_n\to0. The jitter density ρ\rho of a smooth quantile passage family is the real function specified by (4), required to be smooth, integrable, strictly positive, and of integral one. Put v=c+Vv=c+V. The continuous phases HnH_n and speeds σn\sigma_n are specified by (8), with Tn=THn,σnT_n=T_{H_n,\sigma_n}. Hn(xmod⁡1)=n(∫0xv(t)dt+Δnx)mod⁡1,σn(a)=1+J(a)/nnmod⁡1(n>0),σ0=0.H_n(x\bmod1)= n\left(\int_0^xv(t)\,dt+\Delta_nx\right)\bmod1, \qquad \sigma_n(a)=\frac{1+J(a)/n}{n}\bmod1\quad(n>0),\qquad\sigma_0=0.(8) Such a family is required to have smooth Tn,Tn−1T_n,T_n^{-1} preserving μ\mu for every nn, and the lift identities in (9) for n>0n>0. Tn±1∘π=π∘Φ±1v,J0((n−1,Δn);⋅).T_n^{\pm1}\circ\pi =\pi\circ\Phi_{\pm1}^{v,J_0}\bigl((n^{-1},\Delta_n);\cdot\bigr).(9)

The oscillatory passage amplitude gm,H,Ag_{m,H,A} of a packet fm,Af_{m,A} and a circle map HH is the complex function on 𝕋2\mathbb T^2 given by gm,H,A(t,a)=A(t,a)e(mH(t))g_{m,H,A}(t,a)=A(t,a)e(mH(t)).✦ For the oscillatory passage amplitude gm,Hn,Ag_{m,H_n,A}, a circle cutoff Ψ\Psi, and a real frequency cutoff χ(s,a)\chi(s,a), the exact passage Fourier coefficients cn,jc_{n,j} are complex functions on the circle, the retained angular amplitudes An,j+A^+_{n,j} are complex functions on 𝕋2\mathbb T^2, and the circle weights wn,jw_{n,j} are real functions on the circle, all specified by (10). cn,j(a)=∫𝕋gm,Hn,A(t,a)e(−jt)dt,An,j+(u,a)=Ψ(u)χ(j/n,u)cn,j(u)e(ma),wn,j(u)=Ψ(u)2χ(j/n,u)2|cn,j(u)|2.\begin{aligned} c_{n,j}(a)&=\int_{\mathbb T}g_{m,H_n,A}(t,a)e(-jt)\,dt,\\ A^+_{n,j}(u,a)&=\Psi(u)\chi(j/n,u)c_{n,j}(u)e(ma),\\ w_{n,j}(u)&=\Psi(u)^2\chi(j/n,u)^2|c_{n,j}(u)|^2. \end{aligned}(10) For a finite retained set S⊂ℤS\subset\mathbb Z, the retained passage packet is Rn(θ,u,a)=∑j∈SAn,j+(u,a)e(jθ)R_n(\theta,u,a)=\sum_{j\in S}A^+_{n,j}(u,a)e(j\theta).✦ More generally, retained packet indices are pairs (i,j)(i,j) with j∈S(i)j\in S(i).✦ For passage packet data GG on these pairs, let pi,jGp^G_{i,j} denote the packet spectral density of the packet indexed by (i,j)(i,j), as defined by (12); the output packet sum and output spectral density at the old index ii are Fi+=∑j∈S(i)fG.mi,j,G.Ai,jF_i^+=\sum_{j\in S(i)}f_{G.m_{i,j},G.A_{i,j}} and pi+=∑j∈S(i)pi,jGp_i^+=\sum_{j\in S(i)}p^G_{i,j}, respectively.

For a real branch map J*J_* on I⊂ℝI\subset\mathbb R, an inverse branch gg, a Jacobian D*D_*, and a weight qq, the inverse branch density ℬq\mathcal Bq and scaled branch kernel KqK_q are defined by (11). The stationary frequency density of an amplitude A*A_* is ℬ(|A*|2)\mathcal B(|A_*|^2) with g=(J*|I)−1g=(J_*|_I)^{-1} and D*=J*′D_*=J_*'.✦ ℬI,J*,g,D*q(y)=𝟏J*(I)(y)q(g(y))|D*(g(y))|,Kq(s,y)=|s|−1ℬI,J*,g,D*q(y/s).\mathcal B_{I,J_*,g,D_*}q(y) =\mathbf1_{J_*(I)}(y)\frac{q(g(y))}{|D_*(g(y))|}, \qquad K_q(s,y)=|s|^{-1}\mathcal B_{I,J_*,g,D_*}q(y/s).(11) The stationary spectral weight of packet ii uses I=[l−R,u+R]I=[l-R,u+R], J*(x)=miV(x)J_*(x)=m_iV(x), and A*(x)=A0,i(0,x)A_*(x)=A_{0,i}(0,x).✦ Its packet spectral density pip_i is the folded density in (12), where z̃∈(0,1]\widetilde z\in(0,1] represents zz.✦ Folding uses the lattice periodization of the real density. pi(z)=Fold⁡(ℬ|A*|2)(z),Fold⁡q(z)=∑k∈ℤq(z̃+k).p_i(z)=\operatorname{Fold}(\mathcal B|A_*|^2)(z), \qquad \operatorname{Fold}q(z)=\sum_{k\in\mathbb Z}q(\widetilde z+k).(12)

For a circle weight w:𝕋→ℝw:\mathbb T\to\mathbb R and a real cut origin oo, its interval representative is the real function equal to w(xmod⁡1)w(x\bmod1) on (o,o+1](o,o+1] and zero elsewhere. For a circle weight w:𝕋→ℝw:\mathbb T\to\mathbb R, its interval representative q0q_0 at cut origin zero is the real function equal to w(xmod⁡1)w(x\bmod1) on (0,1](0,1] and zero elsewhere. For a jitter branch J0J_0 on II, take g=(J0|I)−1g=(J_0|_I)^{-1} and D*=J0′D_*=J_0' in (11). The unfolded mode density Ln,jL_{n,j} and exact retained packet density pnretp^{\mathrm{ret}}_n are (13), using the interval representative q0,n,jq_{0,n,j} of wn,jw_{n,j}. Ln,j(t)=Knq0,n,j(j/n,nt−j),pnret=Fold⁡(∑j∈SLn,j).L_{n,j}(t)=K_{nq_{0,n,j}}(j/n,nt-j), \qquad p^{\mathrm{ret}}_n=\operatorname{Fold}\left(\sum_{j\in S}L_{n,j}\right).(13) The three-mode circle signal QP,γ,B,d,mQ_{P,\gamma,B,d,m} is a real function on the spectral circle with modulus PP, strength γ\gamma, frequency scale BB, frequency amplitude dd, and old mode mm.

The Claim

Write q̌(t)=∫ℝq(R)e(tR)dR\check q(t)=\int_{\mathbb R}q(R)e(tR)\,dR for the positive Fourier transform of a real integrable function, and write ∥q∥L1(dz)=∫𝕋|q(z)|dz\|q\|_{L^1(dz)}=\int_{\mathbb T}|q(z)|\,dz. Derivatives DeD^e of passage lifts below are derivatives of order e≥0e\ge0 on ℝ3\mathbb R^3.

Theorem 1. Let b>0b>0, let D∈ℝD\in\mathbb R, and let d:ℝ→ℂd:\mathbb R\to\mathbb C be smooth and compactly supported, with d(−v)=d(v)¯d(-v)=\overline{d(v)}, d(0)=0d(0)=0, and supp⁡d⊂[−D,D]\operatorname{supp}d\subset[-D,D]. There are 0<α<b/20<\alpha<b/2, an integer L≥2L\ge2 with D/b<LD/b<L, and γ0>0\gamma_0>0 such that, for P=PLP=P_L and every fixed 0<γ≤γ00<\gamma\le\gamma_0, each qhq_h is smooth and integrable and satisfies (14); its last identity holds for |x|≥b|x|\ge b, every residue m∈ℤ/Pℤm\in\mathbb Z/P\mathbb Z, and every integer ℓ≥1\ell\ge1. ∫ℝqh(R)dR=P−1,qα(R)2P≤qh(R)≤3qα(R)2P,2γ|d(v)|≤12,∑h∈ℤ/Pℤq̌h(ℓx)e(ℓmh/P)={γd(x),ℓ=1 and m∈(ℤ/Pℤ)×,0,otherwise.\begin{gathered} \int_{\mathbb R}q_h(R)\,dR=P^{-1},\qquad \frac{q_\alpha(R)}{2P}\le q_h(R)\le\frac{3q_\alpha(R)}{2P},\qquad 2\gamma|d(v)|\le\tfrac12,\\ \sum_{h\in\mathbb Z/P\mathbb Z}\check q_h(\ell x)e(\ell mh/P) =\begin{cases} \gamma d(x),&\ell=1\text{ and }m\in(\mathbb Z/P\mathbb Z)^\times,\\ 0,&\text{otherwise}. \end{cases} \end{gathered}(14)✦

Assume also b<Db<D. Each of the following three assertions admits a choice of α,L,γ0\alpha,L,\gamma_0 with the same bounds, independent of the finite packet family and of BB, and with 2γ|d(v)|≤1/22\gamma|d(v)|\le1/2 for every vv and 0<γ≤γ00<\gamma\le\gamma_0. Let FF be finite smooth passage packet data with distinct modes and spare coordinate norm invariance. Fix B>0B>0 with |mi/B|≥b|m_i/B|\ge b for every ii, and fix 0<γ≤γ00<\gamma\le\gamma_0.

For positive ε,τ\varepsilon,\tau, every sufficiently large integer cc admits a smooth quantile passage family and smooth compactly supported frequency cutoffs χi\chi_i satisfying (15). 0≤χi≤1,χi(0)=0,supp⁡clχi∩supp⁡clχj=⌀(i≠j),s∈supp⁡clχi⇒there is x∈Ci with |s−mi(c+V(x))|<τ.\begin{gathered} 0\le\chi_i\le1,\qquad \chi_i(0)=0,\qquad \operatorname{supp}_{\mathrm{cl}}\chi_i\cap \operatorname{supp}_{\mathrm{cl}}\chi_j=\varnothing\quad(i\ne j),\\ s\in\operatorname{supp}_{\mathrm{cl}}\chi_i \quad\Longrightarrow\quad \text{there is }x\in C_i\text{ with }|s-m_i(c+V(x))|<\tau. \end{gathered}(15)✦

For positive ε,η\varepsilon,\eta, there is a smooth quantile passage family and finite sets Sn(i)⊂ℤS_n(i)\subset\mathbb Z, with n>0n>0 for every sufficiently large nn, for which (16) holds for every real rr, compact K⊂ℝ3K\subset\mathbb R^3, and e≥0e\ge0, where v=c+Vv=c+V. supx∈K∥DeΦrv,J0((n−1,Δn);x)−De[x↦(x0+rv(x1),x1,x2)](x)∥→0.\sup_{x\in K}\left\| D^e\Phi_r^{v,J_0}\bigl((n^{-1},\Delta_n);x\bigr) -D^e\bigl[x\mapsto(x_0+rv(x_1),x_1,x_2)\bigr](x) \right\|\longrightarrow0.(16)✦

Suppose in addition that FF has a reflection τ\tau. For positive ε,η,κ\varepsilon,\eta,\kappa, there is a smooth quantile passage family with finite retained sets Sn(i)S_n(i) and the properties in (17). $$0<\delta<\min\{1/8,\kappa\},\qquad n>0\ \text{for all sufficiently large }n,\qquad \text{the convergence in \eqref{eq:claim3} holds for every }r,K,e.$$(17)✦

In the paper, the result reads: for fixed b,D0,db,D_0,d there are an integer P≥2P\ge2 and γ0>0\gamma_0>0 such that every fixed 0<γ≤γ00<\gamma\le\gamma_0 works independently of BB; if 𝒮\mathcal S is an intermediate map with old speed φ\varphi and Fr=∑m∈MrAr,m(u,a)e(mθ)F_r=\sum_{m\in M_r}A_{r,m}(u,a)e(m\theta) are finite smooth packets on its good arc, with mutually disjoint finite nonzero mode sets MrM_r, spare-independent coefficient moduli, and b≤|m|/B≤D0b\le|m|/B\le D_0, then for any ε,δx,δz>0\varepsilon,\delta_x,\delta_z>0 and positive physical L2L^2 error bounds one passage gives output packet sums within those bounds and with ∑r,m∥pr,m+−pr,m(z)[1+2γ𝟏gcd⁡(m,P)=1Re⁡(d(m/B)e(−Nz))]∥L1(dz)<ε\sum_{r,m}\|p^+_{r,m}-p_{r,m}(z)[1+2\gamma\mathbf1_{\gcd(m,P)=1} \operatorname{Re}(d(m/B)e(-Nz))]\|_{L^1(dz)}<\varepsilon, new scale B+=NcBB^+=NcB and speed σN\sigma_N, every retained support point at mode jj having an old support point at mode mm, with slow coordinates u′u' and uu, satisfying |j/B+−m/B|<δx|j/B^+-m/B|<\delta_x and dist⁡𝕋(jσN(u′),mφ(u))<δz\operatorname{dist}_{\mathbb T}(j\sigma_N(u'),m\varphi(u))<\delta_z, while output moduli remain spare-independent, distinct old modes have disjoint output mode supports, conjugation symmetry is preserved, the map and its inverse may lie in any prescribed physical smooth neighborhood, and the new good-arc complement may be arbitrarily small, with choices ordered as cc, bounded jitter and cutoffs, then NN, every sufficiently large final integer being allowed.

What Is Checked

The positive-law assertion is stated over smooth compactly supported frequency amplitudes with conjugation symmetry and zero value at the origin; the passage-family assertions are stated over finite smooth packet data with distinct nonzero modes and spare coordinate norm invariance, with packet reflection in the final assertion.