OctoInformBeta

OpenAI preprint, September 2026 Β· Section 5. Preparation, whitening, and the smooth limit

Section 5: whitening, including the zero set: Exact distance identities and finite Laurent approximation for reflected packet families

Whitening packet spectra within a good arc.

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

Whitening seeks a packet vector with flat spectral measure inside a prescribed good arc, while retaining finite Laurent approximation of an earlier packet prediction. Write 𝕋=ℝ/β„€\mathbb T=\mathbb R/\mathbb Z, with normalized Haar measures mm, m2m_2 and m3m_3 on 𝕋\mathbb T, 𝕋2\mathbb T^2 and 𝕋3\mathbb T^3. The coordinates of the product torus are (ΞΈ,u,a)(\theta,u,a), with (u,a)(u,a) the slow coordinates. For a smooth real 11-periodic speed vv, let Ο†(u)=v(uΜƒ)mod⁡1\varphi(u)=v(\widetilde u)\bmod 1, where uΜƒ\widetilde u is any real representative of uu. For a function HH on 𝕋3\mathbb T^3, the twist and its bilateral Koopman orbit use the conventions in (1). e(t)=exp⁡(2Ο€it),Tk(ΞΈ,u,a)=(ΞΈ+kΟ†(u),u,a),UTkH=H∘Tk(kβˆˆβ„€).e(t)=\exp(2\pi i t),\qquad T^k(\theta,u,a)=(\theta+k\varphi(u),u,a),\qquad U_T^kH=H\circ T^k\quad(k\in\mathbb Z).(1)

A smooth passage packet family β„±\mathcal F indexed by II consists of nonzero integer frequencies mim_i, smooth amplitudes Ai:𝕋2β†’β„‚A_i:\mathbb T^2\to\mathbb C and smooth lifts A0,i:𝕋×ℝ→ℂA_{0,i}:\mathbb T\times\mathbb R\to\mathbb C, with a cut origin oo, support endpoints l,ul,u and a buffer radius R>0R>0. Smoothness of the lifts and amplitudes is in both real covering coordinates. The support and matching conditions, together with the condition on the speed, are (2). o≀l≀u≀o+1,Ai(xmod⁡1,a)=A0,i(a,x)(o<x≀o+1),supp⁡A0,i(a,β‹…)βŠ‚(l,u],vβ€²(x)>0(lβˆ’2R≀x≀u+2R).\begin{aligned} &o\le l\le u\le o+1,\qquad A_i(x\bmod1,a)=A_{0,i}(a,x)\quad(o<x\le o+1),\\ &\operatorname{supp}A_{0,i}(a,\cdot)\subset(l,u],\qquad v'(x)>0\quad(l-2R\le x\le u+2R). \end{aligned}(2) A smooth passage packet family β„±\mathcal F with cut origin oo, support endpoints l,ul,u and speed vv has an increasing good arc of width Ξ΄\delta when o=0o=0, Ξ΄<l≀u<1βˆ’Ξ΄\delta<l\le u<1-\delta, and vβ€²(x)>0v'(x)>0 for every x∈(Ξ΄,1βˆ’Ξ΄)x\in(\delta,1-\delta).

The angular packet PiP_i of frequency mim_i and amplitude AiA_i, and the finite packet sum Hβ„±H_{\mathcal F} when II is finite, are defined in (3). For a positive integer nn and a coloring c:Iβ†’{1,…,n}c:I\to\{1,\ldots,n\}, the same display defines the color packet sum GrG_r and the total color vector GG. Pi(ΞΈ,u,a)=Ai(u,a)e(miΞΈ),Hβ„±=βˆ‘i∈IPi,Gr=βˆ‘c(i)=rPi,G=βˆ‘r=1nGr.\begin{aligned} P_i(\theta,u,a)&=A_i(u,a)e(m_i\theta),& H_{\mathcal F}&=\sum_{i\in I}P_i,\\ G_r&=\sum_{c(i)=r}P_i,&G&=\sum_{r=1}^nG_r. \end{aligned}(3) A colored packet state C=(β„±,c,s)C=(\mathcal F,c,s) is a triple consisting of a smooth passage packet family β„±\mathcal F with finite index set II and pairwise distinct frequencies, a coloring cc of II, and a positive scale ss, subject to spare coordinate norm invariance, the condition in (4). |A0,i(a,x)|=|A0,i(0,x)|,|Ai(u,a)|=|Ai(u,0)|.|A_{0,i}(a,x)|=|A_{0,i}(0,x)|,\qquad |A_i(u,a)|=|A_i(u,0)|.(4)

Packet conjugation symmetry of β„±\mathcal F under a map Ο„:Iβ†’I\tau:I\to I is the condition in (5). A colored packet state C=(β„±,c,s)C=(\mathcal F,c,s) has colored reflection symmetry under a map ρ\rho of its colors when there is a map Ο„:Iβ†’I\tau:I\to I satisfying (5) and c(Ο„i)=ρ(c(i))c(\tau i)=\rho(c(i)).✦ mΟ„i=βˆ’mi,AΟ„i(u,a)=Ai(u,a)Β―,A0,Ο„i(a,x)=A0,i(a,x)Β―.m_{\tau i}=-m_i,\qquad A_{\tau i}(u,a)=\overline{A_i(u,a)},\qquad A_{0,\tau i}(a,x)=\overline{A_{0,i}(a,x)}.(5)✦

Spectral Densities and Packet Orbits

The inverse branch density associated to a real set KK and real functions J,g,D,qJ,g,D,q is Ξ²K,J,g,D,q\beta_{K,J,g,D,q} in (6). For a frequency map JJ and amplitude BB, its stationary frequency density uses an inverse branch on KK, the Jacobian Jβ€²J' and the weight |B|2|B|^2. Ξ²K,J,g,D,q(y)={q(g(y))/|D(g(y))|,y∈J(K),0,yβˆ‰J(K).\beta_{K,J,g,D,q}(y)= \begin{cases} q(g(y))/|D(g(y))|,&y\in J(K),\\ 0,&y\notin J(K). \end{cases}(6)✦ For the passage packets, take K=[lβˆ’R,u+R]K=[l-R,u+R], Ji(x)=miv(x)J_i(x)=m_iv(x) and an inverse branch gig_i of JiJ_i on KK. The stationary spectral weight wiw_i, its integer periodization Per⁡wi\operatorname{Per}w_i and its folded circle density pip_i are (7), where zΜƒβˆˆ(0,1]\widetilde z\in(0,1] represents zβˆˆπ•‹z\in\mathbb T. The color spectral density prp_r is the state color density, and pp is the total block density in the same display. wi=Ξ²K,Ji,gi,Jiβ€²,|A0,i(0,β‹…)|2,Per⁡wi(t)=βˆ‘kβˆˆβ„€wi(t+k),pi(z)=Per⁡wi(zΜƒ),pr(z)=βˆ‘c(i)=rpi(z),p(z)=βˆ‘r=1npr(z).\begin{aligned} w_i&=\beta_{K,J_i,g_i,J_i',\,|A_{0,i}(0,\cdot)|^2},& \operatorname{Per}w_i(t)&=\sum_{k\in\mathbb Z}w_i(t+k),\\ p_i(z)&=\operatorname{Per}w_i(\widetilde z),& p_r(z)&=\sum_{c(i)=r}p_i(z),\qquad p(z)=\sum_{r=1}^np_r(z). \end{aligned}(7)

A measurable packet family β„‹\mathcal H indexed by a finite set II consists of pairwise distinct nonzero integer frequencies mim_i and measurable amplitudes AiA_i in L2(𝕋2,m2)L^2(\mathbb T^2,m_2).✦ A measurable packet family β„‹\mathcal H is supported in [Ξ±,Ξ²][\alpha,\beta] when each of its amplitudes has support contained in ([Ξ±,Ξ²]mod⁡1)×𝕋([\alpha,\beta]\bmod1)\times\mathbb T.✦ A packet reflection symmetry of a measurable packet family β„‹\mathcal H is an involution of its index set that reverses its frequencies and conjugates its amplitudes.✦ The packet spectral measure Οƒβ„‹Ο†\sigma_{\mathcal H}^{\varphi} of a measurable packet family β„‹\mathcal H is the measure given by the sum in (8); each summand is the packet spectral measure of the corresponding angular packet, and ΞΆi(u,a)=miΟ†(u)\zeta_i(u,a)=m_i\varphi(u). Οƒβ„‹Ο†=βˆ‘i∈I(ΞΆi)*(|Ai|2m2).\sigma_{\mathcal H}^{\varphi} =\sum_{i\in I}(\zeta_i)_*\bigl(|A_i|^2m_2\bigr).(8)✦

For finitely supported complex coefficients b=(bk)kβˆˆβ„€b=(b_k)_{k\in\mathbb Z} and functions H,f,gH,f,g in L2(𝕋3,m3)L^2(\mathbb T^3,m_3), the Koopman Laurent polynomial β„’bH\mathcal L_bH and the squared physical L2L^2 distance β„°(f,g)\mathcal E(f,g) are defined in (9). The state Laurent polynomial is β„’bG\mathcal L_bG; for a finite packet sum Hβ„‹H_{\mathcal H}, β„’bHβ„‹\mathcal L_bH_{\mathcal H} is its packet Koopman Laurent polynomial. β„’bH=βˆ‘k∈supp⁡bbkUTkH,β„°(f,g)=βˆ«π•‹3|fβˆ’g|2dm3.\mathcal L_bH=\sum_{k\in\operatorname{supp}b}b_kU_T^kH,\qquad \mathcal E(f,g)=\int_{\mathbb T^3}|f-g|^2\,dm_3.(9)

Energy and the Zero Set

For a complex measurable function HH on a measure space (X,ΞΌ)(X,\mu), its energy density is |H(x)|2|H(x)|^2, and its energy measure is |H|2ΞΌ|H|^2\mu.✦ Let (Y,Ξ½)(Y,\nu) be another measure space, ΞΆ:Xβ†’Y\zeta:X\to Y a measurable map, h:Y→ℝh:Y\to\mathbb R a nonnegative measurable function, and F,B:Xβ†’β„‚F,B:X\to\mathbb C measurable functions. The whitening coefficient QQ is the function in (10), using BB on the inverse image of the zero set of hh. Q(x)={B(x),h(ΞΆ(x))=0,F(x)/h(ΞΆ(x)),h(ΞΆ(x))β‰ 0.Q(x)= \begin{cases} B(x),&h(\zeta(x))=0,\\ F(x)/\sqrt{h(\zeta(x))},&h(\zeta(x))\ne0. \end{cases}(10)✦

The Claim

Theorem 1. For the measurable data above, suppose that the energy pushforwards satisfy (11), so that BB is flat over Ξ½\nu. ΞΆ*(|F|2ΞΌ)=hΞ½,ΞΆ*(|B|2ΞΌ)=Ξ½.\zeta_*(|F|^2\mu)=h\nu,\qquad \zeta_*(|B|^2\mu)=\nu.(11) The whitening distance is given by (12), and the error energy pushforward is (13). ∫X|Qβˆ’F|2dΞΌ=∫Y(hβˆ’1)2dΞ½.\int_X|Q-F|^2\,d\mu=\int_Y(\sqrt h-1)^2\,d\nu.(12)✦ ΞΆ*(|Qβˆ’F|2ΞΌ)=(hβˆ’1)2Ξ½.\zeta_*\bigl(|Q-F|^2\mu\bigr)=(\sqrt h-1)^2\nu.(13)✦

Let CC be a colored packet state with colored reflection symmetry under an involutive color map ρ\rho. Suppose that Ξ΄>0\delta>0 and its passage packets have an increasing good arc of width Ξ΄\delta. There are numbers Ξ΄<Ξ±<Ξ²<1βˆ’Ξ΄\delta<\alpha<\beta<1-\delta and a measurable packet family 𝒲\mathcal W supported in [Ξ±,Ξ²][\alpha,\beta], with packet reflection symmetry, satisfying (14); for every finitely supported bb and every Ξ΅>0\varepsilon>0, the final inequality holds for some finitely supported dd. σ𝒲φ=m,β„°(H𝒲,G)=βˆ«π•‹pdm+1βˆ’2βˆ«π•‹pdm,β„°(β„’bG,β„’dH𝒲)<Ξ΅2.\begin{aligned} \sigma_{\mathcal W}^{\varphi}&=m,\\ \mathcal E(H_{\mathcal W},G) &=\int_{\mathbb T}p\,dm+1-2\int_{\mathbb T}\sqrt p\,dm,\\ \mathcal E(\mathcal L_bG,\mathcal L_dH_{\mathcal W})&<\varepsilon^2. \end{aligned}(14)✦

In the paper, the result reads: for the predicted real packet vector GG of density pp, put E={p=0}E=\{p=0\}, take a flat real vector ww in the same final good arc, and set fβ€²=(𝟏Ec/p)(UT)G+𝟏E(UT)wf'=(\mathbf1_{E^c}/\sqrt p)(U_T)G+\mathbf1_E(U_T)w, whose density is exactly 11 and whose distance satisfies βˆ₯fβ€²βˆ’Gβˆ₯22=βˆ«π•‹(pβˆ’1)2dm=βˆ₯Gβˆ₯22+1βˆ’2βˆ«π•‹pdm\|f'-G\|_2^2=\int_{\mathbb T}(\sqrt p-1)^2\,dm =\|G\|_2^2+1-2\int_{\mathbb T}\sqrt p\,dm, with the old prediction approximable by finite Laurent polynomials applied to fβ€²f'.

What Is Checked

The claims are stated over measurable energy pushforwards and finite packet families on the normalized product torus, with colored reflection symmetry and an increasing good arc as the hypotheses for the packet existence statement.