OctoInformBeta

OpenAI preprint, September 2026 · §6. A spatial measure with dimension greater than seven quarters

Proposition 6.1: Bounded projected label density gives spatial mass exponent 32/1732/17 after compactification

Radial comparison yields separated spatial witnesses.

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

Rank labels record the position of a point relative to convex chains in successive sweep windows. Retaining a compactified position beside each label gives spatial measures to which separation and mass bounds apply. The analytic norm on the plane ℝ2\mathbb R^2 is ∥x∥∞=max⁡(|x1|,|x2|)\|x\|_\infty=\max(|x_1|,|x_2|), and its closed balls B¯∞(x,r)\overline B_\infty(x,r) are squares.

An NN-point configuration is a map P:{0,…,N−1}→ℝ2P:\{0,\ldots,N-1\}\to\mathbb R^2. An NN-point configuration PP is in general position when it is injective and the orientation determinant det⁡(Pj−Pi,Pℓ−Pi)\det(P_j-P_i,P_\ell-P_i) is nonzero for distinct indices.✦ A generic configuration PP is an NN-point configuration in general position with distinct first coordinates and distinct slopes (Pj,2−Pi,2)/(Pj,1−Pi,1)(P_{j,2}-P_{i,2})/(P_{j,1}-P_{i,1}) for distinct unordered pairs i<ji<j, such that an intersection of the open segments of two distinct pairs lies on no closed segment of a third distinct pair.✦

A system of uniform exchange batches of rank kk and distance aa consists of strictly increasing sample slopes s0,…,sRs_0,\ldots,s_R and kk-element marked sets A0,…,ARA_0,\ldots,A_R of point indices. At each sample the sweep projection q↦q2−sbq1q\mapsto q_2-s_bq_1 of every marked point is smaller than that of every unmarked point.✦ Exactly aa indices leave the marked set at each batch. The leaving and entering sets, exchange edges and participating vertices are those in (1). Lb=Ab\Ab+1,Gb=Ab+1\Ab,ℰ=⋃b=0R−1(Lb×Gb),𝒱=pr⁡1(ℰ)∪pr⁡2(ℰ),E*=Ra2,M=aN.\begin{aligned} L_b&=A_b\setminus A_{b+1},& G_b&=A_{b+1}\setminus A_b,\\ \mathcal E&=\bigcup_{b=0}^{R-1}(L_b\times G_b),& \mathcal V&=\operatorname{pr}_1(\mathcal E)\cup\operatorname{pr}_2(\mathcal E),\\ E_*&=Ra^2,&M&=aN. \end{aligned}(1) A generic batch configuration is a generic configuration with such a system and positive N,R,aN,R,a.✦

A straight-line graph drawing has a finite vertex set and a finite set of ordered edges, each directed to the right, with both endpoints in its vertex set.✦ The exchange graph drawing uses 𝒱\mathcal V and ℰ\mathcal E from (1).✦ Two edges cross when their open segments intersect; crossing edge pairs count each unordered pair once. A chosen crossing point of two crossing edges is a point in their common interior.✦ Relative to a finite set of edges, exterior crossing points are crossing points outside the closed segments of that set.✦

A dyadic time window J=(d,j)J=(d,j) has depth d≥0d\ge0 and 0≤j<2d0\le j<2^d.✦ The carrier |J||J| of a dyadic time window J=(d,j)J=(d,j) is the uniform partition cell between j/2dj/2^d and (j+1)/2d(j+1)/2^d, closed on the left and open on the right except for the final cell, which is closed.✦ Batch bb has midpoint time tb=(b+12)/Rt_b=(b+\tfrac12)/R.✦ The starting and stopping batch indices of a dyadic time window JJ are the numbers of batch midpoint times strictly below the window’s left and right boundaries, respectively. The window batch count of a dyadic time window JJ is the difference between its stopping and starting batch indices. The time window batch system of a dyadic window JJ is the batch system consisting of the consecutive batches determined by its starting and stopping indices; more generally a consecutive batch window starting at b0b_0 with length qq, where b0+q≤Rb_0+q\le R, is the batch system retaining the original slopes and marked sets at indices b0,…,b0+qb_0,\ldots,b_0+q.

Rank Labels and Measures

Enumerations of the leaving and entering sets by the cyclic indices modulo aa define cyclic perfect matchings: matching number jj sends index ii to i+ji+j. Matching-induced transport between consecutive marked sets is the bijection that fixes their shared points and matches the leaving points to the entering points. Cumulative token transport from the initial marked set to a later marked set is the bijection obtained by composing the matching-induced transports between consecutive marked sets. An edge token trace of an exchange edge is a choice of a batch, a matching number and an initial token occupying the edge’s source before that batch and its target afterwards, with the source leaving and the target entering.✦ A chosen trace assigns the edge its batch time and its convex chain owner, the pair consisting of matching number and initial token index.

Convex chain graph extensions are MM continuous convex functions on ℝ\mathbb R, with common lower and upper secant slopes strictly between the first and last sample slopes. Each exchange segment lies on its assigned graph. For a prescribed finite set FF of planar points, every point of FF lying on a graph must belong to an exchange segment assigned to that graph.

A convex graph family of size mm is a collection g0,…,gm−1g_0,\ldots,g_{m-1} of continuous convex functions on ℝ\mathbb R and a scale ε>0\varepsilon>0.✦ The flat family is a convex graph family with all graphs zero and scale one.✦ The vertical graph rank rgr_g and smoothed rank coordinate r̃g,ε\widetilde r_{g,\varepsilon} of the convex graph family g0,…,gm−1g_0,\ldots,g_{m-1} are functions of a planar point pp, defined by (2), where sigmoid is the logistic function and the smoothed rank coordinate is zero when m=0m=0. rg(p)=#{i:gi(p1)<p2},r̃g,ε(p)=1m∑i=0m−1sigmoid⁡(p2−gi(p1)ε).r_g(p)=\#\{i:g_i(p_1)<p_2\},\qquad \widetilde r_{g,\varepsilon}(p) =\frac1m\sum_{i=0}^{m-1} \operatorname{sigmoid}\!\left(\frac{p_2-g_i(p_1)}{\varepsilon}\right).(2) Each dyadic window of a generic batch configuration has a window rank graph family of MM graphs. The unit interval rank coordinate of a window rank graph family is the smoothed coordinate in (2).✦ The window label space is ℒ=[0,1]𝒲\mathcal L=[0,1]^{\mathcal W}, where 𝒲\mathcal W is the set of dyadic time windows.✦ For a generic batch configuration FF, the configuration rank label ζF(p)∈ℒ\zeta_F(p)\in\mathcal L has at coordinate JJ the unit interval rank coordinate of pp for that window’s graph family.✦

A weighted atomic measure on a measurable space is a measure w∑i∈Aδxiw\sum_{i\in A}\delta_{x_i} for a finite index set AA, points xix_i of that space and w≥0w\ge0.✦ For a generic batch configuration FF, the full point measure μF\mu_F assigns weight 1/N1/N to each ζF(Pi)\zeta_F(P_i); the participating-vertex measure μF,J\mu_{F,J} assigns the same weight to the labels of vertices in the time window system. The weighted edge endpoint measure with normalization EE assigns weight E−1E^{-1} to each initial-endpoint label, and is zero for E=0E=0.✦ The edge label and time measure νF\nu_F assigns weight E*−1E_*^{-1} to (ζF(Pe1),te)(\zeta_F(P_{e_1}),t_e) for each exchange edge ee, where e1e_1 is its initial endpoint index and tet_e is its batch time.✦

The time-restricted edge measure νF,J\nu_{F,J} is the restriction of νF\nu_F to ℒ×|J|\mathcal L\times|J|.✦ Window crossing edge pairs have one exchange edge in each of two windows I,JI,J.✦ The window crossing point measure assigns weight M−2M^{-2} to each chosen crossing point; its pushforward by ζF\zeta_F is the crossing label measure σF,IJ\sigma_{F,IJ}.

Limiting rank data consist of finite measures μ,μJ,ν,νJ,σIJ\mu,\mu_J,\nu,\nu_J,\sigma_{IJ} on the corresponding label spaces.✦ Joint convergence of a sequence FnF_n to these data means weak convergence of every family just defined.✦ The edge label marginal is λ=(pr⁡1)*ν\lambda=(\operatorname{pr}_1)_*\nu, and its window version is λJ=(pr⁡1)*νJ\lambda_J=(\operatorname{pr}_1)_*\nu_J.

The spatial compactification e:ℝ2→Qe:\mathbb R^2\to Q, where Q=[0,1]2Q=[0,1]^2, and the map d:Q→ℝ2d:Q\to\mathbb R^2 are defined in (3). e(x)=(sigmoid⁡(x1),sigmoid⁡(x2)),d(u)=(−log⁡(u1−1−1),−log⁡(u2−1−1)).\begin{aligned} e(x)&=(\operatorname{sigmoid}(x_1),\operatorname{sigmoid}(x_2)),\\ d(u)&=\bigl(-\log(u_1^{-1}-1),-\log(u_2^{-1}-1)\bigr). \end{aligned}(3) The enhanced label space is ℒ×Q\mathcal L\times Q.✦ A spatially enhanced rank label of pp under a map ψ:ℝ2→ℝ2\psi:\mathbb R^2\to\mathbb R^2 is (ζF(p),e(ψ(p)))(\zeta_F(p),e(\psi(p))), using (3).✦ The enhanced point and vertex measures are measures assigning weight 1/N1/N to the spatially enhanced rank labels of all points and participating vertices, respectively; the enhanced edge measure for a dyadic window JJ is the measure assigning weight E*−1E_*^{-1} to the spatially enhanced rank labels of initial endpoints in that window. Write 𝒜F,ψ,i\mathcal A_{F,\psi,i} for this enhanced measure family, with countable index set ℐ={point}⊔{vertex(J)}⊔{edge(J)}\mathcal I=\{\mathrm{point}\}\sqcup\{\mathrm{vertex}(J)\}\sqcup \{\mathrm{edge}(J)\}. The original label measure family associated with the enhanced measure family 𝒜F,ψ,i\mathcal A_{F,\psi,i} is αpoint=μ\alpha_{\mathrm{point}}=\mu, αvertex(J)=μJ\alpha_{\mathrm{vertex}(J)}=\mu_J and αedge(J)=λJ\alpha_{\mathrm{edge}(J)}=\lambda_J.✦

Spatial Charts

The reflected rank chart is D−=(0,1)×(−1,0)D_-=(0,1)\times(-1,0). For an open partial homeomorphism ff of the plane write LL for its inverse map. The normalized spatial chart of an open partial homeomorphism ff at p0p_0 with nonzero scale uu is the map x↦f(L(p0)+ux)x\mapsto f(L(p_0)+ux).✦ An affine homeomorphism more generally has the form x↦p+qAxx\mapsto p+qAx, where q≠0q\ne0 and AA is an invertible linear map, with inverse y↦A−1(q−1(y−p))y\mapsto A^{-1}(q^{-1}(y-p)).✦ For a sequence of charts fnf_n with inverse maps LnL_n and positive scales unu_n, write Lˆn(p)=(Ln(p)−Ln(p0))/un\widehat L_n(p)=(L_n(p)-L_n(p_0))/u_n for the normalized inverse map.

The radial comparison ℛ(L;p,h,C,T)\mathcal R(L;p,h,C,T) for a planar map LL is the condition in (4) for every pair v,wv,w on the indicated squares. ∥v−p∥∞=h/T,∥w−p∥∞=h,∥L(w)−L(p)∥∞≤CT∥L(v)−L(p)∥∞,T∥L(v)−L(p)∥∞≤C∥L(w)−L(p)∥∞.\begin{gathered} \|v-p\|_\infty=h/T,\qquad \|w-p\|_\infty=h,\\ \|L(w)-L(p)\|_\infty\le CT\|L(v)-L(p)\|_\infty,\qquad T\|L(v)-L(p)\|_\infty\le C\|L(w)-L(p)\|_\infty. \end{gathered}(4)✦ For a vector vv, the scalar product is v⋅x=v1x1+v2x2v\cdot x=v_1x_1+v_2x_2, and its squared Euclidean norm is v12+v22v_1^2+v_2^2. Let K=B¯∞(0,1)K=\overline B_\infty(0,1) be the closed unit square, and let S(v,t,r)={x:|v⋅x−t|≤r}S(v,t,r)=\{x:|v\cdot x-t|\le r\} and H(v,t)={x:v⋅x=t}H(v,t)=\{x:v\cdot x=t\} be subsets of the plane.

The Claim

Theorem 1 (Spatial selection and enhanced limits). Let μ,θ\mu,\theta be finite measures on a measurable space XX, let E⊂XE\subset X be measurable with θ(E)>0\theta(E)>0, and let π:X→ℝ2\pi:X\to\mathbb R^2 be measurable. Suppose π*(θ|E)\pi_*(\theta|_E) is absolutely continuous with respect to planar Lebesgue measure ℒ2\mathcal L^2. Let fnf_n be open partial homeomorphisms with source ℝ2\mathbb R^2 and target D−D_-. Assume that for θ|E\theta|_E-almost every xx, π(x)∈D−\pi(x)\in D_- and there is Cx>0C_x>0 such that, for every real T≥1T\ge1 and all sufficiently small positive hh, the comparison ℛ(Ln;π(x),h,Cx,T)\mathcal R(L_n;\pi(x),h,C_x,T) holds for all sufficiently large nn. There are a measurable set GG, an integer m≥0m\ge0, a point p0p_0, positive scales unu_n, a strictly increasing subsequence map φ\varphi, a map l:π(G)→ℝ2l:\pi(G)\to\mathbb R^2 and c>0c>0 with the properties in (5). G⊂E,θ(G)>0,π*(μ|G)≤mℒ2,∥l(p)∥∞≤1,limn→∞q→pLˆφ(n)(q)=l(p)(p∈π(G)),c∥p−q∥∞17/16≤∥l(p)−l(q)∥∞(p,q∈π(G)).\begin{gathered} G\subset E,\qquad \theta(G)>0,\qquad \pi_*(\mu|_G)\le m\mathcal L^2,\\ \|l(p)\|_\infty\le1,\qquad \lim_{\substack{n\to\infty\\q\to p}} \widehat L_{\varphi(n)}(q)=l(p)\quad(p\in\pi(G)),\\ c\|p-q\|_\infty^{17/16}\le\|l(p)-l(q)\|_\infty \quad(p,q\in\pi(G)). \end{gathered}(5)✦

For any jointly convergent generic batch sequence FnF_n and any planar maps ψn\psi_n, there are a strictly increasing map τ\tau and finite enhanced measures ηi\eta_i satisfying (6) for every i∈ℐi\in\mathcal I. 𝒜Fτ(n),ψτ(n),i⇒ηi,(pr⁡1)*ηi=αi.\mathcal A_{F_{\tau(n)},\psi_{\tau(n)},i}\ \Longrightarrow\ \eta_i, \qquad (\operatorname{pr}_1)_*\eta_i=\alpha_i.(6)✦

Let XX be a metric space with its Borel sets, let ηn,η\eta_n,\eta be finite measures on X×QX\times Q, and suppose ητ(n)⇒η\eta_{\tau(n)}\Longrightarrow\eta along a strictly increasing τ\tau. Let π:X→ℝ2\pi:X\to\mathbb R^2 be continuous and ψn\psi_n planar maps such that u=e(ψn(π(x)))u=e(\psi_n(\pi(x))) for ηn\eta_n-almost every (x,u)(x,u), for each nn. For a set G⊂XG\subset X and a map l:π(G)→ℝ2l:\pi(G)\to\mathbb R^2, suppose ψn(q)→l(p)\psi_n(q)\to l(p) jointly as n→∞n\to\infty and q→pq\to p, for each p∈π(G)p\in\pi(G). Then the support satisfies (7). (x,u)∈supp⁡η,x∈G⇒u=e(l(π(x))).(x,u)\in\operatorname{supp}\eta,\quad x\in G \quad\Longrightarrow\quad u=e(l(\pi(x))).(7)✦

Consider a measure η\eta on X×QX\times Q, where XX is a metric space with its Borel sets, GG is Borel and π\pi is measurable. Suppose its support has the graph property (7), that π*(((pr⁡1)*η)|G)≤mℒ2\pi_*(((\operatorname{pr}_1)_*\eta)|_G)\le m\mathcal L^2 for some integer m≥0m\ge0, and that ll takes values in KK and has the separation bound of (5) with c>0c>0. Let A=(G×Q)∩supp⁡ηA=(G\times Q)\cap\operatorname{supp}\eta be a subset of X×QX\times Q, and let β=(d∘pr⁡2)*(η|A)\beta=(d\circ\operatorname{pr}_2)_*(\eta|_A) be a measure on ℝ2\mathbb R^2. The assertions in (8) hold for some C≥0C\ge0. A is Borel,d(u)∈K((x,u)∈A),β(B¯∞(x,r))≤Cr32/17(x∈ℝ2,r>0).\begin{gathered} A\text{ is Borel},\qquad d(u)\in K\quad((x,u)\in A),\\ \beta(\overline B_\infty(x,r))\le Cr^{32/17} \quad(x\in\mathbb R^2,\ r>0). \end{gathered}(8)✦

Let ρ\rho be a planar measure satisfying ρ(B¯∞(x,r))≤Crd\rho(\overline B_\infty(x,r))\le Cr^d for every xx and r>0r>0, where C≥0C\ge0 and d∈ℝd\in\mathbb R. For every vv with v12+v22=1v_1^2+v_2^2=1, every t∈ℝt\in\mathbb R and 0<r≤10<r\le1, the strip estimate is (9). ρ(K∩S(v,t,r))≤13C3drd−1.\rho(K\cap S(v,t,r))\le13C\,3^d r^{d-1}.(9)✦ If d>1d>1, the same square mass hypothesis gives (10) for every such vv and tt. ρ(K∩H(v,t))=0.\rho(K\cap H(v,t))=0.(10)✦

In the paper, the result reads: in the common Euclidean normalization, let Ln:D=(0,1)2→ℝ2L_n:D=(0,1)^2\to\mathbb R^2 be inverse rank homeomorphisms and let EE be Borel with λK0(E)>0\lambda_{K_0}(E)>0 for a selected window K0K_0, with π*(λ|E)\pi_*(\lambda|_E) absolutely continuous and λ≪μ\lambda\ll\mu; suppose that at every selected p∈π(E)p\in\pi(E) there is a finite CpC_p independent of TT such that, for each integer T≥2T\ge2 and every sufficiently small fixed h>0h>0, eventually in nn the radial maximum UnU_n and minimum VnV_n satisfy Vn(p,h/T)≥Un(p,h)/(CpT)V_n(p,h/T)\ge U_n(p,h)/(C_pT) and Un(p,h/T)≤CpVn(p,h)/TU_n(p,h/T)\le C_pV_n(p,h)/T; then there are a Borel set G⊂EG\subset E with λK0(G)>0\lambda_{K_0}(G)>0, a subsequence, and translations and uniform rescalings of the normalized planes for which the limiting graph over every p∈π(G)p\in\pi(G) has a unique position l(p)l(p) in the closed Euclidean unit disk, with |l(p)−l(q)|≥cG|p−q|17/16|l(p)-l(q)|\ge c_G|p-q|^{17/16}, and, after position is attached as a compactified label and a weak vertex limit is taken, its restriction to old labels in GG has position marginal βG\beta_G satisfying βG(B(x,r))≤CGr32/17\beta_G(B(x,r))\le C_Gr^{32/17} for every xx and 0<r<10<r<1, every strip of width rr has mass at most CG′r15/17C'_Gr^{15/17}, and every line has zero mass.

What Is Checked

The spatial selection is stated over finite measures on a measurable space and charts with full planar source and reflected target; the enhancement and support statements concern the batch measures and Borel measures on metric spaces. Every exchange edge admits an edge token trace.✦ The unit interval rank coordinate of a positive convex graph family lies strictly between zero and one.