OctoInformBeta

OpenAI preprint, September 2026 · §3. Rank coordinates and limiting measures

§3 (setting): vertex, edge, crossing and conditional-time measures: Subsequences, conditional edge times and crossing densities for generic exchange batches

Rank labels admit joint measure limits.

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 exchange batch joins points leaving a rank set to points entering it during a planar sweep. Rank labels assign each point one coordinate for every dyadic time window. The associated measures record all points, the vertices participating in a window, edges with their batch times, and crossing points.

A configuration P:{0,…,N−1}→ℝ2P:\{0,\ldots,N-1\}\to\mathbb R^2 is in general position if it is injective and no three of its points are collinear✦; the orientation of three points p,q,rp,q,r is given by the determinant det⁡(q−p,r−p)\det(q-p,r-p).✦ A generic configuration P:{0,…,N−1}→ℝ2P:\{0,\ldots,N-1\}\to\mathbb R^2 is a configuration in general position with distinct first coordinates and distinct slopes (q2−p2)/(q1−p1)(q_2-p_2)/(q_1-p_1) for distinct unordered point pairs, represented by i<ji<j. It also satisfies the following condition: an intersection of the open segments of two distinct pairs belongs to no closed segment of a third distinct pair.

A uniform exchange system SS of rank kk and distance aa on PP consists of strictly increasing slopes s0<⋯<sRs_0<\cdots<s_R and marked index sets QbQ_b of size kk. Writing the coordinates of each point PiP_i as xi,yix_i,y_i, projection separation means that yi−sbxi<yj−sbxjy_i-s_bx_i<y_j-s_bx_j whenever i∈Qbi\in Q_b and j∉Qbj\notin Q_b.✦ The leaving and entering sets, and the exchange graph, are defined by (1); the distance condition is |Lb|=a|L_b|=a. Lb=Qb\Qb+1,Gb=Qb+1\Qb,E=⋃b=0R−1Lb×Gb,V={u:(u,v)∈E}∪{v:(u,v)∈E}.L_b=Q_b\setminus Q_{b+1},\qquad G_b=Q_{b+1}\setminus Q_b,\qquad E=\bigcup_{b=0}^{R-1}L_b\times G_b,\qquad V=\{u:(u,v)\in E\}\cup\{v:(u,v)\in E\}.(1) The exchange graph drawing places VV at its images under PP and draws each edge as a straight segment, oriented from smaller to larger first coordinate.✦ Two edges cross when their open segments intersect✦; the crossing pairs count each unordered pair once✦, and a chosen crossing point lies in both open segments.✦

The sweep projection of a point p=(x,y)∈ℝ2p=(x,y)\in\mathbb R^2 at a real slope ss is the real number y−sxy-sx.✦

For a straight-line graph drawing DD and a finite set E0E_0 of ordered pairs of point indices, the exterior edge crossing points form the finite set of crossing points of DD exterior to E0E_0.✦

A generic batch configuration FF consists of a generic configuration PP and a uniform exchange system SS on PP, with N,R,a>0N,R,a>0.✦ Its edge and crossing normalizations are respectively (2). E*=Ra2,M=aN.E_*=Ra^2,\qquad M=aN.(2)

Windows and Rank Labels

A dyadic window J=(d,j)J=(d,j) is a pair of integers with d≥0d\geq0 and 0≤j<2d0\leq j<2^d.✦ Its carrier is the cell [j2−d,(j+1)2−d)[j2^{-d},(j+1)2^{-d}) in [0,1][0,1], with the right endpoint included for the last cell.✦ Write 𝒲\mathcal W for the set of these windows. The batch midpoint time tbt_b, the number cR(x)c_R(x) of earlier batch times, and the window’s starting and stopping indices are defined by (3). tb=b+1/2R,cR(x)=#{b:tb<x},rJ=cR(j2−d),uJ=cR((j+1)2−d),ℓJ=uJ−rJ.\begin{aligned} t_b&=\frac{b+1/2}{R},& c_R(x)&=\#\{b:t_b<x\},\\ r_J&=c_R(j2^{-d}),& u_J&=c_R((j+1)2^{-d}),& \ell_J&=u_J-r_J. \end{aligned}(3) The time window system SJS_J is the consecutive subsystem starting at rJr_J and containing ℓJ\ell_J batches.✦ A window AA precedes a window BB when the upper boundary of AA is at most the lower boundary of BB.✦

For finite sets L,GL,G of size aa, enumerations l:L→ℤ/aℤl:L\to\mathbb Z/a\mathbb Z and g:G→ℤ/aℤg:G\to\mathbb Z/a\mathbb Z define the cyclic perfect matching u↦g−1(l(u)+j)u\mapsto g^{-1}(l(u)+j) for each shift jj.✦ Applying this to the leaving and entering enumerations gives a batch perfect matching.✦ The induced transport Qb→Qb+1Q_b\to Q_{b+1} fixes shared points and matches the remaining points; its inverse fixes shared points and uses the inverse matching.✦ For each matching number jj and stage bb, the cumulative token transport is the bijection from Q0Q_0 to QbQ_b obtained by composing the successive batch transports for jj, starting with the identity at stage zero; its value at an initial token is that token’s state at stage bb. An edge token trace specifies a batch bb, a matching number jj and an initial token whose states before and after bb are the leaving and entering endpoints of the edge.✦ For an exchange edge ee, write b(e)b(e) for the batch of a chosen edge token trace. The convex chain owner of an exchange edge is its trace’s matching number together with its initial token index.✦

A convex graph family is a finite family of continuous convex functions gi:ℝ→ℝg_i:\mathbb R\to\mathbb R with a positive scale ε\varepsilon.✦ The flat family has every gi=0g_i=0 and scale 11.✦ The vertical graph rank and the smoothed rank coordinate of p=(x,y)p=(x,y) are given by (4), where mm is the number of functions in the convex graph family and sigmoid\operatorname{sigmoid} is the sigmoid function. slot⁡g(x,y)=#{i:gi(x)<y},rg,ε(x,y)=1m∑i=0m−1sigmoid⁡(y−gi(x)ε),m>0.\operatorname{slot}_g(x,y)=\#\{i:g_i(x)<y\},\qquad r_{g,\varepsilon}(x,y)= \frac1m\sum_{i=0}^{m-1} \operatorname{sigmoid}\!\left(\frac{y-g_i(x)}{\varepsilon}\right), \qquad m>0.(4) The unit interval rank coordinate rg,εr_{g,\varepsilon} takes values in [0,1][0,1]. For each window JJ, the system assigns a rank graph family of MM functions. Write its rank coordinate as ρJ(p)\rho_J(p). The window rank label of a point is the element z(p)z(p) of the label space ZZ defined by (5). Z=[0,1]𝒲,z(p)(J)=ρJ(p).Z=[0,1]^{\mathcal W},\qquad z(p)(J)=\rho_J(p).(5)

A full convex chain graph extension consists of MM continuous convex functions, indexed by the chain owners, with secant slopes between bounds s0<α≤β<sRs_0<\alpha\leq\beta<s_R. Each exchange segment lies on its owner’s graph. For a specified finite set H⊂ℝ2H\subset\mathbb R^2, a full convex chain graph extension is clean on HH when any point of HH on one of its graphs lies on an exchange segment with that graph’s owner.

Finite Measures and Limiting Data

A weighted atomic measure on a finite indexed family xix_i assigns weight w≥0w\geq0 to each index, so it is w∑iδxiw\sum_i\delta_{x_i}.✦ Let VJV_J be the vertex set of SJS_J. The full point label measure μF\mu_F, the window vertex measure μF,J\mu_{F,J}, and the edge label and time measure νF\nu_F are defined by (6). An edge’s batch time is tb(e)t_{b(e)}✦, and its label is taken at its source endpoint. μF=1N∑i=0N−1δz(Pi),μF,J=1N∑i∈VJδz(Pi),νF=1E*∑e∈Eδ(z(Pe1),tb(e)).\begin{aligned} \mu_F&=\frac1N\sum_{i=0}^{N-1}\delta_{z(P_i)},& \mu_{F,J}&=\frac1N\sum_{i\in V_J}\delta_{z(P_i)},\\ \nu_F&=\frac1{E_*}\sum_{e\in E} \delta_{(z(P_{e_1}),t_{b(e)})}. \end{aligned}(6) The time-restricted edge measure νF,J\nu_{F,J} is the restriction of νF\nu_F to Z×JZ\times J, using the window’s carrier.✦ Let CABC_{AB} be the window crossing edge pairs for A,BA,B, and let pcp_c be the chosen crossing point of c∈CABc\in C_{AB}. The window crossing label measure is defined by (7). σF,AB=1M2∑c∈CABδz(pc).\sigma_{F,AB}=\frac1{M^2}\sum_{c\in C_{AB}}\delta_{z(p_c)}.(7)✦

Limiting measure data LL consist of finite measures μ\mu, μJ\mu_J and σAB\sigma_{AB} on ZZ, together with ν\nu and νJ\nu_J on Z×[0,1]Z\times[0,1], for every window JJ and every pair A,BA,B.✦ For a sequence (Fq)(F_q) of generic batch configurations with point counts NqN_q and exchange distances aqa_q, joint convergence Fq⇒LF_q\Longrightarrow L means the simultaneous weak convergences in (8) for every dyadic window and every pair of dyadic windows. μFq⇀μ,μFq,J⇀μJ,νFq⇀ν,νFq,J⇀νJ,σFq,AB⇀σAB.\mu_{F_q}\rightharpoonup\mu,\qquad \mu_{F_q,J}\rightharpoonup\mu_J,\qquad \nu_{F_q}\rightharpoonup\nu,\qquad \nu_{F_q,J}\rightharpoonup\nu_J,\qquad \sigma_{F_q,AB}\rightharpoonup\sigma_{AB}.(8)✦ The edge label marginal λ\lambda and the window edge marginal λJ\lambda_J are the pushforwards in (9), where πZ(z,t)=z\pi_Z(z,t)=z. The conditional edge time kernel hzh_z is the conditional kernel of ν\nu given its label coordinate.✦ For windows A,BA,B, write πAB(z)=(z(A),z(B))\pi_{AB}(z)=(z(A),z(B)). λ=(πZ)#ν,λJ=(πZ)#νJ.\lambda=(\pi_Z)_\#\nu,\qquad \lambda_J=(\pi_Z)_\#\nu_J.(9)

The Claim

Theorem 1. Every sequence (Fq)(F_q) of generic batch configurations has a strictly increasing map φ:ℕ→ℕ\varphi:\mathbb N\to\mathbb N and limiting measure data LL satisfying (10). Fφ(q)⇒L(q→∞).F_{\varphi(q)}\Longrightarrow L\qquad(q\to\infty).(10)✦ For any jointly convergent sequence Fq⇒LF_q\Longrightarrow L whose batch counts RqR_q tend to infinity, every dyadic window JJ and measurable set U⊂ZU\subset Z satisfy (11). λJ(U)=∫Uhz(J)dλ(z).\lambda_J(U)=\int_U h_z(J)\,d\lambda(z).(11)✦ If also Mq=aqNq→∞M_q=a_qN_q\to\infty, then every pair of windows AA preceding BB satisfies (12), where ℒ2\mathcal L^2 is planar Lebesgue measure. (πAB)#σAB≤ℒ2|[0,1]2.(\pi_{AB})_\#\sigma_{AB} \ \leq\ \mathcal L^2\big|_{[0,1]^2}.(12)✦

In the paper, the result reads: under the preceding extremal setup, weak limits of vertex weights 1/N1/N, participating-vertex weights 1/N1/N, edge weights 1/E*1/E_* with batch time, and crossing weights 1/M21/M^2 are μ\mu, μJ\mu_J, ν\nu and σIJ\sigma_{IJ}; with sJ=μJ(Z)s_J=\mu_J(Z), λ=(πZ)#ν\lambda=(\pi_Z)_\#\nu and λJ(U)=ν(U×J)\lambda_J(U)=\nu(U\times J), the measures μ,ν\mu,\nu are probabilities, μJ≤μ\mu_J\leq\mu, the time marginal of ν\nu is Lebesgue measure, λJ(Z)=|J|\lambda_J(Z)=|J|, a conditional-time probability hzh_z satisfies λJ=hz(J)λ\lambda_J=h_z(J)\lambda, and for disjoint I,JI,J the (z(I),z(J))(z(I),z(J)) projection of σIJ\sigma_{IJ} is dominated by planar Lebesgue measure.

What Is Checked

The claims are stated over sequences of generic planar configurations equipped with uniform exchange batch systems, with positive point counts, batch counts and exchange distances, and over their five families of finite measures and limiting measure data. The disintegration identity assumes that the batch counts tend to infinity; the crossing bound also assumes that the chain normalizations tend to infinity and that the first window precedes the second. Every exchange edge has an edge token trace.✦ Both endpoints of every exchange edge belong to the exchange graph’s vertex set.✦