OctoInformBeta

OpenAI preprint, September 2026 · §3. Arithmetic and nested partitions · §3

§3 (setting): the additive overlap identity and the scales W₂, A, W: One coordinate at a time and for the total cell mass, which also defines the overlap, potential and small-cell overlap scales

Pair overlaps split into a constant and two potentials.

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

Each point xx of a finite set VV carries two coordinates Z1(x)Z_1(x) and Z2(x)Z_2(x) in a number field KK, each coordinate injective on VV. The closeness of two distinct points xx, yy at a place vv of KK, for the coordinate ZjZ_j, is measured by the raw depth −log⁡|Zj(y)−Zj(x)|v-\log|Z_j(y)-Z_j(x)|_v, which is large when the coordinates are close at vv; at the archimedean places it is measured through random dyadic grids of the complex plane instead.

For each place and coordinate, the levels of this measure cut VV into nested cells, and weighting the cells over all places and both coordinates puts a mass on the subsets of VV. The question is how the mass of the subsets containing two given points depends on the pair, and what it is on average over a random pair of distinct points.

Random Dyadic Grids

An axis sequence is a real sequence (ωi)i≥0(\omega_i)_{i\ge 0}. The axis law is the probability law on axis sequences under which the terms are independent, ω0\omega_0 is uniform on (0,1](0,1], and ωn+1\omega_{n+1} equals 00 or 2n2^n with probability 1/21/2 each.✦ At a scale m∈ℤm\in\mathbb{Z} the sequence gives the offset sm=ω0+…+ωms_m=\omega_0+\dots+\omega_m when m≥0m\ge 0 and sm=ω0s_m=\omega_0 when m<0m<0✦, and the real line is cut into the half-open intervals of length 2m2^m with an endpoint at sms_m✦.

Two independent axis sequences, one for each axis, give the random dyadic grid of the plane ℝ2=ℂ\mathbb{R}^2=\mathbb{C}: two points share a square at scale mm when on each axis their coordinates lie in one interval of that scale✦. Write ω\omega for the two axis sequences. For points pp and qq of the plane, mω(p,q)m_\omega(p,q) is the infimum of the scales at which pp and qq share a square, and the grid depth of pp and qq is gω(p,q)=−mω(p,q)log⁡2g_\omega(p,q)=-m_\omega(p,q)\log 2.

A sample ω\omega is a phase (ξ,θ)(\xi,\theta) together with two axis sequences.✦ The sample law is the probability law on samples under which ξ\xi is uniform on (0,log⁡2](0,\log 2], θ\theta is uniform on (0,2π](0,2\pi], each axis sequence follows the axis law, and the four components are independent; 𝔼ω\mathbb{E}_\omega is the mean under this law.✦ The hierarchy depth of two complex numbers zz and ww under ω\omega is dω(z,w)=gω(e−ξ+iθz,e−ξ+iθw)−ξ.d_\omega(z,w)=g_\omega\bigl(e^{-\xi+i\theta}z,\;e^{-\xi+i\theta}w\bigr)-\xi .(1)✦

The error mean κ\kappa is the constant given by the mean excess of the grid depth of 00 and e−ξ+iθe^{-\xi+i\theta} over their raw depth ξ\xi: κ=𝔼ω[gω(0,e−ξ+iθ)−ξ].\kappa=\mathbb{E}_\omega\bigl[g_\omega(0,e^{-\xi+i\theta})-\xi\bigr].(2)✦

Depths at the Places of the Number Field

An ultrametric depth on VV is a symmetric function d:V×V→ℝd:V\times V\to\mathbb{R} such that d(x,z)≥min⁡{d(x,y),d(y,z)}d(x,z)\ge\min\{d(x,y),d(y,z)\} for all pairwise distinct xx, yy, zz in VV.✦

Fix an injective coordinate Z:V→KZ:V\to K. For an archimedean place vv of KK, given by an embedding ιv:K→ℂ\iota_v:K\to\mathbb{C}, a sample ω\omega is good at vv when the hierarchy depths dω(ιvZ(x),ιvZ(y))d_\omega(\iota_vZ(x),\iota_vZ(y)) satisfy the ultrametric inequality on every triple of distinct points of VV.✦ The archimedean depth at vv under ω\omega is the function dv,ωd_{v,\omega} on V×VV\times V given by dv,ω(x,y)={dω(ιvZ(x),ιvZ(y))if ω is good at v and x≠y,0otherwise.d_{v,\omega}(x,y)= \begin{cases} d_\omega\bigl(\iota_vZ(x),\iota_vZ(y)\bigr) & \text{if $\omega$ is good at $v$ and $x\ne y$,}\\ 0 & \text{otherwise.} \end{cases}(3)✦ An archimedean place vv carries the weight σv=mult(v)/[K:ℚ]\sigma_v=\mathrm{mult}(v)/[K:\mathbb{Q}], where mult(v)\mathrm{mult}(v) is 11 for a real place and 22 for a complex one.✦

For a finite place vv of KK, the finite depth at vv is the function dvd_v on V×VV\times V that vanishes on the diagonal and equals the raw depth −log⁡|Z(x)−Z(y)|v-\log|Z(x)-Z(y)|_v at distinct points xx, yy; a finite place carries the weight 1/[K:ℚ]1/[K:\mathbb{Q}].✦ The bad places of ZZ form the finite set BZB_Z of finite places vv at which |Z(x)−Z(y)|v≠1|Z(x)-Z(y)|_v\ne 1 for some distinct xx, yy in VV.✦

Cells and Their Masses

Fix an ultrametric depth dd on VV and a level r∈ℝr\in\mathbb{R}. The cell of a point xx at level rr is the set of points yy of VV with y=xy=x or d(x,y)≥rd(x,y)\ge r✦, and the cells at level rr are the cells of the points of VV at that level✦. A giant at level rr is a cell at level rr with more than |V|/2|V|/2 points.✦ The transformed cells at level rr form the family TrT_r, counted with multiplicity, of the cells at level rr with at most |V|/2|V|/2 points and the complements V\CV\setminus C of the giants CC at level rr.✦

The cell mass of a subset CC of VV is md(C)=∫ℝnr(C)drif |C|≥2,md(C)=0if |C|≤1,m_d(C)=\int_{\mathbb{R}}n_r(C)\,dr\ \ \text{if } |C|\ge 2,\qquad m_d(C)=0\ \ \text{if } |C|\le 1,(4)✦ where nr(C)n_r(C) is the number of times CC occurs in TrT_r.

The overlap length of two distinct points xx, yy of VV and the potential of a point xx are Ld(x,y)=∫ℝ#{C∈Tr:x,y∈C}dr,L_d(x,y)=\int_{\mathbb{R}}\#\{C\in T_r: x,y\in C\}\,dr,(5)✦ Sd(x)=∫ℝ𝟏[some giant exists at level r and x lies in none]dr,S_d(x)=\int_{\mathbb{R}}\mathbf{1}\bigl[\text{some giant exists at level $r$ and $x$ lies in none}\bigr]\,dr,(6)✦ The last giant level MdM_d of dd is the supremum of the levels rr at which a giant exists.✦

Sums Over Places and Coordinates

The global cell mass of the injective coordinate ZZ is the function μZ\mu_Z on subsets of VV given by μZ(C)=∑v∣∞σv𝔼ωmdv,ω(C)+1[K:ℚ]∑v∈BZmdv(C).\mu_Z(C)=\sum_{v\mid\infty}\sigma_v\,\mathbb{E}_\omega\,m_{d_{v,\omega}}(C) +\frac{1}{[K:\mathbb{Q}]}\sum_{v\in B_Z}m_{d_v}(C).(7)✦ The global overlap length LZ(x,y)L_Z(x,y), last giant level MZM_Z and potential SZ(x)S_Z(x) of ZZ are given by the right side of (7) with md(C)m_d(C) replaced by Ld(x,y)L_d(x,y), MdM_d and Sd(x)S_d(x) respectively.

For two injective coordinates Z1Z_1 and Z2Z_2, write μj\mu_j, MjM_j and SjS_j for the global cell mass, last giant level and potential of ZjZ_j. The total cell mass is μ*=μ1+μ2\mu^*=\mu_1+\mu_2✦, the total potential is S*=S1+S2S^*=S_1+S_2✦, and the base constant is b*=2κ−M1−M2b_*=2\kappa-M_1-M_2✦. The overlap of two points xx, yy of VV is the total cell mass of the subsets containing both: L*(x,y)=∑C⊆V,x,y∈Cμ*(C).L^*(x,y)=\sum_{C\subseteq V,\ x,y\in C}\mu^*(C).(8)✦

The uniform law α\alpha on VV gives each point the mass 1/|V|1/|V|✦, and the uniform law π\pi on ordered pairs of distinct points of VV gives each such pair the mass 1/(|V|(|V|−1))1/(|V|(|V|-1))✦; 𝔼α\mathbb{E}_\alpha and 𝔼π\mathbb{E}_\pi denote the means under α\alpha and π\pi, and for a subset CC of VV, π(C×C)\pi(C\times C) is the probability that a pair drawn from π\pi lies in C×CC\times C✦.

The Claim

Theorem 1. Let KK be a number field and VV a finite set of at least two points. If Z:V→KZ:V\to K is injective, then for distinct points xx, yy of VV LZ(x,y)=κ−MZ+SZ(x)+SZ(y).L_Z(x,y)=\kappa-M_Z+S_Z(x)+S_Z(y).(9)✦ If Z1,Z2:V→KZ_1,Z_2:V\to K are injective, then for distinct points xx, yy of VV L*(x,y)=b*+S*(x)+S*(y).L^*(x,y)=b_*+S^*(x)+S^*(y).(10)✦

Definition 1. Let KK be a number field, VV a finite set with injective coordinates Z1,Z2:V→KZ_1,Z_2:V\to K, and cc a real number. The overlap scale W2W_2, the potential scale AA and the small-cell overlap scale W2,smW_{2,\mathrm{sm}} of VV are the real numbers W2=∑C⊆Vμ*(C)π(C×C),W_2=\sum_{C\subseteq V}\mu^*(C)\,\pi(C\times C),(11)✦ A=∑x∈Vα(x)S*(x),A=\sum_{x\in V}\alpha(x)\,S^*(x),(12)✦ W2,sm=∑C⊆V,|C|≤|V|1−cμ*(C)π(C×C).W_{2,\mathrm{sm}}=\sum_{C\subseteq V,\ |C|\le|V|^{1-c}}\mu^*(C)\,\pi(C\times C).(13)✦

In the paper, the result reads: L*(x,y)=b*+S*(x)+S*(y)L^*(x,y)=b_*+S^*(x)+S^*(y) for x≠yx\ne y, with W2=𝔼πL*W_2=\mathbb{E}_\pi L^*, A=𝔼αS*A=\mathbb{E}_\alpha S^* and W=W2+AW=W_2+A.

What Is Checked

The identities (9) and (10) and the scales (11), (12) and (13) are stated over a finite set VV with injective coordinates in a number field KK: the archimedean depths (3), built from rotated and dilated random dyadic grids, and the finite depths at the bad places, with their weights and the error mean κ\kappa; the cells, giants and transformed cells of each depth, with the cell masses, overlap lengths, potentials and last giant levels they define; and the weighted sums of these over the places, among them the global cell mass (7).

The library proves that each archimedean depth and each finite depth is symmetric and satisfies the ultrametric inequality on triples of distinct points of VV.