OctoInformBeta

OpenAI preprint, September 2026 · §8. Dimension bounds and exact embeddings

§8.3 (setting and proof of Theorem 1.1): disjointness and the grid obstruction: A signed grid configuration gives the exact logarithmic exponent 22

Exact dimension is quadratic for p>1p>1, p≠2p\ne2..

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 exact embedding f:X→Yf:X\to Y of a finite subset XX of a real LpL_p space into a metric space YY is a map that preserves all pairwise distances up to one positive scale. The coordinate dimension needed for such embeddings is constrained by the supports of the coordinate vectors.

For a real exponent pp and a finite set FF, the power sum Pp(x)P_p(x) of a vector x∈ℝFx\in\mathbb R^F and the coordinate distance ρp,F(x,y)\rho_{p,F}(x,y) between x,y∈ℝFx,y\in\mathbb R^F are the real quantities specified by (1). Pp(x)=∑a∈F|xa|p,ρp,F(x,y)=(∑a∈F|xa−ya|p)1/p.P_p(x)=\sum_{a\in F}|x_a|^p, \qquad \rho_{p,F}(x,y)=\left(\sum_{a\in F}|x_a-y_a|^p\right)^{1/p}.(1) The support supp⁡(x)\operatorname{supp}(x) of a vector x∈ℝFx\in\mathbb R^F is the set of coordinates at which xx is nonzero. The scalar power defect Δp(a,b)\Delta_p(a,b) of a,b∈ℝa,b\in\mathbb R is the quantity in (2). Δp(a,b)=|a+b|p+|a−b|p−2|a|p−2|b|p.\Delta_p(a,b)=|a+b|^p+|a-b|^p-2|a|^p-2|b|^p.(2)✦

For a set II, the row indicator ri∈ℝI×Ir_i\in\mathbb R^{I\times I} of i∈Ii\in I and the column indicator cj∈ℝI×Ic_j\in\mathbb R^{I\times I} of j∈Ij\in I are the functions in (3). ri(a,b)={1,a=i,0,a≠i,cj(a,b)={1,b=j,0,b≠j.r_i(a,b)=\begin{cases}1,&a=i,\\0,&a\ne i,\end{cases} \qquad c_j(a,b)=\begin{cases}1,&b=j,\\0,&b\ne j.\end{cases}(3) The grid index set ΓI\Gamma_I is the disjoint union in (4), where RR and CC distinguish row labels from column labels. ΓI={0}⊔({R,C}×I×{−1,1}).\Gamma_I=\{0\}\sqcup\bigl(\{R,C\}\times I\times\{-1,1\}\bigr).(4)✦ The signed grid configuration g:ΓI→ℝI×Ig:\Gamma_I\to\mathbb R^{I\times I} is the map specified by (5). g0=0,g(R,i,σ)=σri,g(C,j,σ)=σcj.g_0=0,\qquad g_{(R,i,\sigma)}=\sigma r_i,\qquad g_{(C,j,\sigma)}=\sigma c_j.(5)✦

For a real exponent pp, write p+=max⁡{p,0}p_+=\max\{p,0\}. An admissible coordinate dimension dd for pp, a cardinality n∈ℕn\in\mathbb N and a distortion bound D∈ℝD\in\mathbb R is a natural number such that every injective family x1,…,xnx_1,\ldots,x_n in Lp+(Ω,μ)L_{p_+}(\Omega,\mu), for every measure space (Ω,μ)(\Omega,\mu), admits vectors y1,…,yn∈ℝdy_1,\ldots,y_n\in\mathbb R^d and a scale s>0s>0 satisfying (6). s∥xi−xj∥Lp≤ρp,{1,…,d}(yi,yj)≤Ds∥xi−xj∥Lp(1≤i,j≤n).s\|x_i-x_j\|_{L_p}\leq\rho_{p,\{1,\ldots,d\}}(y_i,y_j) \leq Ds\|x_i-x_j\|_{L_p}\qquad(1\leq i,j\leq n).(6)✦ For a real exponent pp, a cardinality n∈ℕn\in\mathbb N and a distortion bound D∈ℝD\in\mathbb R, the set 𝒜p(n,D)\mathcal A_p(n,D) consists of the admissible coordinate dimensions. The universal coordinate dimension dp(n,D)d_p(n,D) is the infimum in the natural numbers of 𝒜p(n,D)\mathcal A_p(n,D), as in (7). dp(n,D)=infℕ𝒜p(n,D).d_p(n,D)=\inf_{\mathbb N}\mathcal A_p(n,D).(7)✦

The Claim

Theorem 1. For p>2p>2 and nonzero scalars a,b∈ℝa,b\in\mathbb R, the positive defect assertion (8) holds: Δp(a,b)>0.\Delta_p(a,b)>0.(8)✦ When 0<p<20<p<2, nonzero scalars a,b∈ℝa,b\in\mathbb R satisfy (9). Δp(a,b)<0.\Delta_p(a,b)<0.(9)✦ For p>0p>0 with p≠2p\ne2 and arbitrary a,b∈ℝa,b\in\mathbb R, the zero defect criterion is (10). Δp(a,b)=0⇔a=0orb=0.\Delta_p(a,b)=0\quad\Longleftrightarrow\quad a=0\ \text{or}\ b=0.(10)✦

Let FF be a finite set and let u,v∈ℝFu,v\in\mathbb R^F. For p>0p>0 with p≠2p\ne2, the disjointness criterion is (11). Pp(u+v)+Pp(u−v)=2Pp(u)+2Pp(v)⇔supp⁡(u)∩supp⁡(v)=⌀.P_p(u+v)+P_p(u-v)=2P_p(u)+2P_p(v) \quad\Longleftrightarrow\quad \operatorname{supp}(u)\cap\operatorname{supp}(v)=\varnothing.(11)✦ For p>1p>1, vectors u,v∈ℝFu,v\in\mathbb R^F satisfying Pp(v)=Pp(u)P_p(v)=P_p(u) and Pp(u−v)=2pPp(u)P_p(u-v)=2^pP_p(u) have the antipodal conclusion (12). v=−u.v=-u.(12)✦

Let I,J,EI,J,E be finite sets and let (ri)i∈I(r_i)_{i\in I} and (cj)j∈J(c_j)_{j\in J} be families in ℝE\mathbb R^E with pairwise disjoint supports within each family. If the support of every rir_i meets the support of every cjc_j, their cardinalities satisfy (13). |I||J|≤|E|.|I|\,|J|\leq |E|.(13)✦

For finite sets I,EI,E and an exponent p>1p>1 with p≠2p\ne2, suppose f:ΓI→ℝEf:\Gamma_I\to\mathbb R^E satisfies f(0)=0f(0)=0 and Pp(f(α)−f(β))=Pp(gα−gβ)P_p(f(\alpha)-f(\beta))=P_p(g_\alpha-g_\beta) for every α,β∈ΓI\alpha,\beta\in\Gamma_I. The grid obstruction is (14). |I|2≤|E|.|I|^2\leq |E|.(14)✦

Fix p>1p>1 with p≠2p\ne2. For an integer n≥9n\geq9, every admissible exact dimension d∈𝒜p(n,1)d\in\mathcal A_p(n,1) satisfies (15). ⌊n−14⌋2≤d.\left\lfloor\frac{n-1}{4}\right\rfloor^2\leq d.(15)✦ For the same range of nn, the universal dimension has the bounds and admissibility assertion (16). dp(n,1)∈𝒜p(n,1),⌊n−14⌋2≤dp(n,1)≤(n2).d_p(n,1)\in\mathcal A_p(n,1),\qquad \left\lfloor\frac{n-1}{4}\right\rfloor^2 \leq d_p(n,1)\leq\binom n2.(16)✦ Along the natural numbers, its logarithmic limit is (17). limn→∞log⁡dp(n,1)log⁡n=2.\lim_{n\to\infty}\frac{\log d_p(n,1)}{\log n}=2.(17)✦

In the paper, the result reads: for 1<p<∞1<p<\infty with p≠2p\ne2, Δp(a,b)\Delta_p(a,b) vanishes exactly when ab=0ab=0 and otherwise has the sign of p−2p-2, so ∥u+v∥pp+∥u−v∥pp=2∥u∥pp+2∥v∥pp\|u+v\|_p^p+\|u-v\|_p^p=2\|u\|_p^p+2\|v\|_p^p holds exactly when u,vu,v have disjoint supports, while for k≥2k\geq2 the 4k+14k+1 points consisting of zero and the signed row and column indicators in ℓpk2\ell_p^{k^2} require at least k2k^2 coordinates for an exact embedding, and taking k=⌊(n−1)/4⌋k=\lfloor(n-1)/4\rfloor gives the exact lower bound for n≥9n\geq9.

What Is Checked

The statements concern real scalar defects and vectors on arbitrary finite coordinate sets, the signed grid map (5), and admissible dimensions (6) for injective finite families in real LpL_p spaces over arbitrary measure spaces.