OctoInformBeta

OpenAI preprint, September 2026 · §5. Contractions into ℓ₁

Proposition 5.1: An explicit m(log⁡m)5/2\sqrt{m}(\log m)^{5/2} integral bound on the uniform cube and its numerical constant inequality under chart-metric hypotheses

Average ℓ1\ell_1 displacement bound for chart cell functions.

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

Let mm be a positive integer dimension, let U=(−2,2)mU = (-2, 2)^m and T=(−1,1)mT = (-1, 1)^m, and for any finite set AA equip ℝA\mathbb{R}^A with the ℓ1\ell_1 norm ∥y∥1=∑a∈A|ya|\|y\|_1 = \sum_{a \in A} |y_a| (denoted ℓ1(A)\ell_1(A)). From m3m^3 families of m6m^6 directions in ℝm\mathbb{R}^m, one builds a vertex set VV of rounded coordinate cells on UU and equips ℝV\mathbb{R}^V with the ℓ2\ell_2 norm (denoted ℓ2(V)\ell_2(V)). Given a map Z:V→ℓ2(V)Z : V \to \ell_2(V) whose squared Euclidean distances ∥Z(v)−Z(w)∥2\|Z(v) - Z(w)\|^2 approximate a reference chart distance on VV and a map F:V→ℓ1(A)F : V \to \ell_1(A) satisfying ∥F(v)−F(w)∥1≤∥Z(v)−Z(w)∥2\|F(v) - F(w)\|_1 \le \|Z(v) - Z(w)\|^2, the question is how to bound the expected ℓ1\ell_1 distance between values of a chart cell function for FF at two independent uniform points of TT, and how to combine the explicit error terms into a single constant and weight in mm.

Mollifier and Metric Constants

For a real number xx and a natural number nn, the inverse power p(x,n)p(x, n) is the real number 1/xn1/x^n; in particular, p(m,2000)=m−2000p(m, 2000) = m^{-2000}, p(m,8)=m−8p(m, 8) = m^{-8}, and p(m,2)=m−2p(m, 2) = m^{-2}.✦ The smooth bump function ψ:ℝ→ℝ\psi : \mathbb{R} \to \mathbb{R} is a C∞C^\infty bump function centered at 00 with inner radius 1/21/2 and outer radius 11, equal to 11 on [−1/2,1/2][-1/2, 1/2] and supported in (−1,1)(-1, 1).✦ The standard mollifier ρ:ℝ→ℝ\rho : \mathbb{R} \to \mathbb{R} is the normalization of ψ\psi to unit Lebesgue integral, so that ρ\rho is smooth, nonnegative, supported in (−1,1)(-1, 1), and satisfies ∫ℝρ(t)dt=1\int_{\mathbb{R}} \rho(t)\,dt = 1.✦ The mollifier derivative integral HH is the real number ∫ℝ|ρ′(t)|dt\int_{\mathbb{R}} |\rho'(t)|\,dt✦, and at smoothing width λ∈ℝ\lambda \in \mathbb{R} the mollifier Lipschitz constant L(λ)L(\lambda) is the real number λ−1∫ℝ|ρ′(t)|dt\lambda^{-1}\int_{\mathbb{R}} |\rho'(t)|\,dt.✦

The profile function ϕ:ℝ→ℝ\phi : \mathbb{R} \to \mathbb{R} is the function defined by ϕ(t)=1−exp⁡(−1/(2t2))\phi(t) = 1 - \exp(-1/(2t^2))✦, and the profile constant c*c^* is the real number 2∫0∞ϕ(t)dt2\int_0^\infty \phi(t)\,dt.✦ In terms of c*c^* and HH, the source metric constant C0C_0 is the real number 9+2c*9 + 2c^*✦ and the contraction constant CC is the real number 2(c*+C0)+π(1.1×109+C0H)+12(c^* + C_0) + \pi(1.1 \times 10^9 + C_0 H) + 1.✦ For any natural number mm, the contraction weight w(m)w(m) is the real number m(log⁡m)2log⁡m\sqrt{m}(\log m)^2\sqrt{\log m}.✦

For natural numbers NN and SS and real parameters q,k,ε,a,b,R,M,Rposq, k, \varepsilon, a, b, R, M, R_{\mathrm{pos}}, the macro cost macroCost(N,S,q,k,ε,a,b,R,M,Rpos)\mathrm{macroCost}(N, S, q, k, \varepsilon, a, b, R, M, R_{\mathrm{pos}}) is the real number given by (1): macroCost(N,S,q,k,ε,a,b,R,M,Rpos)=q(2a+(2M)2b)+4εSN(1+kRε)2log⁡(ba)+2c*qRpos.\mathrm{macroCost}(N, S, q, k, \varepsilon, a, b, R, M, R_{\mathrm{pos}}) = q\left(2a + \frac{(2M)^2}{b}\right) + 4\varepsilon S N \left(1 + \frac{kR}{\varepsilon}\right)^2 \log\left(\frac{b}{a}\right) + 2 c^* q R_{\mathrm{pos}}.(1) For a natural number mm, the macro cost error err(m)\mathrm{err}(m) is the real number macroCost(m6,m3,m,p(m,5),p(m,200),p(m,40),m40,p(m,1997),3m,0)\mathrm{macroCost}(m^6, m^3, \sqrt{m}, p(m, 5), p(m, 200), p(m, 40), m^{40}, p(m, 1997), 3\sqrt{m}, 0).

Directions and Frame Charts

Identify ℝm\mathbb{R}^m with its Euclidean inner product ⟨⋅,⋅⟩\langle \cdot, \cdot \rangle and norm ∥⋅∥\|\cdot\|, indexing coordinates by 0≤j<m0 \le j < m. For a natural number NN, a tuple g=(gi)i=1N∈(ℝm)Ng = (g_i)_{i=1}^N \in (\mathbb{R}^m)^N, and vectors v,w∈ℝmv, w \in \mathbb{R}^m, the empirical covariance Cov(g,v)\mathrm{Cov}(g, v) of gg along vv is the real number 1N∑i=1N⟨v,gi⟩2\frac{1}{N}\sum_{i=1}^N \langle v, g_i \rangle^2✦, and the empirical sine transform Sin(g,w)\mathrm{Sin}(g, w) of gg at ww is the vector 1N∑i=1Nsin⁡(⟨gi,w⟩)gi\frac{1}{N}\sum_{i=1}^N \sin(\langle g_i, w \rangle) g_i in ℝm\mathbb{R}^m.✦ The bad projection set Bad(q,v)⊆(ℝm)N\mathrm{Bad}(q, v) \subseteq (\mathbb{R}^m)^N for a real threshold qq and a vector v∈ℝmv \in \mathbb{R}^m consists of all tuples x=(xi)i=1N∈(ℝm)Nx = (x_i)_{i=1}^N \in (\mathbb{R}^m)^N such that |⟨v,xi⟩|>q|\langle v, x_i \rangle| > q for some ii. Given a natural number SS, a subset 𝒜⊆(ℝm)N\mathcal{A} \subseteq (\mathbb{R}^m)^N, and a family y=(ys)s=1Sy = (y_s)_{s=1}^S in (ℝm)N(\mathbb{R}^m)^N, the number of bad samples badCount(𝒜,y)\mathrm{badCount}(\mathcal{A}, y) is the count of indices 1≤s≤S1 \le s \le S with ys∈𝒜y_s \in \mathcal{A}.✦

For a natural number mm, a direction family ff consists of a family g=(gs)s=1m3g = (g_s)_{s=1}^{m^3} of tuples gs=(gs,i)i=1m6g_s = (g_{s,i})_{i=1}^{m^6} in ℝm\mathbb{R}^m satisfying 12m≤∥gs,i∥≤2m\frac{1}{2}\sqrt{m} \le \|g_{s,i}\| \le 2\sqrt{m} for all ss and ii, nonparallelism gs,a≠tgs,bg_{s,a} \ne t g_{s,b} for all ss, a≠ba \ne b, and t∈ℝt \in \mathbb{R}, and covariance bounds 12≤Cov(gs,v)≤2\frac{1}{2} \le \mathrm{Cov}(g_s, v) \le 2 for every ss and unit vector v∈ℝmv \in \mathbb{R}^m, together with the Fourier bound ∥Sin(gs,w)−e−∥w∥2/2w∥≤2/m\|\mathrm{Sin}(g_s, w) - e^{-\|w\|^2/2}w\| \le 2/\sqrt{m} for every ss and every w∈ℝmw \in \mathbb{R}^m with ∥w∥≤m42\|w\| \le m^{42}, and the goodness bound badCount(Bad(11log⁡m,v),g)≤m3/2\mathrm{badCount}(\mathrm{Bad}(11\sqrt{\log m}, v), g) \le m^3/2 for every unit vector v∈ℝmv \in \mathbb{R}^m.✦ A pivoted direction family ff in dimension mm is a direction family satisfying the pivot condition |gs,i,j|>m−30|g_{s,i,j}| > m^{-30} for all ss, ii, and 0≤j<m0 \le j < m.✦ Associated with a direction family ff are the scaled directions us,i=gs,i/m∈ℝmu_{s,i} = g_{s,i}/\sqrt{m} \in \mathbb{R}^m✦, while the frame lower bound constant α_(m)\underline{\alpha}(m) is the real number m6/(2m)m^6/(2m).✦

Given a continuous linear map J:W1→W2J : W_1 \to W_2 between real inner-product spaces with adjoint J*:W2→W1J^* : W_2 \to W_1 and a constant a>0a > 0 satisfying a∥x∥2≤∥Jx∥2a\|x\|^2 \le \|Jx\|^2 for every x∈W1x \in W_1, the Gram operator J*J:W1→W1J^* J : W_1 \to W_1 is a bijective continuous linear operator✦, and the left pseudoinverse J+:W2→W1J^+ : W_2 \to W_1 is the continuous linear map (J*J)−1J*(J^* J)^{-1} J^*.✦ Equipping ℝN\mathbb{R}^N with the ℓ2\ell_2 norm (denoted ℓ2N\ell_2^N, with N=m6N = m^6), for a direction family ff in dimension mm with scaled directions us,iu_{s,i} and a chart index 1≤s≤m31 \le s \le m^3, the frame chart operator Js:ℝm→ℓ2NJ_s : \mathbb{R}^m \to \ell_2^N is the continuous linear map with components (Jsθ)i=⟨us,i,θ⟩(J_s \theta)_i = \langle u_{s,i}, \theta \rangle✦, and the chart pseudoinverse Js+:ℓ2N→ℝmJ_s^+ : \ell_2^N \to \mathbb{R}^m is the left pseudoinverse of JsJ_s.✦

Rounded Charts and Cell Functions

For a real radius RR, the cube QRQ_R of radius RR in ℝm\mathbb{R}^m is the open cube {θ∈ℝm:|θj|<R for all 0≤j<m}\{\theta \in \mathbb{R}^m : |\theta_j| < R \text{ for all } 0 \le j < m\}, so that Q1=TQ_1 = T and Q2=UQ_2 = U.✦ Let ν\nu be the uniform probability measure on (−1,1)(-1, 1), defined as Lebesgue measure restricted to (−1,1)(-1, 1) scaled by 1/21/2✦; the cube uniform measure μ\mu on ℝm\mathbb{R}^m is the mm-fold product of ν\nu across the coordinates of ℝm\mathbb{R}^m.✦

Given a natural number NN, a family u=(ui)i=1Ninℝmu = (u_i)_{i=1}^N in \mathbb{R}^m, and a threshold scale τ∈ℝ\tau \in \mathbb{R}, a point θ∈ℝm\theta \in \mathbb{R}^m is a regular chart angle if ⟨ui,θ⟩≠kτ\langle u_i, \theta \rangle \ne k\tau for all 1≤i≤N1 \le i \le N and k∈ℤk \in \mathbb{Z}✦, and the integer chart label Xu(θ)∈ℤNX_u(\theta) \in \mathbb{Z}^N of θ\theta is the tuple with entries Xu(θ)i=⌊⟨ui,θ⟩/τ⌋X_u(\theta)_i = \lfloor \langle u_i, \theta \rangle / \tau \rfloor.✦ For natural numbers SS and NN and SS families u=(us,i)u = (u_{s,i}) in ℝm\mathbb{R}^m, the rounded chart vertex set V(u,τ)V(u, \tau) is the set of all pairs (s,x)(s, x) with 1≤s≤S1 \le s \le S and x∈ℤNx \in \mathbb{Z}^N for which some regular chart angle θ∈Q2\theta \in Q_2 satisfies Xus(θ)=xX_{u_s}(\theta) = x✦; for a direction family ff with scaled directions uu, the source vertex set VV is the vertex set V(u,m−2000)V(u, m^{-2000}).✦ The rounded chart coordinate map coord⁡:V(u,τ)→ℓ2N\operatorname{coord} : V(u, \tau) \to \ell_2^N is the map sending each vertex v=(s,x)v = (s, x) to τx\tau x.✦

When m≥1m \ge 1, set Z=ℝm−1Z = \mathbb{R}^{m-1} with the Euclidean norm. The continuous coordinate split σ:ℝm→ℝ×Z\sigma : \mathbb{R}^m \to \mathbb{R} \times Z is the continuous linear isomorphism sending x=(x0,x1,…,xm−1)x = (x_0, x_1, \dots, x_{m-1}) to (x0,(x1,…,xm−1))(x_0, (x_1, \dots, x_{m-1})), with inverse prepending tt to the coordinates of zz for each (t,z)∈ℝ×Z(t, z) \in \mathbb{R} \times Z.✦ For u∈ℝmu \in \mathbb{R}^m with u0≠0u_0 \ne 0, the non-pivot linear functional restu:Z→ℝ\mathrm{rest}_u : Z \to \mathbb{R} is the continuous linear functional z↦⟨u,σ−1(0,z)⟩z \mapsto \langle u, \sigma^{-1}(0, z)\rangle✦, and the hyperplane slope functional βu:Z→ℝ\beta_u : Z \to \mathbb{R} is the continuous linear functional −u0−1restu-u_0^{-1}\mathrm{rest}_u.✦ Given a finite index set II and a family u=(ui)i∈Iu = (u_i)_{i \in I} in ℝm\mathbb{R}^m, the chart pivot coordinate ωu(i)\omega_u(i) is the real number ui,0u_{i,0}✦ and the hyperplane slope family βu\beta_u is the family (βui)i∈I(\beta_{u_i})_{i \in I}.✦

For a continuous linear functional β:Z→ℝ\beta : Z \to \mathbb{R}, the slice linear form formβ:ℝ×Z→ℝ\mathrm{form}_\beta : \mathbb{R} \times Z \to \mathbb{R} is defined by formβ(t,z)=t−β(z)\mathrm{form}_\beta(t, z) = t - \beta(z).✦ Given a finite index set II, a family β=(βi)i∈I\beta = (\beta_i)_{i \in I} of continuous linear functionals on ZZ, and pivot scalars ω:I→ℝ\omega : I \to \mathbb{R}, the pivoted slice functional ℓβ,ω,i:ℝ×Z→ℝ\ell_{\beta, \omega, i} : \mathbb{R} \times Z \to \mathbb{R} at i∈Ii \in I is the continuous linear functional ωiformβi\omega_i\,\mathrm{form}_{\beta_i}.✦ For a finite set KK, real offsets q:I×K→ℝq : I \times K \to \mathbb{R}, and a function GG assigning real numbers to subsets of I×KI \times K, the slice cell function cellβ,q,G:ℝ×Z→ℝ\mathrm{cell}_{\beta, q, G} : \mathbb{R} \times Z \to \mathbb{R} is the function sending (t,z)(t, z) to G({(i,k)∈I×K:formβi(t,z)+q(i,k)<0})G(\{(i, k) \in I \times K : \mathrm{form}_{\beta_i}(t, z) + q(i, k) < 0\}).✦

On any real normed space WW, given a finite index set II, continuous linear functionals ℓ=(ℓi)i∈I\ell = (\ell_i)_{i \in I} on WW, a scale τ∈ℝ\tau \in \mathbb{R}, and a subset Ω⊆W\Omega \subseteq W, a point x∈Wx \in W is an ℓ\ell-regular cell point if ℓi(x)≠kτ\ell_i(x) \ne k\tau for all i∈Ii \in I and k∈ℤk \in \mathbb{Z}✦, and its cell integer label labelℓ,τ(x)\mathrm{label}_{\ell,\tau}(x) is the function from II to ℤ\mathbb{Z} with entries labelℓ,τ(x)i=⌊ℓi(x)/τ⌋\mathrm{label}_{\ell,\tau}(x)_i = \lfloor \ell_i(x)/\tau \rfloor✦; the cell label vertex set Vℓ,τ(Ω)V_{\ell,\tau}(\Omega) consists of all functions l:I→ℤl : I \to \mathbb{Z} equal to labelℓ,τ(x)\mathrm{label}_{\ell,\tau}(x) for some ℓ\ell-regular point x∈Ωx \in \Omega.✦ Given pivot scalars ω:I→ℝ\omega : I \to \mathbb{R} and a finite set K⊂ℤK \subset \mathbb{Z}, the cell sign pattern Pℓ,ω,τ,K(x)P_{\ell,\omega,\tau,K}(x) of x∈Wx \in W is the subset {(i,k)∈I×K:(ℓi(x)−kτ)/ωi<0}\{(i, k) \in I \times K : (\ell_i(x) - k\tau)/\omega_i < 0\}✦, and for any set AA, the cell pattern extension extℓ,ω,τ,K,Ω\mathrm{ext}_{\ell,\omega,\tau,K,\Omega} is the map assigning to each map F:Vℓ,τ(Ω)→ℝAF : V_{\ell,\tau}(\Omega) \to \mathbb{R}^A a function extℓ,ω,τ,K,Ω(F)\mathrm{ext}_{\ell,\omega,\tau,K,\Omega}(F) from subsets of I×KI \times K to ℝA\mathbb{R}^A.✦ For a subset P⊆I×KP \subseteq I \times K, if some ℓ\ell-regular point x∈Ωx \in \Omega has cell sign pattern Pℓ,ω,τ,K(x)=PP_{\ell,\omega,\tau,K}(x) = P, then extℓ,ω,τ,K,Ω(F)(P)\mathrm{ext}_{\ell,\omega,\tau,K,\Omega}(F)(P) equals F(labelℓ,τ(x))F(\mathrm{label}_{\ell,\tau}(x)) (which is independent of the choice of xx), and otherwise extℓ,ω,τ,K,Ω(F)(P)=0\mathrm{ext}_{\ell,\omega,\tau,K,\Omega}(F)(P) = 0.

In ℝ×Z\mathbb{R} \times Z, the split cube domain Uˆ⊂ℝ×Z\widehat{U} \subset \mathbb{R} \times Z is the preimage σ(Q2)\sigma(Q_2) of Q2Q_2 under σ−1\sigma^{-1}✦, and at threshold scale τ\tau the chart label range KτK_\tau is the integer interval [−κ,κ]∩ℤ[-\kappa, \kappa] \cap \mathbb{Z} with κ=⌈4m+1/τ+1⌉\kappa = \lceil 4\sqrt{m+1}/\tau + 1 \rceil.✦ Given a finite index set II, a family u=(ui)i∈Iu = (u_i)_{i \in I} in ℝm\mathbb{R}^m, scale τ\tau, a finite set K⊂ℤK \subset \mathbb{Z}, a subset Ω⊆ℝ×Z\Omega \subseteq \mathbb{R} \times Z, a set AA, and a map F:Vℓβu,ωu,τ(Ω)→ℝAF : V_{\ell_{\beta_u,\omega_u},\tau}(\Omega) \to \mathbb{R}^A, the cell pattern values map valu,τ,K,Ω(F)\mathrm{val}_{u,\tau,K,\Omega}(F) is the function extℓβu,ωu,ωu,τ,K,Ω(F)\mathrm{ext}_{\ell_{\beta_u,\omega_u},\omega_u,\tau,K,\Omega}(F)✦, and for each a∈Aa \in A, the Euclidean chart function Ψu,τ,K,Ω(F,a):ℝm→ℝ\Psi_{u,\tau,K,\Omega}(F, a) : \mathbb{R}^m \to \mathbb{R} is the composition cellβu,qu,Ga∘σ\mathrm{cell}_{\beta_u, q_u, G_a} \circ \sigma (with offsets qu(i,k)=−kτ/ωu(i)q_u(i, k) = -k\tau/\omega_u(i) and Ga(P)=valu,τ,K,Ω(F)(P)aG_a(P) = \mathrm{val}_{u,\tau,K,\Omega}(F)(P)_a).✦ For natural numbers SS and NN, SS families u=(us,i)u = (u_{s,i}) in ℝm\mathbb{R}^m with us,i,0≠0u_{s,i,0} \ne 0, a threshold scale τ\tau, and a chart index 1≤s≤S1 \le s \le S, the cell-to-chart vertex map ιs:Vℓβus,ωus,τ(Uˆ)→V(u,τ)\iota_s : V_{\ell_{\beta_{u_s},\omega_{u_s}},\tau}(\widehat{U}) \to V(u, \tau) is the map sending each label ll to (s,l)(s, l), and for any map F:V(u,τ)→ℝAF : V(u, \tau) \to \mathbb{R}^A and index a∈Aa \in A, the concrete cell function Φu,τ,s,a(F):ℝm→ℝ\Phi_{u,\tau,s,a}(F) : \mathbb{R}^m \to \mathbb{R} is the function Ψus,τ,Kτ,Uˆ(F∘ιs,a)\Psi_{u_s, \tau, K_\tau, \widehat{U}}(F \circ \iota_s, a).✦

The Claim

Theorem 1. Let m≥3000m \ge 3000 be an integer satisfying log⁡m≥1\log m \ge 1 and H≤mH \le m, and let ff be a pivoted direction family in dimension mm with scaled directions u=(us,i)u = (u_{s,i}) and source vertex set V=V(u,m−2000)V = V(u, m^{-2000}). Suppose Z:V→ℓ2(V)Z : V \to \ell_2(V) satisfies the macroscopic approximation bound |∥Z(v)−Z(w)∥2−c*m∥Jsv+(coord⁡(v))−Jsw+(coord⁡(w))∥|≤err(m)+8m−8\bigl|\|Z(v) - Z(w)\|^2 - c^* \sqrt{m}\|J_{s_v}^+(\operatorname{coord}(v)) - J_{s_w}^+(\operatorname{coord}(w))\|\bigr| \le \mathrm{err}(m) + 8 m^{-8} for all v=(sv,xv)v = (s_v, x_v) and w=(sw,xw)w = (s_w, x_w) in VV, as well as the local step bound ∥Z(v)−Z(w)∥2≤5×107m−2000m−5(log⁡m)2\|Z(v) - Z(w)\|^2 \le 5 \times 10^7 m^{-2000} m^{-5} (\log m)^2 whenever sv=sws_v = s_w and ∑i=1m6|coord⁡(v)i−coord⁡(w)i|=m−2000\sum_{i=1}^{m^6} |\operatorname{coord}(v)_i - \operatorname{coord}(w)_i| = m^{-2000}. Let AA be a finite set and F:V→ℝAF : V \to \mathbb{R}^A be a map satisfying ∑a∈A|F(v)a−F(w)a|≤∥Z(v)−Z(w)∥2\sum_{a \in A} |F(v)_a - F(w)_a| \le \|Z(v) - Z(w)\|^2 for all v,w∈Vv, w \in V. Then for every chart index 1≤s0≤m31 \le s_0 \le m^3, the average ℓ1\ell_1 difference of the concrete cell function under μ⊗μ\mu \otimes \mu satisfies (2): ∫ℝm×ℝm∑a∈A|Φu,m−2000,s0,a(F)(θ)−Φu,m−2000,s0,a(F)(η)|d(μ⊗μ)(θ,η)≤Cw(m).\int_{\mathbb{R}^m \times \mathbb{R}^m} \sum_{a \in A} \bigl|\Phi_{u, m^{-2000}, s_0, a}(F)(\theta) - \Phi_{u, m^{-2000}, s_0, a}(F)(\eta)\bigr|\,d(\mu \otimes \mu)(\theta, \eta) \le C \, w(m).(2)✦ Furthermore, for every integer m≥1m \ge 1 satisfying log⁡m≥1\log m \ge 1, the constants satisfy the numerical inequality (3): 2(c*m(m−2m)+C0m−8)+π(11log⁡mm⋅m6⋅2(5×107m−2000m−5(log⁡m)2)m−2000+m(C0m−8)L(m−2))≤Cw(m).2\bigl(c^* \sqrt{m}(m^{-2}\sqrt{m}) + C_0 m^{-8}\bigr) + \pi \left( \frac{11\sqrt{\log m}}{\sqrt{m}} \cdot m^6 \cdot \frac{2\bigl(5 \times 10^7 m^{-2000} m^{-5} (\log m)^2\bigr)}{m^{-2000}} + m (C_0 m^{-8}) L(m^{-2}) \right) \le C \, w(m).(3)✦

In the paper, the result reads: for the metric space (V,d)(V, d) and chart maps vs:U→Vv_s : U \to V from Proposition 4.2, there is an absolute constant C′C' such that every map F:V→ℓ1DF : V \to \ell_1^D (for every finite DD) satisfying ∥F(v)−F(w)∥1≤d(v,w)\|F(v) - F(w)\|_1 \le d(v, w) obeys 𝔼θ,η∈T∥F(v1(θ))−F(v1(η))∥1≤C′m(log⁡m)5/2\mathbb{E}_{\theta, \eta \in T} \|F(v_1(\theta)) - F(v_1(\eta))\|_1 \le C' \sqrt{m} (\log m)^{5/2}, where T=(−1,1)mT = (-1, 1)^m and θ,η\theta, \eta are independent and uniform, and here mm, UU, TT, VV, and FF stand for the paper’s symbols while ℓ1(A)\ell_1(A) stands for ℓ1D\ell_1^D. Rather than using the chart metric dd and the composition F∘v1F \circ v_1 on chart 11, (2) holds for any chart s0s_0 via a map ZZ satisfying the macroscopic and local chart bounds and the concrete cell function of FF, accompanied by the numerical inequality (3).

What Is Checked

The integral bound (2) is stated over an integer m≥3000m \ge 3000 satisfying log⁡m≥1\log m \ge 1 and H≤mH \le m, a pivoted direction family ff in dimension mm, a map Z:V→ℓ2(V)Z : V \to \ell_2(V) obeying the macroscopic approximation bound against c*m∥Jsv+(coord⁡(v))−Jsw+(coord⁡(w))∥c^*\sqrt{m}\|J_{s_v}^+(\operatorname{coord}(v)) - J_{s_w}^+(\operatorname{coord}(w))\| and the local unit-step bound for vertices v=(sv,xv)v = (s_v, x_v) and w=(sw,xw)w = (s_w, x_w) in VV, a finite set AA with a map F:V→ℝAF : V \to \mathbb{R}^A whose ℓ1\ell_1 increments are dominated by ∥Z(v)−Z(w)∥2\|Z(v) - Z(w)\|^2, and any chart index s0s_0, while the numerical bound (3) is stated over any integer m≥1m \ge 1 satisfying log⁡m≥1\log m \ge 1. For any positive integer mm and any pivoted direction family ff in dimension mm, the frame lower bound constant α_(m)\underline{\alpha}(m) is strictly positive✦, each frame chart operator JsJ_s satisfies the energy lower bound α_(m)∥x∥2≤∥Jsx∥2\underline{\alpha}(m)\|x\|^2 \le \|J_s x\|^2 for all x∈ℝmx \in \mathbb{R}^m, and every scaled direction coordinate satisfies |us,i,j|>1/(m30m)|u_{s,i,j}| > 1/(m^{30}\sqrt{m}) for all 1≤s≤m31 \le s \le m^3, 1≤i≤m61 \le i \le m^6, and 0≤j<m0 \le j < m.✦