OctoInformBeta

OpenAI preprint, September 2026 · Appendix A. An explicit map with positive displacement

A (proof of Proposition A.1): correction and normalization: Radial cutoffs and tangent rotations produce a Lipschitz map before negative radial normalization

A correction bounded away from zero.

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

The question is how to correct the identity in a thin curve tube so that negative radial normalization is defined throughout a Hilbert ball. Let EE be a real Hilbert space. The closed unit ball BEB_E of the real Hilbert space EE, the concrete Hilbert space HH, and the closed unit ball BB of HH are defined by (1): BE={x∈E:∥x∥≤1},H=L2([0,1];ℝ2),B=BH.B_E=\{x\in E:\|x\|\leq1\},\qquad H=L^2([0,1];\mathbb R^2),\qquad B=B_H.(1) Here ℝ2\mathbb R^2 is the Euclidean plane, and the measure on [0,1][0,1] is Lebesgue measure restricted to that closed interval. Elements of HH are functions identified when they agree almost everywhere.

For real zz, the circle parametrization p(z)=(cos⁡z,sin⁡z)p(z)=(\cos z,\sin z) and its tangent p′(z)=(−sin⁡z,cos⁡z)p'(z)=(-\sin z,\cos z) define the oscillator W(z)∈HW(z)\in H and oscillator velocity W′(z)∈HW'(z)\in H by the representatives in (2), for w∈[0,1]w\in[0,1]: W(z)(w)=p(zw),W′(z)(w)=wp′(zw).W(z)(w)=p(zw),\qquad W'(z)(w)=w p'(zw).(2) For every real zz, the oscillator representative w↦p(zw)w\mapsto p(zw) is square-integrable with respect to Lebesgue measure on [0,1][0,1]. For every real zz, the oscillator velocity representative w↦wp′(zw)w\mapsto wp'(zw) is square-integrable with respect to Lebesgue measure on [0,1][0,1].

The Oscillating Curve

The radial profile a:ℝ→ℝa:\mathbb R\to\mathbb R and angular profile θ:ℝ→ℝ\theta:\mathbb R\to\mathbb R are the functions in (3): a(t)={t/8,t≤2,5/16−(3−t)2/16,2<t<3,5/16,t≥3,θ(t)={0,t≤1,(t−1)2/2,1<t<2,t−3/2,t≥2.\begin{aligned} a(t)&=\begin{cases} t/8,&t\leq2,\\ 5/16-(3-t)^2/16,&2<t<3,\\ 5/16,&t\geq3, \end{cases} &\theta(t)&=\begin{cases} 0,&t\leq1,\\ (t-1)^2/2,&1<t<2,\\ t-3/2,&t\geq2. \end{cases} \end{aligned}(3) The radial derivative a′:ℝ→ℝa':\mathbb R\to\mathbb R of aa and the angular derivative θ′:ℝ→ℝ\theta':\mathbb R\to\mathbb R of θ\theta are the functions in (4): a′(t)={1/8,t≤2,(3−t)/8,2<t<3,0,t≥3,θ′(t)={0,t≤1,t−1,1<t<2,1,t≥2.\begin{aligned} a'(t)&=\begin{cases} 1/8,&t\leq2,\\ (3-t)/8,&2<t<3,\\ 0,&t\geq3, \end{cases} &\theta'(t)&=\begin{cases} 0,&t\leq1,\\ t-1,&1<t<2,\\ 1,&t\geq2. \end{cases} \end{aligned}(4) The radial direction v:ℝ→Hv:\mathbb R\to H, radial oscillating curve γ:ℝ→H\gamma:\mathbb R\to H, its velocity γ′\gamma', and unit tangent uu are defined by (5): v(t)=W(θ(t)),γ(t)=a(t)v(t),γ′(t)=a′(t)v(t)+a(t)θ′(t)W′(θ(t)),u(t)=∥γ′(t)∥−1γ′(t).\begin{gathered} v(t)=W(\theta(t)),\qquad \gamma(t)=a(t)v(t),\\ \gamma'(t)=a'(t)v(t)+a(t)\theta'(t)W'(\theta(t)),\qquad u(t)=\|\gamma'(t)\|^{-1}\gamma'(t). \end{gathered}(5)

Regular separated curve data C=(c,w,τ,σ)C=(c,w,\tau,\sigma) in EE consist of functions c,w,τ:ℝ→Ec,w,\tau:\mathbb R\to E and a real constant σ\sigma satisfying (6) for real s,ts,t: c′(t)=w(t),∥τ(t)∥=1,w(t)=∥w(t)∥τ(t),∥w(t)∥≥1/8,∥w(t)−w(s)∥≤|t−s|,∥τ(t)−τ(s)∥≤16|t−s|,σ>0,∥c(t)−c(s)∥≥σwhen |t−s|≥1/32.\begin{gathered} c'(t)=w(t),\qquad \|\tau(t)\|=1,\qquad w(t)=\|w(t)\|\tau(t),\qquad \|w(t)\|\geq1/8,\\ \|w(t)-w(s)\|\leq|t-s|,\qquad \|\tau(t)-\tau(s)\|\leq16|t-s|,\qquad \sigma>0,\\ \|c(t)-c(s)\|\geq\sigma\quad\text{when }|t-s|\geq1/32. \end{gathered}(6)✦ For the oscillating curve, choose a separation constant σγ\sigma_\gamma at parameter scale 1/321/32, and write Cγ=(γ,γ′,u,σγ)C_\gamma=(\gamma,\gamma',u,\sigma_\gamma) for its curve data.

Tube Coordinates and Rotations

For regular separated curve data CC, the admissible radius ρC\rho_C and open curve tube TC⊆ET_C\subseteq E are defined by (7): ρC=min⁡{1/1024,σ/8},TC={x∈E:dist⁡(x,c(ℝ))<ρC}.\rho_C=\min\{1/1024,\sigma/8\},\qquad T_C=\{x\in E:\operatorname{dist}(x,c(\mathbb R))<\rho_C\}.(7) A nearest curve parameter for x∈Ex\in E is a real number ss satisfying (8): ∥x−c(s)∥≤∥x−c(t)∥(t∈ℝ).\|x-c(s)\|\leq\|x-c(t)\|\qquad(t\in\mathbb R).(8)✦ For x∈TCx\in T_C, let tC(x)t_C(x) be its unique nearest parameter.✦ The normal residual eC(x)∈Ee_C(x)\in E is the vector defined in (9): eC(x)=x−c(tC(x)).e_C(x)=x-c(t_C(x)).(9)✦ Write ρ=ρCγ\rho=\rho_{C_\gamma} for the oscillating curve tube radius.✦

For vectors ξ,ζ∈E\xi,\zeta\in E, with II denoting the identity operator, the skew rank-two operator Kξ,ζK_{\xi,\zeta} and the rotation Qξ,ζQ_{\xi,\zeta} are the continuous real-linear operators in (10); the reciprocal is assigned value zero when its denominator is zero: Kξ,ζx=⟨ξ,x⟩ζ−⟨ζ,x⟩ξ,Qξ,ζ=I+Kξ,ζ+(1+⟨ξ,ζ⟩)−1Kξ,ζ2,R(t)=Qu(t),v(t).\begin{aligned} K_{\xi,\zeta}x&=\langle\xi,x\rangle\zeta-\langle\zeta,x\rangle\xi,\\ Q_{\xi,\zeta}&=I+K_{\xi,\zeta} +(1+\langle\xi,\zeta\rangle)^{-1}K_{\xi,\zeta}^{\,2},\qquad R(t)=Q_{u(t),v(t)}. \end{aligned}(10) Here R(t)R(t) is the tangent rotation on HH.

The Radial Correction

The radial coefficient b:ℝ→ℝb:\mathbb R\to\mathbb R is defined by (11): b(t)=min⁡{a(t),−1/8}.b(t)=\min\{a(t),-1/8\}.(11)✦ The radial correction q:ℝ→ℝq:\mathbb R\to\mathbb R is the function in (12): q(t)=−max⁡{a(t)+1/8,0}.q(t)=-\max\{a(t)+1/8,0\}.(12)✦ Define the inner cutoff β:ℝ2→ℝ\beta:\mathbb R^2\to\mathbb R by (13): β(ρ0,r)=min⁡{1,max⁡{0,2−4r/ρ0}}.\beta(\rho_0,r)=\min\{1,\max\{0,2-4r/\rho_0\}\}.(13)✦ The outer cutoff χ:ℝ2→ℝ\chi:\mathbb R^2\to\mathbb R is defined by (14): χ(ρ0,r)=min⁡{1,max⁡{0,2−2r/ρ0}}.\chi(\rho_0,r)=\min\{1,\max\{0,2-2r/\rho_0\}\}.(14)✦ In these cutoff definitions division by zero has value zero.

Radial correction data D=(α,ν,δ,ℛ)D=(\alpha,\nu,\delta,\mathcal R) along CC consist of functions α,δ:ℝ→ℝ\alpha,\delta:\mathbb R\to\mathbb R and ν:ℝ→E\nu:\mathbb R\to E, and a family ℛ(t)\mathcal R(t) of continuous real-linear operators on EE, satisfying (15) for real s,ts,t and vectors e∈Ee\in E: c(t)=α(t)ν(t),α(0)=0,α(t)≤5/16,α(t)≥1/8(t>1),δ(t)=min⁡{α(t),−1/8}−α(t),|δ(t)|≤7/16,|δ(t)−δ(s)|≤|t−s|/8,∥ν(t)∥=1,∥ν(t)−ν(s)∥≤|t−s|,∥c(t)−c(s)∥≤(7/16)|t−s|,∥ℛ(t)e∥=∥e∥,⟨ℛ(t)e,ν(t)⟩=0when ⟨e,τ(t)⟩=0,∥ℛ(t)e−ℛ(s)e∥≤238|t−s|∥e∥,ℛ(t)=I,τ(t)=ν(t)(t≤1).\begin{gathered} c(t)=\alpha(t)\nu(t),\qquad \alpha(0)=0,\qquad \alpha(t)\leq5/16, \qquad \alpha(t)\geq1/8\quad(t>1),\\ \delta(t)=\min\{\alpha(t),-1/8\}-\alpha(t),\qquad |\delta(t)|\leq7/16,\qquad |\delta(t)-\delta(s)|\leq|t-s|/8,\\ \|\nu(t)\|=1,\qquad \|\nu(t)-\nu(s)\|\leq|t-s|,\qquad \|c(t)-c(s)\|\leq(7/16)|t-s|,\\ \|\mathcal R(t)e\|=\|e\|,\qquad \langle\mathcal R(t)e,\nu(t)\rangle=0 \quad\text{when }\langle e,\tau(t)\rangle=0,\\ \|\mathcal R(t)e-\mathcal R(s)e\|\leq238|t-s|\|e\|,\\ \mathcal R(t)=I,\qquad \tau(t)=\nu(t)\quad(t\leq1). \end{gathered}(15)✦ The oscillating curve correction data are denoted by Dγ=(a,v,q,R)D_\gamma=(a,v,q,R).✦

For x∈TCx\in T_C, put t=tC(x)t=t_C(x), e=eC(x)e=e_C(x) and r=∥e∥r=\|e\|. The additive tube correction ΔD:TC→E\Delta_D:T_C\to E is the function in (16): ΔD(x)=β(ρC,r)δ(t)ν(t)+χ(ρC,r)(ℛ(t)e−e).\Delta_D(x)=\beta(\rho_C,r)\delta(t)\nu(t) +\chi(\rho_C,r)(\mathcal R(t)e-e).(16)✦ The corrected tube map PD:TC→EP_D:T_C\to E, glued correction GD:E→EG_D:E\to E, and unit ball correction G:B→HG:B\to H are the functions defined in (17): PD(x)=x+ΔD(x),GD(x)={PD(x),x∈TC,x,x∉TC,G=GDγ|B.P_D(x)=x+\Delta_D(x),\qquad G_D(x)=\begin{cases}P_D(x),&x\in T_C,\\x,&x\notin T_C,\end{cases} \qquad G=G_{D_\gamma}|_B.(17) The correction Lipschitz constant LCL_C is the real number in (18): LC=max⁡{1,max{30+74ρC+3808ρC,5+78ρC}}.L_C=\max\left\{1,\max\left\{ 30+\frac7{4\rho_C}+3808\rho_C,\ 5+\frac7{8\rho_C}\right\}\right\}.(18)✦ The negative radial normalization N:E→EN:E\to E is defined by (19): N(z)={−z/∥z∥,z≠0,0,z=0.N(z)=\begin{cases}-z/\|z\|,&z\ne0,\\0,&z=0.\end{cases}(19)✦

The Claim

Theorem 1. Let CC be regular separated curve data in a real Hilbert space EE, and let DD be radial correction data along CC. For x,y∈Ex,y\in E, the glued correction has the estimate (20): ∥GD(x)−GD(y)∥≤LC∥x−y∥.\|G_D(x)-G_D(y)\|\leq L_C\|x-y\|.(20)✦ For the oscillating curve correction and x,y∈Bx,y\in B, the norm estimate is (21) and the difference estimate is (23); under the additional assumption ∥x∥≥1/2\|x\|\geq1/2, its value is given by (22): ∥G(x)∥≥ρ/4.\|G(x)\|\geq\rho/4.(21)✦ G(x)=x.G(x)=x.(22)✦ ∥G(x)−G(y)∥≤LCγ∥x−y∥.\|G(x)-G(y)\|\leq L_{C_\gamma}\|x-y\|.(23)✦ Let ε>0\varepsilon>0, and let z,z′∈Ez,z'\in E have norms at least ε\varepsilon. The normalization estimate is (24): ∥N(z)−N(z′)∥≤2ε∥z−z′∥.\|N(z)-N(z')\|\leq\frac2\varepsilon\|z-z'\|.(24)✦ For P:BE→EP:B_E\to E with ∥P(x)∥≥ε\|P(x)\|\geq\varepsilon for every x∈BEx\in B_E, the normalized nonvanishing map fP:BE→BEf_P:B_E\to B_E is the function defined by (25): fP(x)=N(P(x)),∥fP(x)∥=1(x∈BE).f_P(x)=N(P(x)),\qquad \|f_P(x)\|=1\quad(x\in B_E).(25)✦

In the paper, the result reads: with radius ρ\rho and rotations R(t)R(t), put b(t)=min⁡{a(t),−1/8}b(t)=\min\{a(t),-1/8\} and q(t)=b(t)−a(t)q(t)=b(t)-a(t), let β(ρ,r)\beta(\rho,r) equal 11 up to ρ/4\rho/4, decrease linearly to 00 at ρ/2\rho/2, and vanish thereafter, and let χ(ρ,r)\chi(\rho,r) equal 11 up to ρ/2\rho/2, decrease linearly to 00 at ρ\rho, and vanish thereafter; set G(x)=xG(x)=x outside the tube and, inside it, with nearest parameter t=t(x)t=t(x), residual e=x−γ(t)e=x-\gamma(t) and r=∥e∥r=\|e\|, set G(x)=x+β(ρ,r)q(t)v(t)+χ(ρ,r)(R(t)−I)eG(x)=x+\beta(\rho,r)q(t)v(t)+\chi(\rho,r)(R(t)-I)e; then GG is globally Lipschitz on BB, has ∥G(x)∥≥ρ/4\|G(x)\|\geq\rho/4, and satisfies G(x)=xG(x)=x when ∥x∥≥1/2\|x\|\geq1/2, while negative radial normalization f(x)=−G(x)/∥G(x)∥f(x)=-G(x)/\|G(x)\| yields Proposition A.1.

What Is Checked

The ambient estimate is stated over regular separated curves and radial correction data in a real Hilbert space; the ball estimates use the oscillating curve in HH, and the normalization construction uses a positive uniform lower bound on the norm of a map on BEB_E. The oscillating curve data CγC_\gamma satisfy the regular separated curve conditions in (6), and the oscillating curve correction data DγD_\gamma satisfy the radial correction conditions in (15).