OctoInformBeta

OpenAI preprint, September 2026 · §5. Uniform control of returning geodesics

Lemma 5.2 [Returning time and the strict margin]: The squared height velocity obeys a chord bound, and the timing estimate holds in either endpoint frame

Deep returns have a strict timing margin.

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

Consider a geodesic entering the neck and returning to the same endpoint. Write s∈[0,L]s\in[0,L] for geodesic time, L>0L>0 for the return duration, and t(s)t(s) for its inward height, with height velocity v(s)v(s). The initial velocity and the duration are compared through the cometric in the chosen endpoint frame. Transverse vectors, including the conserved momentum p=(p0,p1)p=(p_0,p_1), belong to ℝ2\mathbb R^2, with coordinates indexed by 0,10,1; transverse matrices are real 2×22\times2 matrices.

For a neck parameter δ∈ℝ\delta\in\mathbb R, the one-sided cutoff aa and its integrated cutoff AA are the real functions defined in (1): a(u)={exp⁡(−1/(2/3−u)),u<2/3,0,u≥2/3,A(u)=∫0ua(y)dy.a(u)= \begin{cases} \exp\!\bigl(-1/(2/3-u)\bigr),&u<2/3,\\ 0,&u\geq2/3, \end{cases} \qquad A(u)=\int_0^u a(y)\,dy.(1) Let η\eta denote the standard smooth transition from zero to one. The shear transition χ\chi is the real function in (2); the diagonal coefficients d1,d2d_1,d_2 and cometric shear σ\sigma are the real functions of neck height tt and parameter δ\delta defined there: χ(t)=η(6(t−5/12)),σ(δ,t)=δ2χ(t),d1(δ,t)=1+δA(t),d2(δ,t)=1+δA(1−t).\begin{aligned} \chi(t)&=\eta\bigl(6(t-5/12)\bigr),& \sigma(\delta,t)&=\delta^2\chi(t),\\ d_1(\delta,t)&=1+\delta A(t),& d_2(\delta,t)&=1+\delta A(1-t). \end{aligned}(2) The cometric matrix Pδ(t)P_\delta(t) is the real 2×22\times2 matrix of (3), where the coefficients on the right are evaluated at (δ,t)(\delta,t): Pδ(t)=(d1+d2σ2d2σd2σd2).P_\delta(t)= \begin{pmatrix} d_1+d_2\sigma^2&d_2\sigma\\ d_2\sigma&d_2 \end{pmatrix}.(3)✦

Write τ\tau for inward endpoint height and choose r=0r=0 for the left frame or r=1r=1 for the right frame. The endpoint frame matrix C2(δ)C_2(\delta), terminal cometric Pend,δP_{\mathrm{end},\delta} and coordinate cometric Pr,δP_{r,\delta} are the matrix and matrix-valued functions in (4): C2(δ)=(011−δ2),Pend,δ(τ)=C2(δ)Pδ(1−τ)C2(δ)𝖳,Pr,δ(τ)={Pδ(τ),r=0,Pend,δ(τ),r=1.\begin{aligned} C_2(\delta)&=\begin{pmatrix}0&1\\1&-\delta^2\end{pmatrix},\\ P_{\mathrm{end},\delta}(\tau) &=C_2(\delta)P_\delta(1-\tau)C_2(\delta)^{\mathsf T},\\ P_{r,\delta}(\tau)&= \begin{cases} P_\delta(\tau),&r=0,\\ P_{\mathrm{end},\delta}(\tau),&r=1. \end{cases} \end{aligned}(4)

For a transverse matrix BB, its matrix bilinear pairing ℬ\mathcal B and associated quadratic form qBq_B are the real-valued maps in (5), with u,w,p∈ℝ2u,w,p\in\mathbb R^2: ℬ(B,u,w)=u𝖳Bw,qB(p)=p𝖳Bp.\mathcal B(B,u,w)=u^{\mathsf T}Bw, \qquad q_B(p)=p^{\mathsf T}Bp.(5)

For fixed r,δ,pr,\delta,p, the momentum potential VpV_p and reduced force FpF_p are the real functions of inward height defined by (6); primes on Pr,δP_{r,\delta} denote differentiation in height: Vp(τ)=ℬ(Pr,δ(τ),p,p),Fp(τ)=ℬ(Pr,δ′(τ),p,p).V_p(\tau)=\mathcal B(P_{r,\delta}(\tau),p,p), \qquad F_p(\tau)=\mathcal B(P'_{r,\delta}(\tau),p,p).(6)

Cometric concavity is the condition (7) on the coordinate cometric: q−Pr,δ″(u)(q)≥0(u∈[0,1],q∈ℝ2).q_{-P''_{r,\delta}(u)}(q)\geq0 \qquad (u\in[0,1],\ q\in\mathbb R^2).(7)

The turning constant a0a_0 and comparison constant β\beta are the real numbers in (8): a0=e−3/2,β=1+e−9/262.a_0=e^{-3/2},\qquad \beta=\frac{1+e^{-9/26}}2.(8)

The Claim

Theorem 1 (Returning time and the strict margin). Let L>0L>0 and v0≥0v_0\geq0. Suppose v:ℝ→ℝv:\mathbb R\to\mathbb R is continuous on [0,L][0,L], convex and nonnegative on [0,L/2][0,L/2], with v(0)=v0v(0)=v_0 and v(L/2)=0v(L/2)=0. Assume v(L−s)=−v(s)v(L-s)=-v(s) for s∈[0,L]s\in[0,L]. Then ∫0Lv(s)2ds≤Lv023.\int_0^L v(s)^2\,ds\leq\frac{Lv_0^2}{3}.(9)✦

Let v0∈ℝv_0\in\mathbb R, T>1/4T>1/4 and L>0L>0. Suppose t:ℝ→ℝt:\mathbb R\to\mathbb R is continuous on [0,L][0,L], takes values in [0,T][0,T] there, and is concave on [0,L/2][0,L/2], with t(0)=0t(0)=0, t(L/2)=Tt(L/2)=T and t(L−s)=t(s)t(L-s)=t(s) for s∈[0,L]s\in[0,L]. Let F1:ℝ→ℝF_1:\mathbb R\to\mathbb R be continuous and nonincreasing on [0,T][0,T]. Suppose v:ℝ→ℝv:\mathbb R\to\mathbb R is differentiable at every s∈[0,L]s\in[0,L], with v′(s)=−F1(t(s))/2v'(s)=-F_1(t(s))/2, v(0)=v0v(0)=v_0 and v(L)=−v0v(L)=-v_0. Then 4v0≤L2(F1(0)+F1(1/8)).4v_0\leq \frac L2\bigl(F_1(0)+F_1(1/8)\bigr).(10)✦

Fix either endpoint frame, δ∈ℝ\delta\in\mathbb R, p∈ℝ2p\in\mathbb R^2 and L>0L>0. Assume cometric concavity (7). Let t,v:ℝ→ℝt,v:\mathbb R\to\mathbb R be differentiable at every s∈[0,L]s\in[0,L], with t′(s)=v(s)t'(s)=v(s) and v′(s)=−Fp(t(s))/2v'(s)=-F_p(t(s))/2, and put T=t(L/2)T=t(L/2).

Suppose t(0)=t(L)=0t(0)=t(L)=0, 0<t(s)<10<t(s)<1 for 0<s<L0<s<L, and Vp(t(s))+v(s)2=1V_p(t(s))+v(s)^2=1 for s∈[0,L]s\in[0,L]. Then $$\begin{aligned} &v(0)>0,\qquad v(L/2)=0,\qquad 0<T<1, \qquad t([0,L])\subseteq[0,T],\\ &t(L-s)=t(s),\qquad v(L-s)=-v(s) &&(0\leq s\leq L),\\ &F_p(u)>0 &&(0\leq u\leq T),\\ &v\text{ is convex and }t\text{ is concave on }[0,L/2],\\ &\text{\eqref{eq:claim1} holds with }v_0=v(0). \end{aligned}$$(11)✦

Suppose 1/4<T<11/4<T<1, t(0)=0t(0)=0, t([0,L])⊆[0,T]t([0,L])\subseteq[0,T], and tt is concave on [0,L/2][0,L/2]. Assume t(L−s)=t(s)t(L-s)=t(s) and v(L−s)=−v(s)v(L-s)=-v(s) for s∈[0,L]s\in[0,L]. Then 4v(0)≤βδa0p02L.4v(0)\leq\beta\,\delta a_0\,p_0^2 L.(12)✦ The turning comparison constant satisfies β<1.\beta<1.(13)✦

What Is Checked

The scalar assertions (9), (10) and (13) concern real functions and the explicit numerical constant β\beta; the dynamics and timing assertions (11) and (12) concern the coordinate cometric with its reduced equations and cometric concavity.