OctoInformBeta

OpenAI preprint, September 2026 · §4. A localized coefficient estimate

Lemma 3: Circle actions and seam-compatible fields with uniformly bounded spatial derivatives

Localized bounds for angular coefficient energy.

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

Angular coefficients on the billiard double have geodesic and transverse energies. Localization singles out a measurable set of base points outside which the geodesic spatial derivative vanishes.

A nondegenerate triangle QQ consists of three affinely independent vertices qi∈ℂq_i\in\mathbb C, with indices taken modulo three.✦ For a nondegenerate triangle QQ with vertices qiq_i, the open billiard table Ω\Omega is the planar domain, the open sides sis_i are the segments, and the side tangents eie_i are the vectors given by (1): Ω=int⁡conv⁡{q0,q1,q2},si=(qi,qi+1),ei=qi+1−qi.\Omega=\operatorname{int}\operatorname{conv}\{q_0,q_1,q_2\},\qquad s_i=(q_i,q_{i+1}),\qquad e_i=q_{i+1}-q_i.(1) Here (qi,qi+1)(q_i,q_{i+1}) is the open line segment. For a nonzero complex tangent ee, specular reflection refl⁡e:ℂ→ℂ\operatorname{refl}_e:\mathbb C\to\mathbb C, affine side reflection Ri:ℂ→ℂR_i:\mathbb C\to\mathbb C, and reflected unit velocity ρi:S1→S1\rho_i:S^1\to S^1 are the maps in (2): refl⁡e(v)=ee¯v¯,Ri(x)=qi+refl⁡ei(x−qi),ρi(v)=refl⁡ei(v).\operatorname{refl}_e(v)=\frac{e}{\overline e}\,\overline v,\qquad R_i(x)=q_i+\operatorname{refl}_{e_i}(x-q_i),\qquad \rho_i(v)=\operatorname{refl}_{e_i}(v).(2)

The phase space is Phase=ℂ×S1\mathrm{Phase}=\mathbb C\times S^1, recording position and physical unit velocity.✦ The double phase space is 𝒟=Phase×ℤ/2ℤ\mathcal D=\mathrm{Phase}\times\mathbb Z/2\mathbb Z, with sheet coordinate bb.✦ The uniform angular probability σ\sigma, normalized area-angular probability μ\mu, and double probability ν\nu are the measures in (3), where dAdA is planar area and each sheet has mass 1/21/2: σ=(θ↦eiθ)*(𝟏[0,2π)dθ2π),μ=𝟏ΩdAA(Ω)⊗σ,ν=μ⊗δ0+δ12.\sigma=(\theta\mapsto e^{i\theta})_* \left(\frac{\mathbf1_{[0,2\pi)}\,d\theta}{2\pi}\right),\qquad \mu=\frac{\mathbf1_\Omega\,dA}{A(\Omega)}\otimes\sigma,\qquad \nu=\mu\otimes\frac{\delta_0+\delta_1}{2}.(3) The double Hilbert space is HQ=L2(𝒟,ν;ℂ)H_Q=L^2(\mathcal D,\nu;\mathbb C), with functions identified almost everywhere.✦

A complete regular billiard trajectory through (x,v)(x,v) is a bi-infinite sequence of collision times τn\tau_n, collision points pn∈sinp_n\in s_{i_n}, outgoing unit velocities vnv_n, and side labels ini_n. The times strictly increase, are unbounded in both directions, and satisfy τ0<0<τ1\tau_0<0<\tau_1. Its flights lie in Ω\Omega strictly between collisions and satisfy (4), including the initial state and the specular collision law: pn+1=pn+(τn+1−τn)vn,vn=ρin(vn−1),x=p0−τ0v0,v=v0.\begin{gathered} p_{n+1}=p_n+(\tau_{n+1}-\tau_n)v_n,\qquad v_n=\rho_{i_n}(v_{n-1}),\\ x=p_0-\tau_0v_0,\qquad v=v_0. \end{gathered}(4) The billiard flow Φt\Phi_t follows the unique complete regular trajectory, using its outgoing velocity at a collision, and fixes a state with no such trajectory.✦ For a regular state zz, let n(t,z)n(t,z) be the index of the flight reached at time tt, with the initial flight indexed by zero.✦ On regular states the double geodesic flow Ψt\Psi_t is given by (5), together with the oriented tangent rotation ℛθ\mathcal R_\theta; sheet addition is modulo two: Ψt(z,b)=(Φtz,b+n(t,z)),ℛθ((x,v),b)=((x,e(−1)biθv),b).\Psi_t(z,b)=\bigl(\Phi_tz,b+n(t,z)\bigr),\qquad \mathcal R_\theta((x,v),b) =\bigl((x,e^{(-1)^b i\theta}v),b\bigr).(5)

Flows and Angular Projections

A strongly continuous unitary flow WW on a complex Hilbert space HH is a family of unitary operators with W0=IW_0=I, Ws+t=WsWtW_{s+t}=W_sW_t, and norm-continuous orbits.✦ For a strongly continuous unitary flow WW on a complex Hilbert space HH and real numbers a,ba,b, the integrated flow Ja,bJ_{a,b} is the bounded operator on HH in (6); for u,v∈Hu,v\in H, the generator relation Au=vAu=v means that the identity there holds at every real time: Ja,bu=∫abWtudt,Au=v⇔Wtu−u=J0,tv(t∈ℝ).J_{a,b}u=\int_a^b W_tu\,dt,\qquad Au=v\quad\Longleftrightarrow\quad W_tu-u=J_{0,t}v\quad(t\in\mathbb R).(6)

For a family FtF_t preserving a measure λ\lambda, its Koopman operator at time tt is the complex-linear isometry of L2(λ;ℂ)L^2(\lambda;\mathbb C) given by u↦u∘Ftu\mapsto u\circ F_t.✦ On HQH_Q the geodesic unitary flow is Gtu=u∘ΨtG_tu=u\circ\Psi_t✦, and the angular unitary flow is Vθu=u∘ℛθV_\theta u=u\circ\mathcal R_\theta.✦ Conjugating GG by VV at angle α\alpha gives the flow with time tt operator VαGtV−αV_\alpha G_tV_{-\alpha}.✦ The transverse flow is this conjugate at α=π/2\alpha=\pi/2; write XX and YY for the geodesic and transverse generators.✦

A strongly continuous circle action of period T>0T>0 on HH is a family of unitary operators VθV_\theta, indexed by ℝ/Tℤ\mathbb R/T\mathbb Z, with V0=IV_0=I, Vθ+φ=VθVφV_{\theta+\varphi}=V_\theta V_\varphi, and norm-continuous orbits.✦ For a strongly continuous circle action VV of period TT on HH, the Fourier projection PjP_j is the bounded linear operator on HH in (7), where mTm_T is normalized Haar measure: Pju=∫ℝ/Tℤe−2πijθ/TVθudmT(θ),j∈ℤ.P_ju=\int_{\mathbb R/T\mathbb Z} e^{-2\pi i j\theta/T}V_\theta u\,dm_T(\theta),\qquad j\in\mathbb Z.(7)✦ On the billiard double, the angular unitary flow VV is the tangent circle action of period 2π2\pi.✦ On either sheet use the oriented angular coordinate θ=(−1)barg⁡v\theta=(-1)^b\arg v and write Pju=uj(x)eijθP_ju=u_j(x)e^{ij\theta}, with a separate coefficient on each sheet.

Spatial Derivatives and Seam Compatibility

For a spatially differentiable function f:𝒟→ℂf:\mathcal D\to\mathbb C, put fv,b(x)=f((x,v),b)f_{v,b}(x)=f((x,v),b). The geodesic and oriented transverse spatial derivatives fX,fYf_X,f_Y of a spatially differentiable function f:𝒟→ℂf:\mathcal D\to\mathbb C, together with their Cauchy-Riemann combinations C−f,C+fC_-f,C_+f, are the functions on 𝒟\mathcal D given by (8); the transverse velocity is iviv on sheet zero and −iv-iv on sheet one: fX((x,v),b)=Dfv,b(x)[v],fY((x,v),b)=Dfv,b(x)[(−1)biv],C−f=12(fX−ifY),C+f=12(fX+ifY).\begin{gathered} f_X((x,v),b)=Df_{v,b}(x)[v],\qquad f_Y((x,v),b)=Df_{v,b}(x)[(-1)^biv],\\ C_-f=\tfrac12(f_X-if_Y),\qquad C_+f=\tfrac12(f_X+if_Y). \end{gathered}(8) The derivatives are real Fréchet derivatives in the base variable; the planar convention is ∂=(∂x1−i∂x2)/2\partial=(\partial_{x_1}-i\partial_{x_2})/2 and ∂‾=(∂x1+i∂x2)/2\bar\partial=(\partial_{x_1}+i\partial_{x_2})/2.

A function f:𝒟→ℂf:\mathcal D\to\mathbb C is seam-compatible if, for every side ii, point x∈six\in s_i, velocity v∈S1v\in S^1, and sheet bb, there is a neighborhood of xx on which the gluing identity (9) holds: f((Ri(y),ρi(v)),b+1)=f((y,v),b).f((R_i(y),\rho_i(v)),b+1)=f((y,v),b).(9)✦

For a measurable base set E⊂ℂE\subset\mathbb C, the restriction operator RER_E on HQH_Q is multiplication by its spatial indicator, as in (10): (REu)((x,v),b)=𝟏E(x)u((x,v),b).(R_Eu)((x,v),b)=\mathbf1_E(x)u((x,v),b).(10)✦

The Claim

Theorem 1. Let VV be a strongly continuous circle action on a complex Hilbert space HH, and let a,b∈Ha,b\in H. Suppose R:H→HR:H\to H is bounded, complex-linear and self-adjoint, commutes with every VθV_\theta, and satisfies R(a+b)=a+bR(a+b)=a+b. Suppose also that ∥Pk+2a∥=∥Pkb∥\|P_{k+2}a\|=\|P_kb\| for every integer kk. Then, for every integer jj, ∥Pj+1a∥2≤∥a+b∥∥R(a−b)∥.\|P_{j+1}a\|^2\leq\|a+b\|\,\|R(a-b)\|.(11)✦

Let QQ be a nondegenerate triangle, and let f:𝒟→ℂf:\mathcal D\to\mathbb C be measurable, seam-compatible, and square-integrable for ν\nu. Suppose each fv,bf_{v,b} is smooth on ℂ\mathbb C and there are constants C,D≥0C,D\geq0 such that ∥Dfv,b(x)∥≤C\|Df_{v,b}(x)\|\leq C and ∥D2fv,b(x)∥≤D\|D^2f_{v,b}(x)\|\leq D for every x,v,bx,v,b. Then, for every integer jj, ∥Pj+1C−f∥2=∥Pj−1C+f∥2.\|P_{j+1}C_-f\|_2=\|P_{j-1}C_+f\|_2.(12)✦ If E⊂ℂE\subset\mathbb C is measurable and fX((x,v),b)=0f_X((x,v),b)=0 for ν\nu-almost every state with x∉Ex\notin E, then PjfP_jf belongs to the domains of both XX and YY, and ∥XPjf∥22≤4∥fX∥2∥REfY∥2,∥YPjf∥22≤4∥fX∥2∥REfY∥2.\begin{aligned} \|XP_jf\|_2^2&\leq4\|f_X\|_2\,\|R_Ef_Y\|_2,\\ \|YP_jf\|_2^2&\leq4\|f_X\|_2\,\|R_Ef_Y\|_2. \end{aligned}(13)✦ All norms in (12) and (13) are taken in HQH_Q.

In the paper, the result reads: on a base MM with unit tangent phase space SMSM and spatial gradient ∇\nabla, if w∈L2(SM)w\in L^2(SM), with angular coefficients wjw_j, has compact spatial support in MM, its distributional derivatives Xw,YwXw,Yw belong to L2(SM)L^2(SM), and XwXw vanishes over M\EM\setminus E for a measurable base set EE, then ∥∇wj∥22≤4∥Xw∥2∥𝟏E∇w∥2\|\nabla w_j\|_2^2\leq4\|Xw\|_2\|\mathbf1_E\nabla w\|_2 for every integer jj.

What Is Checked

The estimate (11) is stated over a circle action and an equivariant self-adjoint operator, while (12) and (13) concern measurable seam-compatible fields on the measured double with globally bounded first and second spatial derivatives.

The double geodesic flow and every oriented tangent rotation preserve ν\nu. Complete regular trajectories are measurably identified with the initial states that admit them.✦

For a measurable field with differentiable spatial sections and a uniform bound on their first derivatives, both fXf_X and fYf_Y belong to HQH_Q.

The Fourier projections in (7) and the restriction operator in (10) have operator norm at most one. The integrated flow in (6) satisfies ∥Ja,bu∥≤|b−a|∥u∥\|J_{a,b}u\|\leq|b-a|\,\|u\|.