OctoInformBeta

OpenAI preprint, September 2026 · §2. The double and angular coefficients

§2 (setting): horizontal generators and angular Fourier modes: Rotation-covariant derivatives for the horizontal unitary flows

Horizontal generators and angular modes on the double.

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

Horizontal motion on the oriented double of a triangular billiard has a geodesic direction and a transverse direction determined by the sheet. Angular Fourier modes describe the interaction of these directions on functions in L2L^2.

A nondegenerate triangle QQ has three vertices qi∈ℂq_i\in\mathbb C that are affinely independent over ℝ\mathbb R, with indices taken cyclically modulo three.✦ The open billiard table Ω\Omega, open sides sis_i and side tangent vectors eie_i of the triangle QQ are the set, line segments and vectors given in (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) Specular reflection in a nonzero tangent ee is the real-linear map ρe\rho_e; reflection in side ii and reflection of a unit direction are, respectively, σi\sigma_i and κi\kappa_i, as defined by (2): ρe(w)=ee¯w¯,σi(x)=qi+ρei(x−qi),κi(v)=ρei(v).\rho_e(w)=\frac{e}{\overline e}\,\overline w, \qquad \sigma_i(x)=q_i+\rho_{e_i}(x-q_i), \qquad \kappa_i(v)=\rho_{e_i}(v).(2)

The phase space is 𝖯𝗁𝖺𝗌𝖾=ℂ×S1\mathsf{Phase}=\mathbb C\times S^1, whose coordinates record position and unit direction.✦ The double phase space is 𝖣=𝖯𝗁𝖺𝗌𝖾×ℤ/2ℤ\mathsf D=\mathsf{Phase}\times\mathbb Z/2\mathbb Z, with sheet coordinate bb.✦ Let mS1m_{S^1} be the pushforward of dθ/(2π)d\theta/(2\pi) on [0,2π)[0,2\pi) under θ↦eiθ\theta\mapsto e^{i\theta}.✦ The normalized area-angular probability μ\mu and double Liouville probability ν\nu are the product measures in (3), and the complex Hilbert space of the double is HQ=L2(𝖣,ν;ℂ)H_Q=L^2(\mathsf D,\nu;\mathbb C): μ=𝟏ΩdAarea⁡(Ω)⊗mS1,ν=μ⊗δ0+δ12.\mu=\frac{\mathbf 1_\Omega\,dA}{\operatorname{area}(\Omega)}\otimes m_{S^1}, \qquad \nu=\mu\otimes\frac{\delta_0+\delta_1}{2}.(3)

Collision data c=(τn,pn,vn,ℓn)n∈ℤc=(\tau_n,p_n,v_n,\ell_n)_{n\in\mathbb Z} consist of real collision times, positions in ℂ\mathbb C, outgoing unit directions and side labels.✦ Collision data cc through an initial state z=(x,v)z=(x,v) are valid if their times are strictly increasing and unbounded in both directions, with τ0<0<τ1\tau_0<0<\tau_1, and they satisfy (4) for every nn. pn∈sℓn,pn+1=pn+(τn+1−τn)vn,pn+pn+12∈Ω,x=p0−τ0v0,v=v0,vn=κℓn(vn−1).\begin{gathered} p_n\in s_{\ell_n},\qquad p_{n+1}=p_n+(\tau_{n+1}-\tau_n)v_n,\qquad \frac{p_n+p_{n+1}}2\in\Omega,\\ x=p_0-\tau_0v_0,\qquad v=v_0,\qquad v_n=\kappa_{\ell_n}(v_{n-1}). \end{gathered}(4) A complete regular billiard trajectory cc through an initial state zz is a collection of valid collision data through zz whose entire open flights lie in Ω\Omega. Valid data specify such a trajectory with the same collision sequences.✦ Its state c(t)c(t) uses the outgoing direction at a collision time.

The space 𝒞Q\mathcal C_Q consists of pairs (z,c)(z,c) with cc valid through zz; its initial-state map is I(z,c)=zI(z,c)=z. An extension to phase space of the inverse of II on its image assigns collision data to an initial state.✦ For a trajectory cc, the flight index ι(c,t)\iota(c,t) selects the flight containing tt.✦

The billiard flow Φt\Phi_t sends a phase state along its unique complete regular trajectory, and fixes a state admitting no such trajectory.✦ For a regular state zz, the collision index Nt(z)N_t(z) is the label of the flight reached at time tt, with the initial flight labeled zero.✦ The geodesic flow on the double is (5), where the second component is reduced modulo two: Ψt(z,b)=(Φt(z),b+Nt(z)).\Psi_t(z,b)=\bigl(\Phi_t(z),\,b+N_t(z)\bigr).(5)✦

Put 𝜍0=1\varsigma_0=1 and 𝜍1=−1\varsigma_1=-1. Oriented tangent rotation RαR_\alpha turns the direction oppositely on the two sheets, and the oriented transverse velocity v⟂v_\perp is given in (6). The Koopman operator KFK_F of a measure-preserving map FF is the composition operator given by KFu=u∘FK_Fu=u\circ F; write Θα=KRα\Theta_\alpha=K_{R_\alpha} for the angular Koopman operators. Rα((x,v),b)=((x,ei𝜍bαv),b),v⟂(v,b)=i𝜍bv.R_\alpha((x,v),b)=((x,e^{i\varsigma_b\alpha}v),b), \qquad v_\perp(v,b)=i\varsigma_bv.(6) The angular unitary flow on HQH_Q is the strongly continuous unitary flow α↦Θα\alpha\mapsto\Theta_\alpha given by pullback under oriented tangent rotation.

Unitary Flows and Fourier Projections

A strongly continuous unitary flow WW on a complex Hilbert space ℋ\mathcal H is a family of complex-linear isometries satisfying (7): W0=Id,Ws+t=WsWt,t↦Wtu is norm-continuous for every u∈ℋ.W_0=\mathrm{Id},\qquad W_{s+t}=W_sW_t,\qquad t\longmapsto W_tu\text{ is norm-continuous for every }u\in\mathcal H.(7) The integrated flow ℐa,bW\mathcal I^W_{a,b} of a strongly continuous unitary flow WW on ℋ\mathcal H is the bounded operator defined by the oriented interval integral in (8). ℐa,bWu=∫abWsuds.\mathcal I^W_{a,b}u=\int_a^b W_su\,ds.(8) The conjugate WA,αW^{A,\alpha} of a unitary flow WW by a unitary flow AA at angle α\alpha is the unitary flow given in (9). WtA,α=AαWtA−α.W^{A,\alpha}_t=A_\alpha W_tA_{-\alpha}.(9)✦

A strongly continuous unitary circle action CC of period TT on ℋ\mathcal H is a strongly continuous action of ℝ/Tℤ\mathbb R/T\mathbb Z by complex-linear isometries.✦ Let mTm_T be normalized Haar measure on this circle and χj\chi_j its jjth Fourier character. The Fourier projection PjCP_j^C of a unitary circle action CC on ℋ\mathcal H is the bounded linear operator on ℋ\mathcal H defined by (10); for period 2π2\pi, χj(α)=eijα\chi_j(\alpha)=e^{ij\alpha}. PjCu=∫ℝ/Tℤχ−j(α)CαudmT(α),j∈ℤ.P_j^Cu=\int_{\mathbb R/T\mathbb Z} \chi_{-j}(\alpha)C_\alpha u\,dm_T(\alpha),\qquad j\in\mathbb Z.(10)✦

The Claim

Definition 1 (Horizontal generators and angular modes). For a strongly continuous unitary flow WW on ℋ\mathcal H, its strong generator graph 𝒢W⊂ℋ×ℋ\mathcal G_W\subset\mathcal H\times\mathcal H is defined by (11): (u,g)∈𝒢W⇔Wtu−u=ℐ0,tWgfor every t∈ℝ.(u,g)\in\mathcal G_W \quad\Longleftrightarrow\quad W_tu-u=\mathcal I^W_{0,t}g\quad\text{for every }t\in\mathbb R.(11)✦

Let QQ be a nondegenerate triangle. The geodesic unitary flow UXU^X on HQH_Q acts by (12): UtXu=u∘Ψt.U^X_tu=u\circ\Psi_t.(12)✦ The transverse unitary flow UYU^Y on HQH_Q is the angular conjugate of the geodesic unitary flow UXU^X given in (13). UtY=Θπ/2UtXΘ−π/2.U^Y_t=\Theta_{\pi/2}\,U^X_t\,\Theta_{-\pi/2}.(13)✦ Write XX and YY for their generators. For a function f:𝖣→ℂf:\mathsf D\to\mathbb C, let dxf(⋅,v,b)d_xf(\cdot,v,b) denote its real Fréchet derivative in position. The pointwise geodesic directional derivative DXfD_Xf of a function f:𝖣→ℂf:\mathsf D\to\mathbb C is the function on 𝖣\mathsf D given by (14). DXf((x,v),b)=dxf(⋅,v,b)[v].D_Xf((x,v),b)=d_xf(\cdot,v,b)[v].(14)✦ For f:𝖣→ℂf:\mathsf D\to\mathbb C, the pointwise transverse derivative is the function DYf:𝖣→ℂD_Yf:\mathsf D\to\mathbb C given by (15). DYf((x,v),b)=dxf(⋅,v,b)[v⟂(v,b)].D_Yf((x,v),b)=d_xf(\cdot,v,b)[v_\perp(v,b)].(15)✦

A function f:𝖣→ℂf:\mathsf D\to\mathbb C is seam compatible if, for every side ii, point x∈six\in s_i, direction v∈S1v\in S^1 and sheet bb, there is a neighborhood UU of xx in ℂ\mathbb C on which (16) holds: f((σi(y),κi(v)),b+1)=f((y,v),b)(y∈U).f((\sigma_i(y),\kappa_i(v)),b+1)=f((y,v),b) \qquad (y\in U).(16)✦

The angular circle action VV on HQH_Q is given by (17), where [α]∈ℝ/(2πℤ)[\alpha]\in\mathbb R/(2\pi\mathbb Z): V[α]u=Θαu.V_{[\alpha]}u=\Theta_\alpha u.(17)✦ The angular Fourier projections PjP_j of the circle action VV on HQH_Q are the bounded linear operators on HQH_Q defined by (18). Pj=PjV,j∈ℤ.P_j=P_j^V,\qquad j\in\mathbb Z.(18) For every u∈HQu\in H_Q, the net of finite Fourier sums, ordered by inclusion of finite subsets of ℤ\mathbb Z, has the norm limit (19): limF↑ℤ∑j∈FPju=u.\lim_{F\uparrow\mathbb Z}\ \sum_{j\in F}P_ju=u.(19)✦

For two strongly continuous unitary flows WA,WBW^A,W^B on ℋ\mathcal H, let 𝒢A,𝒢B\mathcal G_A,\mathcal G_B be their generator graphs, and let CC be a strongly continuous unitary circle action. A triple (u,a,b)∈ℋ3(u,a,b)\in\mathcal H^3 is rotation-covariant Cauchy-Riemann data for the unitary flows WA,WBW^A,W^B and circle action CC if it satisfies (20) for every angle α\alpha. (Cαu,χ−1(α)Cαa+χ1(α)Cαb)∈𝒢A,(Cαu,i(χ−1(α)Cαa−χ1(α)Cαb))∈𝒢B.\begin{aligned} \bigl(C_\alpha u,\, \chi_{-1}(\alpha)C_\alpha a+\chi_1(\alpha)C_\alpha b\bigr) &\in\mathcal G_A,\\ \bigl(C_\alpha u,\, i(\chi_{-1}(\alpha)C_\alpha a-\chi_1(\alpha)C_\alpha b)\bigr) &\in\mathcal G_B. \end{aligned}(20) For such data, each integer jj satisfies the Fourier mode derivative shifts (21): (PjCu,Pj+1Ca+Pj−1Cb)∈𝒢A,(PjCu,i(Pj+1Ca−Pj−1Cb))∈𝒢B.\begin{aligned} \bigl(P_j^Cu,\,P_{j+1}^Ca+P_{j-1}^Cb\bigr)&\in\mathcal G_A,\\ \bigl(P_j^Cu,\,i(P_{j+1}^Ca-P_{j-1}^Cb)\bigr)&\in\mathcal G_B. \end{aligned}(21)✦

In the paper, the result reads: in an oriented flat chart, ∂=(∂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, with X=eiθ∂+e−iθ∂‾X=e^{i\theta}\partial+e^{-i\theta}\bar\partial and Y=ieiθ∂−ie−iθ∂‾Y=ie^{i\theta}\partial-ie^{-i\theta}\bar\partial; writing Pju=uj(x)eijθP_ju=u_j(x)e^{ij\theta}, a chart change z′=eiφz+cz'=e^{i\varphi}z+c, with rotation angle φ\varphi and translation c∈ℂc\in\mathbb C, gives uj′=e−ijφuju'_j=e^{-ij\varphi}u_j, while (Xu)j=∂uj−1+∂‾uj+1(Xu)_j=\partial u_{j-1}+\bar\partial u_{j+1} and (Yu)j=i∂uj−1−i∂‾uj+1(Yu)_j=i\partial u_{j-1}-i\bar\partial u_{j+1}; coefficient gradients are distributional gradients on the regular surface, with no cone-tip trace condition.

What Is Checked

The constructions are stated for arbitrary nondegenerate triangles and the complex Hilbert spaces HQH_Q; the derivative shifts in (21) concern any two strongly continuous unitary flows and a strongly continuous unitary circle action.

The initial-state map I:𝒞Q→𝖯𝗁𝖺𝗌𝖾I:\mathcal C_Q\to\mathsf{Phase} is a measurable embedding.✦ For every real tt and α\alpha, the transformations Ψt\Psi_t and RαR_\alpha preserve ν\nu.