OctoInformBeta

OpenAI preprint, September 2026 · §1. Introduction

§1 (setting): the triangle, specular flow and invariant measure: Complete collision trajectories and the normalized measure on position and direction

Triangular billiards in phase space.

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 phase description of a triangular billiard records position together with direction. A complete collision sequence describes a trajectory through all real times. Positions belong to ℂ\mathbb C, and Circle\mathrm{Circle} denotes the unit circle in ℂ\mathbb C.

The Definition

Definition 1. A nondegenerate Euclidean triangle QQ is an ordered triple of affinely independent points in the real plane: Q=(v0,v1,v2),v0,v1,v2∈ℂaffinely independent over ℝ.Q=(v_0,v_1,v_2),\qquad v_0,v_1,v_2\in\mathbb C \quad\text{affinely independent over }\mathbb R.(1)✦ Vertex subscripts in (1) and below are read modulo three.

The open billiard table ΩQ⊂ℂ\Omega_Q\subset\mathbb C is the interior of the convex hull of the vertices: ΩQ=int⁡conv⁡ℝ{v0,v1,v2}.\Omega_Q=\operatorname{int}\operatorname{conv}_{\mathbb R}\{v_0,v_1,v_2\}.(2)✦ The open side Si⊂ℂS_i\subset\mathbb C from viv_i to vi+1v_{i+1} is Si={(1−s)vi+svi+1:0<s<1},i∈{0,1,2}.S_i=\{(1-s)v_i+s v_{i+1}:0<s<1\},\qquad i\in\{0,1,2\}.(3)✦ The side tangent ei∈ℂe_i\in\mathbb C associated with SiS_i in (3) is ei=vi+1−vi.e_i=v_{i+1}-v_i.(4)✦ The interior angle αi∈ℝ\alpha_i\in\mathbb R at viv_i is the Euclidean angle between the two incident edge vectors: αi=∠(vi+1−vi,vi+2−vi).\alpha_i=\angle(v_{i+1}-v_i,\,v_{i+2}-v_i).(5)✦ A triangle QQ has an irrational angle when at least one angle in (5) is an irrational multiple of π\pi: ∃i∈{0,1,2},αi/π∉ℚ.\exists i\in\{0,1,2\},\qquad \alpha_i/\pi\notin\mathbb Q.(6)✦

The specular reflection map ReR_e associated with a tangent e∈ℂe\in\mathbb C is Re(v)=ee¯v¯,v∈ℂ.R_e(v)=\frac{e}{\overline e}\,\overline v,\qquad v\in\mathbb C.(7)✦

The ambient phase space Phase\mathrm{Phase} is the space of position and direction pairs: Phase=ℂ×Circle.\mathrm{Phase}=\mathbb C\times\mathrm{Circle}.(8)✦

Let mm be Lebesgue measure on ℝ\mathbb R, and let AA be planar area measure. The uniform angular probability ρ\rho on Circle\mathrm{Circle} is the image of normalized Lebesgue measure under θ↦eiθ\theta\mapsto e^{i\theta}: ρ=(θ↦eiθ)*(12πm|[0,2π)).\rho=(\theta\mapsto e^{i\theta})_* \left(\frac{1}{2\pi}\,m\big|_{[0,2\pi)}\right).(9)✦ The normalized Liouville measure μ\mu on the phase space in (8) is the product of normalized area on (2) and the angular probability in (9): μ=(1A(ΩQ)A|ΩQ)⊗ρ.\mu=\left(\frac{1}{A(\Omega_Q)}\,A\big|_{\Omega_Q}\right)\otimes\rho.(10)✦

A complete billiard trajectory through z∈Phasez\in\mathrm{Phase} is a sequence c=(τn,pn,un,wn)n∈ℤc=(\tau_n,p_n,u_n,w_n)_{n\in\mathbb Z} of collision times τn∈ℝ\tau_n\in\mathbb R, collision points pn∈ℂp_n\in\mathbb C, outgoing directions un∈Circleu_n\in\mathrm{Circle} and side indices wn∈{0,1,2}w_n\in\{0,1,2\} satisfying (11), with the reflection map of (7): τn<τn+1,limn→−∞τn=−∞,limn→+∞τn=+∞,τ0<0<τ1,pn∈Swn,pn+1=pn+(τn+1−τn)un,pn+(t−τn)un∈ΩQ(τn<t<τn+1),z=(p0−τ0u0,u0),un=Rewn(un−1)(n∈ℤ).\begin{aligned} &\tau_n<\tau_{n+1},\qquad \lim_{n\to-\infty}\tau_n=-\infty,\qquad \lim_{n\to+\infty}\tau_n=+\infty,\\ &\tau_0<0<\tau_1,\qquad p_n\in S_{w_n},\\ &p_{n+1}=p_n+(\tau_{n+1}-\tau_n)u_n,\\ &p_n+(t-\tau_n)u_n\in\Omega_Q \qquad(\tau_n<t<\tau_{n+1}),\\ &z=(p_0-\tau_0u_0,u_0),\\ &u_n=R_{e_{w_n}}(u_{n-1}) \qquad(n\in\mathbb Z). \end{aligned}(11)✦ For a complete trajectory cc, its state c(t)∈Phasec(t)\in\mathrm{Phase} at real time tt is the position and direction obtained by following its straight flights, taking the outgoing direction at a collision time.

The billiard flow ΦtQ\Phi_t^Q on Phase\mathrm{Phase} sends an initial state zz to its state at time tt along a complete nonsingular specular trajectory, meaning one that avoids the vertices; when no such trajectory exists, it leaves zz fixed.✦ This is the family in (12): ΦQ:ℝ×Phase→Phase,(t,z)↦ΦtQ(z).\Phi^Q:\mathbb R\times\mathrm{Phase}\longrightarrow\mathrm{Phase},\qquad (t,z)\longmapsto\Phi_t^Q(z).(12)✦

In the paper, the result reads: ΩQ\Omega_Q is the interior of a nondegenerate Euclidean triangle, a unit-speed trajectory reflects at open sides by preserving tangential velocity and reversing normal velocity, the phase measure is dμ=dAdθ/(2πA(ΩQ))d\mu=dA\,d\theta/(2\pi A(\Omega_Q)), and trajectories hitting a vertex in either time direction are discarded.

What Is Checked

The definitions are stated over every nondegenerate triangle in (1) and its ambient phase space (8), with complete trajectories subject to (11).