Non-colliding billiards in the plane

Itai Benjamini, Alexander Shamov, Barak Weiss · 2026

Read Non-colliding billiards in the plane by Itai Benjamini, Alexander Shamov, Barak Weiss (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-23575.txt.

Introduction

Benjamini, Shamov, and Weiss pose an open problem on non-colliding freely moving hard disks in the plane, with partial results on both sides. Stated cleanly, velocities must be pairwise distinct and speeds uniformly bounded away from infinity. Positively, they give a bounded, injective, non-colliding velocity assignment for the integer lattice, with speeds also bounded away from zero after a common shift. Negatively, no bounded continuous vector field on the whole plane serves as a universal assignment for point pairs farther than one apart.

Catalog record

CREATORItai Benjamini, Alexander Shamov, Barak Weiss
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE21690
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.23575

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Non-colliding billiards in the plane Itai Benjamini, Alexander Shamov, Barak Weiss Abstract We present an open problem about non-colliding freely moving hard disks in the Euclidean plane, together with related positive and negative partial results. The open problem is stated in a non-degenerate form: velocities are required to be pairwise distinct and their speeds are required to be uniformly bounded away from infinity. The positive deterministic result gives a bounded, injective, non-colliding velocity assignment for the integer lattice; after a common velocity shift, the speeds are also bounded away from zero. The negative result shows that no bounded continuous vector field on the whole plane can serve as a universal assignment satisfying the same separation inequality for all pairs of points at distance greater than one. We also record a space-time interpretation of the problem, relate it to packings by nonparallel cylinders in three dimensions, and formulate a corresponding topological-dynamical question for cylinder packings. 1 Introduction A set S\subset\mathbb{R}^{2} is called uniformly discrete if \inf_{\begin{subarray}{c}x,y\in S\\ x\neq y\end{subarray}}\|x-y\|>0. By scaling the plane, one may normalize this lower bound to be at least 1 . Consider a uniformly discrete set of initial positions in the Euclidean plane. To each point x assign a fixed velocity vector v(x) , and let all particles evolve simultaneously by x(t)=x+tv(x),\qquad t\in\mathbb{R}. We ask when the moving configuration can remain uniformly discrete for all positive and negative times. Equivalently, if one replaces the points by disks of radius \rho>0 , we ask when the disks can move forever without intersecting. The non-degenerate version considered here imposes two additional requirements: distinct particles should have distinct velocities, and all speeds should lie in a deterministic compact subinterval of (0,\infty) . The upper bound is a bounded-speed assumption. The lower bound is not forced by the collision condition, and it is not invariant under arbitrary Galilean shifts. Its purpose is to exclude static particles and to ask for a genuinely moving configuration. In deterministic examples one can often enforce such a lower bound by adding the same velocity vector to every particle, since only relative velocities affect collisions. In an isotropic random problem, however, adding a fixed nonzero vector would generally destroy rotation invariance, so the lower bound has to be part of the formulation. One may also ask weaker variants in which this lower bound is omitted. The basic ergodic version already seems open. The isotropic version asks for the same properties with an additional rotation-invariance assumption. Open problem (ergodic version). Does there exist a probability measure on marked point configurations (S,v) , where S\subset\mathbb{R}^{2} is uniformly discrete and v:S\to\mathbb{R}^{2} assigns a velocity to each point, with the following properties? • Non-degenerate velocities: there is a deterministic constant 0<M<\infty such that, almost surely, \|v(x)\|\leq M\qquad\text{for every }x\in S, and v(x)\neq v(y)\qquad\text{for every distinct }x,y\in S. • Time invariance: the law is invariant under the time evolution (S,v)\mapsto(S_{t},v_{t}),\qquad S_{t}:=\{x+tv(x):x\in S\}, where the velocity attached to the moved point x+tv(x) is still v(x) . • Translation invariance: the law is invariant under translations of the plane, acting on the positions and leaving the velocity vectors unchanged. • Ergodicity: the law is ergodic with respect to translations. • Non-emptiness: S is almost surely non-empty. • Hard-core condition at all times: almost surely, for every t\in\mathbb{R} , \inf_{\begin{subarray}{c}x,y\in S\\ x\neq y\end{subarray}}\|(x+tv(x))-(y+tv(y))\|\geq 1. Isotropic version. Does such a measure exist if, in addition, its law is invariant under rotations of the plane, where rotations act simultaneously on positions and velocity vectors? The first item is part of the non-degenerate formulation, not an optional strengthening. Without some non-degeneracy condition, a stationary hard-core point process with all velocities equal to zero gives a trivial solution. Pairwise distinct velocities rule out this particular degeneracy, while the lower speed bound additionally rules out particles whose individual motion degenerates to rest. It is useful to keep the following space-time picture in mind. Each particle determines a worldline L_{x}:=\{(x+tv(x),t):t\in\mathbb{R}\}\subset\mathbb{R}^{2}\times\mathbb{R}. The hard-core condition says that every horizontal time slice meets these worldlines in a point configuration with mutual distances at least 1 . Time invariance asks that the law of the sliced marked configuration be unchanged when all worldlines are translated in the time direction. This viewpoint also connects the problem to packings by infinite cylinders in three dimensions. Let e_{1},e_{2},e_{3} denote the standard basis of \mathbb{R}^{3} , let B_{\rho}\subset\operatorname{span}(e_{1},e_{2}) be the closed Euclidean disk of radius \rho , and let Z_{\rho}:=B_{\rho}+\mathbb{R}e_{3} be the closed vertical cylinder of radius \rho . A circular cylinder of radius \rho is an isometric image of Z_{\rho} , and its direction is the image of the vertical direction. Lemma 1.1 (Worldlines give cylinder packings) . Suppose that a moving configuration satisfies the hard-core condition above and that \|v(x)\|\leq M for every x\in S . Then any two distinct worldlines L_{x},L_{y}\subset\mathbb{R}^{3} have Euclidean distance at least \frac{1}{\sqrt{1+M^{2}}}. Consequently, the Euclidean cylinders of any radius 0<\rho\leq\frac{1}{2\sqrt{1+M^{2}}} about the worldlines have pairwise disjoint interiors. If the velocities are pairwise distinct, then these cylinders are pairwise nonparallel. If moreover m\leq\|v(x)\|\leq M , then their directions lie in the compact annulus of directions making angle between \arctan m and \arctan M with the e_{3} -axis. Proof. If M=0 , then all worldlines are vertical and the assertion follows directly from the hard-core condition. Hence assume M>0 . Fix x\neq y and points (x+tv(x),t)\in L_{x} and (y+sv(y),s)\in L_{y} . Put \tau=s-t and h=(x+tv(x))-(y+tv(y)). By the hard-core condition, \|h\|\geq 1 . The difference of the two space-time points is \bigl(h-\tau v(y),-\tau\bigr). If |\tau|\geq 1/M , then its Euclidean norm is at least 1/M , which is at least 1/\sqrt{1+M^{2}} . If |\tau|\leq 1/M , then \bigl\|\bigl(h-\tau v(y),-\tau\bigr)\bigr\|^{2}\geq(1-M|\tau|)^{2}+|\tau|^{2}\geq\frac{1}{1+M^{2}}, since the minimum of (1-Mu)^{2}+u^{2} for u\geq 0 occurs at u=M/(1+M^{2}) and equals 1/(1+M^{2}) . This proves the distance bound. The cylinder conclusion follows immediately. Finally, the direction of L_{x} is the line spanned by (v(x),1) . If two such directions agree, then (v(x),1)=\lambda(v(y),1) for some nonzero scalar \lambda ; comparing the third coordinate gives \lambda=1 , and hence v(x)=v(y) . Thus pairwise distinct velocities give nonparallel cylinders. The angle \theta between (v(x),1) and the e_{3} -axis satisfies \tan\theta=\|v(x)\| , giving the stated annulus of directions. ∎ The cylinder-packing viewpoint places the problem near a classical line of work. Bezdek and W. Kuperberg determined the maximal density of a packing by congruent infinite circular cylinders in \mathbb{R}^{3} ; the value is \pi/\sqrt{12} , as in the planar disk-packing problem, and it is achieved by parallel cylinders whose intersections with a perpendicular plane form an optimal disk packing [ 2 ] . K. Kuperberg constructed nonparallel cylinder packings of positive density [ 6 ] . Further developments include dense nonparallel packings of Ismailescu and Laskawiec [ 5 ] and recent work of Eliyahu on lower and upper densities for nonparallel cylinder packings [ 3 ] . This suggests the following topological-dynamical analogue of the ergodic problem above. Let \mathcal{CL}(\mathbb{R}^{3}) be the space of closed subsets of \mathbb{R}^{3} with the Chabauty-Fell topology. Since \mathbb{R}^{3} is locally compact and second countable, this space is compact and metrizable. For background on the Fell topology and related topologies, see Fell’s original paper [ 4 ] and Beer’s monograph [ 1 ] , and see [ 7 , Appendix] for

Read the complete text