Regular simplices and periodic billiard orbits
aa r X i v : . [ m a t h . D S ] O c t Regular simplices and periodic billiard orbits
Nicolas B´edaride ∗ Michael Rao † ABSTRACT
A simplex is the convex hull of n + 1 points in R n which form anaffine basis. A regular simplex ∆ n is a simplex with sides of the samelength. We consider the billiard flow inside a regular simplex of R n .We show the existence of two types of periodic trajectories. One hasperiod n + 1 and hits once each face. The other one has period 2 n and hits n times one of the faces while hitting once any other face. Inboth cases we determine the exact coordinates for the points where thetrajectory hits the boundary of the simplex. We consider the billiard problem inside a polytope. We start with a pointof the boundary of the polytope and we move along a straight line until wereach again the boundary, where the trajectory is reflected according to themirror law. A famous example of a periodic trajectory is Fagnano’s orbit:we consider an acute triangle and the foot points of the altitudes. Thosepoints define a billiard trajectory which is periodic (see Figure 1). If wecode the sides of the polygon by different letters, a billiard orbit becomes aword on this alphabet. For the case of polygons in the plane some resultsare known. For example, one knows that there exists a periodic orbit ineach rational polygon (the angles are rational multiples of π ), and recentlySchwartz, [Sch06], has proved the existence of a periodic billiard orbit inevery obtuse triangle with angle less than 100 degrees. In [GZ03] it is proventhat through every point of a right angled triangle passes a periodic orbit.A good survey of what is known about periodic orbits can be found in thearticle [GSV92] by Galperin, Stepin and Vorobets or in the book of Masurand Tabachnikov [MT02]. For more recent results, see [Sch06], [HS09] or[DFT11]. ∗ Laboratoire d’Analyse Topologie et Probabilit´es UMR 7353 , Universit´e Aix Marseille,Centre de math´ematiques et informatique 29 avenue Joliot Curie 13453 Marseille cedex,France. [email protected] † Laboratoire de l’Informatique du Parall´elisme, ´equipe MC2, ´Ecole NormaleSup´erieure, 46 avenue d’Italie 69364 Lyon cedex 7. [email protected] boundary points as the intersections of the trajectorywith the boundary of the polytope. In [GKT95] the study of the set ofboundary points associated to a periodic word is made.There is no general result on periodic orbits; the only known result con-cerns the example of the tetrahedron. Stenman [Ste75] shows that a periodicorbit of length four exists in a regular tetrahedron. In [B´e08] the authorproves that this orbit exists for an open set of tetrahedra near the regulartetrahedron. In both cases a boundary point with explicit coordinates isgiven, but the method can not be generalized to any dimension.In this paper we consider the case of regular simplex in R n . A simplex isthe convex hull of n + 1 points in R n which form an affine basis. A regularsimplex ∆ n is a simplex with sides of the same length. We obtain by a shortproof the existence of two periodic billiard trajectories in ∆ n for any n ≥ Theorem.
In a regular simplex ∆ n ⊂ R n , there exists at least two periodicorbits: • One has period n + 1 and hits each face once. • The other has period n and hits one face n times and hits each otherface once. Let P be a polytope of R n , we will call face of P the faces of maximal dimen-sion. A billiard orbit is a (finite or infinite) broken line, denoted x x . . . such that1. Each x i is a point in the interior of a face of P .2. The unit vectors x i + − x i | x i + − x i | and x i + − x i + | x i + − x i + | are symmetric with respectto the orthogonal reflection in the face containing x i + for every integer i . 2or each integer i , x i is called a boundary point of the orbit and x i + − x i | x i + − x i | isthe direction of the orbit at this boundary point.A billiard orbit x . . . x n . . . is called periodic of period n if n is thesmaller integer such that for each integer i ≥ x n + i = x i .A coding of the billiard flow is given via a labelling of the faces of thepolyhedron by elements of a finite set A . Then to each nonsingular billiardorbit there is naturally assigned an infinite word v = a a . . . , where a i ∈ A is the label of the face which contains x i .To each periodic billiard orbit of period n , we can associate an infiniteperiodic word. The word a . . . a n − is called the fundamental period of theword. Consider n + 1 points a , . . . , a n ∈ R n which form an affine basis, thenfor every point m ∈ R n there exists real numbers λ , . . . , λ n such that X ≤ i ≤ n λ i a i = m Throughout this paper ( λ , . . . , λ n ) will be called barycen-tric coordinates of m . We do not assume that the sum of coordinates is1, in order to avoid lengthly formulas. Therefore there is no uniqueness ofcoordinates for a given point. Definition 3.1.
Consider a regular simplex in R n and consider barycentriccoordinates with respect to the vertices. Define three points m , p , r ∈ R n by the following formulas, where π i ( m ) denotes the i -th coordinate of m . • π i ( m ) = − i + ( n + 1) i ≤ i ≤ n . • π i ( p ) = ( i = 0, − n + 1) i + 2( n + 1) i − n ( n + 2) if 0 < i ≤ n . • π i ( r ) = ( n if i = 0,2( n + 1)( n − i + 1)( i −
1) if 0 < i ≤ n . Definition 3.2.
For a point in R n with barycentric coordinates ( x , . . . , x n )the cyclic right shift of these coordinates is the point with coordinates( x n , x , . . . , x n − ). Define the points m i , i = 1 . . . n in R n obtained by cyclicright shift of the coordinates of m . Now the points p i , i = 2 . . . n (resp r i )are obtained by a cyclic right shift of the last n coordinates of p (resp. r ).3 xample 3.3. We have m = ( n, , n, n − . . . , n − p =(0 , n , n , n − n − , . . . , n − n − . We obtain :
Theorem 3.4.
In a regular simplex ∆ n ⊂ R n one has:(1) The word . . . n describes the fundamental period of a periodic wordthat codes an orbit of period n + 1 which passes through every faceonce during the period. The boundary points of this orbit are given by m i , i = 0 . . . n .(2) The word . . . n describes the fundamental period of a periodicword that codes an orbit of period n which passes through one face n times while hitting once time any other face. The n boundary pointson the face labeled have coordinates given by p i , i = 1 . . . n . Theother points are given by r i , i = 1 . . . n . Corollary 3.5.
The boundary points of the periodic trajectories satisfies thefollowing properties: • If n is odd, then m i and p i are situated at the intersection of n +12 hyperplanes for every integer i , when each hyperplane is orthogonal toan edge of the simplex and passes through the middle of this edge. • If n is even, the points lie on a line segment that connects a vertex tothe intersection of n hyperplanes. The corollary can be deduced directly from the properties of the barycen-tric coordinates of the periodic points. A formal proof is left to the reader.
Remark 3.6.
The isometric group of a regular tetrahedron ∆ n is the per-mutation group S n +1 . If v is a periodic billiard word and σ a permutationin S n +1 , then σ ( v ) is also a periodic word. Moreover, if v is a periodic word,then the shift of this word is also a periodic word corresponding to the sameperiodic orbit. Thus our theorem gives the existence of ( n − points of firsttype and (2 n )!2 n ! periodic points of the second type inside the regular simplex. To give concrete example we consider the cases of ∆ , ∆ , ∆ : • m = (0 , ,
2) if n = 2(0 , , ,
3) if n = 3(0 , , , ,
2) if n = 44 • •• ••• ••• Figure 2: Boundary points in a face of ∆ n for the first periodic billiard orbitwith n = 2 , , p = (0 , ,
4) if n = 2(0 , , ,
9) if n = 3(0 , , , ,
16) if n = 4(0 , , , , ,
25) if n = 5 r = (2 , ,
6) if n = 2(3 , , ,
16) if n = 3(4 , , , ,
30) if n = 4(5 , , , , ,
48) if n = 5 • In ∆ the words of first and second type are 012 and 0102. Theboundary point for the first one is the middle of the edge. For thesecond one, every point on the edge is a boundary point. This caseis an extremal case since each point on the edge 0 is a periodic point.Moreover the direction of the billiard orbit is orthogonal to the edges1 and 2 of the simplex (see Figure 1). • In ∆ , the two fundamental periods of the periodic words are 0123and 010203. Each boundary point of the periodic trajectories is on anheight of a triangular face. Indeed the point (3 , ,
4) is the barycenterof one vertex of the triangle and one midpoint of an edge. It lies on asegment joining one vertex to the midpoint of the opposite edge. Thusit is on a height of the simplex (see Figure 2). • In ∆ for the fundamental period 01234, each boundary point is insidea regular tetrahedron, it is on a segment which links the two midpointsof non coplanar edges of the tetrahedron. The point (2 , , ,
2) is thebarycenter of two middles of edges. This segment is orthogonal tothe two edges. The symmetric point (3 , , ,
3) is also on this segment.There are six points obtained by permutation. They are on three edgespassing through the center of the regular simplex (see Figure 2).5
Proof of Theorem 3.4
Let i be an integer in [0 . . . n ]. Barycentric coordinates of the points m i − , m i and m i + are given by Definition 3.1 and Definition 3.2. (Indices aretaken modulo n + 1.) We must show that the image of m i − by the reflexionthrough face i belongs to the line ( m i m i + ). By symmetry we can restrictto the case i = 0.Let m ′ n denote the image of m n under reflection in the hyperface 0.We compute the barycentric coordinates of m ′ n . To do so, consider theorthogonal reflection of vertex P of ∆ n through the hyperface 0. Denoteit P ′ . If m n is barycentric of P . . . P n , then m ′ n is barycentric with samecoefficients of P ′ , P , . . . P n . We obtain P ′ = ( − , /n, /n, . . . , /n ) sincethe middle of the segment [ P P ′ ] is the center of the face labeled 0. Bydefinition: π i ( m ) = i = 0, − i + ( n + 1) i if 0 < i < n , n if i = n .We give an equivalent formula for coordinates of m : π i ( m ) = n if i = 0, − i + ( n + 3) i − n − < i < n ,2 n − i = n . π i ( m n ) = n if i = 0, − i + ( n − i + n if 0 < i < n ,0 if i = n .Since P ′ = ( − , n , . . . , n ), we have: π i ( m ′ n ) = − n if i = 0, − i + ( n − i + n + 2 if 0 < i < n ,2 if i = n .It suffices to remark that m = ( m + m ′ n ) / Remark 4.1.
A periodic billiard orbit is a polygonal path inside the regularsimplex. For the periodic word of fundamental period 012 . . . n we see thateach edge of such a path has the same length. But the n + 1 boundary pointsof such a periodic orbit do not form a regular simplex for each integer n .6 .2 Part (2) By symmetry, we can restrict ourself to two cases. We show that the image r ′ n by reflection through face 0 of the point r n belongs to the line ( p r ),and that the image p ′ by reflection through face 1 of the point p belongsto the line ( p r ). π i ( r ) = ( n if i = 0,2( n + 1)( n − i + 1)( i −
1) if 0 < i ≤ n . π i ( r n ) = ( n if i = 0,2( n + 1)( n − i ) i if 0 < i ≤ n . π i ( r ′ n ) = ( − n if i = 0,2( n + 1)( n − i ) i + 2 if 0 < i ≤ n . π i ( p ) = ( i = 0, − n + 1) i + 2( n + 1) i − n ( n + 2) if 0 < i ≤ n . π i ( p ) = i = 0, n if i = 1, − n + 1)( i − + 2( n + 1) ( i − − n ( n + 2) if 1 < i ≤ n .Since P ′ = ( n , − , n , . . . , n ), we have: π i ( p ′ ) = n if i = 0, − n if i = 1, − n + 1)( i − + 2( n + 1) ( i − − n ( n + 2) + 2 n if 1 < i ≤ n .Thus p = ( r + r ′ n ) / and r = ( p + p ′ ) / . This concludes the proofof Theorem 3.4. This section is not necessary to obtain proof of Theorem 3.4. Neverthelesswe explain our method to find the coordinates of the periodic point involvedin the theorem.Let v be a billiard word, we denote by s v i and S v i the affine and vectorialreflection through the face labeled v i . Then s v , S v are defined as product ofmaps s v i or S v i for i = 0 . . . n . First we recall a proposition of [B´e08] :7 roposition 5.1. Let P a polyhedron. then the following properties areequivalent:(1) Ther exists a word v which is the prefix of a periodic word with period | v | .(2) There exists m ∈ v such that −−−−−→ s v ( m ) m is admissible with boundary point m for vv , and u = −−−−−→ s v ( m ) m is such that S v u = u . Now we explain the algorithm. Let v be a word. To prove that v isa periodic word, we must find a point m in a face of the simplex and adirection u such that the billiard orbit of ( m , u ) is periodic with coding v .Thus we must compute the eigenvector associated to S v for the eigenvalue1, and find m on the axis of s v . In classical coordinates systems, the verticesof a regular simplex have simple formulas, thus we can compute u .Now barycentric coordinates are useful to find m . Indeed the imageof a point m by a reflection through a face of the simplex has barycentriccoordinates which depends linearly over Q of those of m : it is the barycenterwith same coefficients of the vertices of the simplex except the vertex whichdoes not belong to the face. This vertex is replaced by the image of thevertex by the reflection. This image can be expressed as barycenter withrational coefficients. Finally we see that −−−−−→ s v ( m ) m is a linear combination ofcoordinates of m . Thus coordinates of m are easy to find.The last step is to check that m and all other points are interior to thefaces labelled v i , i = 0 . . . . Of course, this can be simplified if there exists acoordinate system where the vertices of the regular simplex are in Q n , forexample in dimension three such a system is given in [B´e08]. In this case u belongs to Q . This is not possible in any dimension. In this section we consider the periodic words of Theorem 3.4. As explainedin Remark 3.6 we can associate to a periodic word several boundary pointsinside the same face. We want to describe the convex hull generated bythese points. Let us denote by Q n this polytope associated to ∆ n . Themotivation of this study is given by the example of lowest dimensions. Thestudy of the polytope Q associated to each periodic word could be a goodway to try to obtain other periodic words. Proposition 6.1.
For the periodic orbit passing one time through each face,we have • In dimension , the polytope Q is one point: the center of the edge. • In dimension , the polytope Q is a regular triangle. Figure 3: Polytopes Q n (5 , , ,
9) (5 , , , • In dimension , the polytope Q is a regular octahedron.Proof. The theorem gives barycentric coordinates of vertices of the boundarypoints of the periodic words. We use combinatorics to find description ofthe convex hull of these points. • In dimension three, the points are (4 , , , , , , • In dimension four the points are the permutations of (2 , , , ↔ n ≥
5, a simple computation shows that Q n is not a regu-lar polytope. Nevertheless the polytope has nice properties. For example indimension five: one vertex has coordinates: (5 , , , , , , ,
8) forma regular octahedron. The permutations of (5 , , ,
9) form a regular tetra-hedron truncated at two symmetric vertices: The faces are either triangleseither hexagons. The permutations of (5 , , ,
9) form a polyhedron with 12vertices, 14 faces and 24 edges. The faces are quadrilateral or triangles.9 roposition 6.2.
Consider the boundary points associated to the secondperiodic billiard word in Theorem 3.4. There is one of the faces which con-tains n + 1 points. The convex hull generated by these points is a polytopesimilar to Q n . A natural question is the existence of periodic words in a non regular simplex.A good method to find such periodic orbits is to use stability. A periodicorbit is said to be stable if it survives in a small perturbation of the polytope.There is a characterization of stable periodic words in dimension two in[GSV92]. In dimension three, a sufficient condition was given in [B´e08].There is no generalization to higher dimensions, so that we can not provethe existence of periodic words in an n -dimensional simplex. Nevertheless,a generalization could appear to be natural and derivable.The condition of stability in [B´e08] is given with the matrix S v definedin Section 5. If this matrix is different from the identity then the periodictrajectory is stable. The billiard paths obtained in ∆ in the theorem arestable. References [B´e08] N. B´edaride. Periodic billiard trajectories in polyhedra.
ForumGeom. , 8:107–120, 2008.[DFT11] Diana Davis, Dmitry Fuchs, and Serge Tabachnikov. Periodictrajectories in the regular pentagon.
Mosc. Math. J. , 11(3):439–461, 629, 2011.[GKT95] G. Gal ′ perin, T. Kr¨uger, and S. Troubetzkoy. Local instability oforbits in polygonal and polyhedral billiards. Comm. Math. Phys. ,169(3):463–473, 1995.[GSV92] G. A. Gal ′ perin, A. M. St¨epin, and Ya. B. Vorobets. Periodicbilliard trajectories in polygons: generation mechanisms. UspekhiMat. Nauk , 47(3(285)):9–74, 207, 1992.[GZ03] G. Galperin and D. Zvonkine. Periodic billiard trajectories inright triangles and right-angled tetrahedra.
Regul. Chaotic Dyn. ,8(1):29–44, 2003.[HS09] W. Patrick Hooper and Richard Evan Schwartz. Billiards in nearlyisosceles triangles.
J. Mod. Dyn. , 3(2):159–231, 2009.[MT02] H. Masur and S. Tabachnikov. Rational billiards and flat struc-tures. In
Handbook of dynamical systems, Vol. 1A , pages 1015–1089. North-Holland, Amsterdam, 2002.10Sch06] R. Schwarz. Obtuse triangular billiards i: Near the (2,3,6) triangle.
Journal of Experimental Mathematics , 15(2), 2006.[Ste75] F. Stenman. Periodic orbits in a tetrahedral mirror.