PUBLICATIONS DE L’INSTITUT MATHÉMATIQUE

Nouvelle série, tome 94 (108) (2013), 17–30 DOI: 10.2298/PIM1308017D

MINKOWSKI PLANE, CONFOCAL CONICS,

AND BILLIARDS

Vladimir Dragović and Milena Radnović

Abstract. Geometry of confocal conics in the Minkowski plane and related

billiard dynamics are studied in details. Periodic trajectories are described
and several new examples are presented. Topological properties of the ellip-
tical billiards are analyzed and the results are formulated in the terms of the
Fomenko graphs.

1. Introduction
Geometry of confocal families of quadrics in pseudo-Euclidean spaces of arbi-
trary dimension d and any signature, and related billiard dynamics have recently
been studied by the authors in [8]. With a goal to give a complete description of
periodic billiard trajectories within ellipsoids, there we introduced a new discrete
combinatorial-geometric structure associated to a confocal pencil of quadrics, by
which the quadrics were decomposed into new relativistic quadrics.
The aim of the present paper is, staying focused to the two-dimensional case,
to explain in more detail constructions from [8] and to provide new results and
examples.
In Section 2 we list necessary introductory notions and definitions for the
Minkowski plane, together with definition and properties of confocal families and
relativistic conics.
In Section 3, we consider elliptical billiards. In Section 3.2, we derive an-
alytic Cayley-type conditions for periodic trajectories of elliptical billiard in the
Minkowski plane. Such conditions, for ellipsoidal billiards in the pseudo-Euclidean

2010 Mathematics Subject Classification: 37J3, 70H06, 70H12.

Key words and phrases: Confocal conics, Poncelet theorem, periodic billiard trajectories,
Minkowski plane, light-like billiard trajectories.
The research which led to this paper was partially supported by the Serbian Ministry
of Education and Science (Project no. 174020: Geometry and Topology of Manifolds and In-
tegrable Dynamical Systems) and by Mathematical Physics Group of the University of Lis-
bon (Project Probabilistic approach to finite and infinite dimensional dynamical systems,
PTDC/MAT/104173/2008 ). M. R. is grateful to the Weizmann Institute of Science (Rehovot,
Israel) and The Abdus Salam ICTP (Trieste, Italy) for their hospitality and support in various
stages of work on this paper.
17
18 DRAGOVIĆ AND RADNOVIĆ

spaces of the arbitrary dimensions, are derived in [8]. Here we give another proof,
adopted to the planar case, see Theorem 3.1. In the next Section 3.3, we study
light-like trajectories of such billiards and derive a periodicity criterion in a simple
form, see Theorem 3.2. That criterion, from [8], is proved here in the algebro-
geometric way. The equivalence of this simple criterion with the more complicated
condition of Cayley’s type is illustrated by several new examples. The equivalence
of the elliptical billiard flow to a certain rectangular billiard flow is stated in Theo-
rem 3.3, while for the proof we refer the reader to [8]. In Section 3.4, we conclude
with a complete description of topological properties of elliptical billiards in the
Minkowski plane. We are using Fomenko invariants, see Theorem 3.4, as we did in
the Euclidean case in [6, 7]. The results form Section 3.4 are new.

2. Confocal conics in the Minkowski plane

We start by giving necessary notions of the Minkowski plane in Section 2.1
and we review properties of confocal families of conics in the Minkowski plane in
Section 2.2. In Section 2.3, relativistic conics are presented, following [1], see also
[8].
2.1. The Minkowski plane. The Minkowski plane is R2 with the Minkowski
scalar product
(2.1) hx, yi = x1 y1 − x2 y2 .
p
The Minkowski distance between points x, y is dist(x, y) = hx − y, x − yi.
Since the scalar product can be negative, notice that the Minkowski distance can
have imaginary values as well. In that case, we choose the value of the square root
with the positive imaginary part.
Let ℓ be a line in the Minkowski plane, and v its vector. ℓ is called:
• space-like if hv, vi > 0;
• time-like if hv, vi < 0;
• and light-like if hv, vi = 0.
Two vectors x, y are orthogonal in the Minkowski plane if hx, yi = 0. Note that a
light-like vector is orthogonal to itself.
2.2. Confocal families of conics. Here, we give a review of basic properties
of families of confocal conics in the Minkowski plane, see [8].
Denote by
x2 y2
(2.2) E : + =1
a b
an ellipse in the plane, with a, b being fixed positive numbers.
The associated family of confocal conics is
x2 y2
(2.3) Cλ : + = 1, λ ∈ R.
a−λ b+λ
The family is shown in Figure 1. We may distinguish the following three sub-
families in the family (2.3):
 MINKOWSKI PLANE, CONFOCAL CONICS, AND BILLIARDS 19

Figure 1. Family of confocal conics in the Minkowski plane.

• for λ ∈ (−b, a), conic Cλ is an ellipse;

• for λ < −b, conic Cλ is a hyperbola with x-axis as the major one;
• for λ > a, it is a hyperbola again, but now its major axis is y-axis.
In addition, there are three degenerated quadrics: Ca , C−b , C∞ , corresponding to y-
axis, x-axis, and √
the line at
the infinity
√ respectively.

√The confocal

family
√ has
three
pairs of foci: F1 a + b, 0 , F2 − a + b, 0 ; G1 0, a + b , G2 0, − a + b ; and
H1 (1 : −1 : 0), H2 (1 : 1 : 0) on the line at the infinity.
We notice four distinguished lines:
√ √
x + y = a + b, x + y = − a + b,
√ √
x − y = a + b, x − y = − a + b.
These lines are common tangents to all conics from the family.
Conics in the Minkowski plane have geometric properties analogous to the
conics in the Euclidean plane. Namely, for each point on conic Cλ , either√sum or
difference of its Minkowski distances from the foci F1 and F2 is equal to 2 a − λ;
either
√ sum or difference of the distances from the other pair of foci G1 , G2 is equal
to 2 −b − λ [8].

2.3. Relativistic conics. In the Minkowkski plane, it is natural to consider

relativistic conics, which are suggested in [1]. In this subsection, we give a brief
account of the related analysis.
√
√

Consider points F1 a + b, 0 and F2 − a + b, 0 .
20 DRAGOVIĆ AND RADNOVIĆ

For a given constant c ∈ R+ ∪ iR+ , a relativistic ellipse is the set of points X

satisfying dist(F1 , X) + dist(F2 , X) = 2c, while a relativistic hyperbola is the union
of the sets given by the following equations:
dist(F1 , X) − dist(F2 , X) = 2c,
dist(F2 , X) − dist(F1 , X) = 2c.
Relativistic conics can be described as follows.
√
0 < c < a + b: The corresponding relativistic conics lie on ellipse Ca−c2
from family (2.3). The ellipse Ca−c2 is split into four arcs by touching
points with the four common tangent lines; thus, the relativistic ellipse
is the union of the two arcs intersecting the y-axis, while the relativistic
hyperbola
√ is the union of the other two arcs.
c > a + b: The relativistic conics lie on Ca−c2 –a hyperbola with x-axis as
the major one. Each branch of the hyperbola is split into three arcs by
touching points with the common tangents; thus, the relativistic ellipse
is the union of the two finite arcs, while the relativistic hyperbola is the
union of the four infinite ones.
c is imaginary: The relativistic conics lie on hyperbola Ca−c2 –a hyperbola
with y-axis as the major one. As in the previous case, the branches are
split into six arcs in total by common points with the four tangents. The
relativistic ellipse is the union of the four infinite arcs, while the relativistic
hyperbola is the union of the two finite ones.
The conics are shown in Figure 2.
Notice that all relativistic ellipses are disjoint one with another, as well as all
relativistic hyperbolas. Moreover, at the intersection point of a relativistic ellipse
which is a part of the geometric conic Cλ1 from the confocal family (2.3) and a
relativistic hyperbola belonging to Cλ2 , it is always λ1 < λ2 .

3. Elliptical billiards
Elliptical billiards in Euclidean plane have been studied intensively, see for
example [13] and [7] and references therein. These billiards serve as a paradigm
of complete integrable discrete systems. They can be seen as discretizations of
geodesics on ellipsoids as well. In the setting of Minkowski geometry, geodesics on
ellipsoidis have been studied recently in [9]. Related billiard systems in pseudo-
Euclidean geometry were studied in [12], and then in [8]. Such billiard systems are
closely related to a recent concept of contact complete integrability, introduced in
[11], see also [10].

3.1. Billiard reflection in the Minkowski plane. Let v be a vector and

p a line in the Minkowski plane. Decompose vector v into the sum v = a + np of
a vector np orthogonal to p and a belonging to p. Then vector v ′ = a − np is the
billiard reflection of v on p. It is easy to see that v is also the billiard reflection of
v ′ with respect to p. Moreover, since hv, vi = hv ′ , v ′ i, vectors v, v ′ are of the same
type.
 MINKOWSKI PLANE, CONFOCAL CONICS, AND BILLIARDS 21

Figure 2. Relativistic conics in the Minkowski plane: relativistic

ellipses are represented by full lines, and hyperbolas by dashed
ones.

Note that v = v ′ if v is contained in p and v ′ = −v if it is orthogonal to p. If

np is light-like, which means that it belongs to p, then the reflection is not defined.
Line ℓ′ is the billiard reflection of ℓ off a smooth curve S if their intersection
point ℓ ∩ ℓ′ belongs to S and the vectors of ℓ, ℓ′ are reflections of each other with
respect to the tangent line of S at this point.
The lines containing segments of a given billiard trajectory within S are all of
the same type: they are all either space-like, time-like, or light-like.
Now, take S to be ellipse E. Then it is possible to extend the reflection mapping
to those points where the tangent lines contain the orthogonal vectors. Namely,
the limit of the motion in the neighborhood of such points is that the billiard
particle, after collision with the boundary, would move exactly in the opposite
direction. Thus, at such points, we define that a vector reflects into the opposite
one, i.e. v ′ = −v and ℓ′ = ℓ. Since close trajectories reflect two times in the
neighborhood, it is natural to count each such reflection twice. For the detailed
explanation, see [12].
Billiard trajectories within ellipses in the Minkowski plane have caustic proper-
ties: each segment of a given trajectory will be tangent to the same conic confocal
with the boundary, see [8].
The famous focal property also holds: if one line containing the initial segment
of a given trajectory within an ellipse of the family (2.3) contains a focus of the
family, say F1 , G1 , or H1 , then the line containing the next segment will pass
22 DRAGOVIĆ AND RADNOVIĆ

through F2 , G2 , or H2 respectively, unless the tangent line to the boundary at the

reflection point is light-like.

3.2. Periodic trajectories of elliptical billiard. Analytic conditions for

existence of closed polygonal lines inscribed in one conic and circumscribed about
another one in the projective plane are derived by Cayley [3,4]. They can be applied
to billiard trajectories within ellipses in the Minkowski plane as well, since each such
trajectory has a caustic among confocal conics. In this section, we shall analyze in
more detail some particular properties related to the Minkowski geometry.
Theorem 3.1. In the Minkowski plane, consider a billiard trajectory T within
ellipse E given by equation (2.2).
The trajectory is periodic with period n = 2m if and only if the following
condition is satisfied:
 
B3 B4 . . . Bm+1
 B4 B5 . . . Bm+2 
(3.1) det 
 ...
 = 0.
... ... ... 
Bm+1 Bm+2 . . . B2m−1
Trajectory T is periodic with period n = 2m + 1 if and only if Cα is an ellipse and
the following condition is satisfied:
 
B3 B4 . . . Bm+2
 ... ... ... ... 
(3.2) det 
Bm+1 Bm+2 . . . B2m  = 0.


Cm+1 Cm+2 . . . C2m

Here:
(a − t)(b + t)(α − t) = B0 + B1 t + B2 t2 + · · · ,
p
r
(a − t)(b + t)
= C0 + C1 t + C2 t2 + · · ·
α−t
are the Taylor expansions around t = 0.
Proof. Each point inside E is the intersection of exactly two ellipses Cλ1 and
Cλ2 from (2.3). Parameters λ1 , λ2 are generalized Jacobi coordinates. Take λ1 < λ2 .
Consider first the case when Cα is a hyperbola. Then along T these coordinates
will take values in segments [−b, 0] and [0, a] respectively with the endpoints of the
segments as the only local extrema. λ1 achieves value −b at the intersections of
T with the x-axis, while λ2 achieves a at the intersections with y-axis. At each
reflection point, one of the coordinates achieves value 0. They can both be equal to
0 only at the points where E has a light-like tangent, and there reflection is counted
twice.
This means that on a closed trajectory the number of reflections is equal to the
number of intersection points with the coordinate axes. Since a periodic trajectory
crosses each of the coordinate axes even number of times, the first part of the
theorem is proved.
 MINKOWSKI PLANE, CONFOCAL CONICS, AND BILLIARDS 23

The condition on T to become closed after n reflections on E, n1 crossings over

x-axis, and n2 over y-axis is that the equality n1 Pa + n2 P−b = nP0 holds on the
elliptic curve s2 = (a − t)(b + t)(α − t), where by Pβ we denoted a point on the
curve correspondig to t = β, and P∞ is taken to be the neutral for the elliptic curve
group.
From the previous discussion, n1 + n2 = n and all three numbers are even.
Pa and P−b are branching points of the curve, thus 2Pa = 2P−b = 2P∞ , so the
condition becomes nP0 = nP∞ , which is equivalent to (3.1).
Now suppose Cα is an ellipse. The generalized Jacobi coordinates take values
in segments [−b, 0], [0, α] or in [α, 0], [0, a], depending on the sign of α. Since both
cases are processed in a similar way, we assume α < 0.
Coordinate λ1 has extrema on T at the touching points with the caustic and
some of the reflection points, while λ2 has extrema at the crossing points with
y-axis and some of the reflection points.
The condition on T to become closed after n reflections on E, with n1 crossings
over y-axis, and n2 touching points with the caustic is n1 Pa + n2 Pα = nP0 , with
n1 + n2 = n and n1 even.
Thus, for n even we get (3.1) in the same manner as for a hyperbola as a
caustic.
For n odd, the condition is equivalent to nP0 = (n − 1)P∞ + Pα . Notice that
one basis of the space L((n − 1)P∞ + Pα ) is
s
1, t, . . . , tm , s, ts, . . . , tm−2 s, .
t−α
Using this basis, as it is shown in [5, 7], we obtain (3.2). 

Example 3.1 (3-periodic trajectories). Let us find all 3-periodic trajectories

within ellipse E given by (2.2) in the Minkowski plane, i.e., all conics Cα from the
confocal family (2.3) corresponding to such trajectories.
The condition is
3a2 b2 + 2a2 bα − 2ab2 α − a2 α2 − 2abα2 − b2 α2
C2 = = 0,
8(ab)3/2 α5/2
which gives the following solutions for the parameter α of the caustic:
ab p
a − b − 2 a2 + ab + b2 ,


α1 = 2
(a + b)
ab p
a − b + 2 a2 + ab + b2 .


α2 = 2
(a + b)
Notice that −b < α1 < 0 < α2 < a so both caustics Cα1 , Cα2 are ellipses.

Example 3.2 (4-periodic trajectories). By Theorem 3.1, the condition is B3= 0.

Since
(−ab − aα + bα)(−ab + aα + bα)(ab + aα + bα)
B3 = ,
16(abα)5/2
24 DRAGOVIĆ AND RADNOVIĆ

we obtain the following solutions:

ab ab ab
α1 = , α2 = , α3 = − .
b−a a+b a+b
Since α1 6∈ (−b, a) and α2 , α3 ∈ (−b, a), conic Cα1 is a hyperbola, while Cα2 , Cα3
are ellipses.
Example 3.3 (5-periodic trajectories). The condition is
 
B3 B4
det = 0.
C3 C4
Taking a = b = 1, we get that this is equivalent to 64α6 − 16α4 − 52α2 + 5 = 0. This
equation has four solutions in R, all four contained in (−1, 1), and two conjugated
solutions in C.
Remark 3.1. For the analytic description of the periodic billiard trajectories
within ellipsoids in the higher-dimensional pseudo-Euclidean spaces, see [8]. For
the Euclidean case, see [7] and references therein. Some aspects have recently been
discussed in [14].
3.3. Light-like trajectories. In this section we consider in more detail light-
like trajectories of elliptical billiard, see Figure 3 for an example of such a trajectory.
We are going to review results from [8] and illustrate them by some examples.

Figure 3. Light-like billiard trajectory.

Periodic light-like trajectories. Let us first notice that segments of light-like

billiard trajectories are alternately parallel to two light-like directions in the plane
(see Figure 3), thus a trajectory can close only after even number of reflections.
The analytic condition for n-periodicity of light-like billiard trajectory within
the ellipse E given by equation (2.2) can be derived as in Theorem 3.1. We get
the condition stated in (3.1),
p with α = ∞, i.e., (Bi ) are coefficients in the Taylor
2
expansion around t = 0 of (a − t)(b + t) = B0 + B1 t + B2 t + · · · .
Now, we are going to derive analytic condition for periodic light-like trajectories
in another way, which will lead to a more compact form of (3.1).
 MINKOWSKI PLANE, CONFOCAL CONICS, AND BILLIARDS 25

Theorem 3.2. Light-like billiard trajectory within ellipse E is periodic with

period n, where n is an even integer if and only if
p n kπ n  n o
(3.3) arctan a/ b ∈ | 1 6 k < , k, =1 .
n 2 2
Proof. The isospectral curve for elliptic billiard with caustic C∞ is the rational
curve y 2 = (x − a)(x + b), with the two infinite points identified. Condition for
n-periodicity of a light-like trajectory is that divisors nP0 and nP−b are equivalent
on this curve with one singular point. Here, P0 is one of the points corresponding
+ −
to the value x = 0, P−b corresponds to x = −b, y = 0. Denote by P∞ , P∞ the
points at infinity corresponding to lines x = y and x = −y respectively. On the
+ −
isospectral curve, P∞ and P∞ are identified.
y

1/2
Introduce the following reparametrization of the curve: t = x+b = x−ax+b .
+
p −
Then t(P0 ) = i a/ b, t(P−b ) = ∞, t(P∞ ) = 1, t(P∞ ) = −1. Up to a constant
factor, there is a unique rational function with the only zero of order n at P0 and
a pole of order n at P−b :
p
n
f = t − i a/ b t−n .
+ −
p
n
We need to check if f has the same values at P∞ and P∞ , i.e., 1 − i a/ b =
p
n p
n
−1 − i a/ b . Since n is even, this can be true only if 1 − i a/ b is real.
Thus, (3.3) follows. 
Remark 3.2. A simpler and more elementary proof of Theorem 3.2 is given in
[8]. The one presented here is interesting and instructive because of the algebro-
geometric tools involved.
As an immediate consequence, we get
Corollary 3.1. For a given even integer n, the number of different ratios of
the axes of ellipses having n-periodic light-like billiard trajectories is equal to:
(
ϕ(n)/2, if n is not divisible by 4,
ϕ(n)/4, if n is divisible by 4.
ϕ is Euler’s totient function, i.e., the number of positive integers not exceeding n
that are relatively prime to n.
Remark 3.3. There are four points on E where the tangents are light-like.
Those points cut four arcs on E. An n-periodic trajectory within E hits each one
of a pair of opposite arcs exactly k times, and n2 − k times the arcs from the other
pair.
Example 3.4 (4-periodic light-like trajectories). It is elementary to see that
only circles allow such a trajectory. However, we can also deduce it from condition
(a − b)(a + b)2
(3.1). For n = 4, this condition reads B3 = 0 with B3 = . Thus, it
16(ab)5/2
is equivalent to a = b.
On the other hand, condition (3.3) states a/ b = tan π4 = 1.
p
26 DRAGOVIĆ AND RADNOVIĆ

Example 3.5 (6-periodic light-like trajectory). For n = 6, the condition (3.1)

is  
B3 B4
det = 0,
B4 B5
which is equivalent to
(a − 3b)(3a − b)(a + b)6
= 0.
(4ab)7
Thus, we get that light-like
√ billiard trajectories are 6-periodic in ellipses with the
ratio of the axes equal to 3.
From condition (3.3), we get the same:
√ 1
r  
a n π πo
∈ tan , tan = 3, √ .
b 6 3 3
A few 6-periodic trajectories are shown in Figure 4.

Figure 4. Light-like billiard trajectories with period 6 in the el-

lipse satisfying a = 3b.

Example 3.6 (8-periodic light-like trajectories). For n = 8, the condition (3.1)

is  
B3 B4 B5
(a − b)(a + b)12 (a2 − 6ab + b2 )
det B4 B5 B6  = √ = 0.
B5 B6 B7 225 (ab)13 · ab

From here, we get that light-like billiard

√ trajectories are 8-periodic in ellipses with
the ratio of the axes equal to 1 + 2.
Condition (3.3) gives:
√ √
r  
a π 3π
∈ tan , tan = 2 − 1, 2 + 1 .
b 8 8
A few of such trajectories are shown in Figure 5.
 MINKOWSKI PLANE, CONFOCAL CONICS, AND BILLIARDS 27

Figure 5. Light-like billiard

√ trajectories with period 8 in the el-
lipse satisfying a = (3 + 2 2)b.

Example 3.7 (10-periodic light-like trajectories). For n = 10, the condition

(3.1) is
 
B3 B4 B5 B6
B4 B5 B6 B7  (a + b)20 (5a2 − 10ab + b2 )(a2 − 10ab + 5b2 )
det 
B5 B6 B7 B8  =
 = 0.
(4ab)22
B6 B7 B8 B9
From here, we get that light-like billiard
p trajectories are 10-periodic in ellipses with
√ p √
the ratio of the axes equal to either 1 + 2/ 5 or 5 + 2 5.
From condition (3.3), we get
r  
a π 2π 3π 4π
∈ tan , tan , tan , tan .
b 10 10 10 10
Since
π p √ 1 1
tan = 1 − 2/ 5 = p √ =
10 5+2 5 tan 4π
10
3π p √ 1 1
tan = 1 + 2/ 5 = p √ = ,
10 5−2 5 tan 2π
10
both conditions give the same result.
A few of such trajectories are shown in Figures 6 and 7.

Example 3.8 (12-periodic light-like trajectories). For n = 12, the condition

(3.1) is  
B3 B4 B5 B6 B7
B4 B5 B6 B7 B8 
 
B5 B6 B7 B8 B9  = 0,
det  
B6 B7 B8 B9 B10 
B7 B8 B9 B10 B11
which can be transformed into
(a − 3b)(a − b)(3a − b)(a + b)30 (a2 − 14ab + b2 )
√ = 0.
(4ab)22 ab
28 DRAGOVIĆ AND RADNOVIĆ

Figure 6. Light-like billiard√ trajectories

with period 10 in the
ellipse satisfying a = 1 + 2/ 5 b.

Figure 7. Light-like billiard

√
trajectories with period 10 in the
ellipse satisfying a = 5 + 2 5 b.

From here, we get that light-like billiard

√ trajectories are 12-periodic in ellipses with
the ratio of the axes equal to 2 + 3.
On the other hand, condition (3.3) gives
√ √
r  
a π 5π 
∈ tan , tan = 2 − 3, 2 + 3 .
b 12 12
A few of such trajectories, obtained by a computer simulation, are shown in
Figure 8.
Light-like trajectories in ellipses and rectangular billiards.
Theorem 3.3. The flow of light-like billiard trajectories within ellipse E is
trajectorially equivalent to the flow of those billiard trajectories within a rectangle
whose angle with the sides is π4 . The ratio of the sides of the rectangle is equal to
π
p − 1.
2 arctan a/ b
Remark 3.4. The flow of light-light billiard trajectories within a given oval in
the Minkowski plane will be trajectorially equivalent to the flow of certain trajecto-
ries within a rectangle whenever invariant measure m on the oval exists such that
 MINKOWSKI PLANE, CONFOCAL CONICS, AND BILLIARDS 29

Figure 8. Light-like billiard

√
trajectories with period 12 in the
ellipse satisfying a = 7 + 4 3 b.

m(AB) = m(CD) and m(BC) = m(AD), where A, B, C, D are points on the oval
where the tangents are light-like.
3.4. Topological properties of elliptical billiards. Here, we are going to
present the topological description of elliptical billiard in the Minkowski plane. In
order to do this, we use Fomenko invariants, see [2] and references therein. The
corresponding analysis for the Euclidean case has been done in [6].1
Theorem 3.4. The isoenergy manifold corresponding to the billiard system
within ellipse E (2.2) in the Minkowski plane is represented by the Fomenko graph
in Figure 9.

A r= 0 A
0 r=
ε= 1
1 r=0 ε=
B B
0 ε = −1 r=
r= 0
ε=
=1 1
A ε A
n=0 n=0

| | | | |

λ = +0 λ=a λ=∞ λ = −b λ = −0

Figure 9. Fomenko graph for elliptical billiard in the Minkowski plane.

1The authors are thankful to Victoria Fokicheva and Dmitry Tonkonog for their useful remark
on the Fomenko graph for elliptical billiard in the Euclidean plane.
30 DRAGOVIĆ AND RADNOVIĆ

Proof. Each level set of the isoenergy manifold corresponds to the billiard
motion with a fixed caustic Cλ , λ ∈ R ∪ {∞}.
For λ 6∈ {0, a, −b}, the level sets are non-degenerate. The level set is a torus
when the caustic is a hyperbola, i.e., for λ ∈ (a, +∞) ∪ (−∞, −b) ∪ {∞}. If
λ ∈ (0, a) ∪ (−b, 0) the caustic is an ellipse, and the level set is a union of two
tori.
If λ = −b, the level set contains one periodic trajectory which is placed along y-
axis and homoclinic trajectories, which are naturally grouped onto two separatices–
each one being projected onto the other side of the y-axis. The same, just with
respect to the x-axis, holds for the level set corresponding to λ = a. Those level
sets are represented by Fomenko atom B.
Let us consider the limit case λ = 0, when the caustic coincides with the billiard
table edge. In this case, the limit motion will take place on the four arcs on the
E whose ends are touching points with four joint tangents of the confocal family
(2.3). More precisely, if λ approaches zero from below, then the limit motion will
take place on the two arcs on the left and on the right side of y-axis; if λ approaches
zero from above, then the limit motion will take place on the two arcs below and
above x-axis. Four periodic trajectories appearing in the limit λ = 0 correspond to
the A-atoms in Figure 9. 

References
[1] G. Birkhoff, R. Morris, Confocal conics in space-time, Am. Math. Monthly 69(1) (1962), 1–4.
[2] A. V. Bolsinov, A. T. Fomenko, Integrable Hamiltonian Systems: Geometry, Topology, Clas-
sification, Chapman and Hall/CRC, Boca Roton, Florida, 2004.
[3] A. Cayley, Developments on the porism of the in-and-circumscribed polygon, Philos. Maga-
zine 7 (1854), 339–345.
[4] , On the porism of the in-and-circumscribed polygon, Philos. Trans. Royal Soc. London
51 (1861), 225–239.
[5] V. Dragović, M. Radnović, A survey of the analytical description of periodic elliptical billiard
trajectories, J. Math. Sci. 135(4) (2006), 3244–3255.
[6] , Integrable billiards and quadrics, Russian Math. Surveys 65(2) (2010), 136–197.
[7] , Poncelet Porisms and Beyond, Springer, Birkhauser, Basel, 2011.
[8] , Ellipsoidal billiards in pseudo-Euclidean spaces and relativistic quadrics, Adv. Math.
231 (2012), 1173-1201.
[9] D. Genin, B. Khesin, S. Tabachnikov, Geodesics on an ellipsoid in Minkowski space, Enseign.
Math. 53 (2007) 307–331.
[10] B. Jovanović, Noncommutative integrability and action-angle variables in contact geometry,
J. Symplectic Geom. (2012), arXiv:1103.3611.
[11] B. Khesin, S. Tabachnikov, Contact complete integrability, Regul. Chaotic Dyn. 15(4-5)
(2010), 504–520.
[12] , Pseudo-Riemannian geodesics and billiards, Adv. Math. 221 (2009), 1364–1396.
[13] V. Kozlov, D. Treshchëv, Billiards, Am. Math. Soc., Providence, RI, 1991,
[14] R. Ramirez-Ros, On Cayley conditions for billiards inside ellipsoids, arXiv:1211.6557, 2012,

Mathematical Institute SANU, Belgrade, Serbia (both authors)

University of Texas, Dallas (V. Dragović)

vladad@mi.sanu.ac.rs
milena@mi.sanu.ac.rs