See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/220654433

On Bicycle Tire Tracks Geometry, Hatchet Planimeter, Menzin's Conjecture, and

Oscillation of Unicycle Tracks

Article  in  Experimental Mathematics · January 2009

DOI: 10.1080/10586458.2009.10128894 · Source: DBLP

CITATIONS READS

17 141

2 authors:

Mark Levi Serge Tabachnikov

Pennsylvania State University Pennsylvania State University
84 PUBLICATIONS   1,385 CITATIONS    232 PUBLICATIONS   2,174 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Resonance View project

All content following this page was uploaded by Serge Tabachnikov on 05 June 2014.

The user has requested enhancement of the downloaded file.

 On bicycle tire tracks geometry, hatchet
planimeter, Menzin’s conjecture and oscillation
arXiv:0801.4396v1 [math.DG] 28 Jan 2008

of unicycle tracks
Mark Levi∗ and Serge Tabachnikov†
February 3, 2008

Abstract
The model of a bicycle is a unit segment AB that can move in
the plane so that it remains tangent to the trajectory of point A (the
rear wheel is fixed on the bicycle frame); the same model describes the
hatchet planimeter. The trajectory of the front wheel and the initial
position of the bicycle uniquely determine its motion and its terminal
position; the monodromy map sending the initial position to the ter-
minal one arises. According to R. Foote’s theorem, this mapping of a
circle to a circle is a Moebius transformation. We extend this result
to multi-dimensional setting. Moebius transformations belong to one
of the three types: elliptic, parabolic and hyperbolic. We prove a 100
years old Menzin’s conjecture: if the front wheel track is an oval with
area at least π then the respective monodromy is hyperbolic. We also
study bicycle motions introduced by D. Finn in which the rear wheel
follows the track of the front wheel. Such a ”unicycle” track becomes
more and more oscillatory in forward direction. We prove that it can-
not be infinitely extended backward and relate the problem to the
geometry of the space of forward semi-infinite equilateral linkages.
∗
Department of Mathematics, Pennsylvania State University, University Park, PA
16802, USA; e-mail: [email protected].
†
Department of Mathematics, Pennsylvania State University, University Park, PA
16802, USA; e-mail: [email protected]

1
1 Introduction
The geometry of bicycle tracks is a rich and fascinating subject. Here is a
sampler of questions:

1. Given the tracks of the rear and front wheel, can you tell which way
the bicycle went?

2. The track of the front wheel is a smooth simple closed curve. Can one
ride the bicycle so that the rear wheel track also closes up?

3. Can one ride a bicycle in such a way that the tracks of the rear and
front wheels coincide1 ?

Our model of a bicycle is an oriented segment, say, AB, of length ` that

can move in the plane in such a way that the trajectory of point A always
remains tangent to the segment. Point A represent the rear wheel, point
B the front wheel; the rear wheel is fixed on the bicycle frame whereas the
front wheel can turn, and this explains the law of motion. (Most often we set
` = 1 – this can be always assumed by making a dilation – but sometimes we
shall consider ` as a parameter and allow it to take very small or very large
values.) Thus the end point of the oriented tangent segment to the trajectory
of the rear wheel traces the trajectory of the front wheel, see [6, 10].
The same mathematical model describes another mechanical device, the
Prytz or hatchet planimeter, see [2, 4, 7]. Various kinds of planimeters were
popular objects of study in the late 19th and early 20th century.
The first of the above questions has the following answer: generically,
one can determine the direction, but in some special cases one cannot: for
example, for concentric circles of radii r and R satisfying r2 + `2 = R2 .
Surprisingly, the problem of describing such “ambiguous” pairs of closed
tracks is equivalent to Ulam’s problem of describing (2-dimensional) bodies
that float in equilibrium in all positions. See [18, 19, 20, 21] for a variety of
results and references.
The content of the present paper has to do with the other two questions.
In Section 2 we place the problem into the framework of contact geometry.
We allow the trajectory of the rear wheel to be a wave front, that is, to have
cusp singularities, but we show that the trajectory of the front wheel remains
1
Other than along a straight line.

2
smooth. We deduce a useful differential equation relating the motions of the
rear and the front wheels.
Fixing a path Γ of the front wheel gives rise to a circle map: the initial
direction of the segment, characterized by a point on the circle, determines
its final direction, see figure 1. We will refer to this map of the circle to it-
self (the two circles are identified by parallel translation) as the monodromy
map.2 It is a beautiful theorem of R. Foote [7] (see also [13]) that, for every
trajectory of the front wheel, the monodromy map is a Möbius transforma-
tion. In Section 3 we reprove this theorem and extend it to bicycle motion
in Euclidean space of any dimension.
Γ

Figure 1: The circle mapping generated by the curve Γ. According to Foote’s

theorem, this mapping is a Möbius transformation.

A (non-trivial) Möbius transformation is of one of the three types: elliptic,

parabolic or hyperbolic. The former have no fixed points and the latter have
exactly two, one attracting and one repelling (parabolic ones have a single,
neutral fixed point). Suppose the trajectory of the front wheel is a closed
curve. Then, up to conjugation, the respective monodromy, and therefore its
type, does not depend on the initial point. In Section 3 we give a necessary
and sufficient condition for the monodromy to be parabolic: the trajectory
of the rear wheel is a closed wave front with the total algebraic arc length
equal to zero (the sign of the arc length changes when passing a cusp).
Still assuming that the trajectory of the front wheel is closed, a fixed
point of the monodromy map corresponds to a closed trajectory of the rear
wheel. Thus, in the hyperbolic case, for a given closed trajectory of the front
wheel, there are exactly two bicycle motions such that the trajectory of the
rear wheel is closed; each of these motions is hyperbolically attracting for
one of the choices of the direction of motion; examples are shown in figure 2
(1) and (4). In contrast, in the elliptic case, no trajectory of the rear wheel
2
T. Tokieda suggested the term “opisthodromy” (literary, rear track).

3
 B

1 2

3 4

Figure 2: Examples 1 and 4 are hyperbolic; 2 and 3 are elliptic. The areas
bounded by the two curves in 1 differ by π`2 .

4
closes after one cycle. It is worth mentioning that, for some trajectories of
the front wheel, the monodromy is the identity: for every bicycle motion the
trajectory of the rear wheel closes up.
A hundred years old conjecture by Menzin [14] states, in our terms, that
if the trajectory of the front wheel is a closed convex curve bounding area
greater than π`2 , then the respective monodromy is of the hyperbolic type.
In Section 4 we prove this conjecture. The main tool is the classical Wirtinger
inequality. Earlier Foote [7] proved Menzin’s conjecture for parallelograms.
Section 5 concerns Finn’s construction of bicycle motion leaving a single
track [6]. Consider a “seed” curve, tangent to the x-axis at points 0 and 1
with all derivatives and oriented to the right (the “fat” curve in figure 16).
This curve is the initial trajectory of the rear wheel; drawing the tangent
segments of length 1 to it yields the next curve which is tangent to the x-
axis at points 1 and 2 with all derivatives. Iterating this process, one obtains
bicycle motion leaving a unicycle track, i.e., a curve which both wheels follow.
Numerical study shows that, unless the seed curve is horizontal, the re-
sulting unicycle track becomes more and more oscillating, figure 16. We
prove that the number of intersections with the x-axis and the number of
extrema of the height function increase at least by one with every iteration
of this construction. As a consequence, the seed curve with finitely many
intersections with the x-axis (or a finite number of extrema) has at most
finitely many preimages under Finn’s construction. This means that the cor-
responding unicycle track cannot extend back indefinitely. We also make a
number of conjectures on the Finn construction strongly supported by nu-
merical evidence.
A unicycle track can be viewed as an integral curve of a direction field in
a certain infinite dimensional space. Specifically, we consider the configura-
tion space of equilateral forward infinite linkages in the plane. We constrain
the velocity of the ith vertex to the direction of the ith link (heuristically,
the ith link is the position of the bike on the i − 1st step of Finn’s construc-
tion). This constraint defines a field of directions. Now, a forward bicycle
motion generating a single track corresponds to a particular integral curve of
this field of directions. This field does not satisfy the uniqueness property:
through every points there pass infinitely many smooth integral curves. We
also generalize Finn’s construction for an arbitrary initial equilateral forward
infinite linkage in which the adjacent links are not perpendicular (the Finn
construction corresponds to a linkage aligned along a line).

5
 Acknowledgments. It is a pleasure to thank M. Kapovich, R. Mont-
gomery, A. Novikov, R. Schwartz, S. Wagon and V. Zharnitsky for their
interetst and help. The first author was supported by an NSF grant DMS-
0605878 and the second one by an NSF grant DMS-0555803.

2.5

1.5

0.5

−0.5

−1

−1.5

−1 0 1 2 3 4

Figure 3: The figure shows first four iterates of the initial seed curve y =
46 x6 (1 − x)6 . Since this curve has only a finite order of contact with the
x–axis, only finitely many iterations are defined.

2 Preliminaries: contact geometric point of

view
We use the notation from Section 1. Denote the trajectory of the rear wheel A
by γ and that of the front wheel B by Γ. We allow γ to have cusp singularities
as in figure 4. A proper prospective is provided by contact geometry, see [1]
or [8].

6
 γ

Figure 4: Cusp of the curve γ

The position of the segment AB is determined by its foot point A(x, y)

and by the angle θ between the x–axis and the segment. The infinitesimal mo-
tions in the configuration space {(x, y, θ)} are restricted by the non–skidding
condition: (ẋ, ẏ) k (cos θ, sin θ). This condition defines a field of tangent 2-
planes in the configuration space. This field of planes is non–integrable and
is defined by the contact 1-form λ = sin θ dx − cos θ dy.
A smooth curve in a contact manifold is called Legendrian if its tangent
line at every point lies in the contact plane. Let p : M → R2 be the projection
taking a contact element to its foot point. The image of a Legendrian curve
is called a wave front; generically it is a piece-wise smooth curve with semi-
cubical cusp singularities. The singularities occur at the points where the
Legendrian curve is tangent to the fibers of the projection p. A wave front
has a well defined tangent line at every point and can be uniquely lifted to a
Legendrian curve in the space of contact elements.
The trajectory of the rear wheel uniquely determines the trajectory of
the front wheel. Denote by T the correspondence γ 7→ Γ which assigns to
point x ∈ γ the end point of the unit tangent segment to γ at x. We assume
that a continuous choice is made between the two orientations of the unit
tangent segments at a point. This amounts to choosing a coorientation of γ:
the frame formed by the coorienting vector and the chosen tangent vector is
positive. When the bicycle segment is not unit and has length `, we denote
by T` the respective transformation and by Γ` its image.
The following two lemmas address the smoothness issue.

Lemma 2.1 If γ is a C k , k ≥ 1 curve, then Γ` is C k−1 curve for all ` > 0.

7
Proof. Let γ be parametrized by its arc length s. By the definition, Γ(s) =
γ(s) + γ 0 (s), and it remains only to make sure that Γ0 = γ 0 + γ 00 6= 0. But
the last two vectors are orthogonal and the first has unit length. 2

Lemma 2.2 Even if γ has cusps, the curve Γ` is smooth for all ` > 0.

Proof. Let p1 : M → R2 take segment AB to point B. The correspon-

dence T` is the composition of the Legendrian lifting of a wave front γ and
the projection p1 . We claim that the fibers of p1 are everywhere transverse
to the contact distribution on M . This would imply the statement of the
lemma since the fibers of the projection are transverse to the Legendrian
curve p−1 (γ).
In terms of the coordinates in M , one has: p1 (x, y, θ) = (x + ` cos θ, y +
` sin θ). The vector field v = ∂θ + ` sin θ ∂x − ` cos θ ∂y is tangent to the fibers
of p1 . One has: λ(v) = `, therefore v is everywhere transverse to the contact
planes, and we are done. 2

Let γ be an oriented and cooriented closed wave front. The Maslov index
µ(γ) is the algebraic number of cusps of γ; a cusp is positive if one traverses
it along the coorientation and negative otherwise.
Let γ be an oriented and cooriented closed wave front. Denote by ρ(γ)
the rotation number, that is, the total (algebraic) number of turns made by
its tangent direction. Let Γ = T (γ).

Lemma 2.3 One has: ρ(Γ) = ρ(γ) + 21 µ(γ).

Proof. Consider the 1-parameter family of curves Γ` . By Lemma 2.2, this

is a continuous family of smooth curves, hence the rotation number is the
same for all `. Consider the case of very small `.
Along smooth arcs of γ, the curve Γ` is C 1 -close to γ. At the cusps,
smoothing occurs, and the rotation of Γ` differs from that of γ by ±π. There
are four cases, depending on the orientation and coorientation, depicted in
figure 5. When one traverses a cusp along the coorientation, the total rotation
of Γ` gains π, and when a cusp is traversed against the coorientation, the
total rotation of Γ` looses π. This implies the result. 2

We introduce the following notations. Let x be arclength parameter along

the curve Γ. The position of the segment AB with B = Γ(x) is determined by

8
 +π −π

−π +π

Figure 5: Cusps of the curve γ and their smoothings Γ`

the angle made by the tangent vector Γ0 (t) and the vector BA. Let this angle
be π − α(x). The function α(x) uniquely determines the curve γ, the locus
of points A. Let κ(x) be the curvature of Γ(x). Denote by t the arclength
parameter on γ and by k the curvature of γ. Note that, at cusps, k = ∞.
The next result is borrowed from [18], see also [6].

k
κ
α

Figure 6: Notations for Lemma 2.4.

Lemma 2.4 The condition T` (γ) = Γ is equivalent to the differential equa-

tion on the function α(x):
dα(x) sin α(t)
+ = κ(x). (1)
dx `
One has:  
 dt 
  = | cos α|, k = tan α .
 dx  `
In particular the cusps of γ correspond to the instances of α = ±π/2.

9
Proof. Let J denote the rotation of the plane by angle π/2. Then the end
point of the segment of length ` making angle π − α(x) with Γ0 (x) is

γ(x) = Γ(x) − `Γ0 (x) cos α(x) + `J(Γ0 (x)) sin α(x). (2)

For T` (γ) = Γ to hold, the tangent direction γ 0 (x) should be collinear with
the respective segment, that is, be parallel to the vector

v(x) := −Γ0 (x) cos α(x) + J(Γ0 (x)) sin α(x).

Differentiate (2), taking into account that Γ00 (x) = κ(x)J(Γ0 (x)), and equate
the cross-product with v(x) to zero to obtain (1).
It is straightforward to calculate that |dγ/dx| = | cos α|, hence |dt/dx| =
| cos α|. The computation of the curvature k is also straightforward. 2

It is convenient to adopt the following convention: the sign of the length

element dt on γ changes at each cusp. This is consistent with Lemma 2.4
since cusps correspond to α = π/2, that is, to sign changes of cos α. With
this convention,
R we have dt = cos α(x) dx. In particular, the signed perimeter
length of γ is Γ cos α(x) dx.

3 Bicycle monodromy map

If γ(t) is arc length parameterized trajectory of the rear bicycle wheel then
the trajectory of the front wheel is Γ(t) = γ(t) ± γ 0 (t) (the sign depends on
the coorientation of γ and changes at its cusps). We extend this definition
to bicycle rides in multi-dimensional space Rn .
On the other hand, if Γ is given then one can recover γ, once the initial
position of the bicycle is chosen. The set of all possible positions of the bicycle
with a fixed position of the front wheel is a unit sphere S n−1 . Thus there
arises the time-x monodromy map Mx which assigns the time-x position of
the bicycle with a prescribed front wheel trajectory to its initial position:
Mx : S n−1 → S n−1 .
Consider the hyperbolic space H n realized as the pseudo-sphere x21 +
· · · + x2n − x20 = −1 in the pseudo-Euclidean space Rn,1 with the metric dx20 −
dx21 − · · · − dx2n . The Möbius group O(n, 1) consists of linear transformations
preserving the metric and acts on H n by isometries. This action extends to
the null cone x21 + · · · + x2n = x20 and to its spherization S n−1 , the sphere at

10
infinity of the hyperbolic space. In particular, we obtain an action of the Lie
algebra o(n, 1) on S n−1 .
The following result is a multi-dimensional generalization of Foote’s the-
orem [7].3 We identify all unit spheres S n−1 along a curve Γ(x) by parallel
translations.

Theorem 3.1 For all x, one has: Mx ∈ O(n, 1).

Proof. Note first that the rear wheel’s velocity is projr v = (r · v)r, where
r = AB. Since Mx is the map of the sphere centered at the front wheel, we
consider the moving frame with the origin at the front wheel. This frame
undergoes parallel translation as the wheel moves with its speed v. In the
moving frame, the rear wheel’s velocity is ω(v) = (−v +(r ·v)r) ⊥ r. We thus
have a vector field on the sphere, and our map Mx is the time x–map of this
vector field. It suffices therefore to show that this vector field corresponds to
an element of the Lie algebra of o(n, 1).
The Lie algebra o(n, 1) consists of the matrices
 
M v
C(M, v) =
v∗ 0
where M ∈ o(n) is an n×n skew-symmetric matrix and v is an n-dimensional
vector, and includes matrices of special form C(0, v) = C(v). We will show
that these special matrices generate the vector field ω(v) mentioned above.
(As a side remark, the Lie algebra o(n, 1) is generated by its n-dimensional
subspace C(0, Rn )).
Let us compute the action of C(v) on the unit sphere S n−1 . For a unit
n-dimensional vector r, consider the point (r, 1) of the null cone at height 1.
Then
     
r r + εv r − εω(v)
(E + εC(v)) = =k + O(ε2 ),
1 1 + εr · v 1
where k = (1 + εr · v). Thus C(v) corresponds to the vector field ω(v) on the
sphere, and the result follows. 2

Remark 3.2 It is quite likely that an analog of Theorem 3.1 holds if Rn is

replaced by either spherical or hyperbolic space. We do not dwell on it here.
3
Foote studies the Prytz planimeter.

11
Remark 3.3 It is interesting to point out possible connection with the so-
called snake charmer algorithm [9] in which the monodromy also takes values
in the Möbius group.

Now we consider corollaries of Theorem 3.1 in the case n = 2. Recall

the classification of orientation preserving isometries of the hyperbolic plane:
an elliptic isometry is a rotation about a point of H 2 , and the correspond-
ing map of the circle at infinity is conjugated to a rotation; a hyperbolic
isometry has two fixed points at infinity, one exponentially attracting and
another repelling; a parabolic isometry has a unique fixed point at infinity
with derivative 1, see, e.g., [3].
Let Γ, the trajectory of the front wheel, be closed. Then the monodromy
map M along Γ is well-defined, up to conjugation; in particular, its type
(elliptic, parabolic, hyperbolic) does not depend on the starting point.
The first corollary concerns the case when M is hyperbolic.

Corollary 3.4 Let M be hyperbolic, and let the trajectory of the rear wheel,
γ, be a generic closed wave front. Then the trajectory of the front wheel, Γ,
is also closed, and there exists a unique other closed trajectory of the rear
wheel γ ∗ with the same front wheel trajectory Γ. The correspondence γ ↔ γ ∗
is an involution. For a fixed orientation of Γ, one of the curves, γ and γ ∗ , is
exponentially stable and another exponentially unstable. The unstable curve
γ is the closed path of the bike ridden backwards.

Proof. Since γ is closed, the monodromy M has a fixed point, and since γ
is generic, M is hyperbolic. Then M has another fixed point, corresponding
to the closed trajectory γ ∗ . One of these fixed points is exponentially stable
and another unstable. 2

Corollary 3.4 is illustrated by figures 7 and 8.

We precede the next observation with a remark: for any Möbius map
with two fixed points, the derivatives at the two fixed points are reciprocal
to each other. This, according to the next theorem, implies that γ and γ ∗
have the same length (up to a sign).
Let γ be a closed wave front (the rear wheel track) and let Γ = T (γ) be
the front wheel track. Let M be the monodromy of the curve Γ and L be
the perimeter length of Γ.

12
 Γ

γ γ∗

Figure 7: The unstable curve is on the right. If the direction of traversal of

figure eight is reversed, the two curves exchange stability.

Γ Γ
γ γ∗

Figure 8: The stable and the unstable rear trajectories for the sharmorck,
just before the bifurcation when γ and γ ∗ coalesce. The shamrock is given
by x = r(t) cos t, y = r(t) cos t with r = 0.94(1 − 0.5 sin 3t).

13
Theorem 3.5 Let M be hyperbolic or parabolic, and let γ be the closed path
of the rear wheel corresponding to a fixed point θ0 of the Möbius circle map
θ 7→ M (θ). Then
M 0 (θ0 ) = e−Length(γ) . (3)

Corollary 3.6 If M is hyperbolic and γ and γ ∗ are the rear tracks corre-
sponding to the two fixed points then the curves γ and γ ∗ have equal lengths.

Proof of the corollary. For the fixed points θ0 , θ0∗ of any Möbius map
one has M 0 (θ0 )M 0 (θ0∗ ) = 1, and the statement follows from Theorem 3.5. 2

Remark 3.7 The case γ = γ ∗ is quite interesting: this is when one cannot
tell which way the bicycle went from closed tire tracks of the front and rear
wheels, see Section 1.

Proof of Theorem 3.5. Using the notation of Section 2, consider equation

(1) (with ` = 1). This equation has an L-periodic solution α(x). Consider
an infinitesimal perturbation α(x) + εβ(x); the derivative of the monodromy
map is given by M 0 (θ0 ) = β(L)/β(0). But β satisfies the linearized equation
β 0 + β cos α = 0, from which we find

β(L) RL
M 0 (θ0 ) = = e− 0 cos α(x)dx .
β(0)
RL
But cos α(x) is the speed of the rear wheel, and thus 0
cos α(x)dx = Length(γ).
2

Remark 3.8 It is interesting that the monodromy may be identical, that is,
there exist closed trajectories of the front wheel for which every trajectory
of the rear wheel is closed. To construct such an example, let Γ be a small
simple closed curve. Then the monodromy M is elliptic, see analysis in [7].
(This also follows from equation R(1): in the limit ` → ∞, the equation
becomes α0 (x) = κ(x), and since κ(x) dx = 2π, the function α(x) cannot
be periodic.) Slightly deforming Γ, if necessary, one may assume that M is
conjugated to a periodic rotation. Then, traversing Γ an appropriate number
of times, the monodromy becomes identical.

14
 In contrast, if Γ is a closed immersed curve (not necessarily simple) and
` is sufficiently small, one has a hyperbolic monodromy. Indeed, in the limit
` → 0, equation (1) becomes sin α = 0 and has two solutions α(x) = 0 and
α(x) = π, corresponding to the forward and backward tangent vectors to
Γ. The two exponentially stable and unstable solutions survive for ` small
enough.
As a limiting case of Theorem 3.5 for the parabolic monodromy we have
the following.
Theorem 3.9 M is parabolic if and only if the total algebraic length of γ is
zero.

Proof. At the fixed point θ0 we have M 0 (θ0 ) = 1; comparison with (3)

shows that Length(γ) = 0. 2

Corollary 3.10 In the parabolic case, the curve γ has cusps.

An example of a wave front γ yielding parabolic monodromy is depicted
in figure 9. The curve γ has total turning number π, so for Γ to close up, one
traverses γ twice. This “doubled” front γ obviously has zero total length.

Figure 9: Curve with turning number π

An example of the saddle–node bifurcation from the hyperbolic to the

elliptic case, as the size of Γ decreases, is shown in figure 10.
Remark 3.11 Computation of the monodromy amounts to multiplying in-
finitely many 2 × 2 matrices corresponding to infinitesimal arcs of the curve
Γ (if Γ is a polygon, one has a finite products of hyperbolic elements in
SL(2, R)). A similar problem concerning the group of isometries of the
sphere SO(3) is treated in [11, 12]; we plan to extend this work to the group
of isometries of the hyperbolic plane.

15
 Γ Γ

γ
γ

Figure 10: A saddle–node bifurcation: the ellipse on the left is (slightly)

larger. The length of the coalesced curve on the right is zero in accordance
with Theorem 3.9. In this particular case this is seen directly: the four arcs
are congruent by symmetry, and their signs alternate.

4 Proof of the Menzin conjecture

Theorem 4.1 If Γ is a closed convex curve bounding area greater than π
then the respective monodromy is hyperbolic.

Proof. By approximation, we may assume that Γ is an oval, that is, a

smooth closed strictly convex curve. We need to prove that if the monodromy
M is elliptic or parabolic then Area(Γ) ≤ π. As we already mentioned, if Γ
is large enough, the monodromy M is hyperbolic. Hence, if M is elliptic, we
can make Γ larger (say, by homothety) and render M parabolic. Therefore
it suffices to prove that if M is parabolic then Area(Γ) ≤ π.
The proof is based on two observations:

• area(Γ) = area(γ)+π, so that Area(Γ) ≤ π is equivalent to area(γ) ≤ 0,

and

• If length(γ) = 0 then area(γ) ≤ 0.

We proceed with the detailed proof. Consider a generic one-parameter family

of decreasing ovals Γt , t ∈ [0, 1], starting with a very large oval Γ0 and ending
with the given oval Γ = Γ1 such that all the monodromies Mt for t ∈ [0, 1)
are hyperbolic. Therefore one has a family of wave fronts γt (the closed
trajectories of the rear wheel). Since Γ0 is large enough, γ0 is also an oval.
The Legendrian liftings of the fronts γt form a continuous family of immersed
Legendrian curves in the space of contact elements. Therefore the Maslov
index of γ1 equals that of γ0 , that is, zero. Likewise, the rotation number

16
ρ(γ1 ) equals one. The number of cusps may change in the family γt , see
figure 11.

Figure 11: Birth/death of a pair of cusps

The following holds due to the convexity of Γ.

Lemma 4.2 The wave front has no inflections.

Proof. Assume γ1 has an inflection point. Note that γ0 is convex. Let τ be

the first value of parameter t for which the curvature of γt vanishes. Then,
for t slightly greater than τ , the curve γt has a “dimple” and Γt is not convex,
see figure 12. 2

Γ
γ

Figure 12: Inflections of γ

Thus γ1 is a wave front made of an even number of convex smooth arcs;

the adjacent arcs form cusps. The total turning of the tangent direction to
γ1 is 2π. The arcs are marked by ±, the sign changes at each cusp. By
Theorem 3.9, the algebraic length of γ1 vanishes: Length(γ1 ) = 0.
Consider a smooth arc of γ1 in the arclength parameterization; abusing
notation, call this arc γ1 (x). The respective arc of Γ1 is Γ1 (x) = γ1 (x)+σγ10 (x)
where σ = ± is the sign of the arc γ1 . Therefore Γ01 = γ10 + σγ100 , hence
Γ1 × Γ01 = γ1 × γ10 + σγ1 × γ100 + σ 2 γ10 × γ100 .
Note that γ1 × γ100 = (γ1 × γ10 )0 and that γ10 × γ100 = k, the curvature of γ1 .
Now we can find the area bounded by Γ1 :
X Z 
0 0
2Area(Γ1 ) = γ1 (x) × γ1 (x) dx + σi ∆i (γ1 × γ1 ) + θi (4)
i

17
where the sum is taken over the smooth arcs of γ1 , where σi is the sign of
ith arc, ∆i is the difference of the momenta γ1 × γ10 at the end points of ith
arc, and θi is the angle of turning of ith arc. P
Note that the sum of integrals in (4) is 2Area(γ
P 1 ). Note0 also that θi =
2π. Finally note that the terms ∆ cancel out: σi ∆i (γ1 ×γ1 ) = 0. Therefore
the inequality Area(Γ1 ) ≤ π is equivalent to Area(γ1 ) ≤ 0.
To prove the latter inequality, let p(ϕ) be the support function of the
front γ1 (the signed distance from the origin to the tangent line to γ1 as a
function of the direction of this line; see, e.g., [17] for the theory of support
functions). The support function exists because γ1 is free from inflections
and makes one full turn. One has the following formulas:
Z 2π
1 2π 2
Z
Length(γ1 ) = p(ϕ) dϕ, Area(γ1 ) = (p (ϕ) − p02 (ϕ) dϕ.
0 2 0

Thus we need to show that if

Z 2π Z 2π Z 2π
p(ϕ) dϕ = 0 then 2
p (ϕ) dϕ ≤ p02 (ϕ) dϕ.
0 0 0

But this is a well known Wirtinger inequality, which concludes the proof. 2

Remark 4.3 Wirtinger inequality is intimately related with the isoperimet-

ric inequality. Consider an oval γ with area A and perimeter length L.
Consider the one-parameter family of equidistant fronts γt inside the oval
(that is, consider γ as a source of light propagating inwards). The support
function of γt is that of γ minus t. One has:

Length(γt ) = L − 2πt, Area(γt ) = A − Lt + πt2 .

By the Wirtinger inequality, when Length(γt ) = 0, one has Area(γt ) ≤ 0.

Therefore if t = L/2π then A − Lt + πt2 ≤ 0, that is, A ≤ L2 /4π, which is
the isoperimetric inequality.

5 Oscillation of unicycle tracks

Recall Finn’s construction described in Section 1. Let γ(t), t ∈ [0, L] be an
arc length parameterized smooth curve in R2 such that the ∞-jets of γ(t)

18
coincide, for t = 0 and t = L, with the ∞-jets of the x-axis at points (0, 0)
and (1, 0), respectively. We use γ as a “seed” trajectory of the rear wheel
of a bicycle. Then Γ = T (γ) = γ + γ 0 is also tangent to the horizontal axis
with all derivatives at its end points (1, 0) and (2, 0). Iterating this procedure
yields a smooth infinite forward bicycle trajectory T such that the tracks of
the rear and the front wheels coincide. We shall study oscillation properties
of T . For starters, we note that the length of each new arc of T increases
compared to the previous one.

Lemma 5.1 The length of Γ equals

Z Lp
1 + k 2 (t) dt > L
0

where k(t) = |γ 00 (t)| is the curvature of γ.

Proof. One has:

Γ0 (t) = γ 0 (t) + γ 00 (t), |Γ0 (t)|2 = 1 + |γ 00 (t)|2 ,

therefore the length of Γ is

Z L Z L p
0
|Γ (t)| dt = 1 + k 2 (t) dt.
0 0

Denote by Z(γ) the number of intersection points of the curve γ(t), t ∈

(0, L) with the x-axis (we exclude the end points); assume that Z(γ) is finite.

Proposition 5.2 One has: Z(Γ) > Z(γ).

Proof. Note that
0

e−t et γ(t) = Γ(t). (5)
Let Z(γ) = n and let t0 = 0 < t1 < · · · < tn < tn+1 = L be the consecutive
moments of intersection of γ(t) with the x-axis. Then ti are also the consec-
utive moments of intersection of the curve ∆(t) := et γ(t) with the x-axis. By
a version of the Rolle theorem, see figure 13, for each i = 0, 1, . . . , n, there
is t ∈ (ti , ti+1 ) for which the curve ∆(t) has a horizontal tangent, i.e., the

19
 ∆
∆

Figure 13: Rolle theorem for curves

vector ∆0 (t) is horizontal. It follows from (5) that Γ(t) lies on the x-axis, and
we are done. 2

Consider the problem of extending the curve T backwards, that is, in-
verting the operator T . It turns out that usually T can be inverted only
finitely many times. Namely, one has the following corollary of Proposition
5.2.

Corollary 5.3 Let Γ be a curve whose end points are unit distance apart and
which is tangent, to all orders, at the end points to the x-axis. Let Z(Γ) = n.
Then for no curve γ whose end points are unit distance apart and which is
tangent, to all orders, at the end points to the x-axis, one has T n+1 (γ) = Γ.

Here is another oscillation property of the the curve T . Let E(γ) be the
(finite) number of locally highest and lowest points of the curve γ. As before,
Γ = T (γ).

Proposition 5.4 One has: E(Γ) > E(γ).

Proof. At a locally highest point of γ, the curve Γ has the downward

direction, and at a locally lowest point – the upward direction, see figure 14.
It follows that, between consecutive locally highest and lowest points of γ,
one has a locally lowest point of Γ, and between consecutive locally lowest
and highest points of γ, one has a locally highest point of Γ. Considering
the end points of γ as local extrema of the height function as well, yields the
result. 2

Conjecture 5.5 It follows from figure 14 that the maximum height of Γ is

greater than that of γ, and likewise for the minimum height. We conjecture
that the amplitude of the curve T is unbounded; in other words, unless γ is a

20
 Γ Γ γ
Γ Γ
γ

Figure 14: Height extrema of the curve γ

segment, T is not contained in any horizontal strip. We also conjecture that,

unless γ is a segment, T is not the graph of a function (i.e., one of the curves
T n (γ) has a vertical tangent line) and, further, fails to be an embedded curve.
One more conjecture: unless T is the horizontal axis, the curvature of T is
unbounded.

Configuration space of equilateral forward infinite linkages. The

construction of bicycle motion generating a single track can be interpreted
as follows. Let M be the space of semi-infinite equilateral linkages {X =
(x0 , x1 , x2 , . . . )} where each xi is a point in the plane and |xi − xi+1 | = 1 for
all i. Denote by vi the unit vector xi xi+1 and by αi the angle between vi−1
and vi . Let M0 be an open subset of M given by the condition ai 6= ±π/2
for all i.

xi αi

Figure 15: Note the change of direction when α > π/2. Only the direction of
motion, and not the speeds, is indicated – the latter becomes large for large
values of i.

Consider the constraint on M defined by the condition that the velocity

of point xi is proportional to vi . If ti is the speed of xi then the condition
that all links remain unit reads

ti = ti+1 cos αi+1 (6)

21
for all i. On M0 , where cos αi 6= 0, all the velocities are uniquely defined,
up to a common factor, and one has a well defined field of directions ξ which
can be normalized to a vector field by setting t0 = 1. If αi = π/2 for some i
then the speeds of all xj with j < i must vanish; in particular, if αi = π/2
for infinitely many values of i then such a configuration has no infinitesimal
motions at all. See [15, 16] for this non-holonomic system in relation to
“Monster Tower” and Goursat flags.

−1

−2

−2 −1 0 1 2 3 4 5

Figure 16: The rear wheel follows the track of the front wheel. The “seed”
curve is shown in heavier stroke. Several consecutive position of the associ-
ated moving linkage are shown. The shape of the linkage is very sensitive to
the position of the starting point on the seed curve.

A forward bicycle motion generating a single track corresponds to a solu-

tion to our system. The above described curve T yields an integral curve of
the field ξ in M0 : indeed, αi = π/2 corresponds to a cusp of the trajectory
of point xi−1 , whereas T is a smooth curve, as follows from Lemma 2.1.

22
 The starting configuration X of the Finn construction consists of non-
negative integers on the horizontal axis, xi = (i, 0), and one has a variety
of integral curves of ξ through X ∈ M0 (of which the simplest one is the
uniform motion along the horizontal axis). Thus one has non-uniqueness of
solutions of the differential equation describing the field ξ.
Finn’s construction can be easily generalized as follows. Let δ be an
infinite jet of a curve at point x0 . Consider the infinite jet T (δ) at point
x1 = T (x0 ), and let γ be a curve smoothly interpolating between δ and
T (δ). Then the concatenation of the curves γ, T (γ), T 2 (γ), etc., is a smooth
unicycle track left by the bicycle motion with the seed curve γ.
The above construction provides a mapping Φ : J ∞ (x0 ) → M0 from the
space of infinite jets off curves at point x0 to unit forward infinite linkages
{(x0 , x1 , . . . )}.

Proposition 5.6 The mapping Φ is a bijection.

Proof. We construct the inverse map Ψ : M0 → J ∞ (x0 ). Let X =

(x0 , x1 , . . . ) ∈ M0 and set: Ci = cos αi 6= 0. Then, according to (6),
1
t0 = 1, tk = ,
Πki=1 Ci

hence the speeds of all points are determined.

(r)
We claim that, for each r ≥ 1, one has: xj = Fj,r (xi , Ci ) where F is
a polynomial in xi and a Laurent polynomial in Ci for i = 0, 1, . . . . This
is proved by induction on r. For r = 1, one has x0j = tj (xj+1 − xj ). If
(r)
xj = Fj,r (xi , Ci ) then

(r+1)
X ∂Fj,r ∂Fj,r 0
xj = x0i + C.
i
∂xi ∂Ci i

The induction step will be completed if we show that Ci0 is also a polynomial
in xi and Ci . Indeed, Ci = (xi − xi−1 ) · (xi+1 − xi ) and hence

Ci0 = (ti (xi+1 − xi ) − ti−1 (xi − xi−1 )) · (xi+1 − xi )+

(xi − xi−1 ) · (ti+1 (xi+2 − xi+1 ) − ti (xi+1 − xi )),

as needed.

23
 (r)
In particular, X determines all the derivatives x0 , that is, the infinite
jet of a curve at x0 . This is Ψ(X). 2

We finish with another question: is a straight line the only real analytic
“unicycle” trajectory?

References
[1] V. Arnold, A. Givental. Symplectic geometry, 1-136. Encycl. of Math.
Sci., Dynamical Systems, 4, Springer-Verlag, 1990.

[2] G. Barnes. Hatchet or hacksaw blade planimeter. Amer. J. Phys. 25

(1957), 25–29.

[3] A. Beardon. The geometry of discrete groups. Springer-Verlag, New

York, 1983.

[4] A. Crathorne, The Prytz Planimeter. Amer. Math. Monthly 15 (1908),

55–57.

[5] S. Dunbar, R. Bosman, S. Nooij. The track of a bicycle back tire. Math.
Mag. 74 (2001), 273–287.

[6] D. Finn. Can a bicycle create a unicycle track? College Math. J.,
September, 2002.

[7] R. Foote. Geometry of the Prytz planimeter. Rep. Math. Phys. 42 (1998),
249–271.

[8] H. Geiges. Contact geometry. Handbook of differential geometry. Vol.

II, 315–382, Elsevier/North-Holland, Amsterdam, 2006.

[9] J.-C. Hausmann, E. Rodriguez. Holonomy orbits of the snake charmer

algorithm. preprint math.DG/0603620.

[10] J. Konhauser, D. Velleman, S. Wagon. Which way did the bicycle go?
... and other intriguing mathematical mysteries. MAA, 1996.

[11] M. Levi. Composition of rotations and parallel transport. Nonlinearity 9

(1996), 413–419.

24
 [12] M. Levi. Geometric phases in the motion of rigid bodies. Arch. Rational
Mech. Anal. 122 (1993), 213–229.

[13] M. Levi, W. Weckesser. Non-holonomic systems as singular limits for

rapid oscillations. Ergod. Th. Dynam. Sys. 22 (2002), 1497–1506.

[14] A. Menzin. The tractigraph, an improved form of Hatchet planimeter.

Eng. News 56 (1906), 131–132.

[15] R. Montgomery, M. Zhitomirskii. Geometric approach to Goursat flags.

Ann. Inst. Poincare Anal. Non Linéare, 12 (2001), 459-493.

[16] R. Montgomery, M. Zhitomirskii. Points and curves in the Monster

Tower. Submitted to Memoirs of the AMS

[17] L. Santalo, Integral Geometry and Geometric Probability. Cambridge

University Press, Cambridge, 2004.

[18] S. Tabachnikov. Tire track geometry: variations on a theme. Israel J.

Math. 151 (2006), 1–28.

[19] F. Wegner. Floating bodies of equilibrium. Stud. Appl. Math. 111 (2003),
167–183.

[20] F. Wegner. Floating bodies of equilibrium. Explicit solution. preprint

physics/0603160.

[21] F. Wegner. Floating bodies of equilibrium in 2D, the tire track problem
and electrons in a parabolic magnetic field. preprint physics/0701241.

25

View publication stats