Journal of Geometry and Physics ( ) –

Contents lists available at ScienceDirect

Journal of Geometry and Physics

journal homepage: www.elsevier.com/locate/jgp

On the bicycle transformation and the filament equation:

Results and conjectures
Serge Tabachnikov
Department of Mathematics, Pennsylvania State University, University Park, PA 16802, USA

article info abstract

Article history: The paper concerns a simple model of bicycle kinematics: a bicycle is represented by an
Received 20 February 2016 oriented segment of constant length in n-dimensional space that can move in such a way
Accepted 23 May 2016 that the velocity of its rear end is aligned with the segment (the rear wheel is fixed on the
Available online xxxx
bicycle frame). Starting with a closed trajectory of the rear end, one obtains the two respec-
tive trajectories of the front end, corresponding to the opposite directions of motion. These
Keywords:
two curves are said to be in the bicycle correspondence. Conjecturally, this correspondence
Bicycle correspondence
Monodromy
is completely integrable; we present a number of results substantiating this conjecture. In
Filament equation dimension three, the bicycle correspondence is the Bäcklund transformation for the fila-
Complete integrability ment equation; we discuss bi-Hamiltonian features of the bicycle correspondence and its
Bäcklund transformation integrals.
© 2016 Elsevier B.V. All rights reserved.

1. Bicycle model, bicycle monodromy, and bicycle correspondence

This paper concerns a simple model of bicycle kinematics and continues the work done in [1–5]. In this model, a bicycle
is represented by an oriented segment of constant length ℓ that can move in such a way that the velocity of its rear end is
aligned with the segment (the rear wheel is fixed on the bicycle frame). The motion takes place in the Euclidean plane but,
equally well, albeit less physically, we consider the motion in Rn (see [6] for bicycle motion in the hyperbolic and elliptic
planes).
The configuration space of oriented segments of fixed length in Rn is the spherization of the tangent bundle ST Rn , and
the non-holonomic constraint that we have imposed on the motion defines an n-dimensional distribution D therein. The
motion of the bicycle is a smooth curve in ST Rn tangent to the distribution D .
The two projections ST Rn → Rn , to the front and to the rear ends of the segment, yield the front and rear bicycle tracks.
The former projection is transverse to D , and the front track is a smooth curve, whereas the direction of the latter projection
is contained in D , and the rear track may have singularities (generically, semi-cubical cusps). This happens when the front
wheel of the bicycle turns 90◦ . We denote the rear track by γ and the front track by Γ .
In this section, we review the relevant results from [2,4,5] that will be used later.
The bicycle segment gives the rear track γ coorientation (the orientation of the normal bundle), well defined even at
cusps. Conversely, given a cooriented curve γ , possibly, with cusps, one uniquely reconstructs the respective curve Γ as the
locus of endpoints of the tangent segments of length ℓ. That is, the rear track and the direction of the motion determine the
front track.
On the other hand, the front track Γ determines the motion of the bicycle, and hence the rear track γ , only after its initial
position is chosen (this is obvious geometrically, and it follows from the fact that the motion of the bicycle with a given front

E-mail address: tabachni@math.psu.edu.

http://dx.doi.org/10.1016/j.geomphys.2016.05.013
0393-0440/© 2016 Elsevier B.V. All rights reserved.
2 S. Tabachnikov / Journal of Geometry and Physics ( ) –

Fig. 1. Bicycle correspondence.

Source: Figures courtesy of R. Perline.

Fig. 2. Discrete bicycle transformation.

track is described by a first order differential equation, see [2] and Eq. (6)). Thus the bicycle monodromy map MΓ ,ℓ arises that
takes the initial position of the bicycle to the terminal one.
Assume that Γ is a closed curve. Then MΓ ,ℓ is a self-map of the sphere S n−1 , well defined up to conjugation (due to the
freedom of choice of the initial point on Γ ). Consider S n−1 as the sphere at infinity of the hyperbolic space Hn . Then the
Möbius group O(n, 1) of isometries of Hn acts on S n−1 . The next result was proved in [2].

Theorem 1. The bicycle monodromy MΓ ,ℓ is a Möbius transformation.

Let γ be a closed rear track. It has two coorientations, and they correspond to two front tracks, Γ1 and Γ2 (one may think
of a forward-facing and a backward-facing bicycle joined together). We say that Γ1 and Γ2 are in the bicycle correspondence
and write B2ℓ (Γ1 , Γ2 ). Equivalently, two points, x1 and x2 , traverse the curves Γ1 and Γ2 in such a way that the distance x1 x2
is equal to 2ℓ, and the velocity of the midpoint of the segment x1 x2 is aligned with x1 x2 (it follows that the speeds of x1 and
x2 are equal). See Fig. 1.
The bicycle monodromy and the bicycle correspondence have discrete versions [7,8,5]. Let P = (P1 , P2 , . . . , Pn ) be an
n-gon in Rn , and let P1 Q1 be a segment of length 2ℓ. Let Q2 be the point such that P1 Q1 P2 Q2 is an isosceles trapezoid. In other
words, one parallel translated the segment P1 Q1 to P2 Q1′ and then reflects Q1′ in the line Q1 P2 , see Fig. 2. Continuing this, one
constructs a polygon Q1 , Q2 , . . . , Qn .
In the limit n → ∞, the two polygons become curves in the bicycle correspondence. The point Qn+1 does not necessarily
coincide with Q1 , and we obtain the discrete bicycle monodromy Q1 → Qn+1 . Theorem 1 holds in this case as well, see [5].
If the polygon Q closes up, we say that the polygons P and Q are in the discrete bicycle correspondence, see Fig. 3.
The next two theorems hold both for closed curves and for polygons, see [5] for proofs in the discrete setting. We
formulate the continuous versions.

Theorem 2. If closed curves Γ1 and Γ2 are in the bicycle correspondence, Bℓ (Γ1 , Γ2 ), then the bicycle monodromies MΓ1 ,λ and
MΓ2 ,λ are conjugated for every value of λ.

Theorem 3 (Bianchi Permutability). Let Γ1 , Γ2 and Γ3 be three closed curves such that Bℓ (Γ1 , Γ2 ) and Bλ (Γ1 , Γ3 ) hold. Then
there exists a closed curve Γ4 such that Bλ (Γ2 , Γ4 ) and Bℓ (Γ3 , Γ4 ) hold.
The bicycle correspondence is not a mapping: for example, the bicycle monodromy of a curve Γ1 may be the identity,
and then there is a whole sphere worth of corresponding curves Γ2 . For instance, this happens when Γ1 is a circle, see Fig. 1
on the left.
According to Theorem 1, in dimension three, the bicycle monodromy is a complex fractional–linear transformation and,
for a generic curve and generic value of the length, it has two fixed points. Therefore the bicycle correspondence is a 2–2
map and, up to a binary choice, one can consider it as a mapping of space curves.
In the plane, the bicycle monodromy is a real fractional–linear transformation, and one has the usual trichotomy: it can
be elliptic, parabolic, or hyperbolic. Let Γ1 be a plane curve such that MΓ1 ,ℓ is hyperbolic, that is, it has two fixed points. Then
 S. Tabachnikov / Journal of Geometry and Physics ( ) – 3

Fig. 3. Two polygons in the discrete bicycle correspondence.

one can choose one of the two respective curves, Γ2 such that B2ℓ (Γ1 , Γ2 ). By Theorem 2, MΓ2 ,ℓ is again hyperbolic, and
there is a unique curve Γ3 , different from Γ1 , with B2ℓ (Γ2 , Γ3 ), and so on. In this way the bicycle correspondence becomes
a mapping of the space of plane curves.
Our ultimate goal is the following conjecture which we formulate in somewhat loose terms.

Conjecture 1. The bicycle correspondence in Rn is Liouville integrable.

This means that the correspondence has an invariant Poisson bracket and an infinite collection of independent integrals
in involution that is, in an appropriate sense, complete.
Theorem 2 is a source of integrals. Let
n+1 
 ∞
det(E − tMΓ ,λ ) = Iij (Γ )t i λj
i=1 j=1

be the Taylor expansion of the characteristic polynomial of the monodromy, considered as an element of the Möbius group
O(n, 1). Then the functions Iij (Γ ) on the space of curves in Rn are conserved by the bicycle correspondence. We call them
the monodromy integrals.

2. Two differential 2-forms

Consider two differential 2-forms on the space of smooth embedded parameterized curves, the first defined in Rn for all
n, and the second in R3 :
 
ω(u, v) = u′ (t ) · v(t ) dt , Ω (u, v) = det(Γ ′ (t ), u(t ), v(t )) dt .

Here Γ (t ) is a curve, and u(t ), v(t ) are vector fields along Γ .

Note that the integrals do not depend on the parameterization. Note also that Ω vanishes on the submanifold of plane
curves. The form ω can be defined on any Riemannian manifold, and the form Ω depends only on the volume form (and not
on the metric).

Proposition 2.1. The 2-forms ω and Ω are closed. The forms are degenerate: the kernel of ω consists of the parallel vector fields
along the curve, and the kernel of Ω consists of the tangent vector fields.

Proof. Consider the following 1-forms:

 
1 1
α(u) = Γ (t ) · u(t ) dt ,
′
β(u) = − det(Γ (t ), Γ ′ (t ), u(t )) dt .
2 3
We claim that

ω = dα, Ω = dβ.
4 S. Tabachnikov / Journal of Geometry and Physics ( ) –

Indeed, let u and v be commuting vector fields in a neighborhood of Γ . One has

dα(u, v) = Lu (α(v)) − Lv (α(u)) − α([u, v]).
The first two terms on the right hand side are easy to compute:
  
Lu (α(v)) = u (t ) · v(t ) dt ,
′
Lv (α(u)) = v (t ) · u(t ) dt = −
′
u′ (t ) · v(t ) dt ,

and the result for ω follows. A similar computation works for Ω .

If u is in the kernel of ω, then u′ (t ) · v(t ) dt = 0 for all vector fields v along Γ . It follows that u′ = 0, that is, u is parallel


along Γ . Likewise, if u is in the kernel of Ω , then det(Γ ′ (t ), u(t ), v(t )) dt = 0 for all vector fields v along Γ . This implies
that u is collinear with Γ ′ . 

Corollary 2.2. The form ω descends on the quotient space of curves modulo parallel translations and is symplectic therein, and
the form Ω descends on the quotient space of curves mod Diff+ (S 1 ), that is, on the space of non-parameterized oriented curves,
and is a symplectic structure therein.
The latter is the celebrated Marsden–Weinstein symplectic structure, see [9–11]. One can identify isotopic equivalence
classes of smooth knots with coadjoint orbits of the group of volume-preserving diffeomorphisms of R3 (the pairing of a
curve with a divergence-free vector field is given by the flux across a film that spans the curve). Under this identification,
the Marsden–Weinstein symplectic structure coincides with the Kirillov–Kostant–Souriau symplectic structure on coadjoint
orbits of a Lie group.
Our main observation is as follows.

Theorem 4. The bicycle correspondence preserves the forms ω and Ω , the former in any dimension, and the latter in dimension
three.
Proof. Consider a smooth arc of the rear track γ between two consecutive cusps and give it arc length parameterization with
parameter t ∈ [a, b]. Let Γ± = γ ±ℓγ ′ be the respective arcs of the two front tracks which are in the bicycle correspondence.
Let u(t ), v(t ) be two vector fields along γ (t ). The differentials of the maps γ → Γ± take u and v to vector fields U and
V along Γ± . We claim that

U = u + ℓ(u′ − (γ ′ · u′ )γ ′ ), (1)
and likewise for V ; here ℓ may be positive or negative, depending on whether one considers Γ+ or Γ− .
To prove (1), let γ1 = γ + ε u be a variation of γ . Then |γ1′ | = 1 + εγ ′ · u′ , and one has:
γ1′
Γ1 = γ1 + ℓ = γ + ℓγ ′ + ε(u + ℓ(u′ − (γ ′ · u′ )γ ′ )),
|γ1′ |
as claimed (all computations are mod ε 2 , and ℓ is signed).
It follows from (1) that
U ′ = u′ + ℓ(u′′ − (γ ′′ · u′ )γ ′ − (γ ′ · u′′ )γ ′ − (γ ′ · u′ )γ ′′ ).
U ′ · Vdt which is odd in ℓ vanishes. The only odd term is linear in ℓ; it is

We want to show that the part of the integral
equal to
 b
[u′ · v ′ + u′′ · v] − [(γ ′ · v ′ )(γ ′ · u′ ) + (γ ′ · v)(γ ′′ · u′ ) + (γ ′ · v)(γ ′ · u′′ ) + (γ ′′ · v)(γ ′ · u′ )] dt .
a

The integral of the first bracket is equal to ℓ(u′ · v)|ba , and of the second to ℓ(γ ′ · v)(γ ′ · u′ )|ba (this would finish the proof if
γ had no cusps).
Now consider the next smooth arc of γ , arc length parameterized by t ∈ [b, c ]. A similar computation applies to this arc,
with ℓ changing the sign. The contribution of the cusp, corresponding to the value of parameter b, is ℓ(u′ · v)(b), computed
for the first arc, and ℓ(u′ · v)(b), computed for the second one. They cancel each other: the sign of u′ also changes to the
opposite because the direction of motion reverses in the cusp. Likewise, one has cancellation of the terms ℓ(γ ′ · v)(γ ′ · u′ ),
and this implies that T̄ℓ∗ (ω) = ω.
for Ω involves det(Γ ′ , U , V )dt . There are two terms odd in ℓ, the linear and the cubic

Similarly, the computation
ones. The linear term is det(γ ′ , u, v)′ dt, again, the integral of a derivative. And again, the total integral of a full derivative

vanishes, taking the sign changes at the cusps into account.
Finally, consider the cubic in ℓ term,

det(γ ′′ , u′ − (γ ′ · u′ )γ ′ , v ′ − (γ ′ · v ′ )γ ′ ) dt .

The determinant is identically zero since all three vectors are orthogonal to γ ′ , and hence linearly dependent. 
 S. Tabachnikov / Journal of Geometry and Physics ( ) – 5

The monodromy integrals descend to both quotient spaces, curves modulo parallel translations and non-parameterized
curves.

Conjecture 2. The monodromy integrals Poisson commute on these quotient spaces with respect to the Poisson bracket induced,
respectively, by the forms ω (in every dimension) and Ω (in dimension three).

3. The bicycle transformation in dimension three and the filament equation

The filament (binormal, localized induction, smoke ring) equation is an evolution on curves in R3

Γ̇ = Γ ′ × Γ ′′
where Γ (x) is arc length parameterized curve, prime is d/dx, and dot is the time derivative.
This system (equivalent to the non-linear Schrödinger equation via the Hasimoto transformation) is completely
integrable in the following sense, see [12,13]. It is a Hamiltonian system with respect to the symplectic structure induced
by Ω on the space of arc length parameterized curves, the Hamiltonian function being the perimeter length of the curve. It
has a hierarchy of Poisson commuting integrals F1 , F2 , . . . that starts with
      
1
1 dx, τ dx, κ 2 dx, κ 2 τ dx, (κ ′ )2 + κ 2 τ 2 − κ 4 dx, . . . (2)
4
where τ is the torsion, and κ is the curvature. One also has a hierarchy of commuting Hamiltonian vector fields X0 , X1 , X2 , . . .
that starts with

κ2 κ3
 
−T , κ B, T + κ N + κτ B, κ τ T + (2κ τ + κτ )N +
′ 2 ′ ′
κτ − κ −
2 ′′
B, . . . (3)
2 2
where T , N , B is the Frenet frame along Γ , and where Xi is the Hamiltonian vector field of Fi . The vector fields Xi satisfy the
recurrence relation

T × Xn = Xn′ −1 . (4)

In dimension three, the bicycle transformation is completely integrable, in the same sense as the filament equation.

Theorem 5. The functions (2) are integrals of the bicycle transformation (independently of the length parameter ℓ). The bicycle
transformation commutes with the vector fields (3).

Proof. We claim that

ω(Xn−1 , ·) = Ω (Xn , ·) = dFn . (5)

Indeed, the first equality follows from (4) and the definitions of the 2-forms, and the second from the fact that Xn is the
Hamiltonian vector field of the function Fn .
Assume that the bicycle transformation preserves the field Xn−1 . By Theorem 4, it preserves ω, and hence by (5), it
preserves dFn . Assume now that the bicycle transformation preserves the differential dFn . By Theorem 4, it preserves Ω ,
and hence the Hamiltonian vector field Xn .
Since the bicycle transformation preserves dFn , it changes Fn by a constant: Fn → Fn + c. To see that this constant is zero,
it suffices to consider a circle of sufficiently large radius. Such a circle is fixed by the bicycle transformation, hence c = 0.
Thus one may argue inductively, starting with the function F1 , the perimeter length, clearly preserved by the bicycle
transformation. 

Theorem 5 is hardly new: the bicycle transformation is the Bäcklund transformation for the filament equation, see,
e.g., [14,15]. One of the reasons to use the term ‘bicycle’ (rather than Bäcklund, or Darboux) transformation is to decouple it
from the filament equation which, unlike the bicycle transformation, is specifically 3-dimensional.

Conjecture 3. In dimensions three and two, the monodromy integrals and the integrals (2) are functionally dependent of each
other.

Specifically, we conjecture that the integrals (2) are the consecutive terms in the Taylor expansion at λ = 0 of the
conjugacy invariant

Tr2 (MΓ ,λ )
det(MΓ ,λ )
where the monodromy is considered as a 2 × 2 complex (in dimension 3) or real (in dimension 2) matrix.
6 S. Tabachnikov / Journal of Geometry and Physics ( ) –

Fig. 4. Examples of Zindler curves from [23].

Remark 3.1. In dimension two, the motion of the bicycle is described by a differential equation

dα(x) sin α(x)

+ = κ(x), (6)
dx ℓ
where α is the steering angle, κ is the curvature of the front track Γ , x is the arc length parameter on the front track Γ , and
ℓ the length of the bicycle segment, see [2]. It is proved in [2] that the derivatives σ1,2 of the monodromy at its fixed points
are equal to exp(± cos α dx), where α(x) is a periodic solution of (6). One has


1 1 Tr2 (MΓ ,λ )
+ = − 2,
σ1 σ2 det(MΓ ,λ )

hence the quantity cos α dx is an integral of the bicycle transformation.


This suggest a method of obtaining integrals: consider a periodic solution α(x) of Eq. (6) as a function of ℓ and expand cos α
as a Taylor series in ℓ. Then the integral of each coefficient of this series is a conserved quantity of the bicycle transformation.
The first three integrals obtained this way are
    
1
1 dx, κ 2 dx, (κ ′ )2 − κ 4 dx,
4
as predicted by Conjecture 3.

Now let M be an oriented 3-dimensional Riemannian manifold. One can define the differential 2-forms ω and Ω and
considers the filament equation on M. One can also define an analog of the bicycle correspondence: the bicycle frame is a
geodesic segment of fixed length. It is natural to ask when one has an analog of the recursion scheme (5) for the filament
equation and an analog of Theorem 5.

Conjecture 4. This holds if and only if M has constant curvature.

4. In the plane. Zindler curves

A plane analog of the filament equation is the planar filament equation

κ2
Γ̇ = T + κ ′N
2
where (T , N ) is the planar Frenet frame. This system is equivalent to the modified Korteweg–de Vries equation, the same
way as the filament equation to the non-linear Schrödinger equation.
It is shown in [16] that the planar filament equation has an infinite hierarchy of integrals, namely, the odd-numbered
ones in the sequence (2) (the torsion τ vanishes for plane curves). Likewise, the odd-numbered vector fields in the list (3)
commute and preserve the planarity of curves. With this modification, Theorem 5 holds for the bicycle transformation in
dimension two.
An interesting ‘bicycle’ problem is to describe the situations when, given closed rear and front tracks, γ and Γ , one cannot
tell which way the bicycle went [17]. A trivial example is a pair of concentric circles; for numerous non-trivial examples,
see [18–23] and Fig. 4.
Such an ‘ambiguous’ front track Γ can be characterized by the property that a chord of fixed length ℓ can traverse the
curve in such a way that the velocity of its mid-point is aligned with the chord. That is, Γ is in the bicycle correspondence
with itself: Bℓ (Γ , Γ ). Following [24,25], we call such curves Zindler curves (see [26]).1 Of course, Zindler curves can be
defined in other dimensions as well.
The 2-dimensional version of Theorem 5 has the following corollary.

1 Incidentally, (the interiors of) Zindler curves are solutions to the 2-dimensional case of Ulam’s problem: which bodies float in equilibrium in all
positions? See [27].
 S. Tabachnikov / Journal of Geometry and Physics ( ) – 7

Corollary 4.1. Let Γ be a Zindler curve, and let Γ1 be the result of evolving Γ along the planar filament flow. Then Γ1 is also a
Zindler curve with the same length parameter ℓ. The same is true if the planar filament field is replaced by any of the odd-numbered
vector fields in (3) or their linear combination.
Proof. As we mentioned earlier, the end points of the segment of fixed length move along a Zindler curve with the
same speed. Let Γ (x) be an arc length parameterization. Then Bℓ (Γ (x), Γ (x + c )) for some constant c. Since the bicycle
transformation commutes with the planar filament flow, and the latter preserves the arc length, we have Bℓ (Γ1 (x), Γ1 (x +
c )), as needed. 
Recall that a soliton of the filament equation is a curve that is a critical point for a linear combination of integrals of
motion.

Conjecture 5. The Zindler curves are solitons of the planar filament equation.
In particular, according to preliminary computations of R. Perline, the Zindrel curves constructed by Wegner [22] are
pressurized elastica, that is, extrema of a linear combination of area, length, and total squared curvature.

5. Other integrals

The bicycle correspondence in Rn has additional integrals.

Theorem 6. The bivector and the vector,

 
A(Γ ) = Γ (t ) ∧ Γ (t ) dt and J (Γ ) =
′
(Γ (t ) · Γ ′ (t )) Γ (t ) dt ,
Γ Γ

are preserved by the bicycle correspondence.

Proof. As before, Γ = γ + ℓγ ′ with t being an arc length parameter along γ . The odd in ℓ part of A(Γ ) is ℓ γ ∧ γ ′′ dt.

Note that γ ∧ γ ′′ = (γ ∧ γ ′ )′ . This would prove that the integral vanishes for a smooth rear track γ . If γ has cusps, we note
that γ ′ and ℓ change sign in each cusp, and we conclude similarly to the proof of Theorem 4.
Likewise, for J (Γ ), the odd in ℓ parts are
 
ℓ ((1 + γ · γ ′′ ) γ + (γ · γ ′ ) γ ′ ) dt and ℓ3 (γ ′ · γ ′′ ) γ ′ dt .

Concerning the first integral, the integrand is ((γ · γ ′ ) γ )′ , and that (γ · γ ′ ) γ changes sign at cusps, along with ℓ. The second
integral vanishes because γ (t ) is an arc length parameterization, and γ ′ · γ ′′ = 0. 
If the ambient dimension is n then the area bivector A(Γ ) provides 2 integrals, the areas of the projections on all the
n
coordinate 2-planes, and the ‘‘centroid’’ vector J (Γ ) provides n integrals. In dimension two, the meaning of these integrals
is as follows: A(Γ ) is twice the area bounded by the curve Γ , and the vector J (Γ ), rotated 90◦ and divided by the area, is the
center of mass of the domain bounded by Γ .
We finish with a remark about the area vector as an integral of the filament equation. Let a be a constant vector field in
R3 (that is in the kernel of the form ω), and let

1 1
Fa = A(Γ ) · a = (Γ (t ) × Γ ′ (t )) · a dt ,
2 2 Γ
the projection of the area vector along a. Likewise, define

Ja = −J (Γ ) · a = − (Γ (t ) · Γ ′ (t )) (Γ (t ) · a) dt .
Γ

Lemma 5.1. One has:

ia Ω = dFa = iΓ ×a ω, iΓ ×a Ω = dJa .

Proof. Let u be a test vector field along Γ . Then

 
1
dFa (u) = [(u × Γ ′ ) + (Γ × u′ )] · a dt = det(Γ ′ , a, u) dt = ia Ω (u),
2 Γ Γ

proving the first equality. The second equality follows from the identity (a × b) · c = det(a, b, c ) for any triple of vectors.
For the last equality, let v be a test vector field along Γ . Then
 
iΓ ×a Ω (v) = det(Γ ′ , Γ × a, v) dt = [(Γ ′ · a)(Γ · v) − (Γ ′ · Γ )(a · v)] dt ,
Γ Γ
8 S. Tabachnikov / Journal of Geometry and Physics ( ) –

where we use the identity

det(a, b × c , d) = (a · c )(b · d) − (a · b)(c · d)
that holds for every quadruple of vectors. On the other hand,

dJa (v) = − [(v · Γ ′ )(Γ · a) + (Γ · v ′ )(Γ · a) + (Γ · Γ ′ )(v · a)] dt
Γ

= [(Γ ′ · a)(Γ · v) − (Γ ′ · Γ )(a · v)] dt ,
Γ

where we used integration by parts. This completes the proof. 

Therefore, a is a Hamiltonian vector field of the function Fa with respect to Ω . The constant vector field a commutes with
the filament flow and all higher flows Xn , and hence Fa , and therefore A(Γ ), is an integral of the filament equation. For a
discrete version of the filament equation, the area vector is a monodromy integral, see [8].

Acknowledgments

I am grateful to J. Langer and R. Perline for introducing me to the beautiful theory of the filament equation and related
topics; many thanks to R. Perline for pointing out the connection between the bicycle model and the filament equation. It
is a pleasure to acknowledge stimulating discussions on the subject of this note that I have had over the years with G. Bor,
R. Foote, T. Hoffmann, B. Khesin, M. Levi, V. Ovsienko, F. Pedit, U. Pinkall, C. Roger, Yu. Suris, W. Thurston, E. Tsukerman, A.
Veselov, and V. Zharnitsky. I was supported by the NSF grants DMS-1105442 and DMS-1510055.

References

[1] R. Foote, M. Levi, S. Tabachnikov, Tractrices, bicycle tire tracks, hatchet planimeters, and a 100-year-old conjecture, Amer. Math. Monthly 120 (2013)
199–216.
[2] M. Levi, S. Tabachnikov, On bicycle tire tracks geometry, hatchet planimeter, Menzin’s conjecture, and oscillation of unicycle tracks, Exp. Math. 18
(2009) 173–186.
[3] M. Levi, ‘‘Bike tracks’’, quasi-magnetic forces, and the Schrödinger equation, SIAM News 47 (June 2014).
[4] S. Tabachnikov, Tire track geometry: variations on a theme, Israel J. Math. 151 (2006) 1–28.
[5] S. Tabachnikov, E. Tsukerman, On the discrete bicycle transformation, Publ. Mat. Urug. 14 (2013) 201–219.
[6] S. Howe, Sean, M. Pancia, V. Zakharevich, Isoperimetric inequalities for wave fronts and a generalization of Menzin’s conjecture for bicycle monodromy
on surfaces of constant curvature, Adv. Geom. 11 (2011) 273–292.
[7] T. Hoffmann, Discrete Hashimoto surfaces and a doubly discrete smoke-ring flow, in: Discrete differential geometry, in: Oberwolfach Semin., 38,
Birkhäuser, Basel, 2008, pp. 95–115.
[8] U. Pinkall, B. Springborn, S. Weissmann, A new doubly discrete analogue of smoke ring flow and the real time simulation of fluid flow, J. Phys. A 40
(2007) 12563–12576.
[9] V. Arnold, B. Khesin, Topological Methods in Hydrodynamics, Springer-Verlag, New York, 1998.
[10] A. Calini, Recent developments in integrable curve dynamics, in: Geometric approaches to differential equations, in: Austral. Math. Soc. Lect. Ser., 15,
Cambridge University Press, Cambridge, 2000, pp. 56–99.
[11] J. Marsden, A. Weinstein, Coadjoint orbits, vortices, and Clebsch variables for incompressible fluids, Physica D 7 (1983) 305–323.
[12] J. Langer, R. Perline, Poisson geometry of the filament equation, J. Nonlinear Sci. 1 (1991) 71–93.
[13] J. Langer, Recursion in curve geometry, New York J. Math. 5 (1999) 25–51.
[14] A. Calini, T. Ivey, Bäcklund transformations and knots of constant torsion, J. Knot Theory Ramifications 7 (1998) 719–746.
[15] C. Rogers, W. Schief, Bäcklund and Darboux Transformations, Cambridge University Press, Cambridge, 2002.
[16] J. Langer, R. Perline, The planar filament equation, in: Mechanics day (Waterloo, ON, 1992), in: Fields Inst. Commun., 7, Amer. Math. Soc., Providence,
RI, 1996, pp. 171–180.
[17] D. Finn, Which way did you say that bicycle went? Math. Mag. 77 (2004) 357–367.
[18] F. Wegner, Floating bodies of equilibrium, Stud. Appl. Math. 111 (2003) 167–183.
[19] F. Wegner, Floating Bodies of Equilibrium I. arXiv:physics/0203061.
[20] F. Wegner, Floating Bodies of Equilibrium II. arXiv:physics/0205059.
[21] F. Wegner, Floating Bodies of Equilibrium. Explicit Solution. arXiv:physics/0603160.
[22] F. Wegner, Floating Bodies of Equilibrium in 2D, the Tire Track Problem and Electrons in a Parabolic Magnetic Field. arXiv:physics/0701241.
[23] F. Wegner, Three problems – one solution. http://www.tphys.uni-heidelberg.de/~wegner/Fl2mvs/Movies.html#float.
[24] J. Bracho, L. Montejano, D. Oliveros, A classification theorem for Zindler carrousels, J. Dyn. Control Syst. 7 (2001) 367–384.
[25] J. Bracho, L. Montejano, D. Oliveros, Carousels, Zindler curves and the floating body problem, Period. Math. Hungar. 49 (2004) 9–23.
[26] K. Zindler, Über konvexe Gebilde II, Monatsh. Math. Phys. 31 (1921) 25–57.
[27] H. Auerbach, Sur un problème de M. Ulam concernant l?èquilibre des corps flottants, Studia Math. 7 (1938) 121–142.