Beruflich Dokumente
Kultur Dokumente
This content has been downloaded from IOPscience. Please scroll down to see the full text.
(http://iopscience.iop.org/1064-5616/205/2/157)
View the table of contents for this issue, or go to the journal homepage for more
Download details:
IP Address: 150.216.68.200
This content was downloaded on 30/09/2014 at 08:47
I. Yu. Beschastnyı̌
Abstract. The problem of a sphere rolling on the plane, with twisting but
without slipping, is considered. It is required to roll the sphere from one
configuration to another in such a way that the minimum of the action is
attained. We obtain a complete parametrization of the extremal trajecto-
ries and analyse the natural symmetries of the Hamiltonian system of the
Pontryagin maximum principle (rotations and reflections) and their fixed
points. Based on the estimates obtained for the fixed points we prove upper
estimates for the cut time, that is, the moment of time when an extremal
trajectory loses optimality. We consider the problem of re-orienting the
sphere in more detail; in particular, we find diffeomorphic domains in the
pre-image and image of the exponential map which are used to construct
the optimal synthesis.
Bibliography: 15 titles.
§ 1. Introduction
We consider a mechanical system consisting of a sphere rolling on a plane with
twisting but without slipping. At every moment of time the state of such a system
is characterized by a point on the plane and the orientation of the sphere in space.
It is required to roll the sphere from a given initial state to a given terminal state in
such a way that the minimum of the action is attained. No slipping means that the
contact point of the sphere and the plane has zero instantaneous velocity; presence
of twisting means that the angular velocity vector of the sphere can have arbitrary
direction.
This problem is a natural modification of the problem of the optimal rolling
of a sphere on a plane without twisting or slipping, which was first stated by
Hammersley [1]. Later Arthurs and Walsh [2] obtained equations for extremal
trajectories in terms of quaternions. Jurdjevic [3] conducted a qualitative study
of these trajectories and showed that for optimal rolling the contact point of the
sphere and the plane moves over Euler elastics. Sachkov initiated the study of
This research was supported by the grant of the Government of the Russian Federation for
state support of scientific research (contract no. 14.B25.31.0029) and by the Russian Foundation
for Basic Research (grant nos. 12-01-00913 and 13-01-91162-GFEN_a).
AMS 2010 Mathematics Subject Classification. Primary 49K15; Secondary 70B10, 93B27.
⃝
c 2014 Russian Academy of Sciences (DoM), London Mathematical Society, Turpion Ltd.
158 I. Yu. Beschastnyı̌
1 t1 2
Z
J[u] = (u1 + u22 + u23 ) dt, (2.2)
2 0
which, up to a constant factor, is the integral of the rotational energy of the sphere.
It is required to roll the sphere from an initial state Q0 to a terminal state Q1 in
such a way that the minimum of the functional J[u] is attained.
This problem can be stated in a natural way as a sub-Riemannian left-invariant
problem on the Lie group G = R2 × SO(3). We recall from [13] that a sub-
Riemannian manifold (M, ∆, g) is by definition a smooth manifold M with a distri-
bution ∆ on which a Riemannian metric g is defined. A Lipschitzian curve is said
to be admissible if it is tangent to the distribution ∆ at almost all of its points. The
sub-Riemannian problem is stated as follows: on the sub-Riemannian manifold it is
required to find a shortest (in the sense of the metric g) admissible curve connecting
two given points Q0 and Q1 .
The group G can be represented as a subgroup of GL(6) using the matrices
R 0
Q= .
1 0 x
0 0 1 y
0 0 1
where A
ei is the following basis of the Lie algebra so(3):
0 0 0 0 0 1 0 −1 0
Ae1 = 0 0 −1 , Ae2 = 0 0 0 , e3 = 1
A 0 0 .
0 1 0 −1 0 0 0 0 0
Then equations (2.1) define a control system on the group G and can be written in
the form
where
X1 = e1 − V2 , X2 = e2 + V1 , X3 = V3 .
160 I. Yu. Beschastnyı̌
The value of the minimizing length functional (2.8) is independent of the para-
metrization of the curve Q(t); therefore we can assume that it has constant velocity,
that is, u21 + u22 + u23 ≡ const. Moreover, it is clear from (2.5) and (2.2) that if
a control u(t), t ∈ [0, t1 ], takes the sphere from the state Q0 to Q1 in time t1 , then
the control u′ (t) = ku(kt), where k is some positive number, takes Q0 to Q1 in
time t1 /k. Then Q(t) goes to Q(kt). This means we can assume without loss of
generality that u21 + u22 + u23 ≡ 1.
Since the multiplication table in the Lie algebra L = span{e1 , e2 , V1 , V2 , V3 } has
the form
ad ei = 0, [V1 , V2 ] = V3 , [V2 , V3 ] = V1 , [V3 , V1 ] = V2 ,
for the vector fields X1 , X2 , X3 we have
[X1 , X2 ] = X3 , [X1 , X3 ] = −V1 , [X2 , X3 ] = −V2 .
Then it is evident that span{X1 , X2 , X3 , [X1 , X3 ], [X2 , X3 ]} = L, and it follows from
the Rashevskiı̌-Chow theorem [13] that the system is completely controllable, that
is, any two points Q0 , Q1 can be connected by an admissible curve. The existence
of optimal trajectories in problem (2.5)–(2.8) follows from Filippov’s theorem [13].
The optimal rolling of a sphere 161
§ 3. Extremal trajectories
To find optimal trajectories for problem (2.5)–(2.7), (2.2) we apply Pontryagin’s
maximum principle (PMP) in an invariant formulation [13]. The Hamiltonian of
the maximum principle is the family of functions on the cotangent bundle T ∗ G
ν
hνu = λ, u1 X1 (Q) + u2 X2 (Q) + u3 X3 (Q) + (u21 + u22 + u23 ),
2
λ ∈ TQ∗ G, u = (u1 , u2 , u3 ) ∈ R3 , ν ∈ R.
We introduce the following basis Hamiltonians:
hi (λ) = ⟨λ, Xi (Q)⟩, i = 1, 2, 3,
h4 (λ) = −⟨λ, V1 (Q)⟩, h5 (λ) = −⟨λ, V2 (Q)⟩.
Since the basis Hamiltonians are functions that are linear on every fibre, they can
be regarded as coordinates in the cotangent space. The cotangent space has dimen-
sion five, and therefore, apart from the three Hamiltonians hi , i = 1, 2, 3, generated
by the vector fields Xi , it is necessary to introduce two more basis Hamiltonians,
h4 and h5 , generated by the two independent vector fields V1 and V2 .
The Hamiltonian of the PMP takes the form
ν
hνu = u1 h1 + u2 h2 + u3 h3 + (u21 + u22 + u23 ). (3.1)
2
Theorem 1 (Pontryagin maximum principle). Let u e(t) and Q(t),
e t ∈ [0, t1 ], be
an optimal control and the corresponding optimal trajectory in problem (2.5)–(2.7),
∗
(2.2). Then a nontrivial Lipschitzian curve λt ∈ TQ(t) e G and a number ν ∈ R exist
such that
(ν, λt ) ̸= 0
and the following conditions hold:
→
− →
− →
− →
−
λ̇t = h νue(t) (λt ) = u
e1 (t) h 1 (λt ) + u
e2 (t) h 2 (λt ) + u
e3 (t) h 3 (λt ), (3.2)
hνue(t) = max hνu (λt ) for almost all t ∈ [0, t1 ], (3.3)
u∈R3
ν 6 0,
→
−
where h νue(t) is the Hamiltonian vector field on the cotangent bundle T ∗ G corres-
ponding to the Hamiltonian hνue(t) .
The curve λt is called an extremal, and its projection onto the base Q(t) =
π(λ(t)) an extremal trajectory. In the coordinates hi , the Hamiltonian system (3.2)
is written in the form
Q̇ = u1 X1 (Q) + u2 X2 (Q) + u3 X3 (Q),
ḣ1 = −u2 h3 − u3 h4 ,
ḣ2 = u1 h3 − u3 h5 ,
(3.4)
ḣ3 = u1 h4 + u2 h5 ,
ḣ4 = −u1 h3 + u3 h5 ,
ḣ5 = −u2 h3 − u3 h4 .
162 I. Yu. Beschastnyı̌
where Ω
e is the skew-symmetric matrix corresponding to the angular velocity vector
→
−
Ω with component Ω3 = 0:
0 0 Ω2
e = 0
Ω 0 −Ω1 .
−Ω2 Ω1 0
Consequently, in the abnormal case the angular velocity vector is a constant hor-
izontal vector and the sphere is rolling uniformly along a straight line without
twisting.
The optimal rolling of a sphere 163
Proof. The proof of part 1) was given before the statement of this proposition.
2) Let Q(t), t ∈ [0, t1 ], be an abnormal extremal trajectory and let t1 > 0. We
now prove that Q(t) is optimal.
By part 1) abnormal curves of constant velocity satisfy the conditions
u1 = ẋ = const, u2 = ẏ = const, u3 ≡ 0.
3.2. Normal trajectories. We now consider the case ν = −1. The Hamiltonian
(3.1) is equal to
1
hu = u1 h1 + u2 h2 + u3 h3 − (u21 + u22 + u23 )
2
The vertical subsystem of the Hamiltonian system (3.4) for normal extremals
has the form
ḣ1 = −h3 (h2 + h4 ),
ḣ2 = h3 (h1 − h5 ),
ḣ3 = h1 h4 + h2 h5 , (3.8)
ḣ4 = −h3 (h1 − h5 ),
ḣ5 = −h3 (h2 + h4 ).
→
− →
− →
−
We set ω = |→−
ω | and Ω 0 = Ω (0) = (Ω01 , Ω02 , Ω03 ). If Ω 0 = λ→
−
ω for some λ ∈ R or
→
−̇ →
−
ω = 0, then we see that Ω = 0, and, consequently, Ω is a constant vector. In this
The optimal rolling of a sphere 165
case the equations for the extremal trajectories have the same form as (3.7); the
only difference is that Ω3 is not necessarily equal to zero, and therefore the sphere
is rolling along straight lines with or without twisting. Therefore the abnormal
extremal trajectories are normal, that is, they are not strictly abnormal.
We integrate equations (3.15) for ω ̸= 0. We obtain
2
ω1 + ω22 cos ωt ω1 ω2 ω2
(1 − cos ωt) sin ωt
ω2 ω2 ω Ω01
Ω1
Ω2 = ω1 ω2 2 2 0
ω2 + ω1 cos ωt ω1 Ω2 . (3.16)
2
(1 − cos ωt) 2
− sin ωt
Ω3 ω ω ω Ω03
ω2 ω1
− sin ωt sin ωt cos ωt
ω ω
Using equations (3.16) we integrate equations (2.1) for x and y:
Ω03 ω1 (Ω01 ω2 − Ω02 ω1 )ω1 ω2 (Ω02 ω2 + Ω01 ω1 )
x= (1 − cos ωt) + sin ωt − t,
ω2 ω3 ω2 (3.17)
Ω0 ω2 (Ω0 ω2 − Ω02 ω1 )ω2 ω1 (Ω02 ω2 + Ω01 ω1 )
y = 3 2 (1 − cos ωt) + 1 3
sin ωt + t.
ω ω ω2
We will now obtain an expression for the matrix R. To do this we consider
equation (3.15). It is easy to see that its solution is given by
Ω(t)
e e 0 e−teω .
= eteω Ω
Substituting Ω
e into (2.1) we obtain
e 0 e−teω .
Ṙ(t) = R(t)eteω Ω
We make the change of variables
Z(t) = R(t)eteω .
Differentiating Z(t) with respect to t we obtain
Z(t) = Z(0)et(eω+Ω0 ) .
e
Substituting in the original variables and taking the fact that Z(0) = Id into account
we obtain the expression for the matrix R
which shows that the motion of the sphere consists of the composition of two uni-
form rotations: about the vectors →
−
ω and →− ⃗
ω + Ω.
Consequently, we have proved the following proposition.
166 I. Yu. Beschastnyı̌
→
−
Proposition 2. If ω ̸= 0 and Ω ̸= λ→ −ω , then the normal extremal trajectories
parametrized by arc length are described by the equations
Ω03 ω1 (Ω0 ω2 − Ω02 ω1 )ω1 ω2 (Ω02 ω2 + Ω01 ω1 )
x= 2
(1 − cos ωt) + 1 3
sin ωt − t,
ω ω ω2 (3.19)
Ω0 ω2 (Ω0 ω2 − Ω02 ω1 )ω2 ω1 (Ω02 ω2 + Ω01 ω1 )
y = 3 2 (1 − cos ωt) + 1 sin ωt + t,
ω ω3 ω2
R(t) = et(eω+Ω0 ) e−teω . (3.20)
e
In the remaining cases the normal extremal trajectories are described by the equa-
tions
x = −Ω2 t, y = Ω1 t, R = eΩt , (3.21)
e
⇐⇒ e˙ s = [e
Ω ωs , Ω
e s ].
Ṙsβ = Rsβ Ω
eβ
s ⇐⇒ eβA3 Ṙs e−βA3 = eβA3 Rs e−βA3 eβA3 Ω
e s e−βA3
⇐⇒ eβA3 Ṙs e−βA3 = eβA3 (Rs Ω
e s )e−βA3 ⇐⇒ Ṙs = Rs Ω
e s.
The optimal rolling of a sphere 167
In the case of equations (3.9), (3.10) we use equation (3.6). It follows from (3.23)
that
e s e−βA3
e βs = eβA3 Ω →
−β →
−
Ω ⇐⇒ Ω s = eβA3 Ω s
β
Ω1s cos β − sin β 0 Ω1s
⇐⇒ Ωβ = sin β cos β 0 Ω2s .
2s
Ωβ3s 0 0 1 Ω3s
ẋβs −Ωβ2s
cos β sin β ẋs cos β sin β −Ω2s
= ⇐⇒ =
ẏsβ Ωβ1s − sin β cos β ẏs − sin β cos β Ω1s
ẋs −Ω2s
⇐⇒ = .
ẏs Ω1s
q
q −1 = .
|q|2
Rq : a 7→ qaq −1 ∈ I.
To every rotation operator Rq there correspond two different quaternions q and −q,
and therefore the map p : q 7→ Rq defines a two-sheeted covering S 3 over SO(3).
The covering map is a homomorphism [15].
The matrix R of a rotation about an arbitrary nonzero vector → −a ∈ R3 , →
−a =
(a1 , a2 , a3 ), through an angle β has a corresponding unit quaternion
β a1 i + a2 j + a3 k β
q = cos + sin . (3.26)
2 |→
−
a| 2
The optimal rolling of a sphere 169
In particular, since →
−
ω = (ω, 0, 0), the quaternions
→
− →
−
|→
−
ω + Ω 0 |t (ω + Ω01 )i + Ω02 j + Ω03 k |→
−
ω + Ω 0 |t
cos + →
− sin ,
2 |→
−ω + Ω 0| 2
ωt ωt
cos − i sin
2 2
correspond to the matrices et(eω+Ω0 ) and e−teω involved in (3.20). Substituting the
e
expressions we have obtained into (3.20) we obtain a law for changing the compo-
nents of the quaternion q corresponding to the matrix R:
→
− →
−
|Ω0 + →
−ω |t ωt Ω01 + ω |Ω0 + → −ω |t ωt
q0 = cos cos + → − sin sin ,
2 2 →
−
|Ω0 + ω | 2 2
→
− →
−
Ω01 + ω tω |Ω0 + →−ω |t |Ω0 + → −ω |t ωt
q1 = → − cos sin − cos sin ,
→
−
|Ω0 + ω | 2 2 2 2
→
− (3.27)
1
ωt ωt
→
−
| Ω 0 + ω |t
q2 = → − Ω02 cos − Ω03 sin sin ,
→
−
|Ω0 + ω | 2 2 2
→
−
|Ω0 + → −
1 0 ωt 0 ωt ω |t
q3 = → − Ω cos + Ω sin sin .
→
−
|Ω0 + ω |
3
2 2
2 2
M
f
/N
g g′
M
f
/ N.
In what follows we shall denote the pair of maps g and g ′ by the one symbol g.
The exponential map is defined to be a map Exp(λ, t) taking a covector in the
cotangent space λ ∈ {λ ∈ TQ∗ 0 G : H(λ) = 1/2} and a moment of time t ∈ R+ to the
end-point of the corresponding normal extremal trajectory. In the problem under
consideration, this is the map
Exp : S 2 × R2 × R+ → R2 × SO(3),
or
→
− −
Exp : ( Ω , →
ω , t) 7→ (x, y, R)
which is defined by equations of the form (3.7) if the projection of the trajectory onto
the plane (x, y) is a straight line, and (3.22), (3.20) if the projection is a sinusoid.
170 I. Yu. Beschastnyı̌
In this section we consider the symmetries of the exponential map and the Hamil-
→
−
tonian system (3.9)–(3.14) that preserve the condition | Ω | = const. As shown ear-
lier, the Hamiltonian system has one continuous symmetry Φβ . Later we shall show
that the exponential map also has seven discrete symmetries εi , i = 1, . . . , 7. For
brevity, we use the notation N = S 2 × R2 × R+ in what follows.
4.1. Discrete symmetries in the pre-image of the exponential map.
Proposition 5. Equations (3.12)–(3.14) have the following symmetries:
and →
−
ωi = →
−
ω.
To prove this we only need to substitute (4.1) into (3.12)–(3.14).
The transformations (4.1) can also be written in the matrix form
ε1 : Ω
es →
7 −I2 Ω
e t−s I2 ,
ε2 : Ω
es →
7 −I3 Ω
e t−s I3 ,
ε3 : Ω
es →
7 I1 Ω
e s I1 ,
ε4 : Ω
es →
7 −I1 Ω
e s I1 , (4.2)
ε5 : Ω
es →
7 I3 Ω
e t−s I3 ,
ε6 : Ω
es →
7 I2 Ω
e t−s I2 ,
ε7 : Ω
es →
7 −Ω
e s,
where
1 0 0 −1 0 0
I1 = 0 −1 0 = eπA1 ∼
= i, I2 = 0 1 0 = eπA2 ∼
= j,
0 0 −1 0 0 −1
−1 0 0
I3 = 0 −1 0 = eπA3 ∼
= k.
0 0 1
The discrete symmetries commute and form a group with the following multipli-
cation table:
The optimal rolling of a sphere 171
ε1 ε2 ε3 ε4 ε5 ε6 ε7
ε1 Id ε3 ε2 ε5 ε4 ε7 ε6
2 1 6 7 4
ε Id ε ε ε ε ε5
ε3 Id ε7 ε6 ε5 ε4
ε4 Id ε1 ε2 ε3
ε5 Id ε3 ε2
6
ε Id ε1
ε7 Id
x5s = xt−s − xt ,
x6s = xt − xt−s ,
5 6
ys = yt−s − yt , ys = yt−s − yt ,
5 6
ε : Rs5 = I3 eteω e−t(eω−Ω0 ) ε : Rs6 = I2 eteω e−t(eω−Ω0 )
e e
×e(t−s)(eω−Ω0 ) ×e(t−s)(eω−Ω0 )
e
e
×e−(t−s)eω I3 , ×e−(t−s)eω I2 ,
7
xs = −xs ,
7
ε : ys7 = −ys ,
Rs = es(eω−Ω0 ) e−seω .
7 e
Proof. Since we use the same method in all cases, we will only prove the assertion
for the symmetry ε1 .
Equations (3.9), (3.10) imply
d
ẋ1s = (xt−s − xt ) = −ẋt−s = Ω2 (t − s) = −Ω12 (s), (4.3)
ds
d
ẏs1 = (yt − yt−s ) = ẏt−s = Ω1 (t − s) = Ω11 (s). (4.4)
ds
We now verify that equation (3.11) is invariant under ε1 :
d
Ṙs1 = (I2 (Rt )−1 Rt−s I2 ) = −I2 (Rt )−1 Ṙt−s I2 = −I2 (Rt )−1 Rt−s Ωt−s I2
ds
= I2 (Rt )−1 Rt−s I2 (−I2 Ωt−s I2 ) = Rs1 Ω1s .
The symmetry ε2 is the reflection of the arc of a sinusoid with respect to the
centre of the chord connecting the beginning and end-point of a trajectory. The
The optimal rolling of a sphere 173
symmetries ε3 , ε4 are the reflections of the arc with respect to the coordinate axes
OY and OX, respectively. All the other symmetries are compositions of these
reflections (Fig. 4).
Explicit expressions for the Rsi in terms of Rs have been obtained only for the
first three, ε1 –ε3 . Therefore in what follows we confine ourselves to studying only
these three reflections.
4.3. Symmetries of the exponential map. A rotation
Φβ : λt 7→ λβt
(see (3.23)–(3.25)) is a symmetry of the Hamiltonian system; therefore its action in
T ∗ G decomposes naturally into a direct sum of actions in N and G:
Φβ : N → N, (λ, t) 7→ (λβ , t),
→
− − →
− −β →
−β →
−
λ = (Ω,→
ω ), λβ = ( Ω β , →
ω ), Ω = eβA3 Ω , →
−
ω β = eβA3 →
−
ω,
where λ = λt |t=0 , and
Φβ : G → G, Q 7→ Qβ ,
Q = (x, y, R), Qβ = (xβ , y β , Rβ ),
β
x cos β − sin β x
= , Rβ = eβA3 Re−βA3 .
yβ sin β cos β y
The following proposition answers the question of when two extremal trajectories
are different in the problem under consideration.
Suppose that Ω3 (s) ̸= Ω3 (0) for some s > 0, that is, the trajectory is not
abnormal. Then we have two cases.
→
−̇ →
−
1. →
−
ω = 0. Then it follows from (3.15) that Ω = 0. Therefore, Ω = const, and
→
−′
by the last assertion of the chain (5.1) we have Ω = const. Consequently,
→
−
ω ′ = 0.
→
− →
−
2. ω ̸= 0. In this case we obtain that the vector Ω s is rotating uniformly
→
− →
− →
−
about ω with velocity ω. But since we must have Ω s ≡ Ω ′s , we find that
→
−′
Ω s must also be rotating about → −
ω with the same angular velocity in the
same direction. Therefore, ω = →
→
− ′ −
ω , and the first part of the assertion is
proved.
→
− →
−
Suppose that Ω s ≡ Ω 0 is a constant horizontal vector, that is, the extremal
→
−
trajectory is abnormal. Then → −ω = λ Ω for an arbitrary λ; but, as follows from
(3.7), the trajectory is independent of ω. Therefore the last assertion of the chain
→
− →
−
(5.1) is sufficient and is equivalent to Ω 0 = Ω ′0 . The proposition is proved.
The optimal rolling of a sphere 175
In what follows we shall assume that q = q0 +q1 i+q2 j +q3 k ∈ S 3 , q(0) = 1, is the
quaternion corresponding to the orientation matrix R, and χ is the angle between
the chord connecting (0, 0) and (x, y) determined from the following equations for
(x, y) ̸= (0, 0):
p
x = ρ cos χ, y = ρ sin χ, ρ = x2 + y 2 .
5.1. Fixed points in the image of the exponential map and the Maxwell
time for a sphere rolling along a straight line. By Proposition 1, if the
angular velocity vector is horizontal for a sphere rolling along a straight line, then
the trajectory is optimal and there are no Maxwell points on it. If, however, the
angular velocity vector is not horizontal, then the following proposition holds.
→
−
Proposition 9. If the sphere is rolling along a straight line in such a way that Ω
is not horizontal, then t = 2π is a Maxwell time. If the sphere is twisting on the
spot, then t = π is a Maxwell time.
Proof. Let Qt be the trajectory described by equations (3.21) with Ω03 ̸= 0. If
Ω01 = 0 and Ω02 = 0, then the sphere stays on the spot and twists; otherwise it rolls
along a straight line.
Consider a trajectory Q′t such that → − ′ ′ ′
ω ′ = 0, Ω10 = Ω01 , Ω20 = Ω02 , Ω30 = −Ω03 . For
a moment of time t to be a Maxwell time, we must have Q′t = Qt . It follows from
(3.21) that the projection of Q′t onto the plane (x, y) coincides with the projection
of Qt , that is, xt = x′t and yt = yt′ for any t. Consequently, the proof reduces to
verifying that q(2π) = ±q ′ (2π) in the first case, and q(π) = ±q ′ (π) in the second,
e′
where q and q ′ are the quaternions corresponding to the matrices eΩ0 t and eΩ0 t ,
e
t t t t
q(t) = cos + (Ω01 i + Ω02 j + Ω03 k) sin , q ′ (t) = cos + (Ω01 i + Ω02 j − Ω03 k) sin .
2 2 2 2
The proposition is proved.
5.2. Fixed points in the image of the exponential map for the case of
rolling along a sinusoid. In what follows we assume that the sphere is rolling
→
−
along a sinusoid, that is, the conditions →
−
ω ̸= 0 and Ω 0 ̸= λ→−ω hold. Then the
exponential map is defined by equations (3.22) and (3.27). In what follows we use
the following lemma.
Lemma 1 (see [4]). Let q ∈ S 3 ⊂ H. Then
1. q 2 = 1 ⇐⇒ q = ±1 ⇐⇒ Im q = 0;
2. q 2 = −1 ⇐⇒ Re q = 0.
We now find the fixed points in the image of the exponential map under the
action of the symmetries Φβ ◦ εi .
Proposition 10. 1. If (x, y) ̸= (0, 0), then the condition Φβ ◦ ε1 (Q) = Q is equiv-
alent to the disjunction
Suppose that (x, y) = (0, 0); then at the first moment of time when the sphere
returns to the origin, q2 will be equal to zero. Indeed, we transform (3.22) to the
form
2 ωt ωt ωt
x(t) = − sin Ω02 cos − Ω03 sin . (5.3)
ω 2 2 2
ωt ωt ωt
sin = 0 or Ω02 cos − Ω03 sin = 0,
2 2 2
but, as is evident from (3.27), when the second equation holds, q2 is equal to zero.
In this case it follows from the first equation (5.2)
β β
−q2 cos + q1 sin = 0
2 2
q3 = 0 or q = ±k.
Proof. By the definition of the action of the symmetry ε2 , this reflection takes the
endpoint of the sinusoid to itself, and therefore β = 0. Next, we have
−1 2 2 Re(qk) = 0,
I3 R I3 = R ⇐⇒ (RI3 ) = Id ⇐⇒ (qk) = ±1 ⇐⇒
qk = ±1
q3 = 0,
⇐⇒
q = ±k
The proposition is proved.
Proposition 12. 1. If (x, y) ̸= (0, 0), then the condition Φβ ◦ ε3 (Q) = Q is equiv-
alent to the disjunction
( (
q1 x + q2 y = 0, q1 y − q2 x = 0,
or
q3 = 0 q0 = 0.
q3 = 0 or q0 = 0,
Suppose that (x, y) ̸= (0, 0). Then similarly to the proof of Proposition 10 we
see that β = 2χ + π. Substituting this value into (5.4) we arrive at the equations
given in the statement of this proposition.
If (x, y) = (0, 0), then q2 = 0. Therefore (5.4) is transformed to the form
β β
q1 sin = 0, −q1 cos = 0,
2 or 2
q = 0 q = 0.
3 0
Hence it is clear that β = 0 in the first case, and β = π in the second. The
proposition is proved.
178 I. Yu. Beschastnyı̌
Therefore,
→
−
ωt ωt |Ω0 + →
−
ω |t
Ω03 cos + Ω02 sin = 0 or
sin = 0.
2 2 2
→
−
We denote the angle between the projection of Ω onto the plane Ω2 OΩ3 and the
axis OΩ2 by ψ,
Ω2 (t) Ω3 (t)
cos ψ(t) = p , sin ψ(t) = p .
1 − (Ω1 (t))2 1 − (Ω1 (t))2
2π
t2max = →− ,
|Ω0 + →
−
ω|
which has the following property. If we substitute t2max into (3.20), then we obtain
→
− →
−
R = e(2π/| Ω 0 + ω |)eω ,
→
−
that is, this is the rotation about the vector →−ω through the angle 2π/| Ω 0 + → −
ω |.
→
−
Hence it is evident that if we fix ω , then all the trajectories for which the values
→
−
|Ω0 + → −
ω | coincide have the same component R at the moment of time t2max . But
→
−
different points Ω 0 taken on the same circle that is a solution of (3.15) correspond
→
−
to such trajectories, since when Ω 0 rotates about →−
ω , the angle between them does
not change.
We now find the solutions of the equation xq1 + yq2 = 0 which describe the fixed
points of the symmetry Φβ ◦ ε1 . Substituting x(t), y(t), q1 (t), and q2 (t) we obtain
→
−
Ω01 |Ω0 + →−
ωt ωt ω |t
Ω02 cos − Ω03 sin →
− sin
2 2 →
−
|Ω0 + ω | 2
→
− →
−
|Ω0 + →− |Ω0 + →
−
0
2 ωt Ω1 + ω ωt ω |t ω |t ωt
− sin cos sin − cos sin = 0.
ω 2 |→ −
Ω0 + →−
ω| 2 2 2 2
The optimal rolling of a sphere 179
π(2k − 1) − 2ψ0
tR = ,
ω
where k ∈ Z is chosen in such a way that the time tR is minimal for a given
trajectory. It is evident from (5.3) that then x(tR ) = 0. However, this moment of
time is not a Maxwell time, since the trajectory is transformed into itself under the
action of the corresponding symmetry.
Indeed, as in the case of the symmetry ε2 , from (4.1) we have
where (
1 if ψ0 ∈ [0, π),
l=
2 if ψ0 ∈ [π, 2π).
It follows from Proposition 10 that the trajectory we have obtained should be
rotated through the angle β = π + 2χ, but since x(tR ) = 0, we have χ = π/2 and
β = 2π, and so we conclude that the trajectory remains fixed.
In spite of what we have said above, the time tR plays an important role in the
problem of re-orientation, which is discussed in the next section. In fact, this is
the moment of the first return of the sphere to the origin when it is rolling along
a segment of a straight line. In order not to be concerned with the choice of k in
the expression for tR , it is convenient to introduce a new variable
π
ψ′ = ψ + mod π.
2
In this case, tR is equal to
2(π − ψ0′ )
tR = . (6.2)
ω
time t = 2π/ω. It is easy to verify that t2max < 2π/ω. This follows from the obvious
inequality
1 1
√ < . (7.1)
ω2 + 1 ω
Therefore a re-orientation can be optimal only if the sphere returns to the origin
at time tR .
Since the return time is fixed, the pre-image of the restricted exponential map is
a semicylinder R+ × S 1 defined by the values of (ω, ψ0′ ). As follows from equations
(3.27), the orientation quaternion q(tR ) takes the form
√ 2
ω +1
q0 = − cos (π − ψ0′ ) cos ψ0′
ω
√ 2
ω ω +1
+√ sin (π − ψ0 ) sin ψ0′ ,
′
ω2 + 1 ω
√ 2
ω ω +1
q1 = − √ sin (π − ψ0′ ) cos ψ0′
ω2 + 1 ω (7.2)
√ 2
ω +1
− cos (π − ψ0′ ) sin ψ0′ ,
ω
q2 = 0,
√ 2
1 ω +1
q3 = √ sin (π − ψ0 ) sin(π + ψ − ψ0′ ).
′
ω2 + 1 ω
Exp′ : R+ × S 1 → S 2 ,
Exp′ (ω, ψ0′ ) = q(tR ). (7.3)
π − ψ0′
ω=p ′ .
ψ0 (2π − ψ0′ )
Proof. Part 2) follows from inequality (7.1). The proof of part 1) is based on
comparing the quantities
2(π − ψ0′ ) 2π
and √ ,
ω 1 + ω2
which reduces to solving quadratic inequalities of the type
′ ′
2 ψ0 (2π − ψ0 )
ω − 1 > 0.
(π − ψ0′ )2
Since a vector cannot have negative length, we obtain the desired inequalities. The
proposition is proved.
For ψ0′ ̸= 0, by Proposition 13 a trajectory can be optimal only if
π − ψ0′
ω>p ′ , (7.4)
ψ0 (2π − ψ0′ )
and for ψ0′ = 0 the inequality tR 6 t2max cannot hold for any →−ω.
The map (w, ψ0 ) 7→ (w′ , ψ0′ ) defines a two-sheeted covering
π π π 3π
R+ × − , ∪ , → R+ × (0, π),
2 2 2 2
( π
ψ0′ = ψ0 + mod π,
2
′
ω = ω,
We now consider a domain D in the space (ω, ψ0′ ) that characterizes the extremal
trajectories for which the inequality (7.4) is strict,
π − ψ0′
′ 2 ′
D = (ω, ψ0 ) ∈ R : ω > p ′ , 0 < ψ0 < π , (7.5)
ψ0 (2π − ψ0′ )
The main result of this section is the proof of the fact that the following maps
are diffeomorphisms:
where the Q′± are the quarters of the sphere S 2 for which q1 > 0 and q3 > 0 or
q3 < 0, respectively.
Below we shall use the following theorem frequently.
Theorem 2 (see [8]). Let F : X → Y be a differentiable map between manifolds X
and Y . If
1) the map F is nondegenerate, that is, its Jacobian does not vanish,
2) F is proper, that is, the pre-image of a compact set is a compact set,
3) X and Y are connected,
4) Y is simply connected,
then F is a diffeomorphism.
We introduce an auxiliary change of variables α = α(ω, ψ0′ ), θ = θ(ω, ψ0′ ) by
means of the expressions
√
ω ω2 + 1
α = arccos √ , θ= (π − ψ0′ ). (7.6)
2
ω +1 ω
Let
2 π
Π= (α, θ) ∈ R : 0 < α < , 0 < θ < π
2
and let f be the map defined by equations (7.6).
Lemma 2. The map f : D → Π is a diffeomorphism.
Proof. We apply Theorem 2. It is obvious that D is connected and Π is simply
connected. We claim that f (D) ⊂ Π. Since
dα 1
=− 2 < 0,
dω ω +1
The optimal rolling of a sphere 183
the function α(ω) is decreasing and its supremum andp infimum must be attained on
the boundary of the set D, that is, at ω = (π − ψ0′ )/ ψ0′ (2π − ψ0′ ) and ω → +∞,
respectively. We have
inf α(ω) = lim α(ω) = 0,
D ω→+∞
π − ψ0′
π
sup α(ω) = sup arccos = .
D ′
ψ0 ∈(0,π/2) π 2
Since the function α(ω) is strictly decreasing, we have 0 < α(ω, ψ0′ ) < π/2 for all
(ω, ψ0′ ) ∈ D.
Now, √
∂θ ω2 + 1
= − < 0,
∂ψ0′ ω
and so for any fixed ω the function θ = θ(ω, ψ0′ ) is a strictly decreasing function
with respect to ψ0′ , and therefore its supremum and infimum are attained on ∂D.
We have
inf θ = inf θ(ω, π) = 0,
D ω∈D
sup θ = sup θ(ω, 0) = lim θ(ω, 0) = π.
D ω∈D ω→+∞
Since the function θ(ω, ψ0′ ) is strictly decreasing for ω = const, we find that
0 < θ(ω, ψ0′ ) < π for all (ω, ψ0′ ) ∈ D.
The map f is nondegenerate on D, since
∂α ∂θ ∂θ ∂α 1
− = √ ̸= 0.
∂ω ∂ψ0′ ∂ω ∂ψ0′ ω ω2 + 1
Thus, to prove the lemma it remains to show that the map f is proper, that
is, the pre-image of a compact set is a compact set. Let Y be a compact set
in Π, and X = f −1 (Y ) its pre-image in D. We will prove that X is compact.
To do this we take any sequence {xi } ⊂ X and prove that it has a subsequence
converging in X. Let {xik } be a subsequence that has a limit (possibly, infinite).
To simplify the notation we denote it by {xi }. Since Y is compact, the sequence
{yi = f (xi )} is separated from the boundary of the set Π; therefore, ε < θi < π − ε
and ε < αi < π/2 − ε for some ε > 0.
1) Suppose that lim xi = ∞. Then ωi → ∞ and therefore, αi → 0, a contra-
diction.
2) Suppose thatplim xi = x ∈ ∂D. If (ψ0′ )i → π, then θi → 0, a contradiction. If
ωi → (π − (ψ0′ )i )/ (ψ0′ )i (2π − (ψ0′ )i ), then θi → π, a contradiction.
Consequently, the map f is proper and is a diffeomorphism by Theorem 2. The
lemma is proved.
Since f is a diffeomorphism, we can invert formulae (7.6) and substitute ω(α, θ)
and ψ0′ (α, θ) into (7.2):
q0 = cos θ cos(θ cos α) + cos α sin θ sin(θ cos α),
q1 = cos α sin θ cos(θ cos α) − cos θ sin(θ cos α), (7.7)
q3 = ± sin α sin θ.
184 I. Yu. Beschastnyı̌
These maps are defined in the standard way and are the diffeomorphisms
q0′ = cos θ,
q1′ = cos α sin θ,
q3′ = ± sin α sin θ,
where q3′ is taken with plus for g+ , and with minus for g− .
Consider the following change of variables:
′
q1 cos−1 q0′
′
q1 cos−1 q0′
′ ′
q0 = q0 cos p + q1 sin p ,
1 − q0′2 1 − q0′2
′
q1 cos−1 q0′
′
q1 cos−1 q0′
(7.9)
q1 = −q0′ sin p + q1
′
cos p ,
1 − q0′2 1 − q0′2
q3 = q3′ .
Lemma 3. The map h defined by (7.9) maps Q′+ and Q′− diffeomorphically to
themselves.
Proof. The sets Q′+ and Q′− are connected and simply connected. It is clear from
(7.9) that the map h is a rotation in the plane q0′ Oq1′ , and, consequently, q3′ does
not change sign. Therefore the pre-image of each of the quarters is contained
in the upper or lower half of the sphere. Therefore, to prove h(Q′+ ) ⊂ Q′+ and
h(Q′− ) ⊂ Q′− we only need to show that q1 > 0.
Now, we set cos α = k. Then it follows from formulae (7.7) that
q1 (k, θ) = k sin θ cos(θk) − cos θ sin(θk)
1
= k sin(θ + θk) + k sin(θ − θk) − sin(θk + θ) + sin(θ − θk)
2
1
= (1 + k) sin(θ(1 − k)) − (1 − k) sin(θ(1 + k)) .
2
We differentiate the function we have obtained with respect to θ:
∂q1 1 − k2
cos(θ(1 − k)) − cos(θ(1 + k)) = (1 − k 2 ) sin θ sin θk.
=
∂θ 2
∂q1
Thus, > 0 on Π, and q1 (α, θ) only has an infimum on ∂Π for θ = 0. It
∂θ
follows from (7.7) that q1 (α, 0) = 0, and therefore q1 (α, θ) > 0 on Π. Consequently,
h(Q′+ ) ⊂ Q′+ .
The Jacobian of the map h is equal to
p
(1 − q0′2 − q1′2 )( 1 − q0′2 − q0′ cos−1 q0′ )
Jh = .
(1 − q0′2 )3/2
The optimal rolling of a sphere 185
It vanishes when
or q
1 − q0′2 − q0′ cos−1 q0′ = 0.
The set {(q0′ , q1′ , q3′ ) ∈ R3 : q3′ = 0} does not intersect either Q′+ or Q′− . Consider
p
the function s(q0′ ) = 1 − q0′2 − q0′ cos−1 q0′ . Its derivative is equal to
ds
= − cos−1 q0′ .
dq0′
Thus, the function attains extremum at the points qs = {q ′ : q0′ = ±1}, and
max s(q0′ ) = s(−1) = π and min s(q0′ ) = s(1) = 0. But qs ∈ / Q′+ and qs ∈/ Q′− ,
and therefore s(q0′ ) ̸= 0 for any q ′ ∈ Q′+ or q ′ ∈ Q′− . Consequently, the map h is
nondegenerate.
We will show that the map h : Q′+ → Q′+ is proper. The proof that h : Q′− → Q′−
is proper is similar.
Let Y ⊂ Q+ be compact. We will prove that X = h−1 (Y ) is also compact. As
in the proof of Lemma 2, we take any sequence {xi } ∈ X and choose a subsequence
{xik } that has a limit. To simplify the notation we denote it by {xi }. Since Y is
compact, (q1 )i > ε and (q3 )i > ε for some ε > 0.
Suppose that lim xi ∈ ∂Q′+ . If (q1′ )i → 0, then (q1 )i → 0, which is a contra-
diction. If (q3′ )i → 0, then (q3 )i → 0, again a contradiction. Consequently, h is
a proper map, and it is a diffeomorphism of Q′+ onto itself by Theorem 2. Similar
arguments imply that h : Q′− → Q′− is also a diffeomorphism. The lemma is proved.
The maps h ◦ g+ and h ◦ g− are defined in coordinates by equations (7.7). There-
fore the following theorem holds.
Theorem 3. The restricted exponential map (7.3) defines the following diffeomor-
phisms:
Exp′ : D+ → Q′+ and Exp′ : D− → Q′− .
Proof. It follows from Lemmas 2 and 3 that the following diagrams are commuta-
tive:
f ◦mod+ f ◦mod−
D+ /Π D− /Π
Exp′ g+ Exp′ g−
Q′+ o Q′− o
h h
Q′+ Q′− .
Since mod± , f , g± and h are diffeomorphisms, it follows that Exp′ is also a diffeo-
morphism. The theorem is proved.
This theorem shows that equations (7.2) have a unique solution for any point
q ∈ Q′+ or q ∈ Q′− . Indeed, suppose that for some q ∈ Q′+ there exist two solutions
of (7.2). Then both points belong to D+ , since D− ∩ D+ = ∅, and points in ∂D±
are mapped into ∂Q± ; this follows from the definition of the set Π and (7.7). But
this contradicts the fact that Exp′ is a diffeomorphism between D+ and Q+ .
186 I. Yu. Beschastnyı̌
→
7.3. Re-orienting the sphere for an arbitrary ω . The results obtained in the
preceding subsections can be combined into the following theorem.
Proof. By Theorem 3 the exponential map factorized by rotations defines the two
diffeomorphisms Exp′ : D+ → Q′+ and Exp′ : D− → Q′− , where Q′+ and Q′− are the
two quarters of the sphere defined in (7.8).
By (3.26) the equations q1′ = 0, q2′ = 0 imply that a1 = 0, a2 = 0. But since the
vector → −
a is a unit vector, it follows that |a3 | = 1. Similarly the equations q3′ = 0,
′
q2 = 0 imply that |a1 | = 1. Since one orientation matrix has two corresponding
quaternions ±q, for → −ω = (ω, 0, 0) there exist two extremal trajectories approaching
the point Q1 = (0, 0, R1 ) if R1 is the matrix of a rotation about a vector → −
a =
′
(a1 , 0, a3 ) such that a1 ̸= 1 and a3 ̸= 1. Therefore the map Exp is a two-sheeted
covering.
We now consider the case when the vector → −
ω ∈ R2 is arbitrary. Under the
β →
−
action of the symmetry Φ the vector ω is rotated through the angle β, and
the orientation quaternion q± = ±(q0 + q1 i + q2 j + q3 k) goes to the quaternion
β
q± = ±(q0 + (q1 cos β − q2 sin β)i + (q1 sin β + q2 cos β)j + q3 k); this is easy to see
using (3.25). Then by (3.26) the vector → −a is rotated through the angle β, that is,
β
q is the rotation quaternion corresponding to the matrix of a rotation about the
vector → −
a β = (a1 cos β − a2 sin β, a1 sin β + a2 cos β, a3 ) through some angle. Since
Φ is a symmetry of the exponential map, Exp′ is also a two-sheeted covering in
β
In order to find the conjugate variables to which the desired optimal trajectory
corresponds, we act on each of the quaternions by the rotation Φβ , where β is
determined from the relations
q1 q2
cos β = ± p 2 , sin β = ∓ p .
q1 + q22 q12 + q22
The values of ω and ψ0′ are found from equations (7.2), that is,
√ 2
ω +1
− cos (π − ψ0′ ) cos ψ0′
ω
√ 2
ω ω +1
+√ sin (π − ψ0 ) sin ψ0′ = ±q0 ,
′
ω2 + 1 ω
√ 2
ω ω +1
−√ sin (π − ψ0′ ) cos ψ0′ (7.10)
ω2 + 1 ω
√ 2
ω +1
− cos (π − ψ0′ ) sin ψ0′ = q1β ,
ω
√ 2
1 ω +1 ′
−√ sin (π − ψ0 ) = |q3 |.
ω2 + 1 ω
By Theorem 3 we obtain a pair of solutions (ω, ψ0′ ), which define two extremal
trajectories satisfying the boundary conditions. We choose the values of (ω, ψ0′ ) to
which the shorter trajectory corresponds by comparing the times tR using formula
(6.2). Next, the values of ψ0 and →
−ω are determined as follows:
π →
−
ψ0 = ψ0′ − sign(q3 ) , ω = (ω cos β, ω sin β, 0),
2
and they define the desired control and optimal trajectory. Thus, if a matrix R1 sat-
isfies the hypothesis of Theorem 4, then solving the problem of optimally re-orient-
ing the sphere is reduced to solving equations (7.9) in the domains D+ and D− .
7.4. The induced metric on SO(3). The length functional (2.4) defines a met-
ric on R2 × SO(3), known as the Carnot-Carathéodory metric. In this case the
distance between points Q0 , Q1 ∈ R2 × SO(3) is the infimum of the lengths of
admissible curves connecting these points.
Let S(R0 , R1 ) denote the set of all extremal curves connecting Q0 = (0, 0, R0 )
with Q1 = (0, 0, R1 ). Then the Carnot-Carathéodory metric induces the following
metric on SO(3):
d(R0 , R1 ) = inf l(Q), (7.11)
Q∈S(R0 ,R1 )
where l(Q) is the length (2.4) of the curve Q(t). We describe some properties of
this metric: maximally distant pairs and the diameter (the supremum of the dis-
tances between pairs of points). This metric is left-invariant by construction. Con-
sequently, we can seek the most distant point from R0 = Id. This problem reduces
to finding an optimal trajectory with the maximal value of tR = 2(π − ψ0′ )/ω.
Consider the closure D (see (7.5)) in the space (ω, ψ0′ ). It follows from Theorem 4
that all the rotations about a unit vector → −a = (a1 , a2 , a3 ) with |a3 | ̸= 1 and
a1 + a2 ̸= 1 correspond to the variables (ω, ψ0′ ) ∈ D. The rotations about →
2 2 −
a =
′
(a1 , a2 , a3 ) with a3 = 0 p
correspond to the variables (ω, ψ0 ) in ∂D. Indeed, we
substitute ω = (π − ψ0′ )/ ψ0′ (2π − ψ0′ ) into (7.2) and find that when it returns to
the origin the sphere turns out to be in the state
Proposition 14. If the terminal state of the sphere in the re-orientation problem
is R1 = eθA1 , θ ∈ [0, 2π),
p then rolling along a segment of a straight line is optimal
only if ω = (π − ψ0′ )/ ψ0′ (2π − ψ0′ ) and ψ0′ 6 π/2.
Proof. The quaternion corresponding to the matrix eθA1 has the form
cos θ sin θ
q=± +i .
2 2
The fact that (ω, ψ0′ ) ∈ ∂D is obvious, since for (ω, ψ0′ ) ∈ D the component a3
is nonzero by Theorem 4. Then, as indicated in § 7.3, we need to solve the two
systems
θ ′ θ ′ θ
cos = cos ψ0 , cos = − cos ψ0 , cos = cos(π − ψ0′ ),
2 2 2
and ⇐⇒
θ θ θ
sin = sin ψ0′ sin = sin ψ0′ sin = sin(π − ψ0′ ).
2 2 2
For any θ each of these systems has a unique solution ψ0′ . Consequently, we
have a pair of extremal trajectories, from which we p need to choose the one with
the smaller value of tR . We substitute ω = (π − ψ0′ )/ ψ0′ (2π − ψ0′ ) into (6.2) and
obtain on ∂D the expression
q
tR (ψ0′ ) = 2 ψ0′ (2π − ψ0′ ). (7.12)
The function tR (ψ0′ ) is an increasing function on [0, π], and therefore the smaller of
the two values will be optimal: either ψ0′ , or π − ψ0′ . This implies the assertion of
the proposition. The proposition is proved.
The hypothesis of Proposition 14 and the fact that the function tR is increasing
imply that for an optimal
√ trajectory on ∂D the time √ tR takes its maximal √ value
at ψ0′ = π/2, ω = 1/ 3 and it is equal to tR = π 3. We claim that π 3 is the
maximal value for all optimal trajectories. √
First we verify that if |a3 | = 1, then the cut time satisfies tcut < π 3. Indeed, it
follows from the description of the set D that its image does not contain points with
q1 = 0 or q2 = 0. Therefore the sphere cannot get into the state eθA3 moving along
a segment of a straight line. Consequently, twisting√on the spot must be optimal.
But it follows from Proposition 9 that tcut 6 π < π 3 for such a motion.
We now consider the case |a3 | ̸= 1. The fact that tR is decreasing (see formula
(6.2)) both with respect to ω and with respect
√ to ψ0′ , together √
with the fact that
√
the maximum time is attained at ω = 1/ 3, imply that √ tR 6 π 3 for ω > 1/ 3.
Consequently, it only remains to verify the case ω < 1/ 3, which is shown in Fig. 7.
√
Proposition 15. In √ the problem of re-orienting the sphere for ω < 1/ 3 the cut
time satisfies tcut < π 3.
Proof. By Proposition 12 the set q0 = 0 is contained in the Maxwell set. Con-
sequently, if a trajectory crosses this set at a moment of time t, then it is not
√ q(t, ω) going out from the point q = 1; we
optimal later. Consider the trajectories
claim that for ω < 1/3√ at time t = π 3 the corresponding trajectories cross the set
q0 = 0, that is, q0 (π 3, ω) < 0.
The optimal rolling of a sphere 189
√
Figure 7. The domain D. In the non-coloured subdomain, tR 6 π 3.
§ 8. Conclusion
In this paper we have studied the problem of the optimal rolling of a sphere on
a plane with twisting but no slipping. We have obtained a complete parametrization
of the extremal trajectories, analysed the symmetries of the exponential map, found
their fixed points, and used them to obtain estimates for the cut time.
We have also considered the problem of re-orienting the sphere: we have found
the terminal states of the sphere for which the parameters of an optimal trajectory
are uniquely determined from equations (7.10). We have described some properties
of the left-invariant metric on the group of rotations of three-dimensional space
related to this problem.
Further studies could be directed to solving the optimal control problem for an
arbitrary terminal position of the sphere Q1 . To do this, first it would be neces-
sary to consider the symmetries of the Hamiltonian system ε4 –ε7 and to construct
the corresponding symmetries of the exponential map, find their fixed points, and
compare the new Maxwell times with those that have already been found in this
paper.
The author thanks Yu. L. Sachkov and A. A. Agrachev for posing the problem
and for some useful discussions during this work.
Bibliography
[10] Yu. L. Sachkov, “Exponential map in the generalized Dido problem”, Mat. Sb.
194:9 (2003), 63–90; English transl. in Sb. Math. 194:9 (2003), 1331–1359.
[11] Yu. L. Sachkov, “Complete description of the Maxwell strata in the generalized
Dido problem”, Mat. Sb. 197:6 (2006), 111–160; English transl. in Sb. Math. 197:6
(2006), 901–950.
[12] S. Popov, Extremal trajectories in the problem of the rolling of a sphere on a plane
with twisting, Diploma Thesis, Moscow State University 2009. (Russian)
[13] A. A. Agrachev and Yu. L. Sachkov, Geometric control theory, Fizmatlit, Moscow
2005. (Russian)
[14] M. M. Postnikov, Lectures in geometry. Semester V. Riemannian geometry,
Faktorial, Moscow 1998; English transl. Geometry VI. Riemannian geometry,
Encyclopaedia Math. Sci., vol. 91, Springer-Verlag, Berlin 2001.
[15] V. I. Arnol’d, Geometry of complex numbers, quaternions, and spins, Moscow
Centre for Continuous Mathematical Education, Moscow 2002. (Russian)