Circular Orbits Inside the Sphere of Death

Kirk T. McDonald
Joseph Henry Laboratories, Princeton University, Princeton, New Jersey 08544
arXiv:physics/0008226v1 [physics.class-ph] 28 Aug 2000

(November 8, 1993)

Abstract
A wheel or sphere rolling without slipping on the inside of a sphere in a uniform grav-
itational field can have stable circular orbits that lie wholly above the “equator”, while a
particle sliding freely cannot.

1 Introduction
In a recent article [1] in this Journal, Abramowicz and Szuszkiewicz remarked on an inter-
esting analogy between orbits above the equator of a “wall of death” and orbits near a black
hole; namely that the centrifugal force in both cases appears to point towards rather than
away from the center of an appropriate coordinate system. Here we take a “wall of death”
to be a hollow sphere on the Earth’s surface large enough that a motorcycle can be driven
on the inside of the sphere. The intriguing question is whether there exist stable orbits for
the motorcycle that lie entirely above the equator (horizontal great circle) of the sphere.
In ref. [1] the authors stated that no such orbits are possible, perhaps recalling the well-
known result for a particle sliding freely on the inside of a sphere in a uniform gravitational
field. However, the extra degrees of freedom associated with a rolling wheel (or sphere)
actually do permit such orbits, in apparent defiance of intuition. In particular, the friction
associated with the condition of rolling without slipping can in some circumstances have an
upward component large enough to balance all other downward forces.
In this paper we examine the character of all circular orbits inside a fixed sphere, for both
wheels and spheres that roll without slipping. The rolling constraint is velocity dependent
(non-holonomous), so explicit use of a Lagrangian is not especially effective. Instead we
follow a vectorial approach as advocated by Milne (Chap. 17) [2]. This approach does utilize
the rolling constraint, a careful choice of coordinates, and the elimination of the constraint
force from the equations of motion, all of which are implicit in Lagrange’s method. The
vector approach is, of course, a convenient codification of earlier methods in which individual
components were explicitly written out. Compare with classic works such as those of Lamb
(Chap. 9) [3], Deimel (Chap. 7) [4] and Routh (Chap. 5) [5].
Once the solutions are obtained in sec. 2 for rolling wheels we make a numerical evaluation
of the magnitude of the acceleration in g’s, and of the required coefficient of static friction
on some representative orbits. The resulting parameters are rather extreme, and the circus
name “sphere of death” seems apt.
The stability of steady orbits of wheels is considered in some detail, but completely
general results are not obtained (because the general motion has four degrees of freedom).

1
All vertical orbits are shown to be stable, as are horizontal orbits around the equator of the
sphere. We also find that all horizontal orbits away from the poles are stable in the limit of
small wheels, and conjecture that the a similar condition holds for “death-defying” orbits of
large wheels above the equator of the sphere. In sec. 4 we lend support to this conjecture
by comparing to the related case of a sphere rolling within a sphere for which a complete
stability analysis can be given.
Discussions of wheels and spheres rolling outside a fixed sphere are given in secs. 3 and
5, respectively.

2 Wheel Rolling Inside a Fixed Sphere

2.1 Generalities
We consider a wheel of radius a rolling without slipping on a circular orbit on the inner
surface of a sphere of radius r > a. The analysis is performed in the lab frame, in which the
sphere is fixed. The z-axis is vertical and upwards with origin at the center of the sphere as
shown in Fig. 1. As the wheel rolls on the sphere, the point of contact traces a path that is
an arc of a circle during any short interval. In steady motion the path forms closed circular
orbits which are of primary interest here. We therefore introduce a set of axes (x′ , y ′, z ′ ) that
are related to the circular motion of the point of contact. If the motion is steady, these axes
are fixed in the lab frame.
The normal to the plane of the circular orbit through the center of the sphere (and also
through the center of the circle) is labeled z ′ . The angle between axes z and z ′ is β with
0 ≤ β ≤ π/2. A radius from the center of the sphere to the point of contact of the wheel
sweeps out a cone of angle θ about the z ′ axis, where 0 ≤ θ ≤ π. The azimuthal angle of the
point of contact on this cone is called φ, with φ = 0 defined by the direction of the x′ -axis,
which is along the projection of the z-axis onto the plane of the orbit, as shown in Fig. 2.
Unit vectors are labeled with a superscript ˆ, so that ŷ′ = ẑ′ × x̂′ completes the definition
of the ′ -coordinate system.
For a particle sliding freely, the only stationary orbits have β = 0 (horizontal circles) or
β = π/2 (vertical great circles). For wheels and spheres rolling inside a sphere it turns out
that β = 0 or π/2 also, as we will demonstrate. However, the friction at the point of contact
in the rolling cases permits orbits with a larger range of θ than in the sliding case. If β = 0
or π/2 were accepted as an assumption the derivation could be shortened somewhat.
We also introduce a right-handed coordinate triad of unit vectors (1̂, 2̂, 3̂) related to the
geometry of the wheel. Axis 1̂ lies along the symmetry axis of the wheel as shown shown in
Fig. 1. Axis 3̂ is directed from the center of the wheel to the point of contact of the wheel
with the sphere. The vector from the center of the wheel to the point of contact is then

a = a3̂. (1)

Axis 2̂ = 3̂ × 1̂ lies in the plane of the wheel, and also in the plane of the orbit (the x′ -y ′
plane). The sense of axis 1̂ is chosen so that the component ω1 of the angular velocity vector

2
 Figure 1: A wheel of radius a rolls without slipping on a circular orbit inside
a fixed sphere of radius r. The orbit sweeps out a cone of angle θ about the
z ′ -axis, which axis makes angle β to the vertical. The x′ -axis is orthogonal
to the z ′ -axis in the z-z ′ plane The angle between the plane of the orbit and
diameter of the wheel that includes the point of contact with the sphere is
denoted by α. A right-handed triad of unit vectors, (1̂, 2̂, 3̂), is defined with
1̂ along the axis of the wheel and 3̂ pointing from the center of the wheel to
the point of contact.

~ of the wheel about this axis is positive. Consequently, axis 2̂ points in the direction of the
ω
velocity of the point of contact, and therefore is parallel to the tangent to the orbit.
Except for axis 1̂, these rotating axes are not body axes, but the inertia tensor is diagonal
with respect to them. We write
I11 = 2kma2 , I22 = kma2 = I33 , (2)
which holds for any circularly symmetric disc according to the perpendicular axis theorem;
k = 1/2 for a wheel of radius a with mass m concentrated at the rim, k = 1/4 for a uniform
disc, etc.
The wheel does not necessarily lie in the plane of the orbit. Indeed, it is the freedom
to “bank” the wheel that makes the “death-defying” orbits possible. The diameter of the
wheel through the point of contact (i.e., axis 3̂) makes angle α to the plane of the orbit.
In general, a wheel can have an arbitrary rotation about the 3̂-axis, but the wheel will roll
steadily along a closed circular orbit orbit only if angular velocity component ω3 is such that
the plane of the wheel intersects the plane of the orbit along the tangent to the orbit at the
point of contact. Hence, for steady motion we will be able to deduce a constraint on ω3 .
The case of a rolling sphere is distinguished by the absence of this constraint, as considered
later.

3
 Figure 2: The azimuth of the point of contact of the wheel with the sphere
to the x′ -axis is φ. The unit vector r̂′ is orthogonal to the z ′ -axis and points
towards the center of the wheel (or equivalently, towards the point of contact).
Unit vector 2̂ = 3̂ × 1̂ = ẑ′ × r̂′ .

Since the wheel lies inside the sphere, as shown in Fig. 3, we can readily deduce the
geometric relation that

θ − π + sin−1 (a/r) < α < θ − sin−1 (a/r). (3)

Figure 3: Geometry illustrating the extremes of angle α.

It is useful to introduce r′ = r ′ r̂′ as the perpendicular vector from the z ′ -axis to the center
of the wheel. The magnitude r ′ is given by

r ′ = r sin θ − a cos α, (4)

4
as shown in Fig. 4. The vector ẑ′ × r̂′ is in the direction of motion of the point of contact,
which was defined previously to be direction 2̂. That is, (r̂′ , 2̂, ẑ′ ) form a right-handed unit
triad, which is related to the triad (1̂, 2̂, 3̂) by
ẑ′ = − cos α1̂ − sin α3̂, (5)
and
r̂′ = 2̂ × ẑ′ = − sin α1̂ + cos α3̂, (6)
as can be seen from Fig. 1.
The length r ′ is negative when the center of the wheel is on the opposite side of the z ′ -axis
from the point of contact. This can occur for large enough a/r when the point of contact is
near the z ′ -axis, such as when θ ≈ 0 and α < 0 or θ ≈ π and α > 0.

Figure 4: Geometry illustrating the vector rcm from the center of the sphere
to the center of the wheel, and the distance r ′ = r sin θ − a cos α from the
z ′ -axis to the center of the wheel.

The force of contact of the sphere on the wheel is labeled F. For the wheel to be in contact
with the sphere the force F must have a component towards the center of the sphere, which
will be verified after the motion is obtained.
The equation of motion of the center of mass of the wheel is
d2 rcm
m = F − mgẑ, (7)
dt2
where g is the acceleration due to gravity. The equation of motion for the angular momentum
Lcm about the center of mass is
dLcm
= Ncm = a × F. (8)
dt
We eliminate the unknown force F in eq. (8) via eqs. (1) and (7) to find
1 dLcm d2 rcm
= g 3̂ × ẑ + 3̂ × . (9)
ma dt dt2
5
 The constraint that the wheel rolls without slipping relates the velocity of the center of
mass to the angular velocity vector ω
~ of the wheel. In particular, the velocity vanishes for
that point on the wheel instantaneously in contact with the sphere:

vcontact = vcm + ω
~ × a = 0, (10)

and hence
drcm
vcm = = a3̂ × ~ω . (11)
dt
Multiplying this equation by 3̂, we find
vcm
~ω = −3̂ × + ω3 3̂. (12)
a
Equations (7)-(12) hold whether the rolling object is a wheel or a sphere.
The strategy now is to extract as much information as possible about the angular velocity
~ω before confronting the full equation of motion (9). The angular velocity can also be written
in terms of the unit vector 1̂ along the symmetry axis of the wheel as
d1̂
~ω = ω1 1̂ + 1̂ × . (13)
dt
This follows on writing ~ω = ω1 1̂ + ~ω⊥ , and noting that the rate of change of the body vector
1̂ is just d1̂/dt = ~ω⊥ × 1̂, so ~ω⊥ = 1̂ × d1̂/dt. Using eq. (2), the angular momentum can now
be written as
~ d1̂
Lcm = I~ · ω
~ = 2kma2 ω1 1̂ + kma2 1̂ × . (14)
dt

2.2 Steady Motion in a Circle

To obtain additional relations we restrict our attention to orbits in which the point of contact
of the wheel with the sphere moves in a closed circle. In such cases the center of mass of the
wheel (and also the coordinate triad (1̂, 2̂, 3̂)) has angular velocity φ̇ about the ẑ′ -axis (and
no other component), where the dot means differentiation with respect to time. Thus

vcm = φ̇ẑ′ × r ′ r̂′ = r ′ φ̇2̂. (15)

Equation (12) can now be evaluated, yielding

ω̂ = (r ′ /a)φ̇1̂ + ω3 3̂. (16)

For steady motion there can be no rotation about axis 2̂; angle α is constant. To find ω3 we
now pursue eq. (13).
As argued above, the angular velocity ~γ of the triad (1̂, 2̂, 3̂) is

~γ = φ̇ẑ′ = −φ̇ cos α1̂ − φ̇ sin α3̂, (17)

using eq. (5). Then,

d1̂
= ~γ × 1̂ = −φ̇ sin α2̂ (18)
dt
6
 d2̂
= ~γ × 2̂ = φ̇ sin α1̂ − φ̇ cos α3̂ = −φ̇r̂′ , (19)
dt
and
d3̂
= ~γ × 3̂ = φ̇ cos α2̂. (20)
dt
It immediately follows that
d1̂
1̂ × = −φ̇ sin α3̂. (21)
dt
Comparing with eq (13) we see that ω3 = −φ̇ sin α and hence from eq. (16) we find

~ω = (r ′ /a)φ̇1̂ − φ̇ sin α3̂. (22)

As anticipated, the rolling constraint specifies how ω1 and ω3 are both related to the angular
velocity φ̇ of the wheel about the ẑ′ -axis.
For use in the equation of motion (9) we can now write

~
L = I~ · ~ω = kma2 [2(r ′/a)φ̇1̂ − φ̇ sin α3̂], (23)

and hence,
1 dL
= 2kr ′φ̈1̂ − k φ̇2 sin α(2r ′ + a cos α)2̂ − kaφ̈ sin α3̂, (24)
ma dt
using eqs. (18-20). Also, by differentiating eq. (15) we find

d2 rcm
= r ′ φ̇2 sin α1̂ + r ′φ̈2̂ − r ′ φ̇2 cos α3̂, (25)
dt2
so that
d2 rcm
3̂ × = −r ′ φ̈1̂ + r ′ φ̇2 sin α2̂. (26)
dt2
Combining (9), (24) and (26), the equation of motion reads

d2 rcm 1 dL
gẑ × 3̂ = 3̂ × −
dt 2 ma dt
= −(2k + 1)r ′ φ̈1̂ + [(2k + 1)r ′ + ka cos α]φ̇2 sin α2̂ + kaφ̈ sin α3̂. (27)

To evaluate ẑ × 3̂, we first express ẑ in terms of the triad (r̂′ , 2̂, ẑ′ ), and then transform to
triad (1̂, 2̂, 3̂). When the point of contact of the wheel (and hence the r̂′ -axis) has azimuth
φ relative to the x̂′ axis, the ẑ axis has azimuth −φ relative to the r̂′ axis. Hence,

ẑ = sin β cos φr̂′ − sin β sin φ2̂ + cos βẑ′ (28)

= −(cos α cos β + sin α sin β cos φ)1̂ − sin β sin φ2̂ − (sin α cos β − cos α sin β cos φ)3̂,

using eqs. (5)-(6). Thus,

ẑ × 3̂ = − sin β sin φ1̂ + (cos α cos β + sin α sin β cos φ)2̂. (29)

7
 The 1̂, 2̂ and 3̂ components of the equation of motion are now

(2k + 1)r ′ φ̈ = g sin β sin φ, (30)

[(2k + 1)r ′ + ka cos α]φ̇2 sin α = g(cos α cos β + sin α sin β cos φ), (31)
and
kaφ̈ sin α = 0. (32)
The cone angle θ enters the equations of motion only through r ′ .

2.2.1 Vertical Orbits

From eq. (32) we learn that for circular orbits either sin α = 0 or φ̈ = 0. We first consider
the simpler case that sin α = 0, which implies that the plane of the wheel lies in the plane
of the orbit. For a wheel inside the sphere with sin α = 0, we must have α = 0 to satisfy the
geometric constraint (3). Then eq. (31) can only be satisfied if cos β = 0; i.e., β = π/2 and
the plane of the orbit is vertical. The remaining equation of motion (30) now reads

(2k + 1)r ′ φ̈ = g sin φ, (33)

with r ′ = r sin θ − a > 0, which integrates to

2k + 1 ′2 2
mr (φ̇ − φ̇20 ) = mgr ′ (1 − cos φ), (34)
2

where φ̇0 is the angular velocity at the top of the orbit at which φ = 0. Equation (34)
expresses conservation of energy. The angular velocity ~ω and the angular momentum Lcm
vary in magnitude but are always perpendicular to the plane of the orbit.
The requirement that the wheel stay in contact with the sphere is that the contact force
F have component F⊥ that points to the center of the sphere. On combining eqs. (7), (25),
(29) and (33) we find

2k
F= mg sin φ2̂ + m(g cos φ − r ′ φ̇2 )3̂. (35)
2k + 1
The contact force is in the plane of the orbit, so the resulting torque about the center of
mass of the wheel changes the magnitude but not the direction of the angular momentum.
On the vertical orbits, axis 2̂ is tangent to the sphere, and axis 3̂ makes angle π/2 − θ to
the radius from the center of the sphere to the point of contact. Hence

F⊥ = −F3 sin θ (36)

is positive and the orbit is physical so long as the angular velocity φ̇0 at the peak of the orbit
obeys
g
φ̇20 > ′ , (37)
r
as readily deduced from elementary considerations as well.

8
 The required coefficient µ of static friction is given by µ = Fk /F⊥ where
q
Fk = F32 cos θ2 + F22 (38)

is the component of the contact force parallel to the surface of the sphere. We see that
q
µ = cot θ 1 + (F2 /F3 cos θ)2 , (39)

which must be greater than cot θ, but only much greater if the wheel nearly loses contact
at the top of the orbit. Hence orbits with π/4 <∼ θ ≤ π/2 are consistent with the friction of
typical rubber wheels, namely µ <∼ 1.
Because a wheel experiences friction at the point of contact, vertical orbits are possible
with θ < π/2. This is in contrast to the case of a particle sliding freely on the inside of a
sphere for which the only vertical orbits are great circles (θ = π/2). The only restriction
in the present case is that the wheel fits inside the sphere, i.e., r sin θ > a, and that the
minimum angular velocity satisfy eq. (37).

2.2.2 Horizontal Orbits

The second class of orbits is defined by φ̈ = 0, so that the angular velocity is constant, say
φ̇ = Ω. From eq. (30) we see that sin β = 0 and hence β = 0 for these orbits, which implies
that they are horizontal. Then eq. (31) gives the relation between the required angular
velocity Ω and the geometrical parameters of the orbit:
g cot α g cot α
Ω2 = ′
= , (40)
(2k + 1)r + ka cos α (2k + 1)r sin θ − (k + 1)a cos α

recalling eq. (4). Compare Ex. 3, sec. 244 of Routh [5] or sec. 407 of Milne [2]. There are
no steady horizontal orbits for which α = 0, i.e., for which the wheel lies in the plane of the
orbit. For such an orbit the angular momentum would be constant, but the torque on the
wheel would be nonzero in contradiction.
In the following we will find that horizontal orbits are possible only for 0 < α < π/2.
First, the requirement that Ω2 > 0 for real orbits puts various restrictions on the param-
eters of the problem. We examine these for the four quadrants of angle α.

1. 0 < α < π/2. Then cot α > 0 so we must have

ka cos α
r′ > − . (41)
2k + 1
This is satisfied by all r ′ > 0 and some r ′ < 0. However, for the wheel to fit inside the
sphere with 0 < α < π/2, we can have r ′ < 0 only for θ > π/2 according to eqs. (3)
and (4).

2. π/2 < α < π. Then cos α < 0 and cot α < 0 so the numerator of (40) is negative and
the denominator is positive. Hence Ω is imaginary and there are no steady orbits in
this quadrant.

9
 3. −π < α < −π/2. Then cos α < 0 but cot α > 0 so Ω2 > 0 and r ′ > 0 and eq. (40)
imposes no to restriction. For the wheel to fit inside the sphere with α in this quadrant
we must have θ < π/2.

4. −π/2 < α < 0. Then cot α < 0 so we must have

ka cos α
r′ < − < 0. (42)
2k + 1
For the wheel to be inside the sphere with r ′ < 0 and α in this quadrant we must have
θ < π/2.
To obtain further restrictions on the parameters we examine under what conditions the
wheel remains in contact with the sphere. The contact force F is deduced from eqs. (7), (25)
and (29) to be

F/m = (−g cos α + r ′ Ω2 sin α)1̂ − (g sin α + r ′ Ω2 cos α)3̂. (43)

It is more useful to express F in components along the r̂ and θ̂ axes where r̂ points away
from the center of the sphere and θ̂ points towards increasing θ. The two sets of axes are
related by a rotation about axis 2̂:

1̂ = − cos(θ − α)r̂ + sin(θ − α)θ̂, 3̂ = sin(θ − α)r̂ + cos(θ − α)θ̂, (44)

so that

F/m = −(r ′ Ω2 sin θ − g cos θ)r̂ − (r ′ Ω2 cos θ + g sin θ)θ̂

= −r ′ Ω2 r̂′ + gẑ. (45)

The second form of eq. (45) follows directly from elementary considerations. The inward
component of the contact force, F⊥ = −Fr , is positive and the orbits are physical provided

r ′ Ω2 > g cot θ. (46)

There can be no orbits with r ′ < 0 and θ < π/2, which rules out orbits in quadrant 4 of
α, i.e., for −π/2 > α < 0.
Using eq. (40) for Ω2 in eq. (46) we deduce that contact is maintained for orbits with
r ′ > 0 only if
cot α > [2k + 1 + k(a/r ′ ) cos α] cot θ. (47)
For r ′ < 0 the sign of the inequality is reversed.
In the third quadrant of α we have cos α < 0, so inequality (47) can be rewritten with
the aid of (4) as
!
k
cot α > 1 + 2k − cot θ > cot θ. (48)
1 + r sin θ/a |cos α|
However, in this quadrant inequality (3) tells us
h i
cot α < cot θ + sin−1 (a/r) < cot θ. (49)

10
Hence there can be no steady orbits with −π < α < −π/2.
Thus steady horizontal orbits are possible only for 0 < α < π/2. Furthermore, since the
factor in brackets of inequality (47) is roughly 2 for a wheel, this kinematic constraint is
somewhat stronger than the purely geometric relation (3). However, a large class of orbits
remains with θ < π/2 as well as θ > π/2.
The coefficient of friction µ at the point of contact must be at least Fk /F⊥ where Fk = |Fθ |
from eq. (45). (For θ > π/2 and α near zero the tangential friction Fθ can sometimes point
in the +θ direction.) Hence we need

|r ′ Ω2 cos θ + g sin θ|
µ≥ . (50)
r ′ Ω2 sin θ − g cos θ

The acceleration of the center of mass of the wheel is r ′ Ω2 , so according to eq. (40) the
corresponding number of g’s is
cot α
. (51)
2k + 1 + k(a/r ′ ) cos α

Table 1 lists parameters of several horizontal orbits for a sphere of size as might be found
in a motorcycle circus. The coefficient of friction of rubber tires is of order one, so orbits
more than a few degrees above the equator involve very strong accelerations. The head of
the motorcycle rider is closer to the vertical axis of the sphere than is the center of the wheel,
so the number of g’s experienced by the rider is somewhat less than that given in the Table.
Figure 5 illustrates the allowed values of the tilt angle α as a function of the angle θ of
the plane of the orbit, for a/r = 0.1 as in Table 1.
90
Excluded
Region Allowed Region

Ω2=0 Ω2=0
α (deg.)

–
45 α = θ – sin 1 ( a / r )
r 'Ω 2 = g cotθ
–
α = θ + sin 1 ( a / r ) – 180 o

Allowed Region

0
0 45 90 135 180
θ (deg.)

Figure 5: The allowed values of the tilt angle α as a function of the angle
θ of horizontal orbits for a/r = 0.1. The allowed region is bounded by three
curves, derived from expressions (3), (40) and (46).

From eq. (40) we see that α = π/2, Ω = 0 is a candidate “orbit” in the lower hemisphere.
On such an “orbit” the wheel is standing vertically at rest, and is not stable against falling
over. We infer that stability will only occur for Ω greater than some minimum value not
revealed by the analysis thus far.

11
 Table 1: Parameters for horizontal circular orbits of a wheel of radius 0.3
m rolling inside a sphere of radius 3.0 m. The wheel has coefficient k = 1/2
pertaining to its moment of inertia. The polar angle of the orbit is θ, so orbits
above the equator of the sphere have θ < 90◦ . The plane of the wheel makes
angle α to the horizontal. The minimum coefficient of friction required to
support the motion is µ. The magnitude of the horizontal acceleration of the
center of mass is reported as the No. of g’s.

θ α µ vcm No. of g’s

(deg.) (deg.) (m/s)

15 5 16.1 4.8 48
30 5 2.82 8.0 53
45 10 2.15 7.0 27
60 10 1.19 7.9 27
60 25 3.45 4.9 10
75 15 0.96 6.8 18
75 30 2.13 4.7 8
90 25 0.96 5.3 10
90 45 2.04 3.7 5
135 60 0.56 2.3 3

2.3 Stability Analysis

A completely general analysis of the stability of the steady circular orbits found above appears
to be very difficult. We give a fairly general analysis for vertical orbits, but for horizontal
orbits we obtain results only for orbits with θ = π/2, i.e., orbits about the equator of the
sphere, and for orbits of “small” wheels.
We follow the approach of sec. 405 of Milne [2] where it was shown how the steady motion
of a disk rolling in a straight line on a horizontal plane is stable if the angular velocity is
great enough. It was also shown that the small oscillatory departures from steady motion
lead to an oscillatory path of the point of contact of the wheel with the plane. Hence in the
present case we must consider perturbations that carry the wheel away from the plane of
the steady orbit.
The difficulty is that there are in general four degrees of freedom for departures from
steady motion: the axis of the wheel can be perturbed in two directions and the angular

12
velocity φ̇ can be perturbed as well as the angle θ to the point of contact. However, the
procedure to eliminate the unknown force of contact from the six equations of motion of a
rigid body leaves only three equations of motion. We will obtain solutions to the perturbed
equations of motions only in special cases where there are in effect just two or three degrees
of freedom. A more general analysis might be possible using the contact force found in
steady motion as a first approximation to the contact force in perturbed motion, but we do
not pursue this here.
A wheel rolling with a steady circular orbit on a plane can suffer only three types of
perturbations and the results of an analysis are reported in Ex. 3, sec. 244 of Routh [5]. For
a sphere rolling within a fixed sphere the direction of what we call axis 3 always points to the
center of the fixed sphere so there are only two perturbations to consider and the solution is
relatively straightforward, as reviewed in sec. 4 below. The stability of horizontal orbits of
rolling spheres lends confidence that stable orbits also exist for wheels.

2.3.1 Vertical Orbits

We define the (x′ , y ′, z ′ ) coordinate system to have the x′ -axis vertical: x̂′ = ẑ. In steady
motion we have

α = 0, 1̂ = −ẑ′ , and 3̂ = r̂′ = x̂′ cos φ + ŷ′ sin φ, (52)

where φ is the azimuth of the center of the wheel from the x̂′ -axis. Thus φ = 0 at the top of
the orbit. To discuss departures from steady motion in which the 1̂-axis is no longer parallel
to the ẑ′ -axis, it is useful to have a unit triad (r̂′ , ŝ′ , ẑ′ ) defined by eq. (52) and

ŝ′ = ẑ′ × r̂′ = −x̂′ sin φ + ŷ′ cos φ, (53)

with φ defined as before. See Fig. 6. The surface of the sphere at the point of contact is
parallel to the s′ -z ′ plane. Axes r̂′ and ẑ′ rotate about the z ′ -axis with angular velocity φ̇,
so that
dr̂′ dŝ′
= φ̇ŝ′ , and = −φ̇r̂′ . (54)
dt dt
The perturbed 1̂-axis can then be written

1̂ = ǫr r̂′ + ǫs ŝ′ − ẑ′ , with |ǫr | , |ǫs | ≪ 1, (55)

where throughout the stability analysis we ignore second-order terms. Writing

3̂ = r̂′ + δs ŝ′ + δz ẑ′ , with |δs | , |δz | ≪ 1, (56)

the condition 1̂ · 3̂ = 0 requires that δz = ǫr . Then to first order,

2̂ = 3̂ × 1̂ = −δs r̂′ + ŝ′ + ǫs ẑ′ . (57)

We expect that vector 2̂ will remain parallel to the surface of the sphere even for large
departure from steady motion, so 2̂ must remain in the s′ -z ′ plane. Hence, δs = 0, and

3̂ = r̂′ + ǫr ẑ′ . (58)

13
 Figure 6: For vertical orbits the x′ -axis is identical with the z axis. The axis
ŝ′ = ẑ′ × r̂′ is in the direction of the unperturbed 2̂-axis.

Also, we can identify α as the tilt angle of the 3̂-axis to the r ′ -s′ plane, so that
α = ǫr . (59)
The analysis proceeds along the lines of sec. 2.1 except that now we express all vectors
in terms of the triad (r̂′ , ŝ′ , ẑ′ ). To the first approximation the angular velocity of the wheel
about the 1̂-axis is still given by ω1 = (r ′ /a)φ̇. From eqs. (54) and (55) we find
d1̂
= (ǫ̇r − ǫs φ̇)r̂′ + (ǫr φ̇ + ǫ̇s )ŝ′ , (60)
dt
d1̂
1̂ × = (ǫr φ̇ + ǫ̇s )r̂′ − (ǫ̇r − ǫs φ̇)ŝ′ , (61)
dt
so that eq. (13) yields
d1̂
~ω = ω1 1̂ + 1̂ × = [(1 + r ′ /a)ǫr φ̇ + ǫ̇s ]r̂′ − [ǫ̇r − (1 + r ′ /a)ǫs φ̇]ŝ′ − (r ′ /a)φ̇ẑ′ . (62)
dt
Then eq. (14) tells us
L d1̂
= 2kaω1 1̂ + ka1̂ × = k[(2r ′ + a)ǫr φ̇ + aǫ̇s ]r̂′ − k[aǫ̇r − (2r ′ + a)ǫs φ̇]ŝ′ − 2kr ′ φ̇ẑ′ , (63)
ma dt
so that to first order of smallness
1 dL
= k[2(r ′ + a)ǫ̇r φ̇ + (2r ′ + a)(ǫr φ̈ − ǫs φ̇2 ) + aǫ̈s ]r̂′
ma dt
−k[aǫ̈r − (2r ′ + a)(ǫr φ̇2 − ǫs φ̈) − 2(r ′ + a)ǫ̇s φ̇]ŝ′
−2k(r ′ φ̈ + ṙ ′ φ̇)ẑ′ . (64)

14
In this we have noted from eq. (4) that ṙ ′ = r θ̇ sin θ to first order, and that θ̇ is small. Next,
drcm
= a3̂ × ~ω ≈ ar̂′ × ~ω = r ′ φ̇ŝ′ − [aǫ̇r − (r ′ + a)ǫs φ̇]ẑ′ . (65)
dt
Then to first order,
d2 rcm
= −r ′ φ̇2 r̂′ + (r ′ φ̈ + ṙ ′ φ̇)ŝ′ − [aǫ̈r − (r ′ + a)(ǫ̇s φ̇ + ǫs φ̈)]ẑ′ , (66)
dt2
so that
d2 rcm ′ ′ d2 rcm
3̂ × = (r̂ + ǫr ẑ ) × =
dt2 dt2
−r ′ ǫr φ̈r̂′ + [aǫ̈r − r ′ ǫr φ̇2 − (r ′ + a)(ǫ̇s φ̇ + ǫs φ̈)]ŝ′ + (r ′ φ̈ + ṙ ′ φ̇)ẑ′ (67)

Also,
3̂ × ẑ = (r̂′ + ǫr ẑ′ ) × (cos φr̂′ − sin φŝ′ ) = ǫr sin φr̂′ + ǫs cos φŝ′ − sin φẑ′ . (68)
The r ′ , s′ and z ′ components of the equation of motion (9) are then

0 = [(2k + 1)r ′ + ka]ǫr φ̈ + 2k(r ′ + a)ǫ̇r φ̇ − gǫr sin φ − k(2r ′ + a)ǫs φ̇2 + kaǫ̈s , (69)

0 = [(2k + 1)r + ka]ǫr φ̇2 − (k + 1)aǫ̈r − gǫr cos φ

+(2k + 1)(r ′ + a)ǫ̇s φ̇ + [(2k + 1)r ′ + (k + 1)a]ǫs φ̈, (70)

0 = (2k + 1)(r ′ φ̈ + ṙ ′ φ̇) − g sin φ. (71)

If the perturbations ǫr , ǫs and ṙ ′ are set to zero eqs. (69) and (70) become trivial while
eq. (71) becomes the steady equation of motion (33).
The general difficulty with this analysis is that there are only three equations, (69-71),
while there are four perturbations, ǫr , ǫs , φ̈ and θ̇. The perturbation θ̇ appears only in
eq. (71) via ṙ ′ ; its effect on r ′ leads only to second-order terms in eqs. (69-70). If we could
neglect the terms in φ̈ in eqs. (69-70) then these two equations would describe only the
perturbations ǫr and ǫs to first order and a solution could be completed.
Therefore we restrict our attention to the top of the orbit, φ = 0, where eq. (71) tells
us that φ̈ = 0 to leading order. The angular velocity φ̇0 at this point is a minimum so the
gyroscopic stability of the wheel is the least here. Hence if the orbit is stable at φ = 0 it will
be stable at all φ.
The forms of eqs. (69) and (70) for φ = 0 indicate that if ǫr and ǫs are oscillatory then
they are 90◦ out of phase. Therefore we seek solutions

ǫr = ǫr cos ωt, ǫs = ǫs sin ωt, (72)

where ω now represents the oscillation frequency. The coupled equations of motion then
yield the simultaneous linear equations

2k(r ′ + a)φ̇0 ωǫr + [kaω 2 + k(2r ′ + a)φ̇20 ]ǫs = 0

{(k + 1)aω 2 + [(2k + 1)r ′ + ka]φ̇20 − g}ǫr + (2k + 1)(r ′ + a)φ̇0 ωǫs = 0. (73)

15
These equations are consistent only if the determinant of the coefficient matrix vanishes,
which leads to the quadratic equation
Aω 4 − Bω 2 − C = 0, (74)
with solutions √
2 B± B 2 + 4AC
ω = , (75)
2A
where
A = k(k + 1)a2 , (76)
′2 2 ′
B = kag + k[(2k + 1)(2r + a ) + (4k + 1)ar ]φ̇20 , (77)
and  
C = k(2r ′ + a) [(2k + 1)r ′ + ka]φ̇20 − g φ̇20 . (78)
Since A and B are positive there are real, positive roots whenever B 2 + 4AC is positive,
i.e., for C > −B 2 /4A. In particular, this is satisfied for positive C, or equivalently for
g
φ̇20 > . (79)
(2k + 1)r ′ + ka
However, this is less restrictive than the elementary result (37) that the wheel stay in contact
with the sphere! All vertical orbits for which the wheel remains in contact with the sphere
are stable against small perturbations. q
The stability analysis yields the formal result that if (φ, φ̇) = (0, 0) then ω = g/(k + 1)a.
We recognize this as the frequency of oscillation of a simple pendulum formed by suspending
the wheel from a point on its rim, the motion being perpendicular to the plane of the wheel.

2.3.2 Horizontal Orbits

We expect the stability analysis of horizontal orbits to be nontrivial since we have identified
steady orbits that are “obviously” unstable.
The spirit of the analysis has been set forth in the preceding sections. For horizontal
orbits the (x′ , y ′, z ′ ) coordinate system can be taken as identical with the (x, y, z) system, so
we drop symbol ′ in this section. We introduce a triad (r̂, ŝ, ẑ) with r̂ being the perpendicular
unit vector from the z-axis toward the center of the wheel. Then ŝ points in the direction of
the motion of the center of the wheel in case of steady motion.
It is also useful to introduce a unit triad that points along the (1̂, 2̂, 3̂) axes for steady
motion. The ŝ axis already points along the 2̂ axis for steady motion, so we only need define
t̂ as being along the direction of 3̂, and û as being along the direction of 1̂ for steady motion,
as shown in Fig. 7. Then, (ŝ, t̂, û) form a right-handed unit triad. The vertical, ẑ, is then
related by
ẑ = − sin α0 t̂ − cos α0 û, (80)
where α0 is the angle of inclination of the wheel to the horizontal in steady motion. The
triad (ŝ, t̂, û) rotates about the ẑ-axis with angular velocity φ̇, so that
dŝ
= φ̇ẑ × ŝ = −φ̇ cos α0 t̂ + φ̇ sin α0 û, (81)
dt
16
 dt̂
= φ̇ cos α0 ŝ, (82)
dt
and
dû
= −φ̇ sin α0 ŝ. (83)
dt

Figure 7: For horizontal orbits of a wheel rolling inside a sphere the (x, y, z)
axes are identical with the (x′ , y ′, z ′ ) axes. The r̂-ŝ plane is horizontal. The
axes û, ŝ and t̂ are along the unperturbed directions of the 1̂, 2̂ and 3̂ axes,
respectively. Axes t̂ and û lie in the vertical plane r̂-ẑ.

We now consider small departures from steady motion. The 1̂-axis deviates slightly from
the û-axis according to
1̂ = ǫs ŝ + ǫt t̂ + û, |ǫs | , |ǫt | ≪ 1. (84)
The 3̂-axis departs slightly from the t-axis, but to the first approximation it remains in a
vertical plane, i.e., the t-u plane. Then we have

2̂ = ŝ − ǫs û, and 3̂ = t̂ − ǫt û. (85)

With the above definitions the signs of angles α and ǫt are opposite:

∆α = −ǫt , α̇ = −ǫ̇t . (86)

To first approximation the component ω1 of the angular velocity of the wheel about its
axis remains ω1 = (r ′ /a)φ̇. Then

d1̂
= (−φ̇ sin α0 + ǫ̇s − ǫt φ̇ cos α0 )ŝ − (ǫs φ̇ cos α0 − ǫ̇t )t̂ + ǫs φ̇ sin α0 û, (87)
dt
d1̂
1̂ × = (ǫs φ̇ cos α0 − ǫ̇t )ŝ − (φ̇ sin α0 + ǫ̇s − ǫt φ̇ cos α0 )t̂ + ǫt φ̇ sin α0 û, (88)
dt

17
so that
d1̂
~ω = ω1 1̂ + 1̂ ×
dt
′
= [(r /a + cos α0 )ǫs φ̇ − ǫ̇t ]ŝ
−[φ̇ sin α0 − ǫ̇s − (r ′ /a + cos α0 )ǫt φ̇]t̂
+(r ′/a + ǫt sin α0 )φ̇û, (89)

and
L d1̂
= 2kaω1 1̂ + ka1̂ ×
ma dt
′
= k[(2r + a cos α0 )ǫs φ̇ − aǫ̇t ]ŝ
−k[aφ̇ sin α0 − aǫ̇s − (2r ′ + a cos α0 )ǫt φ̇]t̂
+k(2r ′ + aǫt sin α0 )φ̇û. (90)

Then to the first approximation

1 dL
= −k[(2r ′ + a cos α0 )φ̇2 sin α0 − 2(r ′ + a cos α0 )ǫ̇s φ̇)
ma dt
− (2r ′ cos α0 + a cos 2α0 )ǫt φ̇2 + aǫ̈t ]ŝ
−k[aφ̈ sin α0 + (2r ′ + a cos α0 )ǫs φ̇2 cos α0 − aǫ̈s − 2(r ′ + a cos α0 )ǫ̇t φ̇)]t̂
+k[2r ′ φ̈ + 2ṙ ′ φ̇ + (2r ′ + a cos α0 )ǫs φ̇2 sin α0 ]û. (91)

Unlike the case of vertical orbits, for horizontal orbits the factor φ̈ has no zeroeth-order
component and we neglect terms like ǫφ̈.
Similarly
drcm
= a3̂ × ~ω = a(t̂ − ǫt û) × ~ω
dt
= r ′ φ̇ŝ − [(r ′ + a cos α0 )ǫs φ̇ − aǫ̇t ]û, (92)

d2 rcm
= [r ′ φ̈ + ṙ ′ φ̇ + (r ′ + a cos α0 )ǫs φ̇2 sin α0 − aǫ̇t φ̇ sin α0 ]ŝ
dt2
−r ′ φ̇2 cos α0 t̂ + [r ′ φ̇2 sin α0 − (r ′ + a cos α0 )ǫ̇s φ̇ + aǫ̈t ]û, (93)

and
d2 rcm d2 rcm
3̂ × = ( t̂ − ǫt û) ×
dt2 dt2
= [r φ̇ sin α0 − (r ′ + a cos α0 )ǫ̇s φ̇ − r ′ ǫt φ̇2 cos α0 + aǫ̈t ]ŝ
′ 2

−[r ′ φ̈ + ṙ ′ φ̇ + (r ′ + a cos α)ǫs φ̇2 sin α0 − aǫ̇t φ̇ sin α0 ]û. (94)

We also need

3̂ × ẑ = (t̂ − ǫt û) × (− sin α0 t̂ − cos α0 û) = −(cos α0 + ǫt sin α0 )ŝ. (95)

18
 The s, and t and u components of the equation of motion (9) are

0 = [(2k + 1)r ′ + ka cos α0 ]φ̇2 sin α0 − g cos α0 − (2k + 1)(r ′ + a cos α0 ]ǫ̇s φ̇
−[(2k + 1)r ′ cos α0 + ka cos 2α0 ]ǫt φ̇2 − gǫt sin α0 + (k + 1)aǫ̈t , (96)

0 = kaφ̈ sin α0 + k(2r ′ + a cos α0 )ǫs φ̇2 sin α0 − kaǫ̈s − 2k(r ′ + a cos α0 )ǫ̇t φ̇, (97)
and

0 = (2k + 1)(r ′ φ̈ + ṙ ′ φ̇) + [(2k + 1)r ′ + (k + 1)a cos α0 ]ǫs φ̇2 sin α0 − aǫ̇t φ̇ sin α0 . (98)

The leading terms of these three equations are just eqs. (30)-(32) for β = 0. Therefore we
can write φ̇ = Ω + δ̇ where Ω is the angular velocity of the steady horizontal orbit and δ is
a small correction.
Although the derivative of r ′ ,

ṙ ′ = r θ̇ cos θ0 + aα̇ sin α0 = r θ̇ cos θ0 − aǫ̇t sin α0 , (99)

appears only in eq. (98), in general the perturbation θ̇ is not decoupled from ǫs and ǫt as
was the case for vertical orbits. Thus far, we have found a way to proceed only in somewhat
special cases in which the θ perturbation can be ignored, as described in secs. 2.3.3 and 2.3.4.

2.3.3 Orbits Near the Equator

It appears possible to carry the analysis forward for the special case θ0 = π/2, the orbit on
the equator of the sphere. This case is, however, of interest.
Assuming θ0 = π/2 the equations of motion (96-98) then provide three relations among
the three perturbations δ, ǫs and ǫt . For this we consider only the first-order terms, noting
that φ̇2 ≈ Ω2 + 2Ωδ̇ and

r ′ = r sin θ − a cos α ≈ r0′ + r∆θ cos θ0 + a∆α sin α0 = r0′ − aǫt sin α0 , (100)

where r0 = r − a cos α0 for θ0 = π/2, recalling eq. (86). Also, from the form of eqs. (96-98)
we infer that if the perturbations are oscillatory then δ and ǫs have the same phase which is
90◦ from that of ǫt . Therefore we seek solutions of the form

δ = δ sin ωt, ǫs = ǫs sin ωt, and ǫt = ǫt cos ωt, (101)

where ω is the frequency of oscillation. The first-order terms of the differential equations
(96-98) then yield the algebraic relations

0 = 2Ω sin α0 [(2k + 1)r0′ + ka cos α0 ]ωδ − Ω(2k + 1)(r0′ + a cos α0 )ωǫs

−{Ω2 [(2k + 1)r0′ cos α0 + (k + sin2 α0 )a] − g sin α0 + (k + 1)aω 2 }ǫt ,
0 = −ka sin α0 ω 2 δ + [kΩ2 cos α0 (2r0′ + a cos α0 ) + kaω 2 ]ǫs + 2kΩ(r0′ + a cos α0 )ωǫt , (102)
0 = −(2k + 1)r0′ ω 2 δ + Ω2 sin α0 [(2k + 1)r0′ + (k + 1)a cos α0 ]ǫs + 2(k + 1)Ω sin α0 aωǫt .

19
These equations have the form

A11 ωδ + A12 ωǫs + (A13 + B13 ω 2 )ǫt = 0

A21 ω 2 δ + (A22 + B22 ω 2 )ǫs + A23 ωǫt = 0 (103)
A31 ω 2 δ + A32 ǫs + A33 ωǫt = 0

To have consistency the determinant of the coefficient matrix must vanish, which leads
quickly to the quadratic equation

Aω 4 − Bω 2 − C = 0, (104)

where
A = B13 B22 A31 , (105)
B = A11 B22 A33 + A12 A23 A31 + B13 A21 A32 − A13 B22 A31 − B13 A22 A31 − A12 A21 A33 , (106)
and
C = A11 A22 A33 + A13 A21 A32 − A13 A22 A31 − A11 A23 A32 . (107)
From numerical evaluation it appears that A, B and C are all positive for angular ve-
locities Ω that obey eq. (40). That is, all steady orbits at the equator of the sphere are
stable. There is both a fast and slow oscillation about steady motion for these orbits, an
effect familiar from nutations of a symmetric top.

2.3.4 Small Wheel Inside a Large Sphere

The analysis can also be carried further in the approximation that the radius a of the wheel
is much less than the radius r of the fixed sphere. In this case the perturbation in angle θ
of the orbit is of higher order than the perturbations in azimuth φ and in the angles ǫs and
ǫt related to the axes of the wheel. A solution describing the three first-order perturbations
can then be obtained.
For the greatest simplification we also require that

a ≪ r0′ ≈ r sin θ0 . (108)

Thus we restrict our attention to orbits significantly different from the special cases of motion
near the poles of the fixed sphere.
In the present approximation the first-order terms of the perturbed equations of motion
(96-98) are !
g sin α 0
2Ωδ̇ sin α0 = Ωǫ̇s + Ω2 cos α0 + ǫt , (109)
(2k + 1)r0′
ǫ̇t
ǫs = , (110)
Ω sin α0
and
δ̈ = −ǫs Ω2 sin α0 . (111)

20
Inserting (110) into (111) we can integrate the latter to find

δ̇ = −Ωǫt . (112)

Using this and the derivative of (110) in (109) we find that ǫt obeys

g sin2 α0
" #
2
ǫ̈t + Ω sin α0 (cos α0 + 2 sin α0 ) + ǫt = 0. (113)
(2k + 1)r0′

The the frequency ω of the perturbations is given by

g sin2 α0
ω 2 = Ω2 sin α0 (cos α0 + 2 sin α0 ) + = Ω2 tan α0 (1 + sin 2α0 ), (114)
(2k + 1)r0′

using eqs. (40) and (108).

Thus all orbits for small wheels are stable if condition (108) holds. We conjecture that
orbits for large wheels are also stable if (108) is satisfied.
For steady orbits that lie very near the poles, i.e., those that have r0′ <
∼ a, we conjecture
that the motion is stable only for Ω greater than some minimum value. For a wheel spinning
about its axis on a horizontal plane the stability condition is
g
Ω2 > . (115)
(2k + 1)a

See, for example, sec. 55 of Deimel [4]. However, we have been unable to deduce the gener-
alization of this constraint to include the dependence on r and θ0 for small r sin θ0 .

3 Wheel Rolling Outside a Fixed Sphere

Equations (1)-(32) hold for a wheel rolling outside a sphere as well as inside when the
geometric relation (3) is rewritten as

θ < α < π + θ. (116)

We expect no vertical orbits as the wheel will lose contact with the sphere at some point.
To verify this, note that the condition sin α = 0 (from eq. (32)) implies that α = π when
the wheel is outside the sphere. Then eqs. (34-36) indicate, for example, that if the wheel
starts from rest at the top of the sphere it loses contact with the sphere when
2
cos φ = . (117)
3 + 2k
The result for a particle sliding on a sphere (k = 0) is well known.
For horizontal orbits, eqs. (40-45) are still valid, but the condition that friction have an
outward component is now
r ′Ω2 < g cot θ, (118)

21
and hence
cot α < (2k + 1 + k(a/r ′ ) cos α) cot θ. (119)
Equation (40) can be satisfied for α < π/2 so long at the radius of the wheel is small enough
that (2k + 1)r ′ + ka cos α is positive. We must have θ < π/2 to have α < π/2 since α > θ, so
horizontal orbits exist on the upper hemisphere. A particular solution is α = π/2 for which
Ω = 0; this is clearly unstable.
There is a class of orbits with θ < π/2 and α very near π + θ that satisfy both eqs. (40)
and (119). These also appear to be unstable.
The stability analysis of the preceding section holds formally for wheels outside spheres,
but the restriction there to the case of θ = 90◦ provides no insight into the present case.

4 Sphere Rolling Inside a Fixed Sphere

The case of a sphere rolling on horizontal orbits inside a fixed sphere has been treated by
Milne [2]. For completeness, we give an analysis for orbits of arbitrary inclination to compare
and contrast with the case of a wheel.
Again the axis normal to the orbit is called ẑ′ , which makes angle β to the vertical ẑ. The
polar angle of the orbit about ẑ′ is θ and φ is the azimuth of the point of contact between
the two spheres. The radius of the fixed sphere is r.
The diameter of the rolling sphere that passes through the point of contact must always
be normal to the fixed sphere. That is, the “bank” angle of the rolling sphere is always
θ − π/2 with respect to the plane of the orbit.
The rolling sphere has radius a, mass m and moment of inertia kma2 about any diameter.
The angular momentum is, of course,

Lcm = kma2 ~ω , (120)

where ~ω is the angular velocity of the rolling sphere.

We again introduce a right-handed triad of unit vectors (1̂, 2̂, 3̂) centered on the rolling
sphere. For consistency with the notation used for the wheel, axis 3̂ is directed towards the
point of contact, axis 2̂ is parallel to the plane of the orbit, and axis 1̂ is in the 3̂-ẑ′ plane,
as shown in Fig. 8. In general, none of these vectors are body vectors for the rolling sphere.
The center of mass of the rolling sphere lies on the line joining the center of the fixed sphere
to the point of contact, and so

rcm = (r − a)3̂ ≡ r ′ 3̂, (121)

Equations (7-12) that govern the motion and describe the rolling constraint hold for the
sphere as well as the wheel. Using eqs. (120) and (121) we can write eq. (9) as

d~ω ′ d2 3̂
ka = g 3̂ × ẑ + r 3̂ × 2 . (122)
dt dt
We seek an additional expression for the angular velocity ~ω of the rolling sphere, but we
cannot use eq. (13) since we have not identified a body axis in the sphere. However, with

22
 Figure 8: Geometry illustrating the case of a sphere rolling without slipping
on a circular orbit perpendicular to the ẑ′ -axis inside a fixed sphere. The 3̂-
axis is along the line of centers of the two spheres, and passes through the
point of contact. The 2̂-axis lies in the plane of the orbit along the direction
of motion of the center of the rolling sphere, and axis 1̂ = 2̂ × 3̂ is in the 3̂-ẑ′
plane.

eq. (121) the rolling constraint (12) can be written

r′ d3̂
~ω = − 3̂ × + ω3 3̂. (123)
a dt
We can now see that ω3 = ~ω · 3̂ is a constant by noting that 3̂ · d~ω /dt = 0 from eq. (122),
and also ~ω · d3̂/dt = 0 from eq. (123). The freedom to chose the constant angular velocity
ω3 for a rolling sphere permits stable orbits above the equator of the fixed sphere, just as
the freedom to adjust the bank angle α allows such orbits for a wheel.
Taking the derivative of eq. (123) we find

d~ω r′ d2 3̂ d3̂
= − 3̂ × 2 + ω3 , (124)
dt a dt dt
so the equation of motion (122) can be written

′ d2 3̂ d3̂
(k + 1)r 3̂ × 2 − kaω3 = gẑ × 3̂. (125)
dt dt
Milne notes that this equation is identical to that for a symmetric top with one point fixed
[2], and so the usual extensive analysis of nutations about the stable orbits follows if desired.

23
 We again restrict ourselves to circular orbits, for which the angular velocity of the center
of mass, and of 1̂, 2̂ and 3̂ is φ̇ẑ′ where the z ′ -axis is fixed. Then with

ẑ′ = − sin θ1̂ + cos θ3̂, (126)

we have
d3̂
= φ̇ẑ′ × 3̂ = φ̇ sin θ2̂, (127)
dt
d2 3̂
= φ̇2 sin θẑ′ × 2̂ + φ̈ sin θ2̂ = −φ̇2 sin θ cos θ1̂ + φ̈ sin θ2̂ + φ̇2 sin2 θ3̂, (128)
dt 2

and hence,
d2 3̂
3̂ × 2 = −φ̈ sin θ1̂ − φ̇2 sin θ cos θ2̂. (129)
dt
With these the equation of motion (125) reads

(k + 1)r ′φ̈ sin θ1̂ + [(k + 1)r ′φ̇2 cos θ + kaω3 φ̇] sin θ2̂ = −gẑ × 3̂. (130)

We can use eq. (29) for ẑ × 3̂ if we substitute α = θ − π/2 for the rolling sphere:

ẑ × 3̂ = − sin β sin φ1̂ + (sin θ cos β − cos θ sin β cos φ)2̂. (131)

The components of the equation of motion are then

(k + 1)r ′φ̈ sin θ = sin β sin φ, (132)

[(k + 1)r ′ φ̇2 cos θ + kaω3 φ̇] sin θ = g cos θ sin β cos φ − g sin θ cos β. (133)
The two equations of motion are not consistent in general. To see this, take the derivative
of eq. (133) and substitute φ̈ from eq. (132):

kaω3 sin β sin φ = −3(k + 1)r ′φ̇ cos θ sin β sin φ. (134)

While this is certainly true for β = 0 (horizontal orbits), for nonzero β we must have φ̇ cos θ
constant since ω3 is constant. Equation (134) is satisfied for θ = π/2 (great circles), but for
arbitrary θ we would need φ̇ constant which is inconsistent with eq. (132). Further, on a
great circle eq. (133) becomes kaω3 φ̇ = −g cos β. This is inconsistent with eq. (132) unless
β = π/2 (vertical great circles) and ω3 = 0.
In summary, the only possible closed orbits for a sphere rolling within a fixed sphere are
horizontal circles and vertical great circles.
We remark further only on the horizontal orbits. For these φ̇ ≡ Ω is constant according
to eq. (132). Equation (133) then yields a quadratic equation for Ω:

(k + 1)r ′ Ω2 cos θ + kaω3 Ω + g = 0, (135)

so that there are orbits with real values of Ω provided

(kaω3 )2 ≥ 4(k + 1)gr ′ cos θ. (136)

24
This is satisfied for orbits below the equator (θ > π/2) for any value of the “spin” ω3 of the
sphere (including zero), but places a lower limit on |ω3 | for orbits above the equator. For the
orbit on the equator we must have Ω = −g/(kaω3 ) so a nonzero ω3 is required here as well.
The contact force F is given by

F/m = (g + r ′ Ω2 cos θ) sin θ1̂ − (r ′ Ω2 sin2 θ − g cos θ)3̂, (137)

using eqs. (7) and (134). For the rolling sphere to remain in contact with the fixed sphere
there must be a positive component of F pointing toward the center of the fixed sphere.
Since axis 3̂ is radial outward from the fixed sphere, we require that F3 be negative, and
hence
r ′ Ω2 sin2 θ > g cos θ. (138)
This is always satisfied for orbits below the equator. For orbits well above the equator this
requires a larger value of |ω3 | than does eq. (136). To see this, suppose ω3 is exactly at the
minimum value allowed by eq. (136), which implies that Ω = −kaω3 /(2(k + 1)r ′ cos θ). Then
eq. (138) requires that tan2 θ > k + 1. So for k = 2/5 and at angles θ < 50◦ larger values of
|ω3 | are needed to satisfy eq. (136) than to satisfy eq. (136). However, there are horizontal
orbits at any θ > 0 for |ω3 | large enough.

5 Sphere Rolling Outside a Fixed Sphere

This case has also been treated by Milne [2]. A popular example is spinning a basketball on
one’s fingertip.
Equations eq. (135) and (136) hold with the substitution that r ′ = r + a. The condition
on the contact force becomes
r ′ Ω2 sin2 θ < g cos θ, (139)
which can only be satisfied for θ < π/2. While eq. (136) requires a large spin |ω3 |, if it is
too large eq. (139) can no longer be satisfied in view of the relation (135). For any case in
which the orbit exists a perturbation analysis shows that the motion is stable against small
nutations [2].

6 References
[1] M.A. Abramowicz and E. Szuszkiewicz, The Wall of Death, Am. J. Phys. 61 (1993)
982-991.

[2] E.A. Milne, Vectorial Mechanics, Interscience Publishers (New York, 1948).

[3] H. Lamb, Higher Mechanics, Cambridge U. Press (Cambridge, 1920).

[4] R.F. Deimel, Mechanics of the Gyroscope, Macmillian (1929); reprinted by Dover Pub-
lications (New York, 1950).

25
[5] E.J. Routh, The Advanced Part of a Treatise on the Dynamics of a System of Rigid
Bodies, 6th ed., Macmillan (London, 1905); reprinted by Dover Publications (New York,
1955).

26