Cylinder Rolling on Another Rolling Cylinder

Kirk T. McDonald
Joseph Henry Laboratories, Princeton University, Princeton, NJ 08544
(October 2, 2014; updated March 31, 2016)

1 Problem
Discuss the motion of a cylinder that rolls without slipping on another cylinder, when the
latter rolls without slipping on a horizontal plane. The cylinders have axial moments of
inertia Ii = ki mi ri2 where mi are the masses and ri are the radii of rolling.1

2 Solution
This problem was suggested by Bradley Klee. For the related case of one cylinder rolling
inside another, see [1].
When one cylinder is directly above the other, we dene the line of contact of the lower
cylinder, 1, with the horizontal plane to be the z-axis, at x = y = 0. Then, the condition of
rolling without slipping for the lower cylinder is that when it has rolled (positive) distance
x1, the initial line of contact has rotated through angle 1 = x1/r1 , clockwise with respect
to the vertical, as shown in the gure below. This rolling constraint can be written as
x 1 = r1 1 . (1)

Meanwhile, if the upper cylinder, 2, rolls such that the line of centers (in the x-y plane)
makes angle (positive clockwise) to the vertical, then the initial point of contact of the
upper cylinder has rotated through angle 2, measured counterclockwise from the line of
centers, such that for rolling without slipping the arc lengths are equal between the initial
1
One of the two dimensionless positive constants ki can be greater than 1 for a cylinder in the form
of a bobbin that rolls on a narrow cylinder or track.

1
points of contact of the two cylinders and the new point of contact. This second rolling
constraint can be written as
r1 r1 + r2 r11 r
r2 2 = r1 (1 ) , 2 = 1 = with r r1 + r2 . (2)
r2 r2 r2
where 2 is the angle of the initial point of contact of cylinder 2 to the vertical.
Of course, the center of cylinder 1 is at y1 = r1 , and so long as the two cylinders are
touching, their axes are separated by distance r = r1 + r2 . Altogether there are 4 constraints
on the 6 degree of freedom (of two-dimensional motion) of the system, such that there are
only two independent degrees of freedom, which we take to be the angles 1 and .
Energy E = T + V is conserved, and since neither the kinetic energy T nor the potential
energy V (taken to be zero when = 0 ),

V = m2gr(cos 0 cos ), (3)

depend on coordinate 1 there will be another conserved quantity, the canonical momentum

L T
p1 = = . (4)

1 1

where L = T V is the Lagrangian of the system. However, p1 is not a single angular

momentum.2
Since there are two conserved quantities and two degrees of freedom, there is no need to
evaluate Lagranges equations of motion to determine the motion, so long as the cylinders
remain in contact and roll without slipping.3
The kinetic energy of cylinder 1, whose axis is at (x1, r1 ), is

m1 x 21 I1 1 1 + k1 2
T1 = + = m1r12 1 , (5)
2 2 2
using the rolling constraint (1) and the expression I1 = k1 m1r12 for the moment of inertia I1
in terms of parameter k1 .
The kinetic energy of cylinder 2, whose axis is at (x2, y2), is, using I2 = k2 m2 r22 ,

m2 (x 22 + y22 ) I2( 2 )
2 m2(x 22 + y22 ) k2 m2r22 ( 2 )
2
T2 = + = + , (6)
2 2 2 2
noting that the separation of kinetic energy into energy of the center-of-mass motion plus
energy of rotation about the center of mass requires the angular velocity to be measured
2
An example of a system in which there exists a constant of the motion involving angular velocity and
moments of inertia, but which is not a single angular momentum, has been given in [2]. See also [3].
3
For the implausible case of n cylinders, one on top of another, there are 3n degrees of freedom, n
constraints of touching, and n rolling constraints, leaving n indpendent degrees of freedom. Energy is
conserved, and if we take the n independent coordinates to be angle 1 and the n 1 angles i,i+1 of the
lines of centers of adjacent cylinders, then the energy depends on the i,i+1 but not 1 . Hence, there is one
conserved canonical momentum. For n > 2 it is necessary to use some of Lagrangres equations of motion
to solve for the motion.

2
with respect to a xed direction in an inertial frame. Then, recalling eqs. (1)-(2), we have

x2 = x1 + r sin ,
x 2 = r1 1 + r cos , (7)
y2 = r1 + r cos , y2 =
r sin , (8)
r1 1 r
2 = , (9)
r2
and the kinetic energy of cylinder 2 can be written as
m2 2 2 2
T2 = [r1 1 + 2r1 r cos 1 + r2 ]
2
k2 m2 2 2 2
+ [r1 1 2r1 r 1 + r2 ]
2
1 + k2 2 1 + k2 2
= m2 r12 1 + (cos k2 )m2r1 r 1 + m2r2 . (10)
2 2
The total kinetic energy T1 + T2 is
(1 + k1 )m1 + (1 + k2 )m2 2 2 1 + k2 2
T = r1 1 + (cos k2 )m2 r1r 1 + m2r2 , (11)
2 2
and the conserved canonical momentum is
T
p1 = = [(1 + k1 )m1 + (1 + k2 )m2 ]r12 1 + (cos k2 )m2 r1 r = constant. (12)

1
The total horizontal momentum of the system is, using the rolling constraint (1),

Px = (m1 + m2)x 1 + m2r cos = (m1 + m2)r1 1 + m2r cos ,

(13)

while the angular momentum of the cylinder 1 about its axis is

L1 = k1 m1 r12 1, (14)

and that of cylinder 2 about its axis is, using the constraint (2),

L2 = k2 m2 r22 ( 2 )
= k2 m2r2 (r1 r).
1
(15)

Hence, the conserved canonical momentum (12) can be written as

r1
p1 = r1 Px + L1 + L2 . (16)
r2
Equation (12) for the constant p1 can be rewritten as

(cos k2 )m2r Ar
1 = 0 = 0
(cos k2 ) , (17)
[(1 + k1 )m1 + (1 + k2 )m2 ]r1 r1
 
= Ar (cos k2 ) sin 2 ,
(18)
1
r1
m2
where A = . (19)
(1 + k1 )m1 + (1 + k2 )m2

3
Equation (17) integrates to give, for 0 (t = 0) = 0,
Ar
1 = 0 t (sin k2 ). (20)
r1
A particular solution is that is constant, say 0 with |0 | < /2, while = 0t, in which
case 2 = r1( 0 t 0 )/r2 according to the rolling constraint (2). Here, the two cylinders roll
together, with cylinder 2 at xed angle 0 , but this motion is unstable.4
For k2 < 1 (as for typical cylinders) and motion that starts with 0 = 0 and x1,0 = 1,0 =
0 = 0, after a small perturbation, the motion leads to angles 1 and with opposite signs
until sin = k2 after which the signs are the same (if the cylinders remain in contact).
Similarly, the angular velocities and begin with opposite signs, but the signs become the
same when cos = k2 . For a bobbin-like cylinder with k2 > 1, angles 1 and (and angular
always have the same signs. The figure on p. 1 corresponds to k2 > 1,
velocities 1 and )
in which the system has positive x-momentum, although it started from rest.
From the rolling constraint (2) we now have (for motion starting from rest)
r1 r1 Ar r1
2 = (1 ) = 0 (sin k2 ) . (21)
r2 r2 r2 r2
For k2 < 1, angles 1 and 2 have the same signs at small times, both opposite to that of .
For k2 > 1 the sign of 2 can be the same as that of , but only for a subset of the possible
values for the other parameters of the system.
The constant energy E = T + V can now be expressed as a function only of and ,
with the form

E 2 g
2
= 0 = [1 + k2 A(cos k2 ) ] (1 cos ), (22)
m2 r 2 2 r
for motion that starts from with = 0 = 1 = 2 , and with 0 = 0.5

2.1 Time Dependence

Thus far, we have obtained analytic expressions for angles 1 and 2 in terms of angle ,
and from these analytic expressions for x1 , x2 and y2 can also be obtained as a function of
. However, we do not know the time dependence (t), from which the time dependence of
all other quantities could be inferred.
By dierentiating the energy equation (22), we obtain a second-order time-dierential
equation for ,

2
= g/r A(cos k2 ) sin = 1 + k2 A(cos k2 )(3 2 cos ) g sin . (23)
1 + k2 A(cos k2 )2 [1 + k2 A(cos k2 )2 ]2 r
4
If the upper cylinder is a supercylinder, making elastic bounces o the horizontal surface, during
which bounces the point of contact of the cylinder comes to rest, the motion of the upper cylinder is a series
of pairs of hops, with or without net horizontal motion [4, 5].
5
The case that m1 = m2 , r1 = r2 = a = r/2 and k1 = k2 = 1/2 is considered in ex. 33, p. 492 of [6]. It
2
follows from eq. (22) that = 12(1 cos )/a(17 + 4 cos 4 cos2 ).

4
For the special case that the upper cylinder is a hollow shell, k2 = 1, the equation of motion
for small simplies to

g , (k2 = 1,  1). (24)

2r
which is the (Mathieu) equation for an inverted pendulum (of length l = 2r), for which
solutions are tabulated in, for example, [9].
Numerical methods must be used to deduce t() via either eqs. (23) or (24). Strictly, in-
nite time is required to reach any nite value of if the system starts from rest, so discussions
of such examples usually consider a small, nonzero initial angle or angular velocity. While
(t) is a monotonic function for the present example, if the axis of the lower cylinder were
subject to a periodic horizontal force in the x- (or y-) direction, the system could exhibit
stability at = 0, as discussed, for example, in sec. 30 of [10].

2.2 Constraint Forces

The various forces on the two rolling cylinders are illustrated in the gure below. Here, we
deduce these forces via Newtons equations of motion, plus the knowledge of the motion
obtained above via a variant of Lagranges method.6

2.2.1 Forces at the Horizontal Surface

The system of two cylinders, whose center of mass is at

(m1 + m2)x1 + m2 r sin (m1 + m2 )r1 + m2r cos

xcm = , ycm = , (25)
m1 + m2 m1 + m2
6
Lagranges method was devised to deduce the equations of motion of a system without consideration of
constraint forces that do no work. The method can be extended to include such forces by use of appropriate
additional coordinates in the Lagrangian, and representing the eects of constraints in terms with Lagrange
multipliers. See, for example, sec. 2.4 of [7] and sec. 19 of [8], as well as the Appendix.

5
 + [N1 (m1 + m2)g] y
is subject to the external force F1 x , so the equation of motion of the
center of mass are
 
2
F1 = (m1 + m2 ) x1 + m2 r cos sin
xcm = (m1 + m2)
 
+ m2r cos sin 2 ,
= (m1 + m2 )r1 (26)
1
 
2
ycm = (m1 + m2)g m2r sin + cos ,
N1 = (m1 + m2 )g + (m1 + m2 ) (27)

using the rolling constraint (1). Then, using eqs. (18), (22) and (23) we obtain F1 and N1
as functions of angle .
A single cylinder that rolls without slipping on a horizontal plane has constant horizontal
speed, and hence the force of friction is zero at the line of contact between the cylinder and
plane.
In the present example the horizontal speeds of the two cylinder are not constant, and
the force of friction F1, eq. (26), due to the plane is not zero, such that the x-momentum of
the system is not constant (as in the gure above).

2.2.2 Friction between the Cylinders

The force of friction, F21 = F12 on cylinder 2 due to cylinder 1, can be determined from
the angular acceleration of cylinder 2, using a torque equation and the rolling constraint (2),
I2
F21 = ( ) = k2 m2 r2(
) = k2 m2 (r1
r). (28)
r2 2 2 1

Then, the friction force F1 at the horizontal surface can also be determined from the
angular acceleration of cylinder 1, using the torque equation
= k1 m1r2
(F1 + F12)r1 = (F1 F21)r1 = I1
1 1 1, (29)

such that
+ k2 m2r.
F1 = (k1m1 + k2 m2)r1 (30)
1

This is consistent with eq. (26) in view of the relation (18).

These nonzero frictional forces imply that linear momentum Px and angular momenta
L1 and L2 are not conserved in this example, although there is a conserved quantity (16).

2.2.3 Normal Force between the Cylinders

The normal force N12 = N21 of cylinder 2 on cylinder 1 can be determined two ways,
by consideration of the x- or y-components of the forces on cylinder 1 (or equivalently, on
cylinder 2).
The vertical force components on cylinder 1 sum to zero, which implies that

N12 cos = N1 m1 g + F12 sin

   
2
r) sin
= m2 g r sin + cos + k2 (r1 1 (31)

6
using eqs. (27) and (28). Likewise, The horizontal force components on cylinder 1 sum to
m1x1 , which implies that

N12 sin = F1 m1 x1 F12 cos

   
2

= m2 r1 1 + r cos sin k2 (r1 1 r) cos , (32)

using eqs. (26) and (28). Then,

2
N12 = N12 cos2 + N12 sin2 = g cos + r1 sin 1 r . (33)

When N12 goes to zero, the cylinders separate.

2.3 Angle of Separation

The above analysis holds only so long as the two cylinders remain in contact, and the normal
force N12 between the cylinders is nonzero, i.e., when
2
r = g cos + r1 sin 1. (34)

For a method that does not use the forces to nd the angle s at which the cylinders
separate, we go to the accelerated frame of the lower cylinder, in which there appears to be
an eective acceleration due to gravity of

ge = gy x
= r1
x1 x 1 gy
. (35)

Cylinder 2 loses contact with cylinder 1 when the component of ge along the line of
centers, r = (sin , cos ), of the cylinders equals the instantaneous radial acceleration,
2
r . That is, separation occurs at angle s where7
 
2
= g cos s rA sin s (cos s k2 ) s sin s 2 , (37)
r s = r ge = g cos s + r1 sin s 1 s

using eq. (18). This conrms eq. (34).

Even for the special case of identical cylinders, m1 = m2, r1 = r2 = r/2 and k1 = k2 , the
expression (37) remains intricate.
7 2
When the lower cylinder is xed, ge = g, and eq. (37) reduces to r s = g cos s . The energy expression
2
(22) simplies to (1 + k2 )r /2 = g(1 cos ), for motion that starts with = 0 = 1 = 2 . Separation
2
occurs when r s = g cos s , such that
2
cos s = . (36)
3 + k2
In the limit that the upper cylinder is a line/point, k2 0 and cos s 2/3, as in the well known freshman
physics problem of a bug sliding o a log. For a solid cylinder, k2 = 1/2 and cos s = 4/7, for a solid sphere
k2 = 2/5 and cos s = 10/17, etc.

7
3 Variants
Thus far we have assumed that both cylinders roll without slipping. Variants include the
three cases in which it is assumed instead that there is no friction at one or both lines of
contact, and the cases where either one or two of the coordinates x1, 1 and 2 are held xed
with either no friction anywhere or rolling without slipping where rolling is possible. Here,
we consider only the rst of these examples.
In all cases the potential energy v is given by eq. (3) and the kinetic energy T by a variant
of eq. (11). We only consider systems that start from rest with cylinder 2 directly above
cylinder 1.

3.1 No Friction Anywhere

In the case of no friction anywhere, the cylinders do not rotate. Energy is conserved, and
the conserved total horizontal momentum is always zero. The system has two degrees of
freedom, which we take to be x1, the coordinate of the center of the lower cylinder, and the
angle of the line of centers between the two cylinders.
The total kinetic energy can be obtained from eq. (11) by setting k1 and k2 to zero, and
replacing factors of r1 1 by x 1 (undoing the rolling constraint (1), so to speak),
m1 + m2 2 m2 2 2
T = x 1 + m2r cos x 1 + r . (38)
2 2
The conserved canonical momentum is, for motion starting from rest with 1 = 2 = = 0,
T
px1 = = (m1 + m2)x 1 + m2r cos = Px = 0, (39)
x 1
which is just the total horizontal momentum.8 Using eq. (39) to eliminate x 1 from the kinetic
energy, we obtain the total energy as
2
E m1 + m2 sin2 g
2
=0= (1 cos ). (40)
m2 r m1 + m2 2 r
To nd the angle s at which the cylinders separate, we again go to the accelerated frame
of the lower cylinder, in which there appears to be an eective acceleration due to gravity,
 
m2 r 2
ge = gy
x1 x = cos sin x gy
, (41)
m1 + m2
where and can be deduced in terms of from eq. (40).
Cylinder 2 loses contact with cylinder 1 when the component of ge along the line of
centers, r = (sin , cos ), of the cylinders equals the instantaneous radial acceleration,
2
r . That is, separation occurs at angle s where
 
2 m2 r sin s 2
r s = r ge = g cos s cos s s sin s s . (42)
m1 + m2
8
This example includes other conserved/zero generalized momenta such as the z-component of the linear
momentum, and the angular momentum about the x- and y-axes.

8
After considerable eort, one can verify that eqs. (40) and (42) combine to give

m2 cos 2s = (m1 + m2)(3 cos s 2), (43)

as noted in ex. 6, p. 121 of [12].

When the lower cylinder is xed, we can set k2 = 0 and the result (36) again becomes
cos s = 2/3 as for a point mass sliding on a cylinder/sphere.

3.2 No Friction at the Horizontal Plane

In the case of no friction at the horizontal plane, but rolling without slipping of cylinder 2
on cylinder 1, both cylinders rotate, being torqued by the friction along the line of contact of
the cylinders. Energy is conserved, and the conserved total horizontal momentum is always
zero. The system has three degrees of freedom, which we take to be x1 , 1 and .
The total kinetic energy can be obtained from eq. (11) by replacing factors of r1 1 not
associated with k1 or k2 by x 1 (again undoing the rolling constraint),

m1 + m2 2 k1 m1 + k2 m2 2 2 1 + k2 2
T = x 1 + r1 1 + m2r cos x 1 k2 m2r1 r 1 + m2r2 . (44)
2 2 2
There are now two the conserved canonical momenta,
T
px1 = = (m1 + m2)x 1 + m2r cos = Px = 0, (45)
x 1
which is the total horizontal momentum (for motion starting from rest with 1 = 2 = = 0),
and
T r1
p1 = = (k1 m1 + k2 m2)r12 1 k2 m2r1 r = L1 + L2 = 0. (46)
1 r2

Using eqs. (45)-(46) to eliminate x 1 and 1 from the kinetic energy, we obtain the total
energy as9
  2
E m2 cos2 k22 m2 g
= 0 = 1 + k2 (1 cos ). (47)
m2 r 2 m1 + m2 k1 m1 + k2 m2 2 r

To nd the angle s at which the cylinders separate, we again go to the accelerated frame
of the lower cylinder, in which there appears to be an eective acceleration due to gravity,
 
m2 r 2
ge = gy
x1 x = cos sin x gy
, (48)
m1 + m2

where and can be deduced in terms of from eq. (47).

9
This case is considered in ex. 32, p. 492 of [6] for k1 = k2 = 1/2, where eq. (47) takes the form
2
(3m1 + 2m2 sin2 ) = 4(m1 + m2 )g(1 cos )/r.

9
 Cylinder 2 loses contact with cylinder 1 when the component of ge along the line of
centers, r = (sin , cos ), of the cylinders equals the instantaneous radial acceleration,
2
r . That is, separation occurs at angle s where
 
2 m2 r 2
r s = r ge = g cos s cos s s sin s s sin s . (49)
m1 + m2
This has the same form as eq. (42), but since the energy expressions (40) and (47) are
dierent, the value of s will be dierent.10
When the lower cylinder is xed, we again have cos s = 2/(3 + k2 ) as in eq. (36).

3.3 No Friction between the Cylinders

In the case of no friction between the cylinders, but cylinder 1 rolls without slipping on
the horizontal plane, only cylinder 1 rotates, being torqued by the friction at the horizontal
surface. Energy is conserved, but the total horizontal momentum is not. The system has
two degrees of freedom, which we take to be 1 and angle , using the rolling constraint (1)
to eliminate x1 from the energy, and the rolling constraint (2) to eliminate 2 in favor of 1
and .
The total kinetic energy can be obtained from eq. (11) by setting k2 to zero,
(1 + k1 )m1 + m2 2 2 m2 2 2
T = r1 1 + m2r1 r cos 1 + r , (51)
2 2
and the conserved canonical momentum is, for motion starting from rest with 1 = 2 =
= 0,
T
p1 = = [(1 + k1 )m1 + m2 ]r12 1 + m2r1 r cos = 0. (52)

1
Using eq. (52) to eliminate 1 from the kinetic energy, we obtain the total energy as
  2
E m2 cos2 g
= 0 = 1 (1 cos ). (53)
m2 r 2 (1 + k1 )m1 + m2 2 r
To nd the angle s at which the cylinders separate, we again go to the accelerated frame
of the lower cylinder, in which there appears to be an eective acceleration due to gravity,
 
1 x m2 r 2
ge = gy
x1 x = r1 gy
= cos sin x gy
, (54)
(1 + k1 )m1 + m2
10
An amusing special case has been noted in ex. 5, p. 121 of [12]. Suppose the two cylinders are identical,
m1 = m2 , r1 = r2 = a = r/2, and k1 = k2 = k . Then, eq. (46) becomes 1 = , such that = , and then
1
by the rolling constraint (2), 2 = 1 = 0. That is, the two cylinder roll together as if they were a single
rigid body until they separate.
2
Routh also claims that the cylinders separate at angle related by (k + 1 + sin2 )a = 2g(1 cos ),
where a = r1 = r2 = r/2, and we note that Rouths k 2 equals our ka2 . However, this is just the energy
relation (47), which holds for any angle at which the cylinders touch. If I evaluated eq. (49) correctly,
2[(2 + k) cos s 1] = cos s (1 cos s )2 (1 + cos s ). (50)

10
where and can be deduced in terms of from eq. (53).
Cylinder 2 loses contact with cylinder 1 when the component of ge along the line of
centers, r = (sin , cos ), of the cylinders equals the instantaneous radial acceleration,
2
r . That is, separation occurs at angle s where
 
2 m2r sin s 2
r s = r ge = g cos s cos s s sin s s . (55)
(1 + k1 )m1 + m2
When the lower cylinder is xed, we again have cos s = 2/3 as in sec. 3.1.

A Appendix: Constraint Forces via Lagrange

Multipliers
In general, two rigid bodies, such as the two cylinders of the present example, are to be
described by six coordinates per body (say, the spatial coordinates of the center of mass of
a body, the two angular directions of some xed body axis, and the angle of orientation of
the body about this axis), for a total of twelve coordinates. In the present example, only
two of these twelve coordinates are independent, as there are ten constraints: the axes of
the cylinders lie along the z-axis (4 constraints), the centers of mass of the cylinders are
at z = 0 (2 constraints), the lower cylinder lies on the plane y = 0 (1 constraint), the two
cylinders touch one another (1 constraint), the lower cylinder rolls without slipping on the
plane y = 0 (1 constraint), and the upper cylinder rolls without slipping on the lower cylinder
(1 constraint).
Furthermore, there is no dissipation of energy in this problem.
Given these constraints/conditions, Lagranges method consists of computing the kinetic
energy T and the potential energy V in terms of the independent coordinates (taken above
to be 1 and ). The total energy E(1 , ) = T + V is conserved, so the time derivative
dE/dt = 0 provides one relation between 1 and .From the Lagrangian L = T V we can,
in principle, deduce the equations of motion via Lagranges equations
d L L
= , (56)
dt qi qi
If L/qi = 0 (as for qi = 1 in the present example), then L/ qi is constant (as in
eq. (12) for the present example), and may be called a conserved quantity.11 Thus, it may
be (as in the present example) that there are as many conserved quantities as independent
coordinates, and Lagranges equations (56) are not needed to determine the motion.
In Lagranges method, for examples like the present with no dissipation of energy and
simple constraints on the coordinates, no mention is made of forces. If desired, expressions
for various forces can be deduced from Newtons F = ma with the acceleration a being
obtained from Lagranges equations (56). A subclass of the forces are those associated with
the various constraints on the coordinates of the systems; these are the so-called constraint
forces, which do no work (if no energy is dissipated).
11
Astonishingly, a paper [11] was published claiming that this in not well known to instructors and
students of physics.

11
 We can also deduce the constraint forces via a method in which more than the minimum
number of coordinates are used, as apparently rst proposed by Routh [12, 13] for holonomic
constraints,12 as a special case of a method for problems with nonholonomic constraints given
by Ferrers [15]. See also [16].
In this method, the minimum number n of independent coordinates is augmented with
m additional coordinates, so that the total set of coordinates is qi , i = 1, . . . , n + m, and
for which the m constraint equations fj (qi) = 0, j = 1, . . . , m, are known, but not explicity
enforced initially. Then, we consider the n + m modied Lagrange equations,
m
d L L  fj
= j , (57)
dt qi qi j=1 qi

where the j are so-called Lagrange multipliers (which have the physical signicance of
being the j constraint force if the dimensions of the constraint equation fj = 0 are chosen
appropriately).
In the present example with 12 coordinates, of which only 2 are independent, there are 10
constraint equations. Any number of these can be ignored in an implementation of eq. (57),
so there are 210 = 1024 dierent possible variations of the analysis of the present problem.
Here, we consider the problem to be two dimensional, in which case the rst six constraints
are automatically satised. The remaining four constrains are:

1. That the lower cylinder rolls without slipping on the plane y = 0, eq, (1),

f1 = x1 r1 1 = 0, (58)

2. That the upper cylinder rolls without slipping on the lower cylinder, eq. (2),

f2 = r2 2 r1 (1 ) = 0, (59)

3. That the two cylinders touch,

f3 = r r1 r2 = 0, (60)

where r is the distance in between the axes of the two cylinders,

4. That the lower cylinder touches the plane y = 0,

f4 = y1 r1 = 0. (61)

That is, we consider as many as six coordinates, x1, y1 , 1 , 2 , and r, rather than the
minimal set 1, used in the main body of this note.
We now consider the 15 analyses based on temporarily relaxing various subsets of the
constraints f1 , f2 , f3 and f4 .
12
The term holonomic was introduced by Hertz on p. 91 of [14].

12
A.1 Relax the Rolling Constraint on the Lower Cylinder
If we imagine that the constraint (58) on the lower cylinder is relaxed, then we need three
coordinates, x1 , 1 and to describe the system.
Constraints (59)-(61) are still enforced, so the kinetic energy of the lower cylinder is given
by the rst form of eq. (5), while the kinetic energy of the upper cylinder becomes
2
m2 2 (1 + k2 )m2 2 2 k2 m2 r12 1
T2 = x + m2x 1 r cos + r + k2 m2 r1r 1 ,
(62)
2 1 2 2
and the potential energy is still given by eq. (3).
The Lagrangian L = T1 + T2 V does not depend on x1 or 1 , so it is useful to identify
the canonical momenta
L
px1 = = (m1 + m2)x 1 + m2r cos = Px , (63)
x 1
which is the total horizontal momentum, eq. (13), of the system, and
L
p1 = = (k1 m1 + k2 m2 )r12 1 k2 m2r1 r = k1 m1 r12 1 + r1 [k2 m2(r1 1 k2 m2 r)]


1
r1
= L1 + L2, (64)
r2
where L1 and L2 are the angular momenta, eqs. (14)-(15), of the two cylinders about their
axes.
The derivatives of the constraint equation (58) are
f1 f1 f1
= 1, = r1, = 0. (65)
x1 1
The extended Lagrange method for this case involves a single multiplier 1 associated
with the rolling constraint (58), such that the three Lagrange equations are now
dpx1 f1
= 1 = 1 , (66)
dt x1
dp1 f1
= 1 = r1 1 , (67)
dt 1
d L L f1
= 1 = 0. (68)

dt
Combining eqs. (66) and (67), we have that
 
d p p1 L1 L2
px1 + 1 = 0, px1 + = Px + + = 0, (69)
dt r1 r1 r1 r2
supposing that the system starts with x1 = 1 = = 0, which is eq. (16) divided by r1 .
The force 1 associated with the constraint f1 that the lower cylinder rolls without
slipping on the plane y = 0 is related by
1 dp1 k2 m2 r,
1 = = (k1 m1 + k2 m2)r1 1 (70)
r1 dt
which is the force F1 found in eq. (30).

13
A.2 Relax the Rolling Constraint on the Upper Cylinder
If we imagine that the constraint (59) on the upper cylinder is relaxed, then we need three
coordinates, 1 , 2 and to describe the system.
Constraints (58) and (60)-(61) are still enforced, so the kinetic energy of the lower cylinder
is given by the second form of eq. (5), while the kinetic energy of the upper cylinder is given
by
   
m2 2 2 2 k2 m2r22 2 2
T2 = r1 1 + 2r1 r cos 1 + r2 + 2 2 2 + , (71)
2 2
and the potential energy is still given by eq. (3).
The Lagrangian L = T1 + T2 V does not depend on 1 or 2, so it is useful to identify
the canonical momenta
L
p1 = = [(1 + k1 )m1 + m2 ]r12 1 + m2r1 r cos ,
(72)

1
and
L
p2 = = k2 m2 r22 ( 2 ).
(73)
2
The derivatives of the constraint equation (59) are
f2 f2 f2
= r1 , = r2 , = r1 . (74)
1 2
The extended Lagrange method for this case involves a single multiplier 2 associated
with the rolling constraint (59), such that the three Lagrange equations are now
dp1 f2
= 2 = r1 2 , (75)
dt 1
dp2 f2
= 2 = r2 2 , (76)
dt 2
d L L f2
= 2 = r1 2 . (77)
dt
Combining eqs. (75) and (76), we have that
 
d p1 p2 p1 p2
+ = 0, + =0 (78)
dt r1 r2 r1 r2
supposing that the system starts with x1 = 1 = = 0. After we enforce the rolling
constraint (59), this becomes Px + L1 /r1 + L2 /r2 = 0, as previously noted.
The force 2 associated with the constraint f2 that the upper cylinder rolls without
slipping on the lower cylinder is related by
1 dp2 ) = k2 m2(r1
r),
F2 = 2 = = k2 m2r2 ( 2 1 (79)
r2 dt
which was previously found as F21 in eq. (28),

14
A.3 Relax the Constraint that the Cylinders Touch
If we imagine that the constraint (60) between the cylinders is relaxed, then we need four
coordinates, 1 , 2, and r to describe the system.
Constraints (58) and (61) are still enforced, so the kinetic energy of the lower cylinder is
given by the second form of eq. (5), while the kinetic energy of the upper cylinder is given
by
   
m2 2 2 2 k2 m2r22 2 2
T2 = r1 1 + 2r1 (r cos + r sin ) 1 + r2 + r 2 + 2 2 2 + , (80)
2 2
while the potential energy should now be written as V = m2g(r cos r1 r2) (to be zero
when cylinder 2 sits directly on top of cylinder 1).
The Lagrangian L = T1 + T2 V does not depend on 1 or 2, so it is useful to identify
the canonical momenta
L
p1 = = [(1 + k1 )m1 + m2 ]r12 1 + m2r1 (r cos + r sin ) = r1 Px + L1 + m2r1 r sin ,(81)

1

L
and p2 = = k2 m2r22 ( 2 )
= L2 . (82)

2

The derivatives of the constraint equation (60) are

f3 f3 f3 f3
= 0, = 0, = 0, = 1. (83)
1 2 r

The extended Lagrange method for this case involves a single multiplier 3 associated
with the touching constraint (60), such that the four Lagrange equations are

dp1 f3
= 3 = 0, (84)
dt 1
dp2 f3
= 3 = 0, (85)
dt 2
d L L f3
= 3 = 0, (86)
dt
d L L f3
= 3 = 3 . (87)
dt r r r
The force 3 associated with the constraint f3 that the upper cylinder touches the lower
cylinder is related by

 
d L L + cos m2 r1 cos + r 2 + m2g cos
3 = = m2r1 sin 1 1 1
dt r r  
r 2 + g cos ,
= m2 r1 sin (88)
1

15
on setting r = 0, as this expression makes physical sense only after constraint (60) is enforced.
A case of particular interest is when this force goes to zero, at the angle s of separation,
which is now related by
2
r s = g cos + r1 sin s

1 (89)
 
2
= g cos Ar sin s (cos s k2 )s sin s s ,

using eq. (17). This relation was previously found in eq. (37).

A.4 Relax the Constraint that the Cylinder 1 Touches the Plane
y=0
If we imagine that the constraint (61) is relaxed, then we need three coordinates, y1 , 1 and
to describe the system.
Constraints (58)-(60) are still enforced, so the kinetic and potential energies of the system
are given by eqs. (11) and (3) with the additional terms
m1 + m2 2
T = y1 m2 r sin y1, V = (m1 + m2)g(y1 r1). (90)
2
The Lagrangian L = T1 + T2 V does not depend on 1 , so it is useful to identify the
canonical momentum
L r1
p1 = = [(1 + k1 )m1 + (1 + k2 )m2]r12 1 + (cos k2 )m2 r1r = r1 Px + L1 + L2 . (91)
1 r2
The derivatives of the constraint equation (61) are
f4 f4 f4
= 0, = 0, = 1. (92)
1 y1
The extended Lagrange method for this case involves a single multiplier 4 associated
with the touching constraint (61), such that the four Lagrange equations are
dp1 f3
= 4 = 0, (93)
dt 1
d L L f4
= 4 = 0, (94)
dt
d L L f4
= 4 = 4 . (95)
dt y 1 y1 r
The force 4 associated with the constraint f4 that the lower cylinder touches the plane
y = 0 is related by
 
d L L 2
F4 = 4 = = (m1 + m2) y1 m2 r sin + cos + (m1 + m2 )g. (96)
dt y 1 y1
This makes physical sense only after the constraint (61) is enforced, such that y1 = 0, and
the constraint force is just the normal force upward on cylinder 1,
 
2
F4 = N1 = (m1 + m2)g m2r sin + cos = (m1 + m2)g + m2 y2, (97)

which was previously found as N1 in eq. (27).

16
A.5 Relax All Constraints
If we imagine that all constraints (58)-(61) are relaxed, then we consider the six coordinates
x1, y1 , 1 , 2 and r.
The kinetic energy is now
   
m1 + m2 2 m1k2 r12 2 k2 m2 r22 2 2 m2 2 2
T = (x 1 + y 12) + 1 + 2 2 2 + + r + r2
2 2 2 2
+m2 r(
x 1 sin + y1 cos ) + m2r(x 1 cos y1 sin ), (98)

and the potential energy is

V = m2gr(1 cos ) + (m1 + m2)g(y1 r1). (99)

The Lagrangian L = T V does not depend on coordinates x1, 1, or 2, so we identify the

canonical momenta
L = Px ,
px1 = = (m1 + m2 )x 1 + m2 (r sin + r cos ) (100)
x 1
L
p1 = = m1k1 r12 1 = L1 , (101)

1
L
p2 = = m2k2 r22 ( 2 )
= L2 , (102)
2
The extended Lagrange method for this case involves four multipliers 1 -4 associated
with the four constraints (58)-(61), such that the six Lagrange equations are
dpx1 f1 f2 f3 f4
= 1 + 2 + 3 + 4 = 1 , (103)
dt x1 x1 x1 x1
dp1 f1 f2 f3 f4
= 1 + 2 + 3 + 4 = r11 r1 2 , (104)
dt 1 1 1 1
dp2 f1 f2 f3 f4
= 1 + 2 + 3 + 4 = r2 2 , (105)
dt 2 2 2 2
d L L f1 f2 f3 f4
= 1 + 2 + 3 + 4 = r1 2 , (106)
dt
d L L f1 f2 f3 f4
= 1 + 2 + 3 + 4 = 3 , (107)
dt r r r r r r
d L L f1 f2 f3 f4
= 1 + 2 + 3 + 4 = 4 , (108)
dt y 1 y1 y1 y1 y1 y1
using the derivatives (65), (74), (83) and (92). We can combine eqs. (103)-(105) to nd
   
d p p d L1 L2 L1 L2
px1 + 1 + 2 = Px + = 0, Px + + = 0, (109)
dt r1 r2 dt r1 r2 r1 r2
for a system that starts with 1 = 2 = = 0. This form is suggestive, but its content is
only understandable if one writes it out in detail, as in eq. (12), which integrates to (20).
Then, we have a description of the motion in terms of a single variable, .

17
 We now enforce the constraints, and evaluate the multipliers.
The force 1 associated with the constraint f1 that the lower cylinder rolls without
slipping on the plane y = 0 is related by eq. (103),
 
dpx1 2
1 = x1 + m2 r cos r sin ,
= (m1 + m2) (110)
dt
after setting r = 0, which is the force F1 found in eq. (26).
The force 2 associated with the constraint f2 that the upper cylinder rolls without
slipping on the lower cylinder is related by eq. (105),
1 dp2 ) = k2 m2(r1
r),
F2 = 2 = = k2 m2r2 ( 2 1 (111)
r2 dt
which was previously found as F21 in eq. (28).13
The force 3 associated with the constraint f3 that the upper cylinder touches the lower
cylinder is related by eq. (107),
 
d L L m2 x 1 cos + r 2 + m2 g cos
3 = = m2(
x1 sin + x 1 cos )
dt r r  
2

= m2 x1 sin r + g cos , (112)

on setting r = 0 and y 1 = 0, as this expression makes physical sense only after constraints
(60)-(61) are enforced.
The force 4 associated with the constraint f4 that the lower cylinder touches the plane
y = 0 is related by eq. (108),
 
d L L 2
F4 = 4 = y1 m2 r sin + cos + (m1 + m2 )g. (113)
= (m1 + m2)
dt y 1 y1
This makes physical sense only after the constraint (61) is enforced, such that y1 = 0, and
the constraint force is just the normal force upward on cylinder 1,
 
2
F4 = N1 = (m1 + m2)g m2r sin + cos = (m1 + m2)g + m2 y2, (114)

which was previously found as N1 in eq. (27).

We return to the description of the motion, and note that since the Lagrangian does not
depend on time, energy is conserved. After enforcing the constraints (58)-(61), and using
the integral (20) of the conserved quantity (109), we arrive at the expression (22) for the
(conserved) energy as a function of angle only. The time derivative of this expression14 (as
well as Lagranges equations) provides a second-order dierential equation for , which can
in principle be integrated to describe the motion in detail, as discussed in sec. 2.1.
Thus, the method of relaxing constraints and adding Lagrange multipliers eventually
recovers the description of the motion that was obtained more directly via the basic method
of Lagrange, which utilizes only the minimum number of independent coordinates (2 in this
example).
13
The forces F1 and F12 could also be determined via eqs. (104) and (106).
14
This approach is called the principle of vis viva in sec. 141 of [12].

18
References
[1] K.T. McDonald, Cylinder Rolling inside Another Rolling Cylinder, (Oct. 21, 2014),
http://physics.princeton.edu/~mcdonald/examples/2cylinders_in.pdf

[2] W.K. Robinson and B.P. Watson, A misuse of angular momentum conservation, Am.
J. Phys. 53, 82 (1985),
http://physics.princeton.edu/~mcdonald/examples/mechanics/robinson_ajp_53_82_85.pdf

[3] A. Tort, F.C. Santos and O.M. Ritter, An extra constant of motion for the N-disc
problem, Eur. J. Phys. 10, 217 (1989),
http://physics.princeton.edu/~mcdonald/examples/mechanics/tort_ejp_10_217_89.pdf

[4] R.L. Garwin, Kinematics of an Ultraelastic Rough Ball, Am. J. Phys. 37, 88 (1969),
http://physics.princeton.edu/~mcdonald/examples/mechanics/garwin_ajp_37_88_69.pdf

[5] F.S. Crawford, Superball and time-reversal invariance, Am. J. Phys. 50, 856 (1982),
http://physics.princeton.edu/~mcdonald/examples/mechanics/crawford_ajp_50_856_82.pdf

[6] L.A. Pars, Introduction to Dynamics (Cambridge U. Press, 1953).

[7] H. Goldstein, C.H. Poole and J. Safko, Classical Mechanics, 3rd ed. (Pearson, 2011).

[8] A.L. Fetter and J.D. Walecka, Theoretical Mechanics of Particles and Continua
(McGraw-Hill, 1980).

[9] M. Abramowitz and I.A. Stegun, Handbook of Mathematical Functions (National Bureau
of Standards, 1964), http://physics.princeton.edu/~mcdonald/examples/EM/abramowitz_and_stegun.pdf

[10] L.D. Landau and E.M. Lifshitz, Mechanics, 3rd ed. (Pergamon, 1976).

[11] B.F. Plybon, Conservation Laws for Undergraduates, Am. J. Phys. 39, 1372 (1971),
http://physics.princeton.edu/~mcdonald/examples/mechanics/plybon_ajp_39_1372_71.pdf

[12] E.J. Routh, The Elementary Part of a Treatise on the Dynamics of a System of Rigid
Bodies, 6th ed. (MacMillan, 1897), secs. 400 and 429,
http://physics.princeton.edu/~mcdonald/examples/mechanics/routh_elementary_rigid_dynamics.pdf

[13] E.J. Routh, The Advanced Part of a Treatise on the Dynamics of a System of Rigid
Bodies, 6th ed. (MacMillan, 1905), sec. 47,
http://physics.princeton.edu/~mcdonald/examples/mechanics/routh_advanced_rigid_dynamics.pdf

[14] H. Hertz, Die Prinzipien der Mechanik (Barth, Leipzig, 1894),

http://physics.princeton.edu/~mcdonald/examples/mechanics/hertz_mechanik_94.pdf

[15] N.M. Ferrers, Extension of Lagranges Equations Quart. J. Pure Appl. Math. 12, 1
(1872), http://physics.princeton.edu/~mcdonald/examples/mechanics/ferrers_qjpam_12_1_72.pdf

19
[16] K.T. McDonald, Calculus of Variations, Princeton U. Ph205 Lecture Notes (1980),
http://physics.princeton.edu/~mcdonald/examples/Ph205/ph205l5.pdf
More about Lagranges Equations,
http://physics.princeton.edu/~mcdonald/examples/Ph205/ph205l6.pdf

20