P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
ii
This page intentionally left blank
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
INTERMEDIATE DYNAMICS FOR ENGINEERS
This book has sufﬁcient material for two fulllength semester courses
in intermediate engineering dynamics. For the ﬁrst course a Newton–
Euler approach is used, followed by a Lagrangian approach in the sec
ond. Using some ideas from differential geometry, the equivalence of
these two approaches is illuminated throughout the text. In addition,
this book contains comprehensive treatments of the kinematics and dy
namics of particles and rigid bodies. The subject matter is illuminated
by numerous highly structured examples and exercises featuring a wide
range of applications and numerical simulations.
Oliver M. O’Reilly is a professor of mechanical engineering at the
University of California, Berkeley. His research interests lie in contin
uum mechanics and nonlinear dynamics, speciﬁcally in the dynamics
of rigid bodies and particles, Cosserat and directed continuua, dynam
ics of rods, history of mechanics, and vehicle dynamics. O’Reilly is the
author of more than 50 archival publications and Engineering Dynam
ics: A Primer. He is also the recipient of the University of California
at Berkeley’s Distinguished Teaching Award and three departmental
teaching awards.
i
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
ii
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
Intermediate Dynamics for Engineers
A UNIFIED TREATMENT OF
NEWTON–EULER AND LAGRANGIAN
MECHANICS
Oliver M. O’Reilly
University of California, Berkeley
iii
CAMBRIDGE UNIVERSITY PRESS
Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São Paulo
Cambridge University Press
The Edinburgh Building, Cambridge CB2 8RU, UK
First published in print format
ISBN13 9780521874830
ISBN13 9780511424359
© Oliver M. O’Reilly 2008
2008
Information on this title: www.cambridge.org/9780521874830
This publication is in copyright. Subject to statutory exception and to the provision of
relevant collective licensing agreements, no reproduction of any part may take place
without the written permission of Cambridge University Press.
Cambridge University Press has no responsibility for the persistence or accuracy of urls
for external or thirdparty internet websites referred to in this publication, and does not
guarantee that any content on such websites is, or will remain, accurate or appropriate.
Published in the United States of America by Cambridge University Press, New York
www.cambridge.org
eBook (NetLibrary)
hardback
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
This book is dedicated to my adventurous daughter, Anna
v
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
vi
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
Contents
Preface page xi
PART ONE DYNAMICS OF A SINGLE PARTICLE 1
1 Kinematics of a Particle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1 Introduction 3
1.2 Reference Frames 3
1.3 Kinematics of a Particle 5
1.4 Frequently Used Coordinate Systems 6
1.5 Curvilinear Coordinates 9
1.6 Representations of Particle Kinematics 14
1.7 Constraints 15
1.8 Classiﬁcation of Constraints 20
1.9 Closing Comments 27
Exercises 27
2 Kinetics of a Particle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.1 Introduction 33
2.2 The Balance Law for a Single Particle 33
2.3 Work and Power 35
2.4 Conservative Forces 36
2.5 Examples of Conservative Forces 37
2.6 Constraint Forces 39
2.7 Conservations 45
2.8 Dynamics of a Particle in a Gravitational Field 47
2.9 Dynamics of a Particle on a Spinning Cone 55
2.10 A Shocking Constraint 59
2.11 A Simple Model for a Roller Coaster 60
2.12 Closing Comments 64
Exercises 66
vii
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
viii Contents
3 Lagrange’s Equations of Motion for a Single Particle . . . . . . . . . . . . 70
3.1 Introduction 70
3.2 Lagrange’s Equations of Motion 71
3.3 Equations of Motion for an Unconstrained Particle 73
3.4 Lagrange’s Equations in the Presence of Constraints 74
3.5 A Particle Moving on a Sphere 78
3.6 Some Elements of Geometry and Particle Kinematics 80
3.7 The Geometry of Lagrange’s Equations of Motion 83
3.8 A Particle Moving on a Helix 87
3.9 Summary 91
Exercises 92
PART TWO DYNAMICS OF A SYSTEM OF PARTICLES 101
4 The Equations of Motion for a System of Particles . . . . . . . . . . . . . 103
4.1 Introduction 103
4.2 A System of N Particles 104
4.3 Coordinates 105
4.4 Constraints and Constraint Forces 107
4.5 Conservative Forces and Potential Energies 110
4.6 Lagrange’s Equations of Motion 111
4.7 Construction and Use of a Single Representative Particle 113
4.8 The Lagrangian 118
4.9 A Constrained System of Particles 119
4.10 A Canonical Form of Lagrange’s Equations 122
4.11 Alternative Principles of Mechanics 128
4.12 Closing Remarks 131
Exercises 131
5 Dynamics of Systems of Particles . . . . . . . . . . . . . . . . . . . . . . . . 134
5.1 Introduction 134
5.2 Harmonic Oscillators 134
5.3 A Dumbbell Satellite 140
5.4 A Pendulum and a Cart 143
5.5 Two Particles Tethered by an Inextensible String 147
5.6 Closing Comments 151
Exercises 153
PART THREE DYNAMICS OF A SINGLE RIGID BODY 161
6 Rotation Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
6.1 Introduction 163
6.2 The Simplest Rotation 164
6.3 ProperOrthogonal Tensors 166
6.4 Derivatives of a ProperOrthogonal Tensor 168
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
Contents ix
6.5 Euler’s Representation of a Rotation Tensor 171
6.6 Euler’s Theorem: Rotation Tensors and ProperOrthogonal
Tensors 176
6.7 Relative Angular Velocity Vectors 178
6.8 Euler Angles 181
6.9 Further Representations of a Rotation Tensor 191
6.10 Derivatives of Scalar Functions of Rotation Tensors 195
Exercises 198
7 Kinematics of Rigid Bodies . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
7.1 Introduction 206
7.2 The Motion of a Rigid Body 206
7.3 The Angular Velocity and Angular Acceleration Vectors 211
7.4 A Corotational Basis 212
7.5 Three Distinct Axes of Rotation 213
7.6 The Center of Mass and Linear Momentum 215
7.7 Angular Momenta 218
7.8 Euler Tensors and Inertia Tensors 219
7.9 Angular Momentum and an Inertia Tensor 223
7.10 Kinetic Energy 224
7.11 Concluding Remarks 226
Exercises 226
8 Constraints on and Potentials for Rigid Bodies . . . . . . . . . . . . . . . 237
8.1 Introduction 237
8.2 Constraints 237
8.3 A Canonical Function 241
8.4 Integrability Criteria 243
8.5 Forces and Moments Acting on a Rigid Body 247
8.6 Constraint Forces and Constraint Moments 248
8.7 Potential Energies and Conservative Forces and Moments 256
8.8 Concluding Comments 262
Exercises 263
9 Kinetics of a Rigid Body . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
9.1 Introduction 272
9.2 Balance Laws for a Rigid Body 272
9.3 Work and Energy Conservation 274
9.4 Additional Forms of the Balance of Angular Momentum 276
9.5 MomentFree Motion of a Rigid Body 279
9.6 The Baseball and the Football 285
9.7 Motion of a Rigid Body with a Fixed Point 289
9.8 Motions of Rolling Spheres and Sliding Spheres 294
9.9 Closing Comments 297
Exercises 299
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
x Contents
10 Lagrange’s Equations of Motion for a Single Rigid Body . . . . . . . . . 307
10.1 Introduction 307
10.2 Conﬁguration Manifold of an Unconstrained Rigid Body 308
10.3 Lagrange’s Equations of Motion: A First Form 311
10.4 A Satellite Problem 315
10.5 Lagrange’s Equations of Motion: A Second Form 318
10.6 Lagrange’s Equations of Motion: Approach II 324
10.7 Rolling Disks and Sliding Disks 325
10.8 Lagrange and Poisson Tops 331
10.9 Closing Comments 336
Exercises 336
PART FOUR SYSTEMS OF RIGID BODIES 345
11 Introduction to Multibody Systems . . . . . . . . . . . . . . . . . . . . . . 347
11.1 Introduction 347
11.2 Balance Laws and Lagrange’s Equations of Motion 347
11.3 Two PinJointed Rigid Bodies 349
11.4 A SingleAxis Rate Gyroscope 351
11.5 Closing Comments 355
Exercises 355
APPENDIX: BACKGROUND ON TENSORS . . . . . . . . . . . . . . . . . . . . . . 362
A.1 Introduction 362
A.2 Preliminaries: Bases, Alternators, and Kronecker Deltas 362
A.3 The Tensor Product of Two Vectors 363
A.4 SecondOrder Tensors 364
A.5 A Representation Theorem for SecondOrder Tensors 364
A.6 Functions of SecondOrder Tensors 367
A.7 ThirdOrder Tensors 370
A.8 Special Types of SecondOrder Tensors 372
A.9 Derivatives of Tensors 373
Exercises 374
Bibliography 377
Index 389
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
Preface
The writing of this book started more than a decade ago when I was ﬁrst given
the assignment of teaching two courses on rigid body dynamics. One of these
courses featured Lagrange’s equations of motion, and the other featured the
Newton–Euler equations. I had long struggled to resolve these two approaches to
formulating the equations of motion of mechanical systems. Luckily, at this time,
one of my colleagues, Jim Casey, was examining the elegant works [205, 207, 208]
of Synge and his coworkers on this topic. There, he found a partial resolution to
the equivalence of the Lagrangian and Newton–Euler approaches. He then went
further and showed how the governing equations for a rigid body formulated by use
of both approaches were equivalent [27, 28]. Shades of this result could be seen in
an earlier work by Greenwood [79], but Casey’s work established the equivalence
in an unequivocal fashion. As is evident from this book, I subsequently adapted
and expanded on Casey’s treatment in my courses. My treatment of dynamics
presented in this book is also heavily inﬂuenced by the texts of Papastavridis [169]
and Rosenberg [182]. It has also beneﬁted from my graduate studies in dynamical
systems at Cornell in the late 1980s. There, under the guidance of Philip Holmes,
Frank Moon, Richard Rand, and Andy Ruina, I was shown how the equations
governing the motion of (often simple) mechanical systems featuring particles and
rigid bodies could display surprisingly rich behavior.
There are several manners in which this book differs from a traditional text on
engineering dynamics. First, I demonstrate explicitly how the equations of motion
obtained by using Lagrange’s equations and the Newton–Euler equations are equiv
alent. To achieve this, my discussion of geometry and curvilinear coordinates is far
more detailed than is normally found in textbooks at this level. The second differ
ence is that I use tensors extensively when discussing the rotation of a rigid body.
Here, I am following related developments in continuum mechanics, and I believe
that this enables a far clearer derivation of many of the fundamental results in the
kinematics of rigid bodies.
I have distributed as many examples as possible throughout this book and have
attempted to cite uptodate references to them and related systems as far as fea
sible. However, I have not approached the exhaustive treatments by Papastavridis
xi
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
xii Preface
[169] nor its classical counterpart by Routh [184, 185]. I hope that sufﬁcient citations
to these and several other wonderful texts on dynamics have been placed through
out the text so that the interested reader has ample opportunity to explore this re
warding subject.
Using This Text
This book has been written so that it provides sufﬁcient material for two fulllength
semester courses in engineering dynamics. As such it contains two tracks (which
overlap in places). For the ﬁrst course, in which a Newton–Euler approach is used,
the following chapters can be covered:
1. Kinematics of a Particle (Section 1.5 can be omitted)
2. Kinetics of a Particle
Appendix on Tensors
6. Rotation Tensors
7. Kinematics of Rigid Bodies
8. Constraints on and Potentials for Rigid Bodies
9. Kinetics of a Rigid Body
11. Multibody Systems
The second course, in which a Lagrangian approach is used, could be based on the
following chapters:
1. Kinematics of a Particle
2. Kinetics of a Particle
3. Lagrange’s Equations of Motion for a Single Particle
4. Lagrange’s Equations of Motion for a System of Particles
5. Dynamics of Systems of Particles
Appendix on Tensors
6. Rotation Tensors (with particular emphasis on Section 6.8)
7. Kinematics of Rigid Bodies
8. Constraints on and Potentials for Rigid Bodies
9. Kinetics of a Rigid Body
10. Lagrange’s Equations of Motion for a Single Rigid Body
11. Multibody Systems
In discussing rotations for the second course, time constraints permit a detailed
discussion of only the Euler angle parameterization of a rotation tensor from
Chapter 6 and a brief mention of the examples on rigid body dynamics discussed in
Chapter 9.
Most of the exercises at the end of each chapter are highly structured and are
intended as a selfstudy aid. As I don’t intend to publish or distribute a solutions
manual, I have tailored the problems to provide answers that can be validated.
Some of the exercises feature numerical simulations that can be performed with
Matlab or Mathematica. Completing these exercises is invaluable both in terms of
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
Preface xiii
comprehending why obtaining a set of differential equations for a system is
important and for visualizing the behavior of the system predicted by the model.
I also strongly recommend semester projects for the students during which they
can delve into a speciﬁc problem, such as the dynamics of a wobblestone, the ﬂight
of a Frisbee, or the reorientation of a dualspin satellite, in considerable detail.
In my courses, these projects feature simulations and animations and are usually
performed by students working in pairs who start working together after 7 weeks
of a 15week semester.
Image Credit
The portrait of William R. Hamilton in Figure 4.6 in Subsection 4.11.3 is from the
Royal Irish Academy in Dublin, Ireland. I am grateful to Pauric Dempsey, the Head
of Communications and Public Affairs of this institution, for providing the image.
Acknowledgments
This book is based on my class notes and exercises for two courses on dynamics,
ME170, Engineering Mechanics III, and ME175, Intermediate Dynamics, which
I have taught at the Department of Mechanical Engineering at the University of
California at Berkeley over the past decade. Some of the aims of these courses are
to give senior undergraduate and ﬁrstyear graduate students in mechanical engi
neering requisite skills in the area of dynamics of rigid bodies. The book is also
intended to be a sequel to my book Engineering Dynamics: A Primer, which was
published by SpringerVerlag in 2001.
I have been blessed with the insights and questions of many remarkable students
and the help of several dedicated teaching assistants. Space precludes mention of all
of these students and assistants, but it is nice to have the opportunity to acknowl
edge some of them here: Joshua P. Coaplen, Nur Adila Faruk Senan, David Gulick,
Moneer Helu, Eva Kanso, Patch Kessler, Nathan Kinkaid, Todd Lauderdale, Henry
Lopez, David Moody, Tom Nordenholz, Jeun Jye Ong, Sebasti ´ en Payen, Brian
Spears, Philip J. Stephanou, Meng How Tan, Peter C. Varadi, and St ´ ephane Ver
guet. I am also grateful to Chet Vignes for his careful reading of an earlier draft of
the book.
Many other scholars helped me with speciﬁc aspects of and topics in this book.
Figure 9.1 was composed by Patch Kessler. Henry Lopez (B.E. 2006) helped me with
the rollercoaster model and simulations of its equations of motion. Professor Chris
Hall of Virginia Tech pointed out reference [118] on Lagrange’s solution of a satel
lite dynamics problem. Professor Richard Montgomery of the University of Califor
nia at Santa Cruz discussed the remarkable ﬁgureeight solutions to the threebody
problem with me, Professor Glen Niebur of the University of Notre Dame provided
valuable references on Codman’s paradox, Professor Harold Soodak of the City
College of New York provided valuable comments on the tippe top, and Profes
sors Donald Greenwood and John Papastavridis carefully read a penultimate draft
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
xiv Preface
of this book and generously provided many constructive comments and corrections
for which I am most grateful.
Most of this book was written during the past 10 years at the University of Cali
fornia at Berkeley. The remarkable library of this institution has been an invaluable
resource in my quest to distill more than 300 years of work on the subject matter in
this book. I am most grateful to the library staff for their assistance and the taxpay
ers for their support of the University of California.
Throughout this book, several references to my own research on rigid body
dynamics can be found. In addition to the students mentioned earlier, I have had
the good fortune to work with Jim Casey and Arun Srinivasa on several aspects of
the equations of motion for rigid bodies. The numerous citations to their works are
a reﬂection of my gratitude to them.
This book would not have been published without the help and encouragement
of Peter Gordon at Cambridge University Press and would contain far more er
rors were it not for the editorial help of Victoria Danahy. Despite the assistance of
several other proofreaders, it is unavoidable that some typographical and technical
errors have crept into this book, and they are my unpleasant responsibility alone. If
you ﬁnd some on your journey through these pages, I would be pleased if you could
bring them to my attention.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
PART ONE
DYNAMICS OF A SINGLE PARTICLE
1
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
2
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
1 Kinematics of a Particle
1.1 Introduction
One of the main goals of this book is to enable the reader to take a physical sys
tem, model it by using particles or rigid bodies, and then interpret the results of the
model. For this to happen, the reader needs to be equipped with an array of tools
and techniques, the cornerstone of which is to be able to precisely formulate the
kinematics of a particle. Without this foundation in place, the future conclusions on
which they are based either do not hold up or lack conviction.
Much of the material presented in this chapter will be repeatedly used through
out the book. We start the chapter with a discussion of coordinate systems for a
particle moving in a threedimensional space. This naturally leads us to a discussion
of curvilinear coordinate systems. These systems encompass all of the familiar co
ordinate systems, and the material presented is useful in many other contexts. At
the conclusion of our discussion of coordinate systems and its application to particle
mechanics, you should be able to establish expressions for gradient and acceleration
vectors in any coordinate system.
The other major topics of this chapter pertain to constraints on the motion of
particles. In earlier dynamics courses, these topics are intimately related to judi
cious choices of coordinate systems to solve particle problems. For such problems,
a constraint was usually imposed on the position vector of a particle. Here, we also
discuss timevarying constraints on the velocity vector of the particle. Along with
curvilinear coordinates, the topic of constraints is one most readers will not have
seen before and for many they will hopefully constitute an interesting thread that
winds its way through this book.
1.2 Reference Frames
To describe the kinematics of particles and rigid bodies, we presume on the ex
istence of a space with a set of three mutually perpendicular axes that meet at a
common point P. The set of axes and the point P constitute a reference frame. In
Newtonian mechanics, we also assume the existence of an inertial reference frame.
In this frame, the point P moves at a constant speed.
3
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
4 Kinematics of a Particle
Path of the particle
m
O
A
v
r(t)
r(t +t)
Figure 1.1. The path of a particle moving in E
3
. The position
vector, velocity vector, and areal velocity vector of this particle
at time t and the position vector of the particle at time t +t
are shown.
Depending on the application, it is often convenient to idealize the inertial
reference frame. For example, for ballistics problems, the Earth’s rotation and
the translation of its center are ignored and one assumes that a point, say E,
on the Earth’s surface can be considered as ﬁxed. The point E, along with three
orthonormal vectors that are ﬁxed to it (and the Earth), is then taken to approximate
an inertial reference frame. This approximate inertial reference frame, however,
is insufﬁcient if we wish to explain the behavior of Foucault’s famous pendulum
experiment. In this experiment from 1851, L´ eon Foucault (1819–1868) ingeniously
demonstrated the rotation of the Earth by using the motion of a pendulum.
∗
To
explain this experiment, it is sufﬁcient to assume the existence of an inertial frame
whose point P is at the ﬁxed center of the rotating Earth and whose axes do not
rotate with the Earth. As another example, when the motion of the Earth about the
Sun is explained, it is standard to assume that the center S of the Sun is ﬁxed and to
choose P to be this point. The point S is then used to construct an inertial reference
frame. Other applications in celestial mechanics might need to consider the location
of the point P for the inertial reference frame as the center of mass of the solar sys
tem with the three ﬁxed mutually perpendicular axes deﬁned by use of certain ﬁxed
stars [80].
For the purposes of this text, we assume the existence of a ﬁxed point O and
a set of three mutually perpendicular axes that meet at this point (see Figure 1.1).
The set of axes is chosen to be the basis vectors for a Cartesian coordinate system.
Clearly, the axes and the point O are an inertial reference frame. The space that
this reference frame occupies is a threedimensional space. Vectors can be deﬁned
in this space, and an inner product for these vectors is easy to construct with the dot
product. As such, we refer to this space as a threedimensional Euclidean space and
we denote it by E
3
.
∗
Discussions of his experiment and their interpretation can be found in [62, 138, 207]. Among his
other contributions [215], Foucault is also credited with introducing the term “gyroscope.”
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
1.3 Kinematics of a Particle 5
1.3 Kinematics of a Particle
Suppose a single particle of mass m is in motion in E
3
. The position vector of the
particle relative to a ﬁxed origin O is denoted by r (see Figure 1.1). In mechanics,
this vector is usually considered to be a function of time t: r = r(t).
The velocity v and acceleration a vectors of the particle are deﬁned to be the
respective ﬁrst and second time derivatives of the position vector:
v =
dr
dt
, a =
dv
dt
=
d
2
r
dt
2
.
It is crucial to note that, because r is measured relative to a ﬁxed origin, v and a are
the absolute velocity and acceleration vectors. By deﬁnition, the velocity vector can
be calculated from the following limit:
v(t) = lim
t→0
r (t +t) −r(t)
t
.
We also use an overdot to denote the time derivative: v = ˙ r and a = ¨ r.
Supplementary to the aforementioned kinematical quantities, we also have the
linear momentum G of the particle:
G = mv.
Further, the angular momentum H
O
of the particle relative to Ois
H
O
= r ×mv.
As we now show, this vector is related to the areal velocity vector A.
As used in celestial mechanics, the magnitude of the areal velocity vector is the
rate at which the position vector r of the particle sweeps out an area about the ﬁxed
point O (see, e.g., Moulton [150]). To establish an expression for this vector, we
consider the position vector of the particle at time t and t +t. Then, the area of the
parallelogram deﬁned by these vectors is r(t) ×r (t +t) (see Figure 1.1). This is
twice the area swept out by the particle during the interval t. Taking the limit of
the vector
r(t)×r(t+t)
2t
as t → 0 and using the fact that r(t) ×r(t) = 0, we arrive at
an expression for the areal velocity vector A(t):
A(t) = lim
t→0
r(t) ×r (t +t)
2t
=
1
2
r(t) ×
lim
t→0
r (t +t)
t
=
1
2
r(t) ×
lim
t→0
r (t +t) −r (t)
t
.
That is,
A =
1
2
r ×v. (1.1)
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
6 Kinematics of a Particle
The vector A plays an important role in several mechanics problems in which either
the angular momentum H
O
is constant or a component of H
O
is constant. Several
other examples of its use are discussed in the exercises at the end of this chapter.
Finally, we recall the deﬁnition of the kinetic energy T of the particle:
T =
1
2
mv · v.
The deﬁnitions of the kinematical quantities that have been introduced are inde
pendent of the coordinate system that is used for E
3
. In solving most problems, it is
crucial to have expressions for momenta and energies in terms of the chosen coor
dinate system. It is to this issue that we now turn.
1.4 Frequently Used Coordinate Systems
Depending on the problem of interest, there are several suitable coordinate sys
tems for E
3
. The most commonly used systems are Cartesian coordinates {x = x
1
,
y = x
2
, z = x
3
}, cylindrical polar coordinates {r, θ, z}, and spherical polar coordinates
{R, φ, θ}. All of these coordinate systems can be considered as speciﬁc examples of
a curvilinear coordinate system {q
1
, q
2
, q
3
} for E
3
, which we will discuss later on in
this chapter.
Cartesian Coordinate System
For the Cartesian coordinate system, a set of right–handed orthonormal vectors are
deﬁned: {E
1
, E
2
, E
3
}. Given any vector b in E
3
, this vector has the representation
b =
3
¸
i=1
b
i
E
i
.
For the position vector r, we also have
r =
3
¸
i=1
x
i
E
i
,
where {x
1
, x
2
, x
3
} are the Cartesian coordinates of the particle. Because E
i
are ﬁxed
in both magnitude and direction, their time derivatives are zero:
˙
E
i
= 0.
Cylindrical Polar Coordinates
A cylindrical polar coordinate system {r, θ, z} can be deﬁned by a Cartesian coordi
nate system as follows:
r =
x
2
1
+x
2
2
, θ = tan
−1
x
2
x
1
, z = x
3
,
where θ ∈ [0, 2π). Provided r = 0, then we can invert these relations to ﬁnd that
x
1
= r cos(θ), x
2
= r sin(θ), x
3
= z.
In other words, given (x
1
, x
2
, x
3
), a unique (r, θ, z) exists provided (x
1
, x
2
) = (0, 0).
Otherwise, when r = 0, the coordinate θ is ambiguous.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
1.4 Frequently Used Coordinate Systems 7
r
r
θ
z
O
e
r
e
θ
E
1
E
2
E
3
Figure 1.2. Cylindrical polar coordinates r, θ, and z.
Given a position vector r, we can write
r = x
1
E
1
+x
2
E
2
+x
3
E
3
= r(cos(θ)E
1
+sin(θ)E
2
) +zE
3
= re
r
+zE
3
,
where, as shown in Figure 1.2, e
r
= cos(θ)E
1
+sin(θ)E
2
.
It is convenient to deﬁne the set of unit vectors {e
r
, e
θ
, E
z
}:
e
r
= cos(θ)E
1
+sin(θ)E
2
, e
θ
= cos(θ)E
2
−sin(θ)E
1
, e
z
= E
3
.
We also notice that ˙ e
r
=
˙
θe
θ
, whereas ˙ e
θ
= −
˙
θe
r
. We should also verify that
{e
r
, e
θ
, E
z
} is a righthanded orthonormal basis for E
3
.
∗
Spherical Polar Coordinates
A spherical polar coordinate system {R, φ, θ} can be deﬁned by a Cartesian coordi
nate system as follows:
R =
x
2
1
+x
2
2
+x
2
3
, θ = tan
−1
x
2
x
1
, φ = tan
−1
⎛
⎝
x
2
1
+x
2
2
x
3
⎞
⎠
,
where θ ∈ [0, 2π) and φ ∈ (0, π). Provided φ = 0 or π, we can invert these relations
to ﬁnd
x
1
= Rcos(θ) sin(φ), x
2
= Rsin(θ) sin(φ), x
3
= Rcos(φ).
Given a position vector r, we can now write
r = x
1
E
1
+x
2
E
2
+x
3
E
3
= Rsin(φ)(cos(θ)E
1
+sin(θ)E
2
) +Rcos(φ)E
3
= Re
R
,
where, as shown in Figure 1.3, e
R
= sin(φ) cos(θ)E
1
+sin(φ) sin(θ)E
2
+cos(φ)E
3
.
∗
A basis {p
1
, p
2
, p
3
} is righthanded if p
3
· (p
1
×p
2
) > 0 and is orthonormal if the magnitude of each
of the vectors p
i
is 1 and they are mutually perpendicular: p
1
· p
2
= 0, p
2
· p
3
= 0, and p
1
· p
3
= 0.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
8 Kinematics of a Particle
r
r
θ
φ
O
e
θ
e
R
e
φ
E
1
E
2
E
3
Figure 1.3. The spherical polar coordinates φ and θ.
For future purposes, it is convenient to deﬁne the righthanded orthonormal set
of vectors {e
R
, e
φ
, e
θ
}:
⎡
⎢
⎣
e
R
e
φ
e
θ
⎤
⎥
⎦
=
⎡
⎢
⎣
cos(θ) sin(φ) sin(θ) sin(φ) cos(φ)
cos(θ) cos(φ) sin(θ) cos(φ) −sin(φ)
−sin(θ) cos(θ) 0
⎤
⎥
⎦
⎡
⎢
⎣
E
1
E
2
E
3
⎤
⎥
⎦
.
To establish the relations between these vectors and those deﬁned earlier, we ﬁrst
calculate the intermediate relations
⎡
⎢
⎣
e
r
e
θ
E
3
⎤
⎥
⎦
=
⎡
⎢
⎣
cos(θ) sin(θ) 0
−sin(θ) cos(θ) 0
0 0 1
⎤
⎥
⎦
⎡
⎢
⎣
E
1
E
2
E
3
⎤
⎥
⎦
,
⎡
⎢
⎣
e
R
e
φ
e
θ
⎤
⎥
⎦
=
⎡
⎢
⎣
sin(φ) 0 cos(φ)
cos(φ) 0 −sin(φ)
0 1 0
⎤
⎥
⎦
⎡
⎢
⎣
e
r
e
θ
E
3
⎤
⎥
⎦
. (1.2)
These results enable us to transform among the three distinct sets of basis vectors.
As with the cylindrical polar coordinate system, the basis vectors we deﬁned for
the spherical polar coordinate system vary with the coordinates. Indeed, assuming
that θ and φ are functions of time, a series of long calculations using (1.2) reveals
that
⎡
⎢
⎣
˙ e
R
˙ e
φ
˙ e
θ
⎤
⎥
⎦
=
⎡
⎢
⎣
0
˙
φ
˙
θ sin(φ)
−
˙
φ 0
˙
θ cos(φ)
−
˙
θ sin(φ) −
˙
θ cos(φ) 0
⎤
⎥
⎦
⎡
⎢
⎣
e
R
e
φ
e
θ
⎤
⎥
⎦
. (1.3)
These relations have an interesting form: Notice that the matrix in (1.3) is skew
symmetric. We shall see numerous examples of this later on when we discuss rota
tions and their time derivatives. Our later discussion should allow us to verify (1.3)
rather easily.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
1.5 Curvilinear Coordinates 9
1.5 Curvilinear Coordinates
The preceeding examples of coordinate systems can be considered as speciﬁc ex
amples of a curvilinear coordinate system. The development of the vector calculus
associated with such a system will be the focal point of this section of the book.
Curvilinear coordinate systems have featured prominently in all areas of mechan
ics, and the material presented here has a wide range of applications. Most of our
discussion is based on classical works and can be found in various textbooks on ten
sor calculus. Of these books, the one closest in spirit (and notation) to our treatment
here is that of Simmonds [198]; [139, 201] are also recommended.
Consider a curvilinear coordinate system {q
1
, q
2
, q
3
} that is deﬁned by the
functions
q
1
= ˆ q
1
(x
1
, x
2
, x
3
) ,
q
2
= ˆ q
2
(x
1
, x
2
, x
3
) ,
q
3
= ˆ q
3
(x
1
, x
2
, x
3
) . (1.4)
We assume that the functions ˆ q
i
are locally invertible:
x
1
= ˆ x
1
q
1
, q
2
, q
3
,
x
2
= ˆ x
2
q
1
, q
2
, q
3
,
x
3
= ˆ x
3
q
1
, q
2
, q
3
. (1.5)
This invertibility implies that, given the curvilinear coordinates of any point in E
3
,
there is a unique set of Cartesian coordinates for this point and vice versa. Usually,
the invertibility breaks down at several points in E
3
. For instance, the cylindrical
polar coordinate θ is not uniquely deﬁned when x
2
1
+x
2
2
= 0. This set of points
corresponds to the x
3
axis.
Assuming invertibility, and ﬁxing the value of one of the curvilinear coordi
nates, q
1
say, to equal q
1
0
, we can determine the values of x
1
, x
2
, and x
3
such that the
equation
q
1
0
= ˆ q
1
(x
1
, x
2
, x
3
)
is satisﬁed. The union of all the points represented by these Cartesian coordinates
deﬁnes a surface that is known as the q
1
coordinate surface (cf. Figure 1.4). If we
move on this surface we ﬁnd that the coordinates q
2
and q
3
will vary. Indeed, the
curves on the q
1
coordinate surface that we ﬁnd by varying q
2
while keeping q
3
ﬁxed are known as q
2
coordinate curves.
More generally, the surface corresponding to a constant value of a coordinate
q
j
is known as a q
j
coordinate surface. Similarly, the curve we obtain by varying the
coordinate q
k
while ﬁxing the remaining two curvilinear coordinates is known as a
q
k
coordinate curve.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
10 Kinematics of a Particle
O
q
2
coordinate curve
q
3
coordinate curve
q
1
coordinate surface
a
2
a
3
a
1
S
Figure 1.4. An example of a q
1
coordinate surface S. At a point on this surface, a
1
is normal
to the surface, and a
2
and a
3
are tangent to the surface. The q
1
coordinate surface S is foliated
by curves of constant q
2
and q
3
.
Covariant Basis Vectors
Again assuming invertibility, we can express the position vector r of any point as a
function of the curvilinear coordinates:
r =
3
¸
i=1
ˆ x
i
q
1
, q
2
, q
3
E
i
.
It is also convenient to deﬁne the covariant basis vectors a
1
, a
2
, and a
3
:
a
i
=
∂r
∂q
i
=
3
¸
k=1
∂ ˆ x
k
∂q
i
E
k
.
Mathematically, when we take the derivative with respect to q
2
we ﬁx q
1
and q
3
;
consequently, a
2
points in the direction of increasing q
2
. As a result, a
2
is tangent to
a q
2
coordinate curve. In general, a
i
is tangent to a q
i
coordinate curve.
You should notice that we can express the relationship between the covariant
basis vectors and the Cartesian basis vectors in a matrix form:
⎡
⎢
⎣
a
1
a
2
a
3
⎤
⎥
⎦
=
⎡
⎢
⎢
⎣
∂ ˆ x
1
∂q
1
∂ ˆ x
2
∂q
1
∂ ˆ x
3
∂q
1
∂ ˆ x
1
∂q
2
∂ ˆ x
2
∂q
2
∂ ˆ x
3
∂q
2
∂ ˆ x
1
∂q
3
∂ ˆ x
2
∂q
3
∂ ˆ x
3
∂q
3
⎤
⎥
⎥
⎦
⎡
⎢
⎣
E
1
E
2
E
3
⎤
⎥
⎦
.
It is a good exercise to write out the matrix in the preceding equation for various ex
amples of curvilinear coordinate systems, for instance, cylindrical polar coordinates.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
1.5 Curvilinear Coordinates 11
Contravariant Basis Vectors
Curvilinear coordinate systems also have a second set of associated basis vectors:
{a
1
, a
2
, a
3
}. These vectors are known as the contravariant basis vectors. One method
of deﬁning them is as follows:
a
1
=
3
¸
i=1
∂ ˆ q
1
∂x
i
E
i
, a
2
=
3
¸
i=1
∂ ˆ q
2
∂x
i
E
i
, a
3
=
3
¸
i=1
∂ ˆ q
3
∂x
i
E
i
.
That is,
a
k
= ∇q
k
.
Geometrically, a
i
is normal to a q
i
coordinate surface. However, as in the case of the
covariant basis vectors, the contravariant basis vectors are not necessarily unit vec
tors, nor do they form an orthonormal basis for E
3
. Using the chain rule of calculus,
we can show that
a
i
· a
j
= δ
i
j
,
where δ
i
j
is the Kronecker delta. As discussed in the Appendix, δ
i
j
= 1 if i = j and is
0 otherwise. It is left as an exercise for the reader to show this result.
∗
Covariant and Contravariant Components
As {a
1
, a
2
, a
3
} and {a
1
, a
2
, a
3
} form bases for E
3
, any vector b can be described as
linear combinations of either sets of vectors:
b =
3
¸
i=1
b
i
a
i
=
3
¸
k=1
b
k
a
k
.
The components b
i
are known as the contravariant components, and the compo
nents b
k
are known as the covariant components:
b · a
i
=
3
¸
k=1
b
k
a
k
· a
i
=
3
¸
k=1
b
k
δ
k
i
= b
i
,
b · a
i
=
3
¸
k=1
b
k
a
k
· a
i
=
3
¸
k=1
b
k
δ
i
k
= b
i
.
It is very important to note that b
k
= b · a
k
in general because a
i
· a
k
is not necessar
ily equal to δ
i
k
.
The trivial case in which x
i
= q
i
deserves particular mention. For this case,
r =
¸
3
k=1
x
i
E
i
. Consequently, a
i
= E
i
. In addition, a
i
= E
i
, and the covariant and
contravariant basis vectors are equal.
∗
The starting point for this exercise is to note that
∂x
k
∂x
j
= δ
j
k
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
12 Kinematics of a Particle
y
x
6
2
−4
−4
q
1
= −3
q
1
= −1
q
1
= 1
q
1
= 3
q
2
= −4
q
2
= 2
q
2
= 4
Figure 1.5. Projections of the q
1
and q
2
coordinate surfaces for coordinate system (1.6) on the
x–y plane.
An Example
Although we have met three examples of curvilinear coordinate systems previously,
it is useful to introduce an example that features nonorthogonal basis vectors. Con
sider the following coordinate system for Euclidean threespace:
q
1
= y, q
2
= x −y
2
, q
3
= z. (1.6)
Here, x = x
1
, y = x
2
, and z = x
3
are Cartesian coordinates. Representative projec
tions of the coordinate surfaces for q
1
and q
2
are shown in Figure 1.5.
For this coordinate system, it is straightforward to invert (1.6) to see that x =
q
2
+
q
1
2
and y = q
1
. Thus,
r =
q
2
+
q
1
2
E
1
+q
1
E
2
+q
3
E
3
.
By taking the derivatives of this representation for r with respect to q
1
, q
2
, and q
3
,
we see that
a
1
= 2q
1
E
1
+E
2
, a
2
= E
1
, a
3
= E
3
.
This set of vectors comprises the covariant basis vectors. By taking the gradient of
q
i
, we ﬁnd the contravariant basis vectors:
a
1
= E
2
, a
2
= E
1
−2q
1
E
2
, a
3
= E
3
.
It is interesting to note that a
1
· a
2
= −2q
1
= 0. Further, a
1
and a
2
do not necessarily
have unit magnitudes. By way of illustration, a q
1
coordinate curve is shown in
Figure 1.6. The vector a
1
is tangent to this curve, and a
2
and a
3
are normal to
this curve. To emphasize that a
1
is not necessarily parallel to a
1
, a q
1
coordinate
surface is also shown in the ﬁgure. It is left as an exercise for the reader to
illustrate the tangent vectors a
2
and a
3
to the q
1
coordinate surface shown in the
ﬁgure.
Some Comments on Derivatives
Several partial derivatives of functions (q
1
, q
2
, q
3
, ˙ q
1
, ˙ q
2
, ˙ q
3
, t) play a prominent
role in this book. When taking the partial derivative of this function with respect
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
1.5 Curvilinear Coordinates 13
q
1
coordinate curve
q
1
coordinate surface
a
1
a
1
a
1
a
2
a
3
Figure 1.6. A q
1
coordinate curve and a q
1
coordinate surface showing representative exam
ples of normal (a
2
and a
3
) vectors and tangent (a
1
) vectors to the curve. Note that a
1
is normal
to the q
1
coordinate surface and a
1
is not parallel to a
1
.
to q
2
say, we assume that t, q
1
, q
3
, and ˙ q
k
are constant. A related remark holds
for the partial derivatives with respect to the velocities ˙ q
j
and time t. That is,
∂q
k
∂q
j
= δ
k
j
,
∂ ˙ q
k
∂q
j
= 0,
∂t
∂q
j
= 0, (1.7)
and
∂q
k
∂ ˙ q
j
= 0,
∂ ˙ q
k
∂ ˙ q
j
= δ
k
j
,
∂t
∂ ˙ q
j
= 0. (1.8)
In all these equations, j and k range from 1 to 3. You may have noticed that (1.7)
1
was used in our calculations of a
i
.
It is easy to be confused about the distinction between the derivative
d
dt
and the
derivative
∂
∂t
. The former derivative assumes that q
i
and ˙ q
i
are functions of time,
whereas the latter assumes that they are constant:
˙
=
d
dt
=
3
¸
i=1
∂
∂q
i
dq
i
dt
+
3
¸
k=1
∂
∂ ˙ q
k
d
2
q
k
dt
2
+
∂
∂t
.
For example, consider the function
= q
1
+( ˙ q
3
)
2
+10t.
Then,
∂
∂q
1
= 1,
∂
∂ ˙ q
3
= 2˙ q
3
,
∂
∂t
= 10,
˙
= ˙ q
1
+2˙ q
3
¨ q
3
+10.
It should be clear from this example that
˙
=
∂
∂t
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
14 Kinematics of a Particle
1.6 Representations of Particle Kinematics
We now turn to establishing expressions for the position, velocity, and acceleration
vectors of a particle in terms of the coordinate systems just mentioned. First, for the
position vector we have
∗
r = x
1
E
1
+x
2
E
2
+x
3
E
3
= re
r
+zE
3
= Re
R
=
3
¸
i=1
ˆ x
i
q
1
, q
2
, q
3
E
i
.
Differentiating these expressions, we ﬁnd
v = ˙ x
1
E
1
+ ˙ x
2
E
2
+ ˙ x
3
E
3
= ˙ re
r
+r
˙
θe
θ
+ ˙ zE
3
=
˙
Re
R
+R
˙
φe
φ
+Rsin(φ)
˙
θe
θ
=
3
¸
i=1
˙ q
i
a
i
. (1.9)
Notice the simplicity of the expression for v when expressed in terms of the covariant
basis vectors. For any given curvilinear coordinate system, if we write the position
vector as a function of the coordinates q
1
, q
2
, and q
3
, and then differentiate and
compare the result with v =
¸
3
i=1
˙ q
i
a
i
, we can read off the covariant basis vectors.
For instance, (1.9)
2
implies that a
1
= e
r
, a
2
= re
θ
, and a
3
= E
3
for the cylindrical
polar coordinate system.
A further differentiation yields
a = ¨ x
1
E
1
+ ¨ x
2
E
2
+ ¨ x
3
E
3
= ( ¨ r −r
˙
θ
2
)e
r
+(r
¨
θ +2˙ r
˙
θ)e
θ
+ ¨ zE
3
= (
¨
R−R
˙
φ
2
−Rsin
2
(φ)
˙
θ
2
)e
R
+(R
¨
φ +2
˙
R
˙
φ −Rsin(φ) cos(φ)
˙
θ
2
)e
φ
+(Rsin(φ)
¨
θ +2
˙
R
˙
θ sin(φ) +2R
˙
θ
˙
φcos(φ))e
θ
=
3
¸
i=1
¨ q
i
a
i
+
3
¸
i=1
3
¸
j=1
˙ q
i
˙ q
j
∂a
i
∂q
j
.
We obtain the ﬁnal representation for ¨ r after noting that a
i
depend on the curvilinear
coordinates, which in turn are functions of time: ˙ a
k
=
¸
3
i=1
∂a
k
∂q
i
˙ q
i
.
It is left as an exercise for the reader to establish expressions, using various
coordinate systems, for the linear momentum G and the angular momentum H
O
.
∗
Notice that it is a mistake to assume that r =
¸
3
i=1
q
i
a
i
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
1.7 Constraints 15
The kinetic energy T of the particle has a rather elegant representation using the
curvilinear coordinates:
T =
m
2
v · v
=
m
2
3
¸
i=1
˙ q
i
a
i
·
3
¸
k=1
˙ q
k
a
k
=
3
¸
i=1
3
¸
k=1
m
2
a
ik
˙ q
i
˙ q
k
, (1.10)
where
a
ik
= a
ki
= a
k
· a
i
.
It is also a good exercise to compute a
ik
for a spherical polar coordinate system, and
then, with the help of the representation T =
¸
3
i=1
¸
3
k=1
m
2
a
ik
˙ q
i
˙ q
k
, show that
T =
m
2
˙
R
2
+R
2
˙
φ
2
+R
2
sin
2
(φ)
˙
θ
2
.
The exercises at the end of this chapter feature this result for other coordinate
systems.
1.7 Constraints
A constraint is a kinematical restriction on the motion of the particle. They are
introduced in problems involving a particle in three manners: either as simplifying
assumptions, prescribed motions, or because of rigid connections. The constraints
on the motion of a particle dictate, to a large extent, the coordinate system used
to solve the problem of determining the motion of the particle. In this section, we
examine the simplest class of constraints on the motion of a particle. Later, these
constraints will be classiﬁed as integrable.
Classical Examples
Consider the four mechanical systems shown in Figure 1.7. The ﬁrst system is known
as the spherical pendulum. Here, a particle of mass m is attached by a rigid rod of
length L
0
to a ﬁxed point O. The constraint on the motion of the particle in this
system can be written as
r · e
R
= L
0
.
By differentiating this equation, we see that the velocity vector satisﬁes the relation
v · e
R
= 0. The second system we consider is the planar pendulum. Again, the parti
cle is attached by a rigid rod of length L
0
to a ﬁxed point O, but it is also assumed to
move on a vertical plane. The constraints on the motion of the particle are
r · e
r
= L
0
, r · E
3
= 0.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
16 Kinematics of a Particle
(a)
(b)
(c)
(d)
spinning conical surface
moving plane
m
m
m
m
α
O
O
O
O
L
0
L
0
e
r
E
1
E
1
E
1
E
2
E
2
E
2
E
3
E
3
E
3
Figure 1.7. Four mechanical systems featuring constraints on the motion of a particle: (a) the
spherical pendulum, (b) the planar pendulum, (c) a particle moving on a plane, and (d) a
particle moving on a spinning cone.
After differentiating these equations with respect to time, we observe that the ve
locity vector of the particle has a component only in the e
θ
direction: v · e
r
= 0 and
v · E
3
= 0. The third system involves a particle moving on a horizontal surface that
is moving with a velocity vector
˙
f (t)E
3
. The constraint on the motion of the particle
is
r · E
3
= f (t).
The ﬁnal system of interest consists of a particle moving on a spinning cone. The
constraint on the motion of the particle can be most easily described with the help
of a spherical polar coordinate system:
φ +α(t) −
π
2
= 0.
For all four systems, we have selected a coordinate system in which the constraint(s)
on the motion of the particle is easily described.
A Particle Moving on a Surface
Turning to the more general case, consider a particle constrained to move on a sur
face. With the help of a single smooth function (r, t), we assume that the constraint
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
1.7 Constraints 17
O
Surface = 0
∇
m
r
Figure 1.8. A particle moving on a surface = 0. The particle in this case is subject to a single
constraint.
can be described in a standard (canonical) form:
(r, t) = 0.
At each instant in time, this equation can be interpreted as a single condition on the
three independent Cartesian coordinates of the particle. Thus the condition = 0
deﬁnes a twodimensional surface (see Figure 1.8).
The unit normal vector n to this surface is parallel to ∇ = grad() (see
Figure 1.8). Depending on the coordinate system used, this vector ∇ has
numerous representations:
∇ =
∂
∂r
=
3
¸
i=1
∂
∂x
i
E
i
=
3
¸
i=1
∂
∂q
i
a
i
=
∂
∂r
e
r
+
1
r
∂
∂θ
e
θ
+
∂
∂z
E
3
=
∂
∂R
e
R
+
1
R
∂
∂φ
e
φ
+
1
Rsin(φ)
∂
∂θ
e
θ
. (1.11)
You should notice how simple the expression for the gradient is in curvilinear
coordinates.
∗
A simple differentiation of the function helps to provide the restriction it
imposes on the velocity vector:
˙
=
∂
∂r
· v +
∂
∂t
.
∗
To establish this result, we note that
˙
q
1
, q
2
, q
3
= ∇· v and
˙
q
1
, q
2
, q
3
=
¸
3
i=1
∂
∂q
i
˙ q
i
. Substi
tuting for v and comparing both expressions, we arrive at (1.11)
2
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
18 Kinematics of a Particle
O
Surface
1
= 0
Surface
2
= 0
∇
1
∇
2
m
r
t
Figure 1.9. A particle subject to two constraints. The dotted curve in this ﬁgure corresponds
to the curve of intersection of the surfaces
1
= 0 and
2
= 0, and the vector t is the unit
tangent vector to this curve.
However, if r satisﬁes the constraint, then (r, t) = 0 and
˙
= 0. Consequently, the
constraint = 0 implies that the velocity vector satisﬁes the restriction
∂
∂r
· v +
∂
∂t
= 0.
This result will be important in our discussion of the mechanical power of the con
straint forces.
A Particle Moving on a Curve
We now consider the more complex case of a particle moving on a curve. A curve
can be deﬁned by the intersection of two surfaces. Using the previous developments,
we consider the condition that the particle move on the curve to be equivalent to
two (simultaneous) constraints:
1
(r, t) = 0,
2
(r, t) = 0.
This situation is shown in Figure 1.9. The normal vectors to the two surfaces at a
point of their intersection are assumed not to be parallel: ∇
1
×∇
2
= 0. That is,
the two constraints
1
= 0 and
2
= 0 are assumed to be independent.
Once
1
and
2
are given, then expressions for the two normal vectors to the
curve can be readily established. We also note that deriving the restrictions these
constraints impose on the velocity vector follows from the corresponding results for
a single constraint:
∂
1
∂r
· v +
∂
1
∂t
= 0,
∂
2
∂r
· v +
∂
2
∂t
= 0. (1.12)
If the curve is ﬁxed, then
1
and
2
are not explicit functions of time. In this case,
(1.12) can be used to show the expected result that v is tangent to the curve.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
1.7 Constraints 19
A Particle Whose Motion is Prescribed
The case in which the motion is prescribed can be interpreted as a particle lying at
the intersection of three known surfaces. In other words, the particle is subject to
three constraints:
1
(r, t) = 0,
2
(r, t) = 0,
3
(r, t) = 0.
We assume that these constraints are independent and thus their normal vectors at
their intersection point form a basis for E
3
:
∇
3
· (∇
1
×∇
2
) = 0.
The three conditions
i
= 0 can also be interpreted as three equations for the three
components of r.
The two primary situations in which a particle is subject to three constraints
arise when either the motion of the particle is completely controlled or the particle
is subject to static friction and is therefore in a state of rest relative to a curve or
surface.
Coordinates and Constraints
The constraints we have considered on the motion of the particle have been de
scribed in terms of surfaces that the motion of the particle is restricted to. These
surfaces can be described in terms of the coordinate system used for E
3
. The de
scription is greatly facilitated by a judicious choice of coordinates. For instance, if
a particle is constrained to move on a ﬁxed plane, then we can always choose the
origin O and the Cartesian coordinates such that the constraint is easily described
by the equation x
3
= constant. Similarly, if a particle is constrained to move on a
sphere, then spherical polar coordinates are an obvious choice.
The more sophisticated the surfaces that the particle is constrained to move
on, then the more difﬁcult it becomes to choose an appropriate coordinate system.
Help is at hand: The surfaces (r) = 0 of interest in this book can be described in
an appropriate curvilinear coordinate system by a simple equation, q
3
= constant.
Furthermore, a moving surface (r, t) = 0 can, in principle, be described by the
equation q
3
= f (t), where f is a function of time t. For example, suppose a particle
is moving on a sphere whose radius is a known function R
0
(t). Then the constraint
that the particle move on the sphere is simply described by
R = R
0
(t).
Here, we are choosing the spherical polar coordinate system to be our coordinate
system.
The Classical Examples Revisited
Returning to the four mechanical systems shown in Figure 1.7, you should convince
yourself that the constraint(s) on the motions of the particle in these systems are
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
20 Kinematics of a Particle
individually of the form = 0. Speciﬁcally, for the spherical pendulum,
= r · e
R
−L
0
.
That is, we may imagine the particle in a spherical pendulum as moving on a sphere.
For the planar pendulum, we have
1
= r · e
r
−L
0
,
2
= r · E
3
= 0.
In this case, the particle can be visualized as moving on the intersection of a cylinder
of radius L
0
and a horizontal plane. This intersection deﬁnes a circle. If the rod’s
length L
0
changes with time, then the circle’s radius also changes. For the particle
moving on the horizontal surface,
= r · E
3
− f (t).
Notice that in this example = (r, t).
For the ﬁnal system, the particle moving on a cone, the constraint on the motion
of the particle can be represented by = 0, where
(r) = φ +α −
π
2
.
If the cone were moving in a manner such that α = α(t), then the function =
(r, t). For example, suppose α = α
0
+Asin(ωt); then
(r, t) = φ −
π
2
+α
0
+Asin(ωt).
You should verify that
˙
= 0, but
∂
∂t
= Aωcos(ωt). The spinning of the cone has
purposefully not been mentioned. This motion will feature in any formulation of the
friction forces acting on the particle. Further, in the event that the particle is stuck
to the cone, then the particle will be subject to three constraints. This situation is
discussed in Section 2.9.
1.8 Classiﬁcation of Constraints
All of the constraints discussed so far can be individually written in the form
(r, t) = 0.
Thus they are often known as positional constraints. We now deﬁne a further type
of constraint:
π = 0, (1.13)
where
π = f · v +e,
and f = f(r, t) and e = e(r, t). The constraint π = 0 does not restrict the position of
the particle – it restricts only its velocity vector. Consequently, the constraint π = 0
is often known as a velocity constraint.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
1.8 Classiﬁcation of Constraints 21
As we demonstrated earlier, every constraint of the form (r, t) = 0 can be dif
ferentiated to yield a restriction on the velocity vector:
∂
∂r
· v +
∂
∂t
= 0.
This restriction is of the form (1.13). Thus every constraint (r, t) = 0 provides a
constraint f · v +e = 0. However, the converse is not true.
A constraint π = 0 that can be integrated to yield a constraint of the form
(r, t) = 0 is said to be an integrable (or holonomic) constraint. More precisely,
given a constraint π = 0, if we can ﬁnd an integrating factor k = k(r, t) and a func
tion (r, t), such that
∗
k(f · v +e) = ∇· v +
∂
∂t
,
then the constraint π = 0 is said to be integrable. Otherwise, the constraint π =
0 is said to be nonintegrable (or nonholonomic). The terminology here dates to
Heinrich Hertz [92] (1857–1894). As noted by Lanczos [124], integrable constraints
were further classiﬁed by Ludwig Boltzmann (1844–1906) as rheonomic when =
(r, t) and scleronomic when = (r) (i.e., when is not an explicit function of
time t).
The distinction between integrable and nonintegrable constraints becomes par
ticularily important when rigid bodies are concerned. However, for pedagogical pur
poses, it is desirable to introduce them when discussing single particles. We shall
shortly discuss the forces needed to enforce the constraints: Such forces are known
as constraint forces. To explore the differences between integrable and noninte
grable constraints, it is best to ﬁrst consider some examples. Following such an
exploration, we shall discuss known criteria to determine whether or not a set of
constraints is integrable.
Three Examples
As a ﬁrst example, we suppose that the particle is subject to the constraints
xy −c = 0, z = 0.
That is, the particle is constrained to move on a hyperbola in the x −y plane (see
Figure 1.10). Two points A and B are also shown in this ﬁgure, and it is important
to notice that it is not possible for the particle to move between A and B without
violating the constraint xy −c = 0. The constraints xy −c = 0 and z = 0 imply the
velocity constraints:
(xE
2
+yE
1
) · v = 0, E
3
· v = 0. (1.14)
These conditions imply that v has no component normal to the hyperbola xy =
c. Constraints (1.14) are both clearly of the form f · v +e = 0, where e = 0 and
∗
Further background on integrating factors can be found in most texts on differential equations or
differential forms, see, e.g., [61, 64, 114]. It is well known that integrating factors are not unique.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
22 Kinematics of a Particle
A
B
x = x
1
y = x
2
Figure 1.10. The motion of a particle subject to the con
straints xy = c and z = 0, where c is a positive constant.
The arrows shown on the hyperbolae indicate the possi
ble directions of motion of the particle.
A
B
x = x
1
y = x
2
Figure 1.11. The motion of a particle subject to the
constraints ˙ yx = 0 and z = 0. The arrows indicate the
possible directions of motion of the particle.
f = xE
2
+yE
1
, and e = 0 and f = E
3
, respectively. By construction, constraints
(1.14) are both integrable.
As a second example, let us examine the following constraints:
(xE
2
) · v = 0, z = 0. (1.15)
The motions of the particle that satisfy these constraints are shown in Figure 1.11.
Notice that it is possible to move between any two points Aand B on the x–y plane
without violating the constraint ˙ yx = 0. The restriction this constraint places is that
it restricts howone can go fromany Ato any B. This is in marked contrast to the con
straint xy −c = 0. By multiplying ˙ yx = 0 by
1
x
, for example, we see that, away from
the y axis, this constraint is integrable.
∗
Considering the possible motions shown
in Figure 1.11, it is not surprising to note that we cannot ﬁnd a smooth function
to conclude that the constraint ˙ yx = 0 is integrable throughout the entirety of E
3
.
∗
Here,
1
x
is an example of the integrating factor k(r, t) mentioned earlier in the deﬁnition of a non
integrable constraint.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
1.8 Classiﬁcation of Constraints 23
Instead, we classify the constraint ˙ yx = 0 as a piecewiseintegrable constraint.
∗
We
shall discuss further unusual aspects of this constraint in Section 2.10.
Our third example is the simplest possible nonintegrable constraint on the mo
tion of a particle.
†
The constraint is
(−zE
1
+E
2
) · v = 0. (1.16)
That is, ˙ y −z˙ x = 0. To demonstrate the type of restrictions ˙ y −z˙ x = 0 imposes, we
choose two points Aand Band use the x coordinate to parameterize a path between
them. Choosing
y = f (x), z =
df
dx
,
where f (x) is any sufﬁciently smooth function, we observe that the constraint −z˙ x +
˙ y = 0 is satisﬁed. In order that the particle be able to move between any two points
Ato B, f (x) is subject to the restrictions
y
A
= f (x
A
), z
A
=
df
dx
(x
A
), y
B
= f (x
B
), z
B
=
df
dx
(x
B
),
where r
A
= x
A
E
1
+y
A
E
2
+z
A
E
3
and r
B
= x
B
E
1
+y
B
E
2
+z
B
E
3
. Graphically con
structing a function f (x) that meets these restrictions is not difﬁcult, and some ex
amples are presented in Figure 1.12. A speciﬁc example of f (x) is discussed in Pars
[170], and a slightly modiﬁed version of it is presented here:
f (x) = (3 (y
B
−y
A
) −(x
B
−x
A
) (z
B
+z
A
))
x −x
A
x
B
−x
A
2
− (2 (y
B
−y
A
) −(x
B
−x
A
) (z
B
+z
A
))
x −x
A
x
B
−x
A
3
+c (x −x
A
)
2
(x
B
−x)
2
+dsin
2
π(x −x
A
)
x
B
−x
A
+z
A
(x −x
A
)
x
B
−x
x
B
−x
A
+y
A
, (1.17)
where c and d are arbitrary constants. It is left as an exercise for the reader to verify
that an inﬁnite number of paths between Aand Bare possible without violating the
constraint ˙ y −z˙ x = 0. Shortly, we shall verify that this constraint is indeed noninte
grable.
∗
With the exception of that of Papastavridis [169], this classiﬁcation is not typically mentioned in
the textbooks on classical and analytical mechanics. Further discussion of piecewiseintegrable con
straints can be found in the interesting paper by Ruina [186]. As discussed in his paper, constraints
of this type also arise in many locomotive systems such as passive walking machines that feature
impact.
†
A proof of this statement can be found in Section 163 of Forsyth [64]. Our discussion of constraint
(1.16) is based on the treatments presented in Goursat [75] and Pars [170].
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
24 Kinematics of a Particle
A
B
f
v
O
E
1
E
2
E
3
Figure 1.12. Three possible motions be
tween two given points Aand Bof a par
ticle subject to the constraint ˙ y −z˙ x =
0. The arrows indicate the directions of
motion of the particle and the vector
f = −zE
1
+E
2
. The motions presented
in this ﬁgure were constructed with the
assistance of (1.17).
Integrability Criteria
Suppose a constraint π = 0 is imposed on the motion of the particle. As mentioned
earlier, this constraint is integrable if we can ﬁnd a function (r, t) and an integrat
ing factor k such that
˙
= k(f · v +e) . (1.18)
Otherwise, the constraint f · v +e = 0 is nonintegrable. It is desirable to know if a
constraint is integrable, because we can then, in principle, ﬁnd a coordinate system
{q
1
, q
2
, q
3
} such that f · v +e = 0 is equivalent to the constraint ˙ q
3
+e = 0. The lat
ter constraint in turn is equivalent to the constraint q
3
= g, where ˙ g = e. Using this
coordinate system, the dynamics of the particle is easier to analyze. With this in
mind, several classical integrability criteria are now presented for single and multi
ple constraints.
∗
A SINGLE SCLERONOMIC CONSTRAINT. The ﬁrst criterion we examine pertains to con
straints of the form f · v = 0, where f is not an explicit function of time. Using a
coordinate system{q
1
, q
2
, q
3
}, we can write the constraint π = 0 in the form
f
1
˙ q
1
+ f
2
˙ q
2
+ f
3
˙ q
3
= 0.
A necessary and sufﬁcient condition for f · v = 0 to be integrable is that
†
I
c
= 0 (1.19)
∗
For additional discussion and illustrative examples from mechanics, the texts of Papastavridis [169]
and Rosenberg [182] are recommended. Here, the historical remarks on these criteria are based on
the paper by Hawkins [91].
†
Classical proofs of this result can be found in Section 151 of Forsyth [64], Section 442 of Goursat
[74], and in Papastavridis [169]. A proof featuring differential forms can be found in Flanders [61],
who refers to this result as Frobenius’ integration theorem.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
1.8 Classiﬁcation of Constraints 25
for all possible choices of q
i
. Here,
I
c
= f
1
∂f
3
∂q
2
−
∂f
2
∂q
3
+ f
2
∂f
1
∂q
3
−
∂f
3
∂q
1
+ f
3
∂f
2
∂q
1
−
∂f
1
∂q
2
.
It is convenient to recall at this point the expression for the curl of a vector ﬁeld P
in Cartesian coordinates:
curl(P) =
3
¸
i=1
∂
∂x
i
E
i
×P
=
∂P
3
∂x
2
−
∂P
2
∂x
3
E
1
+
∂P
1
∂x
3
−
∂P
3
∂x
1
E
2
+
∂P
2
∂x
1
−
∂P
1
∂x
2
E
3
, (1.20)
where P
i
= P · E
i
. With the help of this expression, it is easy to see that criterion
(1.19) can also be expressed in the compact form
f · (curl (f)) = 0.
We refer to (1.19) as Jacobi’s criterion after its discoverer Carl G. J. Jacobi (1804–
1851).
Satisfaction of (1.19) does not tell us what (r) or k(r) are; it indicates only that
these functions exist. Further, this criterion is local – it does not tell us if these two
functions are the same for each point in space. For instance, although the constraints
x˙ y +y ˙ x = 0 and x˙ y = 0 trivially satisfy integrability criterion (1.19), only the former
has a continuously deﬁned (r). The function (r) for the latter constraint can be
deﬁned in only a piecewise manner (see Figure 1.11). If we use (1.19) to examine
the constraint ˙ y −z˙ x = 0, then we ﬁnd that
∗
I
c
= −z(0 −0) +1 (−1 −0) +0 (0 −0) .
As I
c
= −1 = 0, the constraint ˙ y −z˙ x = 0 is nonintegrable.
A SINGLE RHEONOMIC CONSTRAINT. It is clearly of interest to present the general
ization of Jacobi’s criterion to rheonomic constraints: f · v +e = 0. The result is very
similar in formto that for a scleronomic constraint, but it is more tedious to evaluate.
To proceed, we express the constraint π = 0 in the form
f
1
˙ q
1
+ f
2
˙ q
2
+ f
3
˙ q
3
+ f
4
= 0,
and deﬁne the variables
U
1
= q
1
, U
2
= q
2
, U
3
= q
3
, U
4
= t.
Clearly, f
4
= e. We next form the functions
I
JKL
= f
J
∂f
L
∂U
K
−
∂f
K
∂U
L
+ f
K
∂f
J
∂U
L
−
∂f
L
∂U
J
+ f
L
∂f
K
∂U
J
−
∂f
J
∂U
K
.
∗
That is, we choose q
1
= x, q
2
= y, and q
3
= z.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
26 Kinematics of a Particle
Here, the integer indices J , K, and L range from 1 to 4. A necessary and sufﬁcient
condition for the constraint π = 0 to be integrable is that the following four equa
tions hold for all q
1
, q
2
, q
3
, t:
I
JKL
= 0, for all J, K, L ∈ {1, 2, 3, 4}, L = J = K, K = L. (1.21)
For a proof of this theorem, the reader is referred to Section 161 of Forsyth [64] or
to Flanders [61].
SYSTEMS OF CONSTRAINTS. When particles are subject to several constraints, their
independence needs to be examined. For the case of two constraints, we ﬁrst express
them both in the form
f
1
· v +e
1
= 0,
f
2
· v +e
2
= 0.
If f
1
×f
2
= 0, then the constraints are said to be independent. For integrable con
straints, this is equivalent to the condition ∇
1
×∇
2
= 0. That is, the normal vec
tors to surfaces
1
= 0 and
2
= 0 are not parallel. The case of three constraints is
similar. We ﬁrst express each of them in the form
f
1
· v +e
1
= 0,
f
2
· v +e
2
= 0,
f
3
· v +e
3
= 0. (1.22)
Then, the condition for their independence is that
f
1
· (f
2
×f
3
) = 0.
If the constraints are integrable, then this condition is equivalent to ∇
1
·
(∇
2
×∇
3
) = 0. Geometrically, this means that the normal vectors at the point
of intersection of the surfaces
1
= 0,
2
= 0, and
3
= 0 form a basis.
The presence of more than one constraint can also imply that a system of con
straints that are individually nonintegrable can become integrable. The most well
known instance occurs when two scleronomic constraints are imposed on a particle
[169]:
f
1
· v = 0,
f
2
· v = 0, (1.23)
where the functions f
1
and f
2
are functions of r, and f
1
×f
2
= 0. In this case, the sys
tem of constraints is integrable. The proof of this result, which is presented in Sec
tion 8.4, uses a criterion that is due to Ferdinand G. Frobenius (1849–1917), which
we postpone discussion of until Chapter 8. This criterion can also be used to show
that if the three constraints (1.22) are imposed on a particle, then the system of con
straints is integrable. Consequently, the motion of the particle is prescribed. Other
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
Exercises 1.1–1.3 27
instances of multiple constraints on the motion of a particle are discussed in the
exercises at the end of this chapter.
1.9 Closing Comments
In this chapter, we have assembled many of the needed kinematical concepts and
tools needed to solve problems in particle dynamics. For most readers, the novel as
pects of the chapter will have been the discussion of curvilinear coordinates and
kinematical constraints. These two topics are intimately related and will feature
prominently in the forthcoming chapters.
EXERCISES
1.1. Consider a particle whose motion is described in Cartesian coordinates as
r(t) = cE
2
+10tE
1
,
where c is a constant. Determine the areal velocity vector A of the particle, and
show that the magnitude of this vector corresponds to the rate at which the particle
sweeps out a particular area. Does the particle sweep out equal areas during equal
periods of time? In your solution, you should also consider the case in which c = 0.
1.2. Consider a particle whose motion is described in cylindrical polar coordinates
as
r(t) = 10e
r
, θ(t) = ωt,
where ω = 0. Determine the areal velocity vector A of the particle. Under which
conditions does the particle sweep out equal areas during equal periods of time?
1.3. Recall that the cylindrical polar coordinates {r, θ, z} are deﬁned in Cartesian
coordinates {x = x
1
, y = x
2
, z = x
3
} by the relations
r =
x
2
1
+x
2
2
, θ = tan
−1
x
2
x
1
, z = x
3
.
Show that the covariant basis vectors associated with the curvilinear coordinate sys
tem, q
1
= r, q
2
= θ, and q
3
= z, are
a
1
= e
r
, a
2
= re
θ
, a
3
= E
3
.
In addition, show that the contravariant basis vectors are
a
1
= e
r
, a
2
=
1
r
e
θ
, a
3
= E
3
.
It is a good exercise to convince yourself with an illustration that a
2
is tangent to a
θ coordinate curve, whereas a
2
is normal to a θ coordinate surface. Finally, for this
coordinate system, show that
T =
m
2
˙ r
2
+r
2
˙
θ
2
+ ˙ z
2
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
28 Exercises 1.4–1.5
1.4. Recall that the spherical polar coordinates {R, φ, θ} are deﬁned in Cartesian
coordinates {x = x
1
, y = x
2
, z = x
3
} by the relations
R =
x
2
1
+x
2
2
+x
2
3
,
θ = tan
−1
x
2
x
1
,
φ = tan
−1
⎛
⎝
x
2
1
+x
2
2
x
3
⎞
⎠
.
Show that the covariant basis vectors associated with the curvilinear coordinate sys
tem, q
1
= R, q
2
= φ, and q
3
= θ, are
a
1
= e
R
, a
2
= Re
φ
, a
3
= Rsin(φ)e
θ
.
In addition, show that the contravariant basis vectors are
a
1
= e
R
, a
2
=
1
R
e
φ
, a
3
=
1
Rsin(φ)
e
θ
.
1.5. In the parabolic coordinate system, the coordinates {u, v, θ} can be deﬁned in
Cartesian coordinates {x = x
1
, y = x
2
, z = x
3
} by the relations
u = ±
x
3
+
x
2
3
+
x
2
1
+x
2
2
,
v = ±
−x
3
+
x
2
3
+
x
2
1
+x
2
2
,
θ = tan
−1
x
2
x
1
.
In addition, the inverse relations can be deﬁned:
x
1
= uv cos(θ), x
2
= uv sin(θ), x
3
=
1
2
(u
2
−v
2
).
(a) In the r–x
3
plane, where r is the cylindrical polar coordinate r =
x
2
1
+x
2
2
,
draw several representative examples of the projections of the u and v co
ordinate surfaces. You should give a sufﬁcient number of examples to con
vince yourself that u, v, and θ can be used as a coordinate system.
(b) In the x
1
–x
2
–x
3
space, draw a u coordinate surface. Illustrate how the v and
θ coordinate curves foliate this surface.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
Exercises 1.5–1.6 29
(c) Show that the covariant basis vectors for the parabolic coordinate system
are
a
1
=
∂r
∂u
= ve
r
+uE
3
,
a
2
=
∂r
∂v
= ue
r
−vE
3
,
a
3
=
∂r
∂θ
= uve
θ
.
Illuminate your results from (a) and (b) by drawing representative exam
ples of these vectors.
(d) Show that the contravariant basis vectors for the parabolic coordinate sys
tem are
a
1
= grad(u) =
1
u
2
+v
2
a
1
,
a
2
= grad(v) =
1
u
2
+v
2
a
2
,
a
3
= grad(θ) =
1
uv
e
θ
.
Again, illuminate your results from (a), (b), and (c) by drawing representa
tive examples of these vectors.
(e) Where are the singularities of the parabolic coordinate system? Verify that,
at these singularities, the contravariant basis vectors are not deﬁned.
(f) For a particle of mass m that is moving in E
3
, establish expressions for the
kinetic energy T and linear momentum G in terms of {u, v, θ} and their time
derivatives.
1.6. A classical problem is to determine the motion of a particle on a circular helix
(see Figure 1.13). In terms of the cylindrical polar coordinates r, θ, z, the equation
of the helix is
r = R
0
, z = αR
0
θ,
where R
0
and α are constants. Here, we use another curvilinear coordinate system
to deﬁne the motion of particle:
q
1
= θ, q
2
= r, q
3
= ν = z −αrθ.
A q
3
coordinate surface is known as a right helicoid.
(a) Show that the covariant basis vectors associated with this coordinate system
are
a
1
= r(e
θ
+αE
3
), a
2
= e
r
+αθE
3
, a
3
= E
3
.
Verify that the covariant basis vectors are not orthonormal.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
30 Exercises 1.6–1.8
Bead of mass m
g
E
1
E
2
E
3
Figure 1.13. A particle moving on a circular helix.
(b) Show that the kinetic energy of the particle has the representation
T =
m
2
(1 +α
2
θ
2
) ˙ r
2
+(1 +α
2
)r
2
˙
θ
2
+ ˙ ν
2
+
m
2
2˙ ν˙ rαθ +2˙ ν
˙
θαr +2˙ r
˙
θα
2
rθ
.
Calculate the contravariant basis vectors associated with the curvilinear co
ordinate system.
1.7. Consider the following curvilinear coordinate system:
q
1
= x
1
sec(α), q
2
= x
2
−x
1
tan(α), q
3
= x
3
,
where α is a constant. This coordinate system is one of the simplest instances
of a nonorthogonal coordinate system. Referring to (1.5), calculate the functions
ˆ x
k
(q
1
, q
2
, q
3
). Draw the coordinate curves and surfaces for the curvilinear coordi
nate system and then show that the covariant and contravariant basis vectors are
a
1
= cos(α)E
1
+sin(α)E
2
, a
2
= E
2
, a
3
= E
3
, (1.24)
and
a
1
= sec(α)E
1
, a
2
= −tan(α)E
1
+E
2
, a
3
= E
3
, (1.25)
respectively. Illustrate these vectors on the coordinate curves and surfaces you pre
viously drew. For which values of α is {a
1
, a
2
, a
3
} not a basis?
1.8. Given a vector
b = 10E
1
+5E
2
+6E
3
,
calculate its covariant b
i
and contravariant b
i
components when the covariant and
contravariant basis vectors are deﬁned by (1.24) and (1.25), respectively. In addition,
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
Exercises 1.8–1.9 31
verify that
b =
3
¸
i=1
b
i
a
i
=
3
¸
i=1
b
i
a
i
.
Furthermore, show that b =
¸
3
i=1
b
i
a
i
and b =
¸
3
i=1
b
i
a
i
. For which values of α is
¸
a
1
, a
2
, a
3
¸
not a basis?
1.9. This exercise illustrates how the covariant and contravariant components of a
vector are related. To start, we deﬁne the following scalars by using the covariant
and contravariant basis vectors:
a
ik
= a
ik
(q
r
) = a
i
· a
k
, a
ik
= a
ik
(q
r
) = a
i
· a
k
.
You should notice that a
ik
= a
ki
and a
ik
= a
ki
. The indices i, k, r, and s in this prob
lem range from 1 to 3.
(a) For any vector b, show that the covariant and contravariant components are
related:
b
i
=
3
¸
k=1
a
ik
b
k
, b
i
=
3
¸
k=1
a
ik
b
k
.
In other words, the covariant components are linear combinations of the
contravariant components and vice versa. Using a matrix notation, these
results can be expressed as
⎡
⎢
⎣
b
1
b
2
b
3
⎤
⎥
⎦
=
⎡
⎢
⎣
a
11
a
12
a
13
a
21
a
22
a
23
a
31
a
32
a
33
⎤
⎥
⎦
⎡
⎢
⎣
b
1
b
2
b
3
⎤
⎥
⎦
,
⎡
⎢
⎣
b
1
b
2
b
3
⎤
⎥
⎦
=
⎡
⎢
⎣
a
11
a
12
a
13
a
21
a
22
a
23
a
31
a
32
a
33
⎤
⎥
⎦
⎡
⎢
⎣
b
1
b
2
b
3
⎤
⎥
⎦
.
(b) By choosing b = a
r
and a
s
, and using the symmetries of a
km
and a
rs
, show
that
3
¸
k=1
a
ik
a
kj
=
3
¸
k=1
a
ki
a
kj
=
3
¸
k=1
a
ki
a
jk
= δ
j
i
.
It might be helpful to realize that, by using matrices, one of the preceding
results has the following representation:
⎡
⎢
⎣
a
11
a
12
a
13
a
21
a
22
a
23
a
31
a
32
a
33
⎤
⎥
⎦
⎡
⎢
⎣
a
11
a
12
a
13
a
21
a
22
a
23
a
31
a
32
a
33
⎤
⎥
⎦
=
⎡
⎢
⎣
1 0 0
0 1 0
0 0 1
⎤
⎥
⎦
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
32 Exercises 1.10–1.11
1.10. With the help of integrability criteria (1.19) and (1.21), show that only one of
the following constraints is integrable:
x ˙ x +y˙ y = −e(t), z˙ y + ˙ x = 0, cos(z)˙ y −sin(z) ˙ x = 0.
In addition, show that the integrable constraint corresponds to a particle moving on
a cylinder whose radius varies with time.
1.11. With the help of integrability criterion (1.21), show that one of the following
constraints is nonintegrable:
z˙ x + ˙ y = −e(t), ˙ z = 0.
If both constraints are imposed on the particle simultaneously, then show that the
system of constraints is integrable. Give a geometric description of the line that the
particle is constrained to move on.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
2 Kinetics of a Particle
2.1 Introduction
In this chapter, the balance law F = ma for a single particle plays a central role.
This law is then used to examine models for several physical systems ranging from
planetary motion to a model for a roller coaster. Our discussion of the behavior
of these systems predicted by the models relies heavily on numerical integration of
the equations of motion provided by F = ma, and it is presumed that the reader is
familiar with the numerical integration of ordinary differential equations.
Two of the most important types of forces featured in many applications are
conservative forces and constraint forces. For the former, the gravitational force
between two particles is the prototypical example, whereas the most common con
straint force in particle mechanics is the normal force. It is crucial to be able to prop
erly formulate and represent conservative and constraint forces, and we will spend
a considerable amount of time discussing them in this chapter. In contrast to most
texts in dynamics, here we consider friction forces to be types of constraint forces.
For most applications, exact (or analytical) solutions are not available and re
course to numerical methods is often the only course of action. In validating these
solutions, any conservations that might be present are crucial. To this end, conser
vations of momentum and energy are discussed at length and we also show (with
the help of two examples) how angular momentum conservation can often be ex
ploited. The examples discussed in this chapter are far from exhaustive. Although
several other examples are included in the exercises, they too do not come close to
encompassing the vast array of solved problems in the mechanics of a single particle.
To this end, it is recommended that the interested reader consult the classical texts
by Routh [185] and Whittaker [228] and the more recent texts by Baruh [14], Moon
[146], and Sheck [190].
2.2 The Balance Law for a Single Particle
Consider a single particle of mass m that is moving in E
3
. As usual, the position
vector of the particle relative to a ﬁxed origin Ois denoted by r. The balance law for
this particle is known as the balance of linear momentum, Newton’s second law, or
33
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
34 Kinetics of a Particle
Euler’s ﬁrst law. The integral (or impulse momentum) form of this law is
G(t) −G(t
0
) =
t
t
0
F(τ)dτ, (2.1)
where F is the resultant force acting on the particle and G = mv is the linear momen
tum. Notice that this form of the balance law does not assume that v is differentiable
with respect to time t. As a result, it is valid in impact problems, among others. An
example of this is discussed in Section 2.10.
If we assume that G is differentiable with respect to time, then we can differ
entiate both sides of the integral form of the balance of linear momentum to obtain
the local form:
F =
˙
G.
Assuming that the mass of the particle is constant, we can write
F = m¨ r. (2.2)
This law represents three (scalar) equations that relate F and the rate of change
of linear momentum of the particle. We refer to F = ma as the balance of linear
momentum.
∗
Our emphasis in this chapter is on the principle F = ma, but it is misleading to
believe that this is the sole accepted principle in dynamics. Indeed, since the creation
of this principle by Newton over 300 years ago, several alternative (and often equiv
alent) principles of dynamics have been proposed, and we shall postpone discussion
of several of them until Section 4.11.
It is convenient to write the balance law as a set of ﬁrstorder ordinary differen
tial equations:
v = ˙ r, F = m˙ v.
In the absence of constraints, these represent six scalar (differential) equations
for the six unknowns r(t) and v(t). To solve these equations, six initial conditions
r(t
0
) and v(t
0
) must be speciﬁed. Alternatively, if the problem is formulated as a
boundaryvalue problem, then a combination of six initial and ﬁnal conditions on
r(t) and v(t) must be prescribed.
If we write F = ma using a Cartesian coordinate system, then we ﬁnd the
following three equations:
m¨ x
1
= F · E
1
,
m¨ x
2
= F · E
2
,
m¨ x
3
= F · E
3
.
∗
Discussions of the historical threads from Newton’s Principia [152] that lead to Euler’s explicit
formulation of (2.2) in [51] can be found in several works by Clifford A. Truesdell (1919–2000); see,
for example, [216, 217].
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
2.3 Work and Power 35
On the other hand, if a cylindrical polar coordinate system is used, we have
m
¨ r −r
˙
θ
2
= F · e
r
,
m
r
¨
θ +2˙ r
˙
θ
= F · e
θ
,
m¨ z = F · E
3
. (2.3)
Finally, if we use a spherical polar coordinate system, we ﬁnd that
m
¨
R−R
˙
φ
2
−Rsin
2
(φ)
˙
θ
2
= F · e
R
,
m(R
¨
φ +2
˙
R
˙
φ −Rsin(φ) cos(φ)
˙
θ
2
) = F · e
φ
,
m(Rsin(φ)
¨
θ +2
˙
R
˙
θ sin(φ) +2R
˙
θ
˙
φcos(φ)) = F · e
θ
. (2.4)
Notice that these equations are different projections of F = ma onto a set of basis
vectors for E
3
.
Establishing the component representations of F = ma for various coordinate
systems can be a laborious task. However, Lagrange’s equations of motion allow us
to do this in a very easy manner. We will examine these equations in Section 3.2.
2.3 Work and Power
The mechanical power P of the force Pacting on a particle of mass mis deﬁned to be
P = P · v.
Clearly, if P is perpendicular to v, then the power of the force is zero.
As shown in Figure 2.1, consider a motion of the particle between two points: A
and B. We suppose that at time t = t
A
the particle is at A: r(t
A
) = r
A
. Similarly, when
t = t
B
, the particle is at B: r(t
B
) = r
B
. During the interval of time that the particle
moves from A to B, we suppose that a force P, among others, acts on the particle.
The work W
AB
performed by P during this time interval is deﬁned to be the integral,
Path of the particle
m
O
A
B
P
v
r
r
A
r
B
Figure 2.1. A force P acting on a particle as it moves
from Ato B.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
36 Kinetics of a Particle
with respect to time, of the mechanical power:
W
AB
=
t
B
t
A
P · vdt.
Notice that this is a line integral, and we are using t to parameterize the path of the
particle. Depending on the choice of coordinate system, the integral in this expres
sion has several equivalent representations.
As an example, suppose that a force P = Pe
θ
acts on a particle, and the mo
tion of the particle is r(t) = Le
αt
(cos(ωt)E
x
+sin(ωt)E
y
), where L, α, and ω =
˙
θ
are constant. A straightforward calculation shows that the power of this force
is
P · v = ωPLe
αt
,
and the work performed by the force is
W
AB
=
t
B
t
A
ωPLe
αt
dt =
ωPL
α
(e
αt
B
−e
αt
A
) ,
where, in evaluating the integral, we have assumed that α = 0.
2.4 Conservative Forces
A force P acting on a particle is said to be conservative if the work done by P during
any motion of the particle is independent of the path of particle. When a result from
vector calculus is used, the path independence implies that P is the gradient of a
scalar function U = U(r):
P = −∇U.
The function Uis known as the potential energy associated with the force P, and the
minus sign in the equation relating P to the gradient of U is a historical convention.
Various representations of the gradient can be found in (1.11).
It is important to notice that, if P is conservative, then its mechanical power is
−
˙
U. To see this, we simply examine
˙
Uand use the deﬁnition of a conservative force:
−
˙
U = −
∂U
∂r
· v = −(−P) · v = P · v.
This result holds for all motions of the particle and is very useful when we
wish to establish expressions for the rate of change of the total energy E of a
particle.
To check if a given force P is conservative, one approach is to ﬁnd a potential
function Usuch that
P · v = −
˙
U
holds for all motions of the particle. This approach reduces to solving a set of cou
pled partial differential equations for U. For example, if cylindrical coordinates
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
2.5 Examples of Conservative Forces 37
are used, one needs to solve the following three partial differential equations for
U(r, θ, z):
P
r
= −
∂U
∂r
, P
θ
= −
1
r
∂U
∂θ
, P
z
= −
∂U
∂z
, (2.5)
where P = P
r
e
r
+P
θ
e
θ
+P
z
E
3
. You might notice that the solution to (2.5) will yield
a potential energy U(r, θ, z) modulo an additive constant. This constant is usually set
by the condition that U = 0 when the coordinates have a certain set of values.
Another approach to ascertain if a given force P is conservative is to examine its
curl. The idea here is based on the identity curl(grad(V)) = 0, where V = V(r) is any
scalar function of r. Clearly, if the given force P is conservative, then curl(P) = 0.
∗
Consequently, if curl(P) = 0, then the Cartesian components of P must satisfy the
following conditions:
∂P
3
∂x
2
=
∂P
2
∂x
3
,
∂P
1
∂x
3
=
∂P
3
∂x
1
,
∂P
2
∂x
1
=
∂P
1
∂x
2
,
where P
i
= P · E
i
.
2.5 Examples of Conservative Forces
The three main types of conservative forces in engineering dynamics are constant
forces, spring forces, and gravitational force ﬁelds.
Constant Forces
All constant forces are conservative. To see this, let C denote a constant force and
let U
c
= −C · r. Now, ∇U
c
= −C, and, consequently, U
c
is the potential energy asso
ciated with C. The most common examples of constant forces are the gravitational
forces −mgE
2
and −mgE
3
, and their associated potentials are mgE
2
· r and mgE
3
· r,
respectively.
Spring Forces
Consider the spring shown in Figure 2.2. One end of the spring is attached to a ﬁxed
point A, and the other end is attached to a particle of mass m. When the spring is
unstretched, it has a length L
0
. Clearly, the stretched length of the spring is r −r
A
,
and the extension/compression of the spring is
= r −r
A
 −L
0
.
The potential energy U
s
associated with the spring is
U
s
= f ()
∗
An expression for the curl of a vector ﬁeld was presented earlier in (1.20).
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
38 Kinetics of a Particle
spring
m
O
A
r
r
A
Figure 2.2. A spring whose ends are attached to a particle and
a ﬁxed point A.
where f is a function of the change in length of the spring. Evaluating the gradient
of U
s
, we ﬁnd the spring force F
s
:
F
s
= −
∂U
s
∂r
= −
∂f
∂
r −r
A
r −r
A

.
To establish this result, we use the identity
∂
∂r
=
∂
∂r
(r −r
A
 −L
0
) =
r −r
A
r −r
A

.
This identity is left as an exercise to establish.
∗
The most common spring in engineering dynamics is a linear spring. For this
spring,
U
s
=
K
2
(r −r
A
 −L
0
)
2
, F
s
= −K(r −r
A
 −L
0
)
r −r
A
r −r
A

.
In words, the potential energy of a linear spring is a quadratic function of its change
in length. Examples of nonlinear springs include those in which f is a polynomial
function in . For instance,
f () = A
2
+B
4
,
where A > 0 and Bare constants. Such a spring is known as hardening if B > 0 and
softening if B < 0.
Newton’s Gravitational Force
Dating to Newton in the late 1600s, the force exerted by a body of mass M on an
other body of mass m is modeled as
F
n
= mg,
where
g = −
GM
r
3
r.
∗
To help you with this, it is convenient to ﬁrst establish that
∂
√
x·x
∂x
=
x
√
x·x
=
x
x
. This result is equiv
alent to showing that the gradient of r is e
R
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
2.6 Constraint Forces 39
The body of mass M is assumed to be located at the origin O, and Gis the universal
gravitation constant. In this model for the gravitational force, both bodies are
modeled as mass particles. Later on, in Sections 4.5 and 8.7, we shall examine gen
eralizations of this force ﬁeld to systems of particles and rigid bodies, respectively.
Because the magnitude of F
n
depends on the distance squared between the two
bodies, Newton’s force ﬁeld is often known as the inversesquare law. The force F
n
is conservative, and its potential energy is
U
n
= −
GMm
r
.
It should be transparent from the expressions for U
n
and F
n
that they have rather
simple representations when a spherical polar coordinate system is used.
Newton’s gravitational force ﬁeld is attractive: It tends to pull m toward M.
Thus, we have the interesting question of what keeps the two bodies from colliding.
The answer, as you know from other courses, is the change of momentum of m. It is
this delicate balance that allows m to steadily orbit M in a circular orbit.
2.6 Constraint Forces
A constraint force F
c
is a force that ensures that a constraint is enforced. Examples
of these forces include reaction forces, normal forces, and tension forces in inexten
sible strings. Given a constraint (r, t) = 0 on the motion of the particle, there is no
universal prescription for the associated constraint force. Choosing the correct pre
scription depends on the physical situation that the constraint represents. However,
when we turn to solving for the motion of the particle by using F = ma, we see that
we need to solve the six equations
˙ r = v, ˙ v =
1
m
F,
subject to the restrictions on r and v,
(r, t) = 0,
∂
∂r
· v +
∂
∂t
= 0.
To close this system of equations, an additional unknown is introduced in the form
of the constraint force F
c
. The prescription of F
c
must be such that F = ma and
(r, t) can be used to determine F
c
and r(t).
There are no unique prescriptions for constraint forces. The prescription most
commonly used, which dates to JosephLouis Lagrange (1736–1813), is referred to
in this book as the Lagrange prescription. As shown in O’Reilly and Srinivasa [162],
this prescription is necessary and sufﬁcient to ensure that the motion of the particle
satisﬁes the constraint. However, freedom is available to include other arbitrary
nonnormal components. This is the reason why prescriptions of constraint forces
featuring frictional components are valid.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
40 Kinetics of a Particle
Surface = 0
m
O
s
1
s
2
∇
n
r
Figure 2.3. Aparticle moving on a surface = 0. The vectors s
1
and s
2
are unit tangent vectors
to this surface at the point of contact of the particle and the surface.
A Single Constraint
Consider the case of a particle subject to a single constraint:
(r, t) = 0.
Referring to Figure 2.3, we recall that the unit normal vector n to this surface is
n =
∇
∇
.
Knowing n, we can construct a unit tangent vector s
1
to the surface. In addition,
by deﬁning another unit tangent vector, s
2
= n ×s
1
, we have constructed a right
handed orthonormal basis {s
1
, s
2
, n} for E
3
. This is not in general a constant set of
vectors; rather, it changes as we move from point to point along the surface. The
ﬁnal ingredient we need is to denote the velocity vector of the point of the surface
(which the particle is in contact with) as v
s
.
With this background in mind, we now consider two prescriptions for the con
straint force F
c
. The ﬁrst prescription is known as the Lagrange prescription:
F
c
= λ∇,
where we must determine λ = λ(t) by using F = ma. In other words, λ is an unde
termined Lagrange multiplier. As discussed in Casey [27], the constraint force F
c
is normal to the surface = 0 and is identical to a virtual work prescription that
is sometimes known as the Lagrange principle or Lagrange–D’Alembert principle.
On physical grounds, Lagrange’s prescription is justiﬁed if the surface on which the
particle is constrained to move is smooth.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
2.6 Constraint Forces 41
s
1
s
2
N = ∇
n
v
rel
F
f
F
c
= N+F
f
Figure 2.4. The constraint force F
c
acting on a particle
moving on a rough surface. The velocity vector v
rel
=
v −v
s
is the velocity vector of the particle relative to the
point of its contact with the surface.
In the event that the surface is rough, an alternative prescription, which is due
to Charles Augustin Coulomb (1736–1806), can be used
∗
:
F
c
= λ∇+F
f
,
where the normal force N = λ∇ and the friction force is
F
f
= −µ
d
λ∇
v −v
s
v −v
s

.
Here, µ
d
is known as the coefﬁcient of dynamic friction. Notice that the tangential
components of F
c
are governed by the behavior of v
rel
= v
rel
1
s
1
+v
rel
2
s
2
and oppose
the motion of the particle relative to the surface (cf. Figure 2.4). The velocity v −v
s

is sometimes known as the slip speed.
The mechanical power of the constraint force F
c
is
F
c
· v = λ∇· v +F
f
· v
= −λ
∂
∂t
+F
f
· v,
where we used the identity
˙
= ∇· v +
∂
∂t
= 0.
For the Lagrange prescription, F
f
= 0, we can now see that, if the surface that the
particle is moving on is ﬁxed, i.e., = (r), then F
c
does no work. Otherwise, this
constraint force is expected to do work because its normal component ensures that
part of the velocity vector of the particle is v
s
. For the Coulomb prescription, ex
cept when v
s
= 0, it is not possible to predict if work is done on the particle by the
constraint force.
As a ﬁrst example, consider a particle moving on a rough sphere of radius L
0
whose center is ﬁxed at the origin O. For this surface, the constraint is (r) = r ·
∗
This prescription is often known as Amontons–Coulomb friction, after Guillaume Amontons (1663–
1705).
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
42 Kinetics of a Particle
e
R
−L
0
= 0. Consequently, ∇ = e
R
. In addition, v −v
s
= L
0
˙
φe
φ
+L
0
sin(φ)
˙
θe
θ
. In
conclusion,
F
c
= λe
R
−µ
d
λ
˙
φe
φ
+sin(φ)
˙
θe
θ
˙
φ
2
+sin
2
(φ)
˙
θ
2
.
In this expression λ is the magnitude of the normal force exerted by the sphere
on the particle. If we now consider the spherical pendulum, then F
c
is given by the
Lagrange prescription: F
c
= λe
R
. In the spherical pendulum, −λ is the tension in the
rod connecting the particle to the ﬁxed point O.
Two Constraints
When a particle is subject to two constraints,
1
(r, t) = 0 and
2
(r, t) = 0, then it can
be considered as constrained to move on a curve. The curve in question is formed
by the instantaneous intersection of the surfaces deﬁned by the constraints.
At each point on the curve there is a unit tangent vector t. We can deﬁne this
vector by ﬁrst observing that ∇
1
and ∇
2
are both normal to the surfaces that the
curve lies on (cf. Figure 1.9). Consequently,
t =
∇
1
×∇
2
∇
1
×∇
2

.
For each instant in time, the point of the curve that is in contact with the particle has
a velocity. We denote this velocity by v
c
. The velocity vector of the particle relative
to the curve is v −v
c
= vt.
We now turn to prescriptions for the constraint force. The ﬁrst prescription is
the Lagrange prescription:
F
c
= λ
1
∇
1
+λ
2
∇
2
,
where λ
1
= λ
1
(t) and λ
2
= λ
2
(t) are both determined by use of F = ma. As in the
case of a single constraint, this prescription is valid when the curve that the particle
moves on is smooth, and it provides a constraint force that is normal to the curve.
For the rough case, we use Coulomb’s prescription:
F
c
= λ
1
∇
1
+λ
2
∇
2
+F
f
,
where the friction force is
F
f
= −µ
d
λ
1
∇
1
+λ
2
∇
2

v −v
c
v −v
c

.
The friction force opposes the motion of the particle relative to the curve and the
normal force N with λ
1
∇
1
+λ
2
∇
2
.
The mechanical power of the constraint force F
c
for this case is
F
c
· v = λ
1
∇
1
· v +λ
2
∇
2
· v +F
f
· v
= −λ
1
∂
1
∂t
−λ
2
∂
2
∂t
+F
f
· v,
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
2.6 Constraint Forces 43
where we again use the identities
˙
1
= ∇
1
· v +
∂
1
∂t
= 0,
˙
2
= ∇ψ
2
· v +
∂
2
∂t
= 0.
For the Lagrange prescription, F
f
= 0, and we can now see that, if the curve that
the particle is moving on is ﬁxed, i.e.,
1
=
1
(r) and
2
=
2
(r), then F
c
does no
work. Otherwise, this constraint force is expected to do work because its normal
components force part of the velocity vector of the particle to be v
c
. As in the case
of a single constraint, for the Coulomb prescription, except when v
c
= 0, it is not
possible to predict if work is done on the particle by this force.
We now consider some examples. Recall that the planar pendulum consists of
a particle of mass m that is attached by a rod of length L
0
to a ﬁxed point O. The
particle is also constrained to move on a vertical plane. In short,
1
= r · e
r
−L
0
= 0
and
2
= r · E
3
= 0. With a little work, we ﬁnd that ∇
1
= e
r
and ∇
2
= E
3
. For
this mechanical system, Lagrange’s prescription is appropriate:
F
c
= λ
1
e
r
+λ
2
E
3
.
For this system, λ
1
e
r
can be interpreted as the tension force in the rod and λ
2
E
3
can
be interpreted as the normal force exerted by the plane on the particle. If we let
L
0
= L
0
(t), the prescription for the constraint force will not change.
A system that is related to the planar pendulum can be imagined as a particle
moving on a rough circle whose radius L
0
= L
0
(t). The particle is subject to the same
constraints as it is in the planar pendulum; however, Lagrange’s prescription is not
valid. Instead, we now have
F
c
= λ
1
e
r
+λ
2
E
3
−µ
d
λ
2
1
+λ
2
2
˙
θ
˙
θ
e
θ
,
where we used the fact that v −v
c
= L
0
˙
θe
θ
.
Three Constraints
The reader may have noticed that our expressions for the constraint force when we
employed Coulomb’s prescription were not valid when the particle was stationary
relative to the surface or curve that it was constrained to move on. This is because
we view this case as corresponding to the motion of the particle subject to three
constraints:
i
(r, t) = 0, i = 1, 2, 3. As mentioned earlier, when a particle is subject
to three constraints, the three equations
i
(r, t) = 0 can in principle be solved to
determine the motion r(t) of the particle. We denote the resulting solution by f(t),
i.e., r(t) = f(t). In other words, the motion is completely prescribed. In this case, the
sole purpose of F = ma is to determine the constraint force F
c
.
For the case in which the particle is subject to three constraints, the Lagrange
prescription and a prescription based on static Coulomb friction are equivalent.
This equivalence holds in spite of the distinct physical situations these prescriptions
pertain to.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
44 Kinetics of a Particle
To examine the equivalence, let us ﬁrst use Lagrange’s prescription:
F
c
= λ
1
∇
1
+λ
2
∇
2
+λ
3
∇
3
.
Here, λ
1
, λ
2
, and λ
3
are functions of time. Because the three constraints are tacitly
assumed to be independent, {∇
1
, ∇
2
, ∇
3
} forms a basis for E
3
. Consequently,
Lagrange’s prescription provides a vector F
c
with three independent components.
Coulomb’s static friction prescription for a particle that is not moving relative to the
curve or surface on which it lies is
F
c
= N+F
f
,
where the magnitude of F
f
is restricted by the static friction criterion:
F
f
≤ µ
s
N ,
where µ
s
is the coefﬁcient of static friction. Again, the Coulomb prescription pro
vides a vector F
c
with three independent components. In other words, both prescrip
tions state that F
c
consists of three independent unknown functions of time.
If we now assume that the resultant force F has the decomposition F = F
c
+F
a
,
where F
a
are the nonconstraint forces, then we see how F
c
is determined from F =
ma:
F
c
= −F
a
+ma = −F
a
+m
¨
f.
This solution F
c
will be the same regardless of whether one uses Lagrange’s pre
scription or Coulomb’s prescription.
Nonintegrable Constraints
Our discussion of constraint forces has focused entirely on the case of integrable
constraints. If a nonintegrable constraint,
f · v +e = 0,
is imposed on the particle, we need to discuss a prescription for the associated con
straint force. To this end, we adopt a conservative approach and use the prescription
F
c
= λf.
The main reason for adopting this prescription is as follows: In the event that the
nonintegrable constraint turns out to be integrable, then the prescription we employ
will agree with Lagrange’s prescription we discussed earlier.
As a further example, suppose the motion of the particle is subject to two con
straints, one of which is integrable:
(r, t) = 0, f · v +e = 0.
Using Lagrange’s prescription, we ﬁnd that the constraint force acting on the parti
cle is
F
c
= λ
1
∇+λ
2
f.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
2.7 Conservations 45
Suppose that the applied force acting on the particle is F
a
; then the equations gov
erning the motion of the particle are
(r, t) = 0, f · v +e = 0,
˙ r = v, ˙ v =
1
m
(F
a
+λ
1
∇+λ
2
f).
This set of equations constitutes eight equations for the eight unknowns: λ
1
, λ
2
, r,
and v.
2.7 Conservations
For a given particle and system of forces acting on the particle, a kinematical quan
tity is said to be conserved if it is constant during the motion of the particle. The
conserved quantities are often known as integrals of motion. The solutions of many
problems in particle mechanics are based on the observation that either a mo
mentum or an energy (or both) is conserved. At this stage in the development of
the ﬁeld, most of these conservations are obvious and are deduced by inspection.
However, for future purposes it is useful to understand the conditions for such
conservations. We shall consider numerous examples of these conservations later
on.
Conservation of Linear Momentum
The linear momentum G of a particle is deﬁned as G = mv. Recalling the integral
form of the balance of linear momentum,
G(t) −G(t
0
) =
t
t
0
F(τ)dτ,
we see that G(t) is conserved during an interval of time (t
0
, t) if
t
t
0
F(τ)dτ = 0. The
simplest case of this conservation arises when F(τ) = 0.
Another form of this conservation pertains to a component of G in the di
rection of a given vector b(t) being conserved. That is,
d
dt
(G· b) = 0. For this to
happen,
˙
G· b =
˙
G· b +G·
˙
b = F · b +G·
˙
b = 0.
In words, if F · b +G·
˙
b = 0, then G· b is conserved.
Examples of conservation of linear momentum arise in many problems. For
example, consider a particle under the inﬂuence of a gravitational force F = −mgE
3
.
For this problem, the E
1
and E
2
components of G are conserved. Another example
is to consider a particle impacting a smooth vertical wall. Then the components of G
in the two tangential directions are conserved. For these two examples, the vector b
is constant.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
46 Kinetics of a Particle
Conservation of Angular Momentum
The angular momentum of a particle relative to a ﬁxed point O is H
O
= r ×G. To
establish how H
O
changes during the motion of a particle, a simple calculation is
needed:
˙
H
O
= v ×G+r ×
˙
G = v ×mv +r ×F = r ×F.
It is important to note that we used F = ma during this calculation. The ﬁnal result
is known as the angular momentum theorem for a particle:
˙
H
O
= r ×F.
In words, the rate of change of angular momentum is equal to the moment of the
resultant force.
Conservation of angular momentum usually arises in two forms. First, the en
tire vector is conserved, and, second, a component, say c(t), is conserved. For
the ﬁrst case, we see from the angular momentum theorem that H
O
is conserved
if F is parallel to r. Problems in which this arises are known as central force
problems. Dating to Newton, they occupy an important place in the history of
dynamics. When the angular momentum theorem is used, it is easy to see that
the second form of conservation, H
O
· c is constant, arises when r ×F · c +H
O
· ˙ c
= 0.
Conservation of Energy
As a prelude to discussing the conservation of energy, we ﬁrst need to discuss the
work–energy theorem. This theorem is a result that is established by use of F = ma
and relates the time rate of change of kinetic energy to the mechanical power of
F:
˙
T = F · v.
This theorem is the basis for establishing conservation of energy results for a single
particle.
The proof of the work–energy theorem is very straightforward. First, recall that
T =
1
2
mv · v. Differentiating T, we ﬁnd
˙
T =
d
dt
1
2
mv · v
=
1
2
(m˙ v · v +mv · ˙ v) = m˙ v · v.
However, we know that m˙ v = F, and so substituting for m˙ v, we ﬁnd that
˙
T = F · v,
as required.
To examine situations in which the total energy of a particle is conserved, we
ﬁrst divide the forces acting on the particle into the sum of a resultant conservative
force P = −
∂U
∂r
and a nonconservative force P
ncon
: F = P +P
ncon
. Here, Uis the sum
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
2.8 Dynamics of a Particle in a Gravitational Field 47
of the potential energies of the conservative forces acting on the particle. From the
work–energy theorem, we ﬁnd
˙
T = F · v
= P · v +P
ncon
· v
= −
∂U
∂r
· v +P
ncon
· v
= −
˙
U+P
ncon
· v.
Deﬁning the total energy E of the particle by
E = T +U,
we see that
˙
E = P
ncon
· v.
This result states that if, during a motion of the particle, the nonconservative forces
do no work, then the total energy of the particle is conserved.
To examine whether energy is conserved, it usually sufﬁces to check whether
P
ncon
· v = 0. To see this, let us consider the example of the spherical pendulum
whose length L
0
= L
0
(t). For this particle,
P = −mgE
3
, P
ncon
= λe
R
.
Consequently,
E = T +mgE
3
· r
and
P
ncon
· v = λe
R
· v =
˙
L
0
λ.
As a result, when the length of the pendulum is constant,
˙
L
0
= 0, then E is con
served. On the other hand, if
˙
L
0
= 0, then the constraint force λe
R
does work by
giving the particle a velocity in the e
R
direction.
2.8 Dynamics of a Particle in a Gravitational Field
The problem of a body of mass m orbiting a body of mass M is one of the center
pieces in Isaac Newton’s Principia.
∗
Over 100 years later, it was also discussed in
wonderful detail in Lagrange’s famous text M´ ecanique Analytique.
†
Newton was
partially motivated to study this problem because of Johannes Kepler’s (1571–
1630) famous three laws of motion for the (then known) planets in the solar
system:
I. The planets move in elliptical paths with the Sun at one of the foci.
∗
See Section III of Book 1 of [152].
†
See Section VII of the Second Part of [121].
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
48 Kinetics of a Particle
x
y
m
O
a
b
a(1 −e)
r
θ
P
A
ˆ
h < 0
ˆ
h > 0
Figure 2.5. Schematic of a particle of mass m moving about a ﬁxed point O in an elliptical
orbit. One of the foci of the ellipse is at O, and the eccentricity e =
1 −
b
2
a
2
of the ellipse is
less than 1, where a and b are the lengths of the semimajor and semiminor axes of the ellipse,
respectively. Point Ais known as the apocenter, and P is known as the pericenter.
II. The vector connecting the Sun to the planet sweeps out equal areas in equal
times.
III. If a denotes the semimajor axis of the elliptical orbit and T denotes the period
of the orbit, then, for any of two planets,
a
3
1
a
3
2
=
T
2
1
T
2
2
.
Concise discussions of Kepler’s laws can be found in [150, 175, 188, 220]. Some of
these authors note that the laws were based on astronomical data taken with the
naked eye.
In our analysis we assume that the body of mass M is ﬁxed. As can be seen in
Exercise 4.6 from Chapter 4, this restriction is easily removed and the results pre
sented can be readily applied to deduce the motions of mand M. Many of the results
presented feature terminology associated with ellipses, and, for convenience, a sum
marization of many of the terms associated with an ellipse, such as its eccentricity e
and axes a and b, is given in Figure 2.5. The area swept out by the particle can be
determined by integrating the areal vector (1.1): A =
1
2m
H
O
.
The third law is remarkable when we note from [188] that the semimajor
axis a and orbital period T for the planet Mercury are 0.387 astronomical units
(AU) and 0.241 years, the Earth’s are 1 AU and 1 year, Jupiter’s are 5.203 AU
and 11.862 years, and Mars’ are 1.524 AU and 1.881 years, respectively. We note
that
0.387
3
0.241
2
= 1.00,
5.203
3
11.862
2
= 1.00,
1.524
3
1.881
2
= 1.00.
All of these results are in agreement with Kepler’s third law.
Here, we start by setting up the coordinates for this problem and establish
ing the equations of motion. Our analysis of the equations of motion then ex
ploits conservation of angular momentum to show that the motion must be planar.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
2.8 Dynamics of a Particle in a Gravitational Field 49
Following this, we reduce the equations of motion to a single secondorder differen
tial equation that we nondimensionalize and integrate numerically.
∗
An alternative
approach, which is used in most textbooks, will also be discussed. These analyses
enable us to classify all ﬁve possible types of trajectories of the particle.
Kinematics
We pick as the origin Othe ﬁxed particle of mass M. Then the position vector of the
particle of mass m is r, and it is convenient to pick cylindrical polar coordinates for
this position vector:
r = re
r
+zE
3
.
Representations for the velocity and acceleration vectors in terms of cylindrical po
lar coordinates were established earlier and we do not rewrite them here.
Equations of Motion
The equations of motion for the particle can be obtained from F = ma, where F is
solely due to Newton’s gravitational force:
F
n
= −
GMm
r
3
r.
You should recall that this force is conservative, and its potential energy is denoted
by U
n
.
Using (2.3), we can write out the component forms of F
n
= ma. The result will
be three differential equations:
m
¨ r −r
˙
θ
2
= −
GMmr
√
r
2
+z
2
3
,
m
r
¨
θ +2˙ r
˙
θ
= 0,
m¨ z = −
GMmz
√
r
2
+z
2
3
. (2.6)
For a given set of six initial conditions,
†
these equations provide r(t), θ(t), and z(t),
and hence can be used to predict the position of the particle of mass m.
Conservations
The solutions of differential equations (2.6) conserve two important kinematical
quantities. First, they conserve the total energy E = T +U, where U = U
n
. Second,
∗
The reduction procedure we use is equivalent to the socalled Routhian or Lagrangian reduction
procedure that is used to incorporate momentum conservation in a variety of mechanical systems
ranging from the problem at hand to Lagrange and Poisson tops. For further details on this proce
dure, the reader is referred to Gantmacher [67], Chapter 2 of Karapetyan and Rumyantsev [108],
and Marsden and Ratiu [138].
†
The six initial conditions needed are r (t
0
), θ (t
0
), z(t
0
), ˙ r (t
0
),
˙
θ (t
0
), and ˙ z(t
0
).
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
50 Kinetics of a Particle
the angular momentum H
O
of the particle is conserved. It is left to the reader
to demonstrate these results by using the work–energy and angular momentum
theorems.
The conservation of angular momentum implies that
r (t) ×v (t) = r (t
0
) ×v (t
0
) .
Now the initial position r (t
0
) and velocity v (t
0
) vectors deﬁne a plane, in general,
and thus the motion of the particle remains on this plane (which is known as the
orbital plane). We also observe that the normal to this plane is parallel to H
O
. If
we allow ourselves the freedom to choose E
3
, then we can pick this vector such that
H
O
=
ˆ
hE
3
, where
ˆ
h = mr
2
˙
θ. Consequently, r and v will have components in only
the E
1
and E
2
directions: z(t) = 0 and ˙ z(t) = 0. We henceforth exploit the fact that
angular momentum is conserved and the motion is planar. Because H
O
and the areal
velocity vector are synonymous, angular momentum conservation for this problem
is often known as the “integrals of area.”
∗
We have tacitly ignored the case in which r (t
0
) v (t
0
). In this case, H
O
is zero
and must remain so. Consequently, the motion of the particle is a straight line. For
convenience, we can choose this line to lie on the E
1
−E
2
plane. It can be shown
that the motion of the particle will eventually lead to a collision with the particle of
mass Mat the origin. Thus, without an initial angular momentum, a collision for this
system would be unavoidable.
Determining the Motion of the Particle
Because the angular momentum is conserved, we choose E
3
such that z(t) = 0 and
˙ z(t) = 0 for the particle. That is, the direction of H
O
is E
3
. As a result, the equations
of motion reduce to
m
¨ r −r
˙
θ
2
= −
GMm
r
2
,
m
r
¨
θ +2˙ r
˙
θ
= 0. (2.7)
Now the second of these equations can be expressed as
d
dt
mr
2
˙
θ
= 0. (2.8)
This equation is equivalent to the conservation of H
O
· E
3
. Using this conservation,
we can eliminate
˙
θ from (2.7) and arrive at a single governing differential equation:
m¨ r −
ˆ
h
2
mr
3
= −
GMm
r
2
. (2.9)
Here,
ˆ
h is determined from the initial position and velocity of the particle:
ˆ
h = H
O
· E
3
= (mr (t
0
) ×v (t
0
)) · E
3
.
∗
See, for example, Section 86 of Moulton [150].
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
2.8 Dynamics of a Particle in a Gravitational Field 51
For a particle with no angular momentum, we see from (2.8) that
˙
θ = 0 and hence
¨ r = −
GM
r
2
. It is not difﬁcult to see that this equation implies that r(t) →0 as t in
creases. This is the collision we discussed earlier.
Given r(t
0
) and v(t
0
), we can determine
ˆ
h and then integrate (2.9) to determine
r(t).
∗
We can then compute the coordinate θ(t) by integrating
˙
θ =
ˆ
h
mr
2
. (2.10)
We can then ﬁnd expressions for x(t) = r(t) cos (θ(t)) and y(t) = r(t) sin(θ(t)) and
construct the orbit of the particle.
The easiest solution of (2.9) to compute is the one for which r is constant: r(t) =
r
0
and ˙ r(t) = 0. In this case, (2.9) is satisﬁed provided
r
0
=
ˆ
h
2
GMm
2
. (2.11)
We also show, using (2.10), that
˙
θ is constant:
˙
θ(t) = ω
K
=
ˆ
h
mr
2
0
=
GM
r
3
0
. (2.12)
The frequency ω
K
is known as the Kepler frequency. Physically, the particle is mov
ing in a circular orbit of radius r
0
at constant speed r
0
ω
K
about the ﬁxed body of
mass M.
In numerically integrating (2.9), the time scale of the integration is very long,
and it is convenient to nondimensionalize the equations of motion. To do this, we
choose the dimensionless variable w =
r
r
0
and time τ = ω
K
t. Now using identities of
the form ˙ r =
dr
dt
=
dτ
dt
dr
dτ
= ω
K
dr
dτ
, we can simplify (2.9) and (2.10) to
d
2
w
dτ
2
=
1
w
3
−
1
w
2
,
dθ
dτ
=
1
w
2
. (2.13)
Notice that we have reduced the problem of determining the motion of the particle
to the integration of two differential equations.
Differential equation (2.13)
1
is a secondorder differential equation for w(τ). As
opposed to exhaustive displays of w(τ) for several sets of initial conditions, a quali
tative method of representing the solutions of (2.13)
1
is to construct what is known
as a phase portrait. In this portrait,
dw
dτ
is plotted as a function of w(τ).
†
Equilibria
of the secondorder differential equation correspond to points in the phase portrait
where
dw
dτ
= 0 and w(τ) is a constant. To ﬁnd such points, we set
dw
dτ
= 0 and
d
2
w
dτ
2
= 0
in the governing secondorder differential equation for w(τ) and solve for the result
ing constant values of w(τ). Later examples in this chapter will feature differential
equations with multiple equilibria.
∗
Differential equation (2.9) has an analytical solution that can be expressed in terms of Jacobi’s
elliptic functions. However, this is beyond our scope here, and the reader is referred to Whittaker
[228] for details on how such an integration can be performed.
†
Phase portraits are a standard method for the graphical representation of solutions to ordinary
differential equations (see, for example, the texts of [10, 18, 81, 229]).
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
52 Kinetics of a Particle
w
dw
dτ
c
e e e
h
h
p
1
1
2
4
−1
−2
Figure 2.6. The phase portrait of (2.13)
1
. The trajectories labeled e and h correspond to ellip
tical and hyperbolic orbits of the particle, the point c corresponds to a circular orbit, and the
trajectory labeled p corresponds to the parabolic orbits. The arrows in this ﬁgure correspond
to the directions of increasing τ.
Returning to the problem at hand, the phase portrait of (2.13)
1
is shown in
Figure 2.6. There, we see an equilibrium point at
w,
dw
dτ
= (1, 0) that corresponds
to a circular orbit of the particle. The closed orbits enclosing this point correspond
to elliptical orbits of the form shown in Figure 2.7.
∗
The remaining orbits shown
in this ﬁgure correspond to hyperbolic orbits of the particle. For these orbits, the
particle circles around the equilibrium once and never returns. The interesting case
in which the particle describes a parabolic orbit is also shown in this ﬁgure. The
trajectory on the phase portrait corresponding to this orbit separates the elliptical
and hyperbolic trajectories.
†
The Orbital Motions
An alternative approach to the one outlined in the previous section is followed in
most textbooks on dynamics. This approach involves solving for the motion of the
particle as a function of θ rather than of time t.
For this approach, we ﬁrst use the chain rule and (2.7)
2
in the form (2.10) to
show that
˙ r = −
ˆ
h
m
d
dθ
1
r
, ¨ r = −
ˆ
h
2
m
2
r
2
d
2
dθ
2
1
r
.
∗
A second integration involving
dθ
dτ
=
1
w
2
is needed to construct these orbits, and this is left as an
exercise.
†
In the parlance of dynamical systems theory, the homoclinic orbit that passes through the point
w,
dw
dτ
= (0.5, 0) connects the ﬁxed point at
w,
dw
dτ
= (∞, 0) and corresponds to the parabolic
orbit of the particle.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
2.8 Dynamics of a Particle in a Gravitational Field 53
x = x
1
x = x
1
x = x
1
x = x
1
x = x
1
y = x
2
y = x
2
y = x
2
y = x
2
y = x
2
O
O O
O O
ˆ
h < 0
ˆ
h > 0
ˆ
h > 0
ˆ
h > 0
r
r
r
r
(a) (b)
(c)
(d) (e)
θ
θ
θ
P
θ
P
θ
P
Figure 2.7. Schematic of the four types of orbits of a particle of mass m moving about a ﬁxed
point O: (a) line, (b) circular orbit (e = 0), (c) elliptical orbit (0 < e < 1), (d) parabolic orbit
(e = 1), and (e) a hyperbolic orbit (e > 1). For each of the orbits (c)–(e), distinct values of θ
p
are considered.
Using the second of these results to rewrite (2.7)
1
, we ﬁnd the differential equation
d
2
dθ
2
1
r
+
1
r
=
1
r
0
. (2.14)
where r
0
was deﬁned earlier [see (2.11)].
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
54 Kinetics of a Particle
Equation (2.14) is a linear ordinary differential equation for
1
r
as a function of θ
that has the exact solution
r = r (θ) = r
0
(1 +e cos (θ −θ
p
))
−1
, (2.15)
where e and θ
p
are constants (which are determined from the initial conditions for
the position and velocity of the particle). We can then integrate
mr
2
(θ)
ˆ
h
dθ = dt to de
termine an analytical expression for θ(t). This is left as an exercise. Solution (2.15)
represents a conic section, and from the theory of conic sections, it is known that,
when e = 0, the orbit (r(θ)) is circular; when 0 < e < 1, the orbit is elliptical; when
e = 1, the orbit is parabolic; and when e > 1, the orbit is hyperbolic. Referring to
Figure 2.5 for the elliptical orbit, it is easy to see that b =
ˆ
h
2
GMm
2
.
To solve for e and θ
p
, let us assume that r (t
0
) and v (t
0
) are given. Then we can
determine H
O
and specify E
3
and
ˆ
h. To compute θ
p
, we ﬁrst calculate ˙ r by using
(2.15) and the chain rule:
˙ r(t) = −
GMm
ˆ
h
e sin(θ(t) −θ
p
) . (2.16)
We also compute the value E
0
of total energy of the particle
E
0
=
m
2
v (t
0
) · v (t
0
) −
GMm
r (t
0
)
. (2.17)
Now as the total energy is conserved, the value of this kinematical quantity when
θ = θ
p
is also equal to E
0
. With some manipulations of (2.15) and (2.16), it can be
shown that
E
0
=
G
2
M
2
m
3
2
ˆ
h
2
e
2
−1
. (2.18)
We now have a method of determining e and θ
p
from a given set of initial conditions.
The procedure is to compute E
0
by use of (2.17) and then use (2.18) to compute e ≥
0. Once e is known, then (2.16) can be used to compute θ
p
. With these values, r(θ) is
speciﬁed and θ(t) can be calculated. Depending on the value of e, the orbits will be
one of four types: a circle, an ellipse, a parabola, and a hyperbola (see Figure 2.7).
For completeness, we could also have nondimensionalized (2.14):
d
2
u
dθ
2
+u = 1, (2.19)
where u =
1
w
=
r
0
r
. The phase portrait of this equation is shown in Figure 2.8. In
contrast to the earlier phase portraint, here the trajectory corresponding to the
parabolic orbit is easily distinguished. However, as in the previous case, to gain
a physical interpretation of the trajectories shown in Figure 2.8, it is necessary to
reconstruct a position vector of the particle corresponding to the particular orbit.
Comments
For this problem, the rare event occurs that a complete classiﬁcation of the motions
of the particle are possible. For many of the problems that are discussed later on in
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
2.9 Dynamics of a Particle on a Spinning Cone 55
u
du
dθ
c
e
e
p
h
h
h
h
1
1
4
2
1
2
Figure 2.8. The phase portrait of (2.19). The trajectories labeled e and h correspond to ellip
tical and hyperbolic orbits of the particle, the point c corresponds to a circular orbit, and the
trajectory labeled p corresponds to the parabolic orbits. In this ﬁgure, the arrow indicates the
direction of increasing θ. For the parabolic trajectory, this angle ranges from−π →π.
this book, this classiﬁcation has not been performed and indeed may not be possible.
As a result, our previous discussion will be a benchmark.
We have not exhausted the literature on this problem, and discussions of related
problems involving escape velocities and transfer orbits can be found in several text
books; see, for example, Baruh [14]. Generalizations of the problem also abound,
and we shall discuss two of them at later stages in this book. Before we leave the
problem for now, we wish to show that the elliptical (and circular) orbits are in
agreement with Kepler’s laws. We ﬁrst note that satisfaction of the ﬁrst law is triv
ial, and the second law is a consequence of angular momentum conservation. To see
that the third law is satisﬁed, we need to compute the period T of the particle exe
cuting an elliptical orbit. We leave it as an exercise to show that T =
2π
√
GM
a
3
2
, and,
consequently, the third law is satisﬁed.
2.9 Dynamics of a Particle on a Spinning Cone
As shown in Figure 1.7(d), a particle of mass m moves on the surface of a cone.
It is attached to the ﬁxed apex O of the cone by a linear spring of stiffness K and
unstretched length L
0
. We assume that the surface of the cone is rough and that the
cone is spinning about its axis of symmetry with an angular speed
0
. Our goal is
to establish the equations of motion for the particle and discuss some features of its
dynamics.
Coordinates, Constraints, and Velocities
As discussed in Section 1.7, when the particle is moving on the surface of the cone, it
is subject to a single constraint = 0. This constraint can be conveniently expressed
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
56 Kinetics of a Particle
by use of a spherical polar coordinate system:
= φ +α −
π
2
.
For future reference, we note that the gradient of is ∇ =
1
R
e
φ
. Because the cone
is rotating with a speed
0
, the velocity of the particle relative to the cone is
v
rel
=
˙
Re
R
+Rcos(α)
˙
θ −
0
e
θ
.
We defer discussion of the case in which the particle is stuck to the cone.
Forces
The particle is under the inﬂuence of a gravitational force −mgE
3
and a spring force:
F
s
= −K(R−L
0
) e
R
,
where K is the stiffness of the spring and L
0
is its unstretched length. Assuming that
the particle is moving relative to the surface of the cone, we ﬁnd that the constraint
force F
c
acting on the particle has the representation
F
c
= N+F
f
,
where the normal force N is parallel to ∇:
N =
λ
R
e
φ
, F
f
= −µ
d
N
v
rel
v
rel

.
Thus the total force on the particle is F = F
c
+F
s
−mgE
3
.
The Equations of Motion
To obtain the equations of motion, we express F = ma in spherical polar coordinates
[see (2.4)] and impose the constraint = 0 to ﬁnd
m
¨
R−Rcos
2
(α)
˙
θ
2
= −K(R−L
0
) +F
f
· e
R
−mg sin(α),
m
Rcos(α)
¨
θ +2
˙
R
˙
θ cos(α)
= F
f
· e
θ
,
−mRsin(α) cos(α)
˙
θ
2
=
λ
R
+mg cos(α). (2.20)
The ﬁrst two of these equations are ordinary differential equations for R and θ, and
the third equation can be solved for λ (and hence the normal force) as a function of
the motion of the particle.
To integrate these equations, it is desirable to nondimensionalize them. We use
L
0
as a measure of length and
L
0
g
as a measure of time:
τ = t
g
L
0
, w =
R
L
0
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
2.9 Dynamics of a Particle on a Spinning Cone 57
With the help of identities of the type
˙
R =
dR
dt
=
dτ
dt
dR
dτ
=
g
L
0
dR
dτ
, we can rewrite
(2.20) in the form
d
2
w
dτ
2
= wcos
2
(α)
dθ
dτ
2
−ω
2
(w −1) −sin(α)
−µ
k
n
dw
dτ
w
2
dθ
dτ
−ω
0
2
+
dw
dτ
2
,
d
dτ
w
2
cos
2
(α)
dθ
dτ
= −µ
k
n(wcos(α))
w
dθ
dτ
−ω
0
w
2
dθ
dτ
−ω
0
2
+
dw
dτ
2
.
(2.21)
In these equations, the constants and dimensionless normal force are
ω
2
=
KL
0
mg
, ω
0
=
L
0
g
, n =
N
mg
= cos(α) +
h
2
w
tan(α),
and we can also show that dimensionless versions of total energy E and angular
momentum H
O
· E
3
are
E
mgL
0
=
1
2
dw
dτ
2
+
dθ
dτ
2
w
2
cos
2
(α)
+
ω
2
2
(w −1)
2
+wsin(α),
h =
H
O
· E
3
m
gL
3
0
= w
2
cos
2
(α)
dθ
dτ
.
Notice that, by nondimensionalizing the equations of motion, we have reduced the
number of parameters by two.
The Static Friction Case
When the particle is stuck to the cone, its velocity vector is v = R
0
0
cos(α)e
θ
. In ad
dition, the particle is subject to three constraints and the friction and normal forces
constitute three undetermined forces that enforce these constraints:
F
c
=
λ
R
0
e
φ
+
λ
1
R
0
cos(α)
e
θ
+λ
2
e
R
.
To determine λ, λ
1
, and λ
2
, we examine F = ma. With some manipulations, we con
clude that
F
f
=
K(R
0
−L
0
) +mg sin(α) −mR
0
cos
2
(α)
2
0
e
R
,
+mR
0
cos(α)
˙
0
e
θ
,
N = −
mg cos(α) +mR
0
2
0
sin(α) cos(α)
e
φ
.
Such a state of the particle is sustained provided sufﬁcient friction is present, and to
check this sufﬁciency we need to examine the static friction criterion F
f
≤ µ
s
N.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
58 Kinetics of a Particle
R
L
0
1
L
0
dR
dτ
x
1
x
1
x
1
x
1
x
2
x
2
x
2
x
2
1 8
20
−20
Figure 2.9. The phase portrait of (2.22) and the corresponding planar projections of the tra
jectories of the particle. For this ﬁgure, we assumed that α = 20
◦
, h = 5, and ω
2
= 10. Con
sequently the equilbrium point corresponding to a circular trajectory of the particle has the
coordinates (w
0
, 0) = (1.58896, 0).
If this criterion holds, then a particle that is stuck on the surface of the cone will
remain stuck on the surface. Otherwise it slips, and the initial direction in which it
slips is parallel to F
f
.
∗
The Smooth Cone
When the particle is moving on a smooth cone, we can simplify (2.21) considerably.
Indeed, as in the earlier particle problem, we can exploit the conservation of angular
momentum to write a single equation for w:
d
2
w
dτ
2
=
h
2
w
3
cos
2
(α)
−ω
2
(w −1) −sin(α). (2.22)
Integrating this equation, we can ﬁnd w(τ). Another integration using
w
2
cos
2
(α)
dθ
dτ
= h provides θ(τ). The equilibrium point at
w,
dθ
dτ
= (w
0
, 0), where
w
0
is the solution of
h
2
w
3
0
cos
2
(α)
−ω
2
(w
0
−1) −sin(α) = 0,
∗
The initial slip direction must be speciﬁed in order that the initial motion of the particle slipping on
the cone can be determined. The prescription of the initial slip direction allows one to specify
v
rel
v
rel

even though v
rel
= 0.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
2.10 A Shocking Constraint 59
corresponds to a circular orbit of the particle that has a radius r = L
0
w
0
cos(α). Some
of the other trajectories of w in the w −
dw
dτ
plane are shown in Figure 2.9. In this
ﬁgure, we have also constructed possible trajectories of the particle corresponding
to w(τ). Unlike the problem of the particle subject to F
n
that we discussed in Section
2.8, here the classiﬁcations of the trajectories for the particle on the cone defy a
simple classiﬁcation.
2.10 A Shocking Constraint
We now return to the constraint ˙ yx = 0, discussed earlier [see (1.15)]. Our interest
is to determine the equations of motion of a particle that is subject to this constraint
and that is also under the inﬂuence of an applied force F
a
= P
1
E
1
+P
2
E
2
.
First, we assume that the constraint force that enforces ˙ yx = 0 has the standard
prescription
F
c
= λxE
y
, (2.23)
where λ is a Lagrange multiplier. We note that, when x = 0, F
c
= 0. From a balance
of linear momentum, we ﬁnd that the equations of motion for the particle are
x˙ y = 0,
m¨ x = P
1
,
m¨ y = P
2
+λx,
m¨ z = 0. (2.24)
The equation for motion in the z direction is trivial to integrate and interpret, and,
for convenience, we henceforth ignore this direction and assume that the motion is
planar.
When some modest restrictions are imposed on P
1
and P
2
, governing equations
(2.24)
1,2,3
have exact solutions that are easy to establish provided the motion is rec
tilinear:
y(t) = y
0
, x(t) = x
0
+ ˙ x
0
t +
t
0
τ
0
P
1
m
dudτ, λ =
F
y
x(t)
(2.25)
and
x = 0, y(t) = y
0
+ ˙ y
0
t +
t
0
τ
0
P
2
m
dudτ. (2.26)
Here x
0
= x(0), ˙ x
0
= ˙ x(0), y
0
= y(0), and ˙ y
0
= ˙ y(0) are initial conditions. From these
solutions to the equations of motion, we observe that λ is not deﬁned when x = 0.
The Shock
Referring to Figure 1.11, we recall that this constraint has the unusual feature that
it can be decomposed into two piecewise integrable constraints:
˙ y = 0 when x = 0, and x = 0.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
60 Kinetics of a Particle
x = x
1
x = x
1
y = x
2
y = x
2
I
c1
I
c2
A A B
B
(a) (b)
Figure 2.10. Two possible motions of a particle subject to a constraint ˙ yx = 0. In (a), the par
ticle moves from A to B and there is no impulse I
c
when x = 0, whereas in (b) the particle
experiences two instances in which v is not continuous.
At the points where the particle makes a transition from one of these integrable
constraints to the other, its velocity vector v will be discontinuous and therefore
its acceleration vector cannot be deﬁned. At such a transition, the prescription for
constraint force (2.23) does not hold, and instead we can calculate only the impulse
I
c
that is due to this force by using (2.1). Supposing that the transition occurs at time
t = T, we will have
I
c
= lim
σ→0
mv (T +σ) −mv (T −σ) −
T+σ
T−σ
F
a
dτ
.
For example, to achieve a motion that goes frompoint Ato point Bin Figure 2.10(b),
the particle needs to perform a motion for which it will possess a discontinuous
velocity vector in at least two locations. That is, the particle will experience a shock.
Impulses I
c1
and I
c2
shown in the ﬁgure enable these shocks. This is in contrast to
the situation shown in Figure 2.10(a), where there is no discontinuity in the motion
of the particle. That is, the shock is absent and consequently I
c
= 0.
If the constraint cannot supply the impulse I
c
, then the particle is effectively
subject to a single holonomic constraint. This constraint is either y = y
0
or x = 0,
depending on the initial position and velocity of the particle. It is left as an exercise
for the reader to imagine a rigid wall placed to the left of the y axis in Figure 1.11 as
a method of realizing the constraint ˙ yx = 0.
2.11 A Simple Model for a Roller Coaster
Imagine being in a cart at the top of a roller coaster. If there is no friction, then the
slightest nudge will set the cart in motion. The presence of Coulomb friction with
stick–slip changes this scenario. It will eventually bring the cart to a halt, and it may
bring the cart to a halt near the top of the roller coaster. Indeed, if there is sufﬁcient
static friction, then the cart can come to a halt at any location on the track, and the
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
2.11 A Simple Model for a Roller Coaster 61
x
1
y
A
B C
E
1
E
2
Figure 2.11. Schematic of a particle on a cosinusoidal path.
chief quantity that governs how fast the halting occurs in this extreme case will be
the dynamic friction coefﬁcient.
Here, a very simple model is presented for the roller coaster that captures its
stick–slip behavior.
∗
First, we establish a differential equation governing the mo
tion of the roller coaster, and then we use numerical integrations to investigate the
dynamics of the roller coaster.
The Equations of Motion
One model for the dynamics of a cart on a roller coaster is to model the cart as a
particle of mass m that is moving on a ﬁxed plane curve: y = f (x
1
), z = 0. That is,
the particle is subject to two constraints,
1
= 0 and
2
= 0, where
1
= y − f (x
1
) ,
2
= z.
These constraints will be enforced by F
c
= N+F
f
. It is a standard exercise to calcu
late the unit tangent e
t
, the unit normal e
n
, and the binormal e
b
vectors to this curve
[159]:
e
t
=
1
1 + f
2
E
1
+ f
E
2
,
e
n
=
sgn
f
1 + f
2
E
1
+ f
E
2
,
e
b
= e
t
×e
n
,
where the prime denotes the derivative with respect to x
1
. These three vectors con
stitute the Frenet triad, and in calculating this triad we assume that ˙ x
1
> 0. A nor
mal force N = Ne
n
+λ
2
E
3
, a friction force F
f
e
t
, and a vertical gravitational force
−mgE
2
act on the cart (see Figure 2.11).
Now, for a particle moving on a curve with a velocity v = ve
t
, the acceleration
vector of the particle is
a = ˙ ve
t
+κv
2
e
n
,
∗
The work presented on this model was performed in collaboration with Henry Lopez [130].
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
62 Kinetics of a Particle
where κ is the curvature of the space curve:
κ =
f
1 + f
2
3
.
Taking the e
t
and e
n
components of F = ma, we can easily calculate the equations
governing the motion of the cart and the normal force:
m
1 + f
2
¨ x
1
+ f
f
˙ x
2
1
= −mgf
−µ
d
1 + f
2
N
˙ x
1
 ˙ x
1

, (2.27)
where
N =
1
1 + f
2
sgn
f
mg +
f
m˙ x
2
1
e
n
, if f
= 0,
=
mg
1 + f
2
e
n
, when f
= 0. (2.28)
These equations apply when the cart is moving and the friction is dynamic. In the
event that the cart is stationary, static friction acts and, provided the static friction
criterion is satisﬁed,
mgf
1 + f
2
≤ µ
s
N , (2.29)
the cart remains stationary. With the help of (2.28), (2.29) can be expressed in the
simple form
f
≤ µ
s
. (2.30)
This equation can be viewed as the basis for the classical experiment to measure
the coefﬁcient of static friction: We place a block on an inclined plane and slowly
increase the angle of inclination until slipping occurs. The tangent of the angle of
inclination is equal to µ
s
. We shall shortly use (2.30) to establish a continuum of
points at which the roller coaster can remain in a state of rest.
We now choose a cosinusoidal track,
f (x
1
) = Acos
πx
1
L
0
. (2.31)
In addition, we employ the following nondimensionalizations:
x =
x
1
L
0
, τ =
g
L
0
t.
Of course, other tracks are possible, and the reader is referred to Shaw and Haddow
[193], where other interesting choices of f (x) can be found. A further interesting
choice would be Euler’s spiral (clothoid) that features in “looptheloop” roller
coasters.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
2.11 A Simple Model for a Roller Coaster 63
x
dx
dτ
−2 2
−1.5
1.5
Figure 2.12. The phase portrait of (2.27)–(2.29) when friction is absent. The points on the
x =
x
1
L
0
axis labeled by · correspond to equilibria of the cart. For this ﬁgure,
A
L
0
= 0.25, and
µ
s
= µ
d
= 0.0.
States of Rest
For a cart on a smooth roller coaster, the motion of the cart will be perpetual, and
a portion of its phase portrait is shown in Figure 2.12. We note the presence of an
equilibrium at
x = 0,
dx
dτ
= 0
. This point corresponds to the cart’s being stationary
at the top of the roller coaster. Referring to Figure 2.13(a), we refer to equilibria
of this type as saddles. The two equilibria at
x = ±1,
dx
dτ
= 0
represent a stationary
cart at the bottom of one of the valleys of the roller coaster. Examining the phase
portrait in Figure 2.13(c), we easily see why equilibria of this type are known as
centers. The equilibria at
x = (−2, 0, 2),
dx
dτ
= 0
correspond to a stationary cart at
one of the crests of the roller coaster.
When stick–slip friction is present, the phase portrait changes dramatically (see
Figure 2.14). First, all of the saddles have split, and between their two split halves
we have what we call a sticking region [see Figure 2.13(b)]. Depending on the value
of x
1
, this region contains either ˙ x
1
> 0 or ˙ x
1
< 0. If the cart’s state enters this region
then the cart will stop. That is, the cart will come to rest near a crest of the roller
coaster. Similarly, the equilibria of the smooth roller coaster at the ﬂoor of its valleys
have now transformed from discrete points to regions surrounding these points [see
Figure 2.13(d)]. These regions are also sticking regions, and if the cart’s state enters
this region, then the cart will stop.
The size of the sticking region is easy to compute by use of (2.30), and a graphi
cal method is shown in Figure 2.15. As µ
s
gets larger, the size of the sticking region
(or sticking states) surrounding the equilibria when µ
d
= 0 grows, and eventually
any point (x, 0) on the
dx
dτ
axis will become an equilibrium, and thus a state of rest
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
64 Kinetics of a Particle
x
x
x x
dx
dτ
dx
dτ
dx
dτ
dx
dτ
(a)
(b)
(c)
(d)
−0.2
−0.2
−0.2
−0.2
−0.2
−0.2
0.2
0.2
0.2
0.2
0.2
0.2
0.8
0.8
1.2
1.2
s
s
Figure 2.13. Expanded views of the phase portraits in the neighborhood of equilibria of (2.27)
with f speciﬁed by (2.31). For (a) and (c), µ
d
= 0 and the roller coaster is smooth, whereas
for (b) and (d), µ
d
= 0.1. For the latter cases, the sticking regions s are shown for the case in
which µ
s
= 0.3.
for the roller coaster.
∗
This phenomenon is also easy to explain physically. We note
for completeness that, for the present choice of f (x), if
µ
s
≥
πA
L
0
,
then it is possible to stick at any location on the roller coaster. This can also be
inferred from the graphical technique shown in Figure 2.15.
2.12 Closing Comments
A vast amount of material has been covered in this chapter, starting with descrip
tions of various forces, discussions of the balance laws, and analyses of various ap
plications. The analyses we employed invariably featured the numerical integration
of an ordinary differential equation and an interpretation of its solutions. Develop
ing physical interpretations of the results provided by the model is one of the most
rewarding aspects of dynamics; however, it can also be the most time consuming.
∗
For the phase portrait shown in Figure 2.13(b), the sticking region s is the interval
[−0.124755, 0.124755], and for situation shown in Figure 2.13(d), the sticking region s is the interval
[1. −0.124755, 1.124755].
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
2.12 Closing Comments 65
x
dx
dτ
−2 2
−1.5
1.5
Figure 2.14. The phase portrait of (2.27)–(2.29). Although it is not evident from the ﬁgure, the
discrete equilibria of the frictionless case shown in Figure 2.12 are now replaced with families
of equilibria that correspond to possible resting (sticking) states for the cart. For this ﬁgure,
A
L
0
= 0.25 and µ
d
= 0.1.
In many of the chapters to follow, several more examples of such interpretations
are presented, and you are strongly encouraged to take the time to do this when
completing the exercises in this book or performing your own research.
x =
x
1
L
0
x =
x
1
L
0
f
µ
s
= 0.3
−2
−2
2
2
0.2
s s s s
s s s s
s
Figure 2.15. A graphical method to compute the possi
ble sticking regions s of the cart on a smooth roller
coaster when µ
s
= 0.3. The method is based on ex
amining (2.30) for the choice (2.31). That is,
f
=
Aπ
L
0
sin
πx
1
L
0
. Examples of the sticking regions can be
seen in Figure 2.13.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
66 Exercises 2.1–2.5
EXERCISES
2.1. Which of the following force ﬁelds are conservative/nonconservative?
P = x
1
E
1
+x
3
E
2
,
P = x
2
E
1
+x
1
E
2
,
P = x
1
x
2
E
1
,
P = −L
0
sin(θ)E
1
+L
0
cos(θ)E
2
,
where L
0
is a constant. For the conservative force ﬁelds, what are the associated
potential energies?
2.2. Consider a particle of mass m that is moving in E
3
. Suppose the only forces
acting on the particle are conservative. Starting from the work–energy theorem,
prove that the total energy E of the particle is conserved.
Suppose during a motion, for which the initial conditions r
0
and v
0
are known, the
position r (t
1
) at some later time t
1
is known. Argue that the conservation of energy
can be used to determine the speed v of the particle. Give three distinct physical
examples of applications of this result.
2.3. In contrast to Exercise 2.2, here consider a particle that is moving on a
smooth ﬁxed surface. The constraint force acting on the particle is prescribed by
use of Lagrange’s prescription, and the applied forces acting on the particle are
conservative. Prove that E is again conserved. In addition, show that the speed of
the particle can be determined at a known position r(t
1
) if the initial position and
velocity vectors are known. Finally, give three distinct physical examples of the
application of this result.
2.4. A particle is free to move on a smooth horizontal surface x
3
= 0. At the same
time, a rough plane propels the particle in the E
1
direction. That is, the constraints
on the motion of the particle are
1
= 0 and
2
= 0, where
1
=
1
(r) = x
3
,
2
=
2
(r, t) = x
1
− f (t).
Give a prescription for the constraint force acting on the particle.
2.5. Suppose a particle of mass m is in motion and has a position vector r and a
velocity vector v.
(a) Show that the areal velocity vector A is conserved if the resultant force F
acting on the particle is a central force.
∗
(b) Show that conservation of angular momentum H
O
is synonymous with
conservation of the areal velocity vector.
(c) Suppose a particle is moving on a horizontal table under the action of a
spring force, a normal force, and a vertical gravitational force −mgE
3
. One
end of the spring is attached to the ﬁxed origin O, and the other is attached
∗
A force P is said to be central if P is parallel to r.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
Exercises 2.5–2.6 67
to the particle. The spring has a stiffness K and unstretched length L
0
.
What can you say about the area swept out by the particle in a given period
of time?
(d) Derive the equations of motion for the particle in (c). Using the nondimen
sionalizations
τ =
K
m
t, x =
r
L
0
,
and conservation of angular momentum, show that the motion of the
particle can be found by integrating the following differential equations:
d
2
x
dτ
2
−
β
2
x
3
= −(x −1) ,
dθ
dτ
=
β
x
2
, (2.32)
where
β =
h
L
2
0
√
Km
and h is a constant that depends on the initial conditions of the motion.
For a selection of values of β, e.g., β = −20, −2, −1, 0, 1, 2, 20, construct
the phase portraits of (2.32)
1
. For a selection of the orbits on each of these
phase portraits, construct images of the motion of the particle.
∗
(e) Verify that the areal velocity vector is conserved for the motions of the
particle you found in (d).
2.6. A particle of mass m is in motion about a ﬁxed planet of mass M. The external
force acting on the body is assumed to be a conservative force P. The potential
energy U
P
associated with this force is a function of r, where r is the position
vector of the particle relative to the ﬁxed center Oof the planet.
(a) Prove that r is parallel to P.
(b) Show that the angular momentum H
O
of the particle is conserved and that
this conservation implies that the motion of the particle is planar. This
plane, which is known as the orbital plane, also contains O. Show that the
particle sweeps out equal areas on its orbital plane in equal times.
(c) Write out the equations of motion of the particle using a spherical polar
coordinate system.
(d) Using the conservation of H
O
, show that the equations of (c) can be
simpliﬁed to
m
¨ r −r
˙
θ
2
= −
∂U
P
∂r
,
mr
2
˙
θ = h, (2.33)
where h is a constant.
∗
Your results will be qualitatively similar to those presented in Section 2.9 for the particle moving on
a smooth cone.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
68 Exercises 2.6–2.7
(e) Show that the solutions to (2.33) conserve the total energy E of the
particle.
2.7. A particle of mass m is free to move on the inner surface of a rough sphere
of constant radius R
0
. The center of the sphere is located at the origin O, and the
particle is attached to a ﬁxed point A whose position vector is aE
x
+bE
y
by a
linear spring of unstretched length L
0
and stiffness K. A vertical gravitational force
−mgE
3
also acts on the particle.
(a) Using a spherical polar coordinate system, r = R
0
e
R
, derive expres
sions for the acceleration vector a and angular momentum H
O
of the
particle.
(b) What is the velocity vector of the particle relative to a point on the surface
of the sphere?
(c) Give a prescription for the constraint force F
c
acting on the particle.
(d) If the particle is moving relative to the surface, show that the equations
governing the motion of the particle are
mR
0
(
¨
φ −sin(φ) cos(φ)
˙
θ
2
) = mg sin(φ) −K(x −L
0
)
x · e
φ
x
−µ
d
N
˙
φ
˙
φ
2
+sin
2
(φ)
˙
θ
2
,
1
R
0
sin(φ)
d
dt
mR
2
0
sin
2
(φ)
˙
θ
= −K(x −L
0
)
x · e
θ
x
−µ
d
N
sin(φ)
˙
θ
˙
φ
2
+sin
2
(φ)
˙
θ
2
,
where x = R
0
e
R
−aE
x
−bE
y
.
(e) Show that the normal force exerted by the surface on the particle is
N =
mgE
3
· e
R
+K(x −L
0
)
x · e
R
x
e
R
−mR
0
(
˙
φ
2
+sin
2
(φ)
˙
θ
2
)e
R
.
(f) For the case in which the particle is not moving relative to the surface,
show that
F
c
= mgE
3
+K(x −L
0
)
x
x
.
What is the static friction criterion for this case?
(g) Show that the total energy of the particle decreases with time if the particle
moves relative to the surface.
(h) If the spring is removed and the surface is assumed to be smooth, prove
that the angular momentum H
O
· E
3
is conserved. Using this conservation,
show that the dimensionless equations governing the motion of the particle
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
Exercises 2.7–2.8 69
simplify to
dθ
dτ
=
h
sin
2
(φ)
,
d
2
φ
dτ
2
= h
2
cos(φ)
sin
3
(φ)
+sin(φ).
In these equations, h =
1
mR
2
0
R
0
g
H
0
· E
3
and τ =
g
R
0
t.
(i) Suppose that the sphere is smooth. Using the fact that the total energy
E of the particle is conserved, show that the criterion for the particle to
remain on the surface of the sphere is
mg (3 cos(φ) −2 cos (φ
0
)) −mR
0
˙
φ
2
0
+sin
2
(φ
0
)
˙
θ
2
0
> 0,
where φ
0
is the value of the initial φ coordinate of the particle and
˙
θ
0
and
˙
φ
0
are the initial velocities. Suppose the particle is placed on top of the
sphere. If the particle is given an initial speed v
0
>
√
gR
0
, show that it
will immediately lose contact with the sphere.
2.8. Consider a particle of mass m whose motion is subject to the following
constraints:
(xE
3
+E
2
) · v = 0, (E
2
) · v +e(t) = 0. (2.34)
(a) Show that one of the constraints is integrable whereas the other is non
integrable. In addition, for the integrable constraint, specify the function
(r, t) = 0.
(b) Suppose that, in addition to the constraint force
F
c
= µ
1
(xE
3
+E
2
) +µ
2
E
2
,
a gravitational force −mgE
3
acts on the particle. With the help of the
balance of linear momentum F = ma, specify the equations governing the
motion of the particle and the constraint forces.
(c) With the help of the work–energy theorem
˙
T = F · v, prove that the total
energy of the particle is not conserved. Give a physical interpretation for
this lack of conservation.
(d) Using the results from (b), determine the motion of the particle and the
constraint force F
c
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
3 Lagrange’s Equations of Motion for a
Single Particle
3.1 Introduction
The balance of linear momentum F = ma for a particle can be traced to Newton
in the late 17th century. As we have seen, this vectorvalued equation yields three
differential equations from which the motion of a particle can be determined. In the
centuries that followed, alternative principles of mechanics were proposed. Some
of them, such as the principle of least action, also yielded equations of motion that
were equivalent to those obtained with F = ma. Others did not, and the equivalence
of, and interrelationships between, the principles of mechanics remains one of the
central issues for any student of dynamics.
At the end of the 18th century, a formulation of the equations of motion for a
single particle appeared in a famous text by Lagrange [121].
∗
Among their attrac
tive features, Lagrange’s equations of motion could easily accommodate integrable
constraints, and they (remarkably) have the same canonical form both for single
particles and systems of particles as well as for systems of rigid bodies.
In this chapter, Lagrange’s equations of motion for a single particle are dis
cussed and several forms of these equations are established. For example [see (3.2)],
d
dt
∂L
∂ ˙ q
i
−
∂L
∂q
i
= F
ncon
· a
i
.
Many of the forms presented can be used with dynamic Coulomb friction and non
conservative forces. One of the most important features of our discussion is the em
phasis on the equivalence of Lagrange’s equations of motion to F = ma. Although
this equivalence is not sufﬁciently discussed in most textbooks, it can be found in
many of the classical texts on dynamics, such as those of Synge and Grifﬁth [207]
and Whittaker [228]. A recent paper by Casey [27] explores this equivalence in a
transparent manner, and we follow many aspects of his exposition in this chapter.
∗
Four editions of Lagrange’s great work, M´ ecanique Analytique, appeared in the years 1789, 1811,
1853, and 1888. The last two of these editions were posthumous. An English translation of the
second edition was recently published [122].
70
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
3.2 Lagrange’s Equations of Motion 71
Casey’s work will also feature later on when we discuss systems of particles and rigid
bodies.
3.2 Lagrange’s Equations of Motion
There are several approaches to deriving Lagrange’s equations of motion that ap
pear in the literature. Among them, a variational principle known as Hamilton’s
principle (or the principle of least action) is arguably the most popular, whereas an
approach based on D’Alembert’s principle was used by Lagrange [121]. Lagrange’s
original developments were in the context of mechanical systems subject to holo
nomic constraints, and his equations were subsequently extended to systems with
nonholonomic constraints by Edward J. Routh (1831–1907) (see Section 24 in Chap
ter IV of [183]) and Aurel Voss (1845–1931) [221].
∗
Here, an approach is used that is rooted in differential geometry and is con
tained in some texts on this subject (see, for example, Synge and Schild [208]). It
probably migrated from there to the classic text by Synge and Grifﬁth [207] and has
been recently revived by Casey [27].
Two Identities
We assume that a curvilinear coordinate systemhas been chosen for E
3
. The velocity
vector v consequently has the representation
v =
3
¸
i=1
˙ q
i
a
i
.
In addition, the kinetic energy has the representations
T =
m
2
v · v =
m
2
3
¸
i=1
3
¸
k=1
a
i
· a
k
˙ q
i
˙ q
k
.
It is crucial to notice that T = T
q
1
, q
2
, q
3
, ˙ q
1
, ˙ q
2
, ˙ q
3
.
We now consider in succession the partial derivatives of T with respect to the
coordinates and their velocities. We wish to establish the following results:
∂T
∂q
i
= mv · ˙ a
i
,
∂T
∂ ˙ q
i
= mv · a
i
.
These two elegant results form the basis for Lagrange’s equations of motion.
First, we start with the derivative of T with respect to a coordinate:
∂T
∂q
i
=
∂
∂q
i
m
2
v · v
= mv ·
∂v
∂q
i
.
∗
For further details on the historical development of Lagrange’s equations, see Papastavridis [167,
169]. Equations of motion (3.8)
2,3
are examples of what could be referred to as the Routh–Voss
equations of motion.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
72 Lagrange’s Equations of Motion for a Single Particle
To proceed with the goal of concluding that
∂T
∂q
i
= mv · ˙ a
i
, we ﬁrst note that v =
¸
3
k=1
˙ q
k
a
k
and that
∂ ˙ q
k
∂q
i
= 0. As a consequence,
∂T
∂q
i
= mv ·
3
¸
k=1
˙ q
k
∂a
k
∂q
i
.
The remaining steps use the fact that a
k
=
∂r
∂q
k
:
∂T
∂q
i
= mv ·
3
¸
k=1
˙ q
k
∂a
k
∂q
i
= mv ·
3
¸
k=1
˙ q
k
∂
2
r
∂q
i
∂q
k
=
3
¸
k=1
˙ q
k
∂
2
r
∂q
k
∂q
i
= mv ·
3
¸
k=1
˙ q
k
∂
∂q
k
∂r
∂q
i
= a
i
= mv · ˙ a
i
.
We achieve the last step by noting that
˙
f =
¸
3
k=1
∂f
∂q
k
˙ q
k
for any function f =
f
q
1
, q
2
, q
3
.
The next result, which is far easier to establish, involves the partial derivative of
T with respect to a velocity. The reason this result is easier to establish is because
the basis vectors a
i
do not depend on ˙ q
k
. Getting on with the proof, we have
∂T
∂ ˙ q
i
= mv ·
3
¸
k=1
∂ ˙ q
k
∂ ˙ q
i
a
k
= mv ·
3
¸
k=1
δ
k
i
a
k
= mv · a
i
.
This completes the proofs of both identities.
A Covariant Form of Lagrange’s Equations
It is crucial to note that Lagrange’s equations are equivalent to F = ma. The form
of Lagrange’s equations of motion discussed here is derived from this balance law
by taking its covariant components, i.e., dotting it with a
i
.
To start, we consider
d
dt
∂T
∂ ˙ q
i
−
∂T
∂q
i
=
d
dt
(mv · a
i
) −mv · ˙ a
i
= ma · a
i
+mv · ˙ a
i
−mv · ˙ a
i
= ma · a
i
= F · a
i
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
3.3 Equations of Motion for an Unconstrained Particle 73
In conclusion, we have a covariant form of Lagrange’s equations of motion:
d
dt
∂T
∂ ˙ q
i
−
∂T
∂q
i
= F · a
i
. (3.1)
We can appreciate some of the beauty of this equation by using it to establish com
ponent forms of F = ma for various curvilinear coordinate systems.
The Lagrangian
Another form of Lagrange’s equations arises when we decompose the force F into
its conservative and nonconservative parts:
F = −∇U+F
ncon
,
where the potential energy U = U(q
1
, q
2
, q
3
). As
∇U =
3
¸
k=1
∂U
∂q
k
a
k
,
∂U
∂ ˙ q
k
= 0,
we ﬁnd that Lagrange’s equations can be rewritten in the form
d
dt
∂T
∂ ˙ q
i
−
∂U
∂ ˙ q
i
−
∂T
∂q
i
−
∂U
∂q
i
= F
ncon
· a
i
.
Introducing the Lagrangian L = T −U, we ﬁnd an alternative form of Lagrange’s
equations:
d
dt
∂L
∂ ˙ q
i
−
∂L
∂q
i
= F
ncon
· a
i
. (3.2)
If there are no nonconservative forces acting on the particle, then the righthand
side of these equations vanishes. In addition, to calculate the equations of motion a
minimal amount of vector calculus is required – it is sufﬁcient to calculate v and U.
3.3 Equations of Motion for an Unconstrained Particle
To illustrate the ease of Lagrange’s equations, we consider the case in which the
curvilinear coordinates chosen are the spherical polar coordinates: q
1
= R, q
2
= φ,
and q
3
= θ. For these coordinates, we have
a
1
= e
R
, a
2
= Re
φ
, a
3
= Rsin(φ)e
θ
,
and
T =
m
2
(
˙
R
2
+R
2
sin
2
(φ)
˙
θ
2
+R
2
˙
φ
2
).
Notice that T does not depend on θ.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
74 Lagrange’s Equations of Motion for a Single Particle
We obtain Lagrange’s equations of motion for the spherical polar coordinate
system by ﬁrst calculating the six partial derivatives of T:
∂T
∂R
= mRsin
2
(φ)
˙
θ
2
+mR
˙
φ
2
,
∂T
∂φ
= mR
2
sin(φ) cos(φ)
˙
θ
2
,
∂T
∂θ
= 0,
∂T
∂
˙
R
= m
˙
R,
∂T
∂
˙
φ
= mR
2
˙
φ,
∂T
∂
˙
θ
= mR
2
sin
2
(φ)
˙
θ.
Using these results, we ﬁnd the covariant form of Lagrange’s equations:
d
dt
∂T
∂
˙
R
= m
˙
R
−
∂T
∂R
= mRsin
2
(φ)
˙
θ
2
+mR
˙
φ
2
= F · e
R
,
d
dt
∂T
∂
˙
φ
= mR
2
˙
φ
−
∂T
∂φ
= mR
2
sin(φ) cos(φ)
˙
θ
2
= F · Re
φ
,
d
dt
∂T
∂
˙
θ
= mR
2
sin
2
(φ)
˙
θ
−
∂T
∂θ
= 0
= F · Rsin(φ)e
θ
. (3.3)
Clearly, these equations were far easier to calculate than an alternative approach
that involves differentiating r = Re
R
twice with respect to t.
Let us now suppose that the only force acting on the particle is gravity:
F = −mgE
3
, U = mgE
3
· r = mgRcos(φ).
For this case, the Lagrangian L is
L = T −U
=
m
2
(
˙
R
2
+R
2
sin
2
(φ)
˙
θ
2
+R
2
˙
φ
2
) −mgRcos(φ).
We can calculate Lagrange’s equations of motion using L,
d
dt
∂L
∂ ˙ q
k
−
∂L
∂q
k
= 0,
or by substituting for F in (3.3). It is left as an exercise to show that both approaches
are equivalent.
3.4 Lagrange’s Equations in the Presence of Constraints
The previous discussion of Lagrange’s equations did not address situations in which
constraints on the motion of the particle were present. It is to this matter that we
now turn our attention. With integrable constraints, whose constraint forces are pre
scribed by use of Lagrange’s prescription, the beauty and power of Lagrange’s equa
tions are manifested. In this case, it is possible to choose the curvilinear coordinates
q
i
such that the equations of motion decouple into two sets. The ﬁrst set describes
the unconstrained motion of the particle, and the second set yields the constraint
forces as functions of the unconstrained motion.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
3.4 Lagrange’s Equations in the Presence of Constraints 75
There are two approaches to obtaining Lagrange’s equations. We refer to them
throughout these sections as Approach I and Approach II. For the novice, we highly
recommended the ﬁrst approach. As in the previous section, our exposition follows
that of Casey [27].
Preliminaries
We assume that the particle is subject to an integrable constraint,
(r, t) = 0,
and a nonintegrable constraint,
f · v +e = 0.
Further, we assume that the curvilinear coordinates are chosen such that the inte
grable constraint has the form
(r, t) = q
3
−d(t) = 0,
and that the constraint forces are prescribed by use of Lagrange’s prescription:
F
c
= λ
1
a
3
+λ
2
3
¸
i=1
f
i
a
i
.
Notice that f
i
= f · a
i
.
Suppose that there is an applied force F
a
acting on the particle. This applied
force can be decomposed into conservative and nonconservative parts:
F
a
= −∇U+F
ancon
.
The resultant force acting on the particle is
F = λ
1
a
3
+λ
2
3
¸
i=1
f
i
a
i
−∇U+F
ancon
.
The total nonconservative force acting on the particle is F
ncon
= F
c
+F
ancon
.
The kinetic energy of the particle is
T =
m
2
v · v =
m
2
3
¸
i=1
3
¸
k=1
a
ik
q
1
, q
2
, q
3
˙ q
i
˙ q
k
,
where a
ik
= a
i
· a
k
. Imposing the integrable constraint on T, we ﬁnd the constrained
kinetic energy:
˜
T =
˜
T
2
+
˜
T
1
+
˜
T
0
, (3.4)
where
˜
T
2
=
m
2
2
¸
i=1
2
¸
k=1
˜ a
ik
˙ q
i
˙ q
k
,
˜
T
1
= m
2
¸
i=1
˜ a
i3
˙ q
i
˙
d,
˜
T
0
=
m
2
˜ a
33
˙
d
2
. (3.5)
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
76 Lagrange’s Equations of Motion for a Single Particle
In these expressions,
˜ a
ik
= ˜ a
ik
(q
1
, q
2
, t) = a
ik
(q
1
, q
2
, q
3
= d(t))
is the constrained metric tensor. Notice that we use a tilde (˜) to denote imposition
of the integrable constraint(s) and that the subscripts on
˜
T
2
,
˜
T
1
and
˜
T
0
refer to the
powers of ˙ q
i
.
A direct calculation shows that
∗
∂T
∂ ˙ q
1
q
3
=d, ˙ q
3
=
˙
d
=
∂
˜
T
∂ ˙ q
1
,
∂T
∂ ˙ q
2
q
3
=d, ˙ q
3
=
˙
d
=
∂
˜
T
∂ ˙ q
2
,
∂T
∂q
1
q
3
=d, ˙ q
3
=
˙
d
=
∂
˜
T
∂q
1
,
∂T
∂q
2
q
3
=d, ˙ q
3
=
˙
d
=
∂
˜
T
∂q
2
,
∂T
∂ ˙ q
3
q
3
=d, ˙ q
3
=
˙
d
=
∂
˜
T
∂ ˙ q
3
= 0,
∂T
∂q
3
q
3
=d, ˙ q
3
=
˙
d
=
∂
˜
T
∂q
3
= 0.
(3.6)
In these relations, the partial derivative of T is evaluated prior to imposing the con
straint q
3
= d(t). These relations imply that we can use
˜
T to obtain the ﬁrst two
Lagrange’s equations of motion, but not the third. Results that are identical in form
to (3.6) pertain to the partial derivatives of L and
˜
L and Uand
˜
U.
Notice that we did not impose the nonintegrable constraint on the kinetic en
ergy T and the Lagrangian L. It is possible to do this, but the result is not useful to
us here.
Approach I
In the ﬁrst approach, we evaluate the partial derivatives in Lagrange’s equations of
motion (3.2) in the absence of any constraints:
d
dt
∂L
∂ ˙ q
k
−
∂L
∂q
k
= F
ancon
· a
k
+(F
c
= 0) · a
k
.
Explicitly, these equations are
d
dt
m
3
¸
i=1
a
i1
˙ q
i
−
3
¸
i=1
3
¸
r=1
m
2
∂a
ir
∂q
1
˙ q
i
˙ q
r
+
∂U
∂q
1
= F
ancon
· a
1
,
d
dt
m
3
¸
i=1
a
i2
˙ q
i
−
3
¸
i=1
3
¸
r=1
m
2
∂a
ir
∂q
2
˙ q
i
˙ q
r
+
∂U
∂q
2
= F
ancon
· a
2
,
d
dt
m
3
¸
i=1
a
i3
˙ q
i
−
3
¸
i=1
3
¸
r=1
m
2
∂a
ir
∂q
3
˙ q
i
˙ q
r
+
∂U
∂q
3
= F
ancon
· a
3
.
Notice that we have not introduced the constraint forces on the righthand side of
these equations. That is, in the preceding equations F = −∇U+F
ancon
.
We now impose the integrable constraint q
3
= d(t) and introduce the nonin
tegrable constraint and the constraint forces. The resulting equations govern the
∗
Suppose g = 10t
2
. Then
∂g
∂t
t=5
= 2(10)(5) = 100. In words, we evaluate the derivative of g with
respect to t and then substitute t = 5 in the resulting function.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
3.4 Lagrange’s Equations in the Presence of Constraints 77
motion of the particle and the constraint forces:
q
3
= d,
˙ q
3
=
˙
d,
f
1
˙ q
1
+ f
2
˙ q
2
+ f
3
˙
d +e = 0,
d
dt
3
¸
i=1
ma
i1
˙ q
i
−
3
¸
i=1
3
¸
r=1
m
2
∂a
ir
∂q
1
˙ q
i
˙ q
r
+
∂U
∂q
1
= λ
2
f
1
+F
ancon
· a
1
,
d
dt
m
3
¸
i=1
a
i2
˙ q
i
−
3
¸
i=1
3
¸
r=1
m
2
∂a
ir
∂q
2
˙ q
i
˙ q
r
+
∂U
∂q
2
= λ
2
f
2
+F
ancon
· a
2
,
d
dt
m
3
¸
i=1
a
i3
˙ q
i
−
3
¸
i=1
3
¸
r=1
m
2
∂a
ir
∂q
3
˙ q
i
˙ q
r
+
∂U
∂q
3
= λ
1
+λ
2
f
3
+F
ancon
· a
3
. (3.7)
We have refrained from ornamenting U, f
i
, a
i
, and a
ik
with a tilde in the last four of
these equations.
It is crucial to notice that if the nonintegrable constraint were absent, then (3.7)
would reduce to two sets of equations. The ﬁrst of these sets, (3.7)
4,5
, would yield
differential equations for the unconstrained motion, q
1
(t), and q
2
(t), of the particle,
whereas the second set, (3.7)
6
, would provide the constraint force F
c
= λ
1
a
3
acting
on the particle.
Approach II
In Approach II, we work directly with
˜
L =
˜
T −
˜
U. Here, as
˜
L =
˜
L
q
1
, q
2
, ˙ q
1
, ˙ q
2
, t
,
the partial derivatives of
˜
L with respect to q
3
and ˙ q
3
are zero. Consequently, using
(3.6) and (3.2), we ﬁnd that there are only two Lagrange’s equations:
d
dt
∂
˜
L
∂ ˙ q
1
−
∂
˜
L
∂q
1
= F
c
· ˜ a
1
+F
ancon
· ˜ a
1
,
d
dt
∂
˜
L
∂ ˙ q
2
−
∂
˜
L
∂q
2
= F
c
· ˜ a
2
+F
ancon
· ˜ a
2
.
Introducing the expression for the constraint force F
c
and the nonintegrable con
straint, we ﬁnd the equations governing λ
2
, q
1
(t), and q
2
(t) are
f
1
˙ q
1
+ f
2
˙ q
2
+ f
3
˙
d +e = 0,
d
dt
∂
˜
L
∂ ˙ q
1
−
∂
˜
L
∂q
1
= λ
2
f
1
+F
ancon
· ˜ a
1
,
d
dt
∂
˜
L
∂ ˙ q
2
−
∂
˜
L
∂q
2
= λ
2
f
2
+F
ancon
· ˜ a
2
. (3.8)
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
78 Lagrange’s Equations of Motion for a Single Particle
Notice that λ
1
does not feature in these equations. In addition, if no nonintegrable
constraint were present, then the differential equations provided by Approach II
are all that are needed to determine q
1
(t) and q
2
(t). Equations (3.8)
2,3
are examples
of the Routh–Voss equations of motion.
3.5 A Particle Moving on a Sphere
To clarify the two approaches just discussed, we now consider the example of a
particle moving on a smooth sphere whose radius R is a known function of time:
R = d(t). The particle is subject to a conservative force −mgE
3
and a nonconserva
tive force DRe
θ
, where D is a constant. Later on, we shall impose a nonintegrable
constraint on the motion of the particle.
For the problem at hand, it is convenient to use a spherical polar coordinate
system:
q
1
= θ, q
2
= φ, q
3
= R.
Using this coordinate system, we can write the integrable constraint R = d(t) in the
form
(r, t) = R−d(t) = 0.
As the sphere is smooth, we can use Lagrange’s prescription,
F
c
= λe
R
.
The kinetic and potential energies of a particle in the chosen coordinate system are
T =
m
2
˙
R
2
+R
2
sin
2
(φ)
˙
θ
2
+R
2
˙
φ
2
, U = mgRcos(φ).
The constrained kinetic and potential energies are
˜
T =
m
2
˙
d
2
+d
2
sin
2
(φ)
˙
θ
2
+d
2
˙
φ
2
,
˜
U = mgdcos(φ). (3.9)
Finally, the covariant basis vectors are
a
1
= Rsin(φ)e
θ
, a
2
= Re
φ
, a
3
= e
R
.
Their constrained counterparts ˜ a
i
are easily inferred from these expressions.
First, we use Approach II to obtain the equations governing θ(t) and φ(t). There
are two equations:
d
dt
∂
˜
L
∂
˙
θ
−
∂
˜
L
∂θ
= F · ˜ a
1
= F
c
· ˜ a
1
+Dde
θ
· ˜ a
1
= Dd
2
sin(φ),
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
3.5 A Particle Moving on a Sphere 79
d
dt
∂
˜
L
∂
˙
φ
−
∂
˜
L
∂φ
= F · ˜ a
2
= F
c
· ˜ a
2
+Dde
θ
· ˜ a
2
= 0.
Evaluating the partial derivatives of the constrained Lagrangian, we ﬁnd that these
equations become
d
dt
(md
2
sin
2
(φ)
˙
θ) = Dd
2
sin(φ),
d
dt
md
2
˙
φ
−md
2
sin(φ) cos(φ)
˙
θ
2
−mgdsin(φ) = 0. (3.10)
Notice that the constraint force λe
R
is absent from these equations.
Alternatively, using Approach I, we start with the unconstrained Lagrangian L
and establish three equations of motion [cf. (3.3)]:
d
dt
∂L
∂
˙
θ
= mR
2
sin
2
(φ)
˙
θ
−
∂L
∂θ
= 0
= F
ancon
· Rsin(φ)e
θ
,
d
dt
∂L
∂
˙
φ
= mR
2
˙
φ
−
∂L
∂φ
= mR
2
sin(φ) cos(φ)
˙
θ
2
+mgRsin(φ)
= F
ancon
· Re
φ
,
d
dt
∂L
∂
˙
R
= m
˙
R
−
∂L
∂R
= mRsin
2
(φ)
˙
θ
2
+mR
˙
φ
2
−mg cos(φ)
= F
ancon
· e
R
.
Next, we impose the integrable constraint and introduce the constraint force F
c
to
ﬁnd the equations of motion:
d
dt
(md
2
sin
2
(φ)
˙
θ) = Dd
2
sin(φ),
d
dt
(md
2
˙
φ) −md
2
sin(φ) cos(φ)
˙
θ
2
−mgdsin(φ) = 0,
d
dt
(m
˙
d) −(mdsin
2
(φ)
˙
θ
2
+md
˙
φ
2
−mg cos(φ)) = λ. (3.11)
Notice that the ﬁrst two of these equations are identical to (3.10), whereas the third
equation is an equation for the constraint force F
c
.
We could now introduce an additional constraint:
f
1
˙
θ + f
2
˙
φ + f
3
˙
R+e = 0.
Using Lagrange’s prescription, we ﬁnd that the total constraint force on the particle
is
F
c
= λe
R
+λ
2
f
1
Rsin(φ)
e
θ
+
f
2
R
e
φ
+ f
3
e
R
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
80 Lagrange’s Equations of Motion for a Single Particle
To obtain the equations of motion for the case in which the nonintegrable constraint
is active, we need to introduce only the constraint force associated with the nonin
tegrable constraint on the righthand side of (3.11) and to append the nonintegrable
constraint to the resulting equations:
f
1
˙
θ + f
2
˙
φ + f
3
˙
d +e = 0,
d
dt
(md
2
sin
2
(φ)
˙
θ) = Dd
2
sin(φ) +λ
2
f
1
,
d
dt
(md
2
˙
φ) −md
2
sin(φ) cos(φ)
˙
θ
2
−mgdsin(φ) = λ
2
f
2
,
d
dt
(m
˙
d) −(mdsin
2
(φ)
˙
θ
2
+md
˙
φ
2
−mg cos(φ)) = λ +λ
2
f
3
.
It is left as an exercise to show what additional simpliﬁcations to these equations
arise if the nonintegrable constraint were integrable with f
1
= 0, f
2
= 1, f
3
= 0, and
e = 0. In this case, one will see that the particle moves on a circle of radius dsin(φ
0
).
3.6 Some Elements of Geometry and Particle Kinematics
As a prelude to our discussion of Lagrange’s equations and their geometrical signif
icance, some material from differential geometry needs to be presented. Our treat
ment is limited to the ingredients we shall shortly need and, as such, it cannot do jus
tice to this wonderful subject. Mercifully, there are several excellent texts that can
be recommended to remedy this: [47, 149, 155, 201]. Reading Chapter 1 of Lanczos
[124] for a related discussion on kinetic energy and geometry and the recent paper
by L¨ utzen [132] for a historical overview of the interaction between geometry and
dynamics in the 19th century is highly recommended.
Here, we are interested in surfaces and curves that are in E
3
. We assume that
these entities are smooth. That is, they are without edges and sharp corners, and
we call them manifolds. In the case of a curve, a single coordinate is needed to
locally parameterize the points P on this manifold, and so it is considered to be
a onedimensional manifold. For a surface, two coordinates are needed to locally
parameterize the points on the surface and so the surface is considered to be a
twodimensional manifold. In an obvious generalization, subsets of E
3
such as solid
spheres and solid ellipsoids are considered to be threedimensional manifolds.
Previously, in Section 1.5, curvilinear coordinates were introduced. For a given
surface (or curve) we used these coordinates both to label points on the manifold
and to deﬁne the manifold. For example, for a sphere of radius R
0
, the spherical
polar coordinates φ and θ label points on the sphere and the coordinate R can
be used to deﬁne the sphere: R = R
0
. Similarly, for a circle, the cylindrical polar
coordinate θ can be used to label points on the circle, and the coordinates r and z can
be used to deﬁne the circle. The curvilinear coordinate system we use to label points
on the manifold is known as a chart. For some manifolds, such as a plane, a straight
line, and a circle, a single chart sufﬁces to enable the labeling of each point on the
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
3.6 Some Elements of Geometry and Particle Kinematics 81
(a)
(b)
P
P
C
T
P
C
S
T
P
S
Figure 3.1. Two examples of manifolds and the tangent spaces to points on them: (a) a curve
C and (b) a sphere S.
manifold. For surfaces such as spheres, for which a set of spherical polar coordinates
will not be deﬁned at the poles, at least two charts are needed. With terminology
borrowed from cartography, the set of all charts for a manifold is known as an atlas.
At each point P of a manifold M we deﬁne a tangent space, and we denote this
space by T
P
M. If the manifold is ndimensional, then T
P
M is also ndimensional.
For example, the tangent space T
P
C is a line for the curve shown in Figure 3.1(a), and
the tangent space T
P
S is a plane for the sphere S shown in Figure 3.1(b). Continuing
with the sphere as an example, if we ﬁx a point P on the sphere, this is equivalent to
ﬁxing the polar coordinates φ = φ
0
and θ = θ
0
. The vectors
e
φ
= e
φ
(φ
0
, θ
0
) = sin(φ
0
) (cos (θ
0
) E
1
+sin(θ
0
) E
2
) +cos (φ
0
) E
3
,
e
θ
= e
θ
(θ
0
) = −sin(θ
0
) E
1
+cos (θ
0
) E
2
,
form a basis for the tangent space T
P
S at P, and any tangent vector to the sphere
at this point can be expressed in terms of these vectors. Related remarks apply at a
point on a curve, but now only a single vector is needed to span the tangent space.
As a ﬁnal example, for a particle that is free to move in M = E
3
, the dimension of
T
P
M is 3.
Returning to the example of a sphere, we choose two points P
1
and P
2
on the
sphere of radius R and consider a path V between them (see Figure 3.2). We wish
to measure the distance one would travel along V. To do this, we ﬁrst parameterize
the curve with a parameter u, where u = u
α
at P
α
. The curve can then be uniquely
described by the functions θ(u) and φ(u).
∗
To determine the distance s traveled
along the curve, one method would be to evaluate the following integral:
s =
u
2
u
1
R
2
sin
2
(φ)
dθ
du
dθ
du
+R
2
dφ
du
dφ
du
du. (3.12)
Referring to (3.4), (3.5), and (3.9), we can express the integrand on the righthand
side of this equation in terms of the kinetic energy of a particle moving on the
∗
For example, if the curve were a segment of the equator, then θ (u) = θ (u
1
) +
(u
2
−u
1
)
2πR
and φ(u) =
π
2
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
82 Lagrange’s Equations of Motion for a Single Particle
P
1
P
2
VVVVV
S
v
rel
Figure 3.2. A curve V connecting two points on a
sphere S. The velocity vector v
rel
of a particle moving
on this curve would lie in the tangent plane T
P
S at each
point P of the curve.
sphere:
s =
u
2
u
1
2
˜
T
2
m
dt
du
du
=
t
2
t
1
2
˜
T
2
m
dt =
t
2
t
1
v
rel
 dt.
For the second integral, we changed variables and parameterized the path by using
t rather than u. Turning to the example of a circle, the reader is invited to show
that a measure of distance corresponding to (3.12) can be established by using the
single polar coordinate θ. For M = E
3
, the measure of distance can be deﬁned in a
standard manner by use of Cartesian coordinates:
s =
u
2
u
1
3
¸
k=1
dx
k
du
dx
k
du
du.
It is easy to see that this expression can be rewritten as
t
2
t
1
2T
m
dt.
From the previous discussion, for an ndimensional manifold M, a measure of
distance similar to that provided by (3.12) can be established
∗
:
s =
u
2
u
1
n
¸
i=1
n
¸
k=1
˜ a
ik
∂q
i
∂u
∂q
k
∂u
du.
Parameterizing the path by using time t instead of a variable u, we ﬁnd
s =
t
2
t
1
n
¸
i=1
n
¸
k=1
˜ a
ik
˙ q
i
˙ q
k
dt. (3.13)
∗
The index n here is either 1, 2, or 3. In later chapters, we shall see that n can range from 1 to 3N for
a system of N particles and from 1 to 6 for a single rigid body.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
3.7 The Geometry of Lagrange’s Equations of Motion 83
Following the advocacy of Hertz [92], we can conveniently imagine a (represen
tative) particle of mass m moving on a manifold M. The manifold is known as
the conﬁguration manifold. The velocity of the particle relative to M is simply
v
rel
=
¸
n
k=1
˙ q
k
˜ a
k
. Thus (3.13) can be expressed as
s =
t
2
t
1
v
rel
 dt. (3.14)
We see from this expression that, in deﬁning s, we have also deﬁned a measure of
the magnitude of a vector v
rel
∈ T
P
M. Such a measure is known as a metric, and a
manifold that is equipped with a metric is known as a Riemannian manifold. The ter
minology here pays tribute to Georg F. B. Riemann (1826–1866) and his remarkable
work [179].
Distance measure (3.13) is clearly intimately related to the kinetic energy of a
particle, and, following Synge [205],
ds =
n
¸
i=1
n
¸
k=1
˜ a
ik
˙ q
i
˙ q
k
dt (3.15)
is known as the kinematical lineelement. This measure of distance has a long his
tory with important contributions by several esteemed ﬁgures such as Jacobi [103]
and Ricci and LeviCivita [178]. Other choices of ds are available (see, for instance,
[123, 124, 149, 205]), and some of them feature prominently in relativistic mechanics.
The freedom of selection is similar to the notion that different measures of distance
such as meters and feet are possible and is intimately related to Einstein’s theory of
relativity.
We remarked earlier on a (representative) particle of mass m moving on M.
Clearly, for a single particle, such a construction can be easily achieved. Indeed, for
a single particle the conﬁguration manifold corresponds to the physical surface or
curve that the particle moves on. However, for a system of particles or rigid bodies
subject to constraints, this is not the case, and the construction of a single represen
tative particle is nontrivial. Indeed, for a system of particles, such a construction was
explicitly recorded only rather recently by Casey [27].
∗
Subsequently, he extended
this construction to a single rigid body [28] and a system of rigid bodies [30].
3.7 The Geometry of Lagrange’s Equations of Motion
Some readers will have gained the perspective that the Lagrange’s equations of mo
tion obtained by use of Approach II are projections of F = ma onto the covariant
basis vectors for the unconstrained coordinates. That is, we are projecting F = ma
onto the basis for T
P
M.
For those who have not yet found this perspective, let us recall the example of
the particle moving on the sphere of radius R = d(t). There, we obtained the two
Lagrange’s equations for the θ and φ by taking the dsin(φ)e
θ
and de
φ
components
∗
We shall examine his construction in Section 4.7.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
84 Lagrange’s Equations of Motion for a Single Particle
of F = ma. These two vectors, dsin(φ)e
θ
and de
φ
, form a basis for the tangent space
T
P
S to a point P of the sphere S. Furthermore, because the constraint force associ
ated with the integrable constraint λe
R
is perpendicular to the sphere, this force did
not appear in the two Lagrange’s equations.
An important feature of nonintegrable constraints is that the constraint force
associated with these constraints is not decoupled from the equations governing
the unconstrained motion. This deﬁciency in Lagrange’s equations of motion can
be removed by use of alternative forms of F = ma that are suited to nonintegrably
constrained systems, but we do not approach this vast subject here. We now delve
a little more deeply into the geometry inherent in Lagrange’s equations of motion.
Our discussion is based on Casey [27], Lanczos [124], and Synge [205].
A Particle Subject to a Single Integrable Constraint
First, let us consider the case in which the particle is subject to a single integrable
constraint:
(r, t) = q
3
−d(t) = 0,
where d
3
(t) is a known function. When considered in E
3
, the constraint = 0
represents a moving twodimensional surface: In this case, a q
3
coordinate surface.
As mentioned earlier, this surface is known as the conﬁguration manifold M (see
Figure 3.3). The velocity of the particle relative to this surface has the representation
v
rel
= ˙ q
1
˜ a
1
+ ˙ q
2
˜ a
2
.
The coordinates q
1
and q
2
in this case are known as the generalized coordinates,
and the number of these coordinates is the number of degreesoffreedom of the
particle. Thus an unconstrained particle has three degreesoffreedom, whereas the
particle constrained to move on the surface has only two.
Recall that at each point P of M, {a
1
, a
2
} evaluated at P is a basis for the tangent
plane T
P
M to M at P. We can also deﬁne a relative kinetic energy T
rel
=
˜
T
2
:
T
rel
=
m
2
v
rel
· v
rel
=
2
¸
i=1
2
¸
k=1
m
2
˜ a
i
· ˜ a
k
˙ q
i
˙ q
k
.
We now consider a particle moving on M. To calculate the distance traveled by the
particle on the surface in a given interval t
1
−t
0
of time t, we integrate the magnitude
of its velocity v
rel
with respect to time:
s(t
1
) −s(t
0
) =
t
1
t
0
√
v
rel
· v
rel
dt
=
t
1
t
0
2T
rel
m
dt.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
3.7 The Geometry of Lagrange’s Equations of Motion 85
O
q
1
coordinate curve
q
2
coordinate curve
q
3
= d(t) coordinate surface
a
1
a
2
a
3
M
Figure 3.3. The conﬁguration manifold M of a particle moving on a surface. Here, the coor
dinates q
1
and q
2
are known as the generalized coordinates and Mis the q
3
= d(t) coordinate
surface.
We can differentiate this result with respect to t to ﬁnd the kinematical lineelement
ds:
ds =
2T
rel
m
dt =
2
¸
i=1
2
¸
k=1
˜ a
i
· ˜ a
k
dq
i
dq
k
. (3.16)
As emphasized previously in Section 3.6, notice that the measure of distance is de
ﬁned by the kinetic energy T
rel
. It is left as an exercise to show that the integral of
(3.16) is none other than (3.12).
With regard to Lagrange’s equations of motion, suppose that the constraint
forces associated with the integrable constraint are prescribed by use of Lagrange’s
prescription: F
c
= λ˜ a
3
. Then F
c
· ˜ a
1
= F
c
· ˜ a
2
= 0, and the constraint force does not
appear in the ﬁrst two Lagrange’s equations. The forces
Q
1
= F · ˜ a
1
, Q
2
= F · ˜ a
2
are known as the generalized forces, and the expressions we use for them are equiv
alent to those used in other texts on dynamics (for example, [14, 80]). In addition,
Approach II can be used to obtain the differential equations governing q
1
(t) and
q
2
(t):
d
dt
∂
˜
L
∂ ˙ q
1
−
∂
˜
L
∂q
1
= F
ancon
· ˜ a
1
,
d
dt
∂
˜
L
∂ ˙ q
2
−
∂
˜
L
∂q
2
= F
ancon
· ˜ a
2
. (3.17)
Imposing a nonintegrable constraint on the motion of the particle will not change
M. Furthermore, this constraint will, in general, introduce constraint forces into
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
86 Lagrange’s Equations of Motion for a Single Particle
Equations (3.17). These forces will destroy the decoupling that Lagrange’s equa
tions achieve for integrable constraints.
A Particle Subject to Two Integrable Constraints
We now turn to the case in which a particle is subject to two integrable constraints:
1
(r, t) = q
3
−d
3
(t) = 0,
2
(r, t) = q
2
−d
2
(t) = 0.
Notice that we have chosen the curvilinear coordinates so that the constraints are
easily represented. At each instant of time, the intersection of the two surfaces
1
= 0 and
2
= 0 in E
3
deﬁnes a curve – in this case a q
1
coordinate curve (see
Figure 3.4). The conﬁguration manifold M in this case corresponds to the q
1
coor
dinate curve. This coordinate is the generalized coordinate for the system.
We can easily represent the velocity vector of the particle relative to M, which
we again denote by v
rel
:
v
rel
= ˙ q
1
˜ a
1
.
It should be clear that ˜ a
1
is tangent to M. Indeed, a
1
evaluated at P is a basis vector
for the onedimensional space T
P
M. In addition, we can associate with v
rel
a relative
kinetic energy:
T
rel
=
m
2
v
rel
· v
rel
=
m
2
˜ a
1
· ˜ a
1
˙ q
1
˙ q
1
.
Paralleling previous developments [see (3.15)], the kinematical lineelement for M
is
ds =
2T
rel
m
dt =
˜ a
1
· ˜ a
1
dq
1
dq
1
.
O
q
2
= d
2
(t) coordinate surface
q
3
= d
3
(t) coordinate surface
˜ a
1
r
m
Figure 3.4. A particle moving on a curve. In this ﬁgure, ˜ a
1
is tangent to the q
1
coordinate curve
corresponding to q
2
= d
2
(t) and q
3
= d
3
(t). That is, the q
1
coordinate curve is the conﬁgura
tion manifold M. The vectors ˜ a
2
and ˜ a
3
, which are not shown, are normal to M.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
3.8 A Particle Moving on a Helix 87
O
increasing s & θ
E
1
E
2
E
3
e
t
e
t
e
n
e
n
e
b
e
b
Figure 3.5. A helix and its associated Frenet triad
{e
t
, e
n
, e
b
}. Here, e
t
is the unit tangent vector, e
n
is the unit principal normal vector, and e
b
= e
t
×
e
n
is the binormal vector.
With regard to Lagrange’s equations of motion, if the constraint forces are pre
scribed by use of Lagrange’s prescription, F
c
= λ
1
˜ a
3
+λ
1
˜ a
2
, then we can easily ﬁnd
the differential equation governing q
1
(t) by using Approach II:
d
dt
∂
˜
L
∂ ˙ q
1
−
∂
˜
L
∂q
1
= F
ancon
· ˜ a
1
.
Here,
˜
L =
˜
L
q
1
, ˙ q
1
, t
and F · ˜ a
1
is the generalized force for this problem.
3.8 A Particle Moving on a Helix
As an illustrative example, we turn our attention to establishing results for a particle
that is in motion on a helix (see Figure 3.5). The helix can be either rough or smooth,
and a variety of applied forces are considered. This example is interesting for sev
eral reasons. First, it is a prototypical problem to illustrate how the Serret–Frenet
formulae and the Frenet triad {e
t
, e
n
, e
b
} help to determine the motion of a parti
cle on a space curve. Second, we can use this example to illustrate a nonorthogonal
curvilinear coordinate system.
∗
Curvilinear Coordinates, Basis Vectors, and Other Kinematics
A helix is deﬁned by the intersection of two surfaces: a cylinder r = Rand a helicoid
z = cθ, where c and Rare constants. To conveniently deﬁne these surfaces, we deﬁne
a curvilinear coordinate system:
q
1
= θ = tan
−1
x
2
x
1
, q
2
= r =
x
2
1
+x
2
2
,
q
3
= η = z −αrθ = x
3
−α
x
2
1
+x
2
2
tan
−1
x
2
x
1
.
∗
This is a coordinate system in which a
i
are not necessarily parallel to a
i
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
88 Lagrange’s Equations of Motion for a Single Particle
It is appropriate to notice that
r = x
1
E
1
+x
2
E
2
+x
3
E
3
= r cos(θ)E
1
+r sin(θ)E
2
+(η +αrθ) E
3
.
Our labeling of the coordinates minimizes subsequent manipulations. You should
note that the curvilinear coordinate system is not deﬁned when r = 0. That is, it has
the same singularities as the cylindrical and spherical polar coordinate systems.
The coordinates θ, r, and η can be used to deﬁne bases for E
3
:
a
1
= re
θ
+αrE
3
, a
2
= e
r
+αθE
3
, a
3
= E
3
.
In addition, using the representation of the gradient in cylindrical polar coordinates,
we ﬁnd that the contravariant basis vectors are
a
1
=
1
r
e
θ
, a
2
= e
r
, a
3
= E
3
−αθe
r
−αe
θ
.
You should notice that a
i
· a
j
= δ
j
i
, as expected. Further, neither the covariant basis
nor the contravariant basis is orthogonal.
You may recall that the Frenet triad for the helix of radius R is (from [159])
e
t
=
1
√
1 +α
2
(e
θ
+αE
3
) , e
n
= −e
r
, e
b
=
1
√
1 +α
2
(E
3
−αe
θ
) .
Furthermore, the torsion τ, curvature κ, and arclength parameter s of the helix are
τ =
α
R(1 +α
2
)
, κ =
1
R(1 +α
2
)
, s = R
1 +α
2
(θ −θ
0
) −s
0
.
These results also apply to a helix for which α and R are functions of time. You
should verify that a
1
is parallel to e
t
and that a
2
and a
3
are in the plane formed by e
n
and e
b
.
For a particle moving freely in E
3
, we have the general representation v =
¸
3
i=1
˙ q
i
a
i
. From this result, we can immediately write
v =
˙
θ(re
θ
+αrE
3
) + ˙ r(e
r
+αθE
3
) + ˙ ηE
3
.
Furthermore, the kinetic energy of the particle is
T =
m
2
˙ r
2
+r
2
˙
θ
2
+(˙ η +α˙ rθ +αr
˙
θ)
2
.
When the particle is in motion on the helix, it is subject to two constraints
1
= 0
and
2
= 0:
1
(r, t) = q
2
−R,
2
(r, t) = q
3
.
In preparation for writing expressions for the constraint forces acting on a particle
moving on the helix, you should calculate the gradient of these two functions. You
might also notice that θ is the generalized coordinate for a particle moving on the
helix that we will be using.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
3.8 A Particle Moving on a Helix 89
Forces
We assume that an applied force F
a
acts on the particle. In addition, we assume that
the friction is of the Coulomb type. Consequently, if the particle is moving relative
to the helix,
F = F
a
+λ
1
˜ a
2
+λ
2
˜ a
3
+F
f
,
where
F
f
= −µ
d
λ
1
˜ a
2
+λ
2
˜ a
3
˙
θ˜ a
1
˙
θ˜ a
1
.
On the other hand, if the particle is not moving relative to the helix, i.e., θ is constant,
then
F = F
a
+λ
1
˜ a
2
+λ
2
˜ a
3
+λ
3
˜ a
1
.
The friction force in this case is subject to the static friction criterion:
F
f
≤ µ
s
N ,
where
F
f
=
(λ
1
˜ a
2
+λ
2
˜ a
3
+λ
3
˜ a
1
) ·
˜ a
1
 ˜ a
1

˜ a
1
 ˜ a
1

,
N = λ
1
˜ a
2
+λ
2
˜ a
3
+λ
3
˜ a
1
−F
f
.
Balance of Linear Momentum and Lagrange’s Equations
For an unconstrained particle moving in E
3
, we have the three Lagrange’s equations:
d
dt
∂T
∂ ˙ q
i
−
∂T
∂q
i
= F · a
i
.
For the present coordinate system θ, r, η, these equations read
d
dt
∂T
∂
˙
θ
= mr
2
˙
θ +mαr(˙ η +α˙ rθ +αr
˙
θ)
−
∂T
∂θ
= mα˙ r(˙ η +α˙ rθ +αr
˙
θ)
= F · (a
1
= re
θ
+αrE
3
),
d
dt
∂T
∂ ˙ r
= m˙ r +mαθ(˙ η +α˙ rθ +αr
˙
θ)
−
∂T
∂r
= mr
˙
θ
2
+mα
˙
θ(˙ η +α˙ rθ +αr
˙
θ)
= F · (a
2
= e
r
+αθE
3
),
d
dt
∂T
∂ ˙ η
= m(˙ η +α˙ rθ +αr
˙
θ)
−
∂T
∂η
= 0
= F · (a
3
= E
3
) .
Equations of Motion for the Particle on the Helix
We obtain the equations of motion for the particle on the helix from the preceding
equations by substituting for the resultant force and imposing the constraints. With
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
90 Lagrange’s Equations of Motion for a Single Particle
some algebra, for the case in which the particle is moving relative to the helix, we
ﬁnd three equations:
d
dt
m(1 +α
2
)R
2
˙
θ
= F
a
· ˜ a
1
−µ
d
λ
1
˜ a
2
+λ
2
˜ a
3
 ˜ a
1

˙
θ

˙
θ
,
d
dt
mα
2
Rθ
˙
θ
−m(1 +α
2
)R
˙
θ
2
= F
a
· ˜ a
2
+λ
1
−µ
d
λ
1
˜ a
2
+λ
2
˜ a
3
˙
θ˜ a
1
· ˜ a
2
˙
θ˜ a
1
,
d
dt
mαR
˙
θ
= F
a
· ˜ a
3
+λ
2
−µ
d
λ
1
˜ a
2
+λ
2
a
3
˙
θ˜ a
1
· ˜ a
3
˙
θ˜ a
1
,
where
˜ a
1
= Re
θ
+αRE
3
, ˜ a
2
= e
r
+αθE
3
, ˜ a
3
= E
3
.
These three equations provide a differential equation for the unconstrained motion
of the particle and two equations for the unknowns λ
1
and λ
2
.
For the case in which the motion of the particle is speciﬁed (i.e., the particle
is not moving relative to the helix), we ﬁnd, from F = ma, three equations for the
three unknowns:
λ
3
= −F
a
· ˜ a
1
, λ
2
= −F
a
· ˜ a
2
, λ
1
= −F
a
· ˜ a
3
.
It remains to use the static friction criterion, but this is left as an easy exercise.
The Particle on a Smooth Helix
In this case,
F = F
a
+λ
1
˜ a
2
+λ
2
˜ a
3
,
and because F
c
· ˜ a
1
= 0, Lagrange’s equations of motion decouple:
d
dt
m(1 +α
2
)R
2
˙
θ
= F
a
· ˜ a
1
,
d
dt
mα
2
Rθ
˙
θ
−m(1 +α
2
)R
˙
θ
2
= F
a
· ˜ a
2
+λ
1
,
d
dt
mαR
˙
θ
= F
a
· ˜ a
3
+λ
2
.
Consequently, the desired differential equation is
m(1 +α
2
)R
2
¨
θ = F
a
· (Re
θ
+αRE
3
),
and the constraint force is
F
c
= λ
1
˜ a
2
+λ
2
˜ a
3
=
mα
2
R
¨
θθ −mR
˙
θ
2
−F
a
· ˜ a
2
˜ a
2
+
mαR
¨
θ −F
a
· ˜ a
3
˜ a
3
.
Once θ as a function of time has been calculated from the ordinary differential equa
tion, then F
c
as a function of time can be determined.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
3.9 Summary 91
To illustrate the previous equations, consider the case in which the applied force
is gravitational, F
a
= −mgE
3
. Then, from the preceding equations,
m(1 +α
2
)R
2
¨
θ = −mgαR. (3.18)
Subject to the initial conditions θ(t
0
) = θ
0
and
˙
θ(t
0
) = ω
0
, this equation has the solu
tion
θ(t) = θ
0
+ω
0
(t −t
0
) −
gα
2R(1 +α
2
)
(t −t
0
)
2
.
Using this result, we ﬁnd that the constraint force is
F
c
=
mgαθ(t)
1 +α
2
−mR
˙
θ
2
(t)
˜ a
2
+
mg
1 +α
2
(E
3
−α(θe
r
+e
θ
)) ,
where θ(t) is as previously given.
Some Observations
Suppose one is interested in determining only the differential equation governing
the unconstrained motion of the particle moving on a smooth helix. In other words,
the constraint forces are of no concern. One can obtain this differential equation by
imposing the constraints on the expression for T:
˜
T =
m
2
R
2
˙
θ
2
+(αR
˙
θ)
2
=
m
2
(1 +α
2
)R
2
˙
θ
2
.
Furthermore,
d
dt
∂
˜
T
∂
˙
θ
−
∂
˜
T
∂θ
= F · ˜ a
1
= F · (Re
θ
+αRE
3
) = F
a
· (Re
θ
+αRE
3
) .
A quick calculation shows that the resulting differential equation is identical to that
obtained previously [(3.18)].
Clearly, Lagrange’s equations calculated with Approach II (i.e., with
˜
T) have
their advantages, but they cannot accommodate dynamic friction forces. It is, how
ever, the standard approach to Lagrange’s equations in the literature and textbooks.
You should note that
∂
˜
T
∂ ˙ r
=
∂
˜
T
∂ ˙ η
=
∂
˜
T
∂r
=
∂
˜
T
∂η
= 0. Consequently, we cannot recover the
other two Lagrange’s equations once we have imposed the constraints.
3.9 Summary
In this chapter, several forms of Lagrange’s equations of motion for a particle were
presented. The most fundamental of these forms is [see (3.1)]
d
dt
∂T
∂ ˙ q
i
−
∂T
∂q
i
= F · a
i
.
In one of the following exercises, we establish two other forms of these equations by
expanding the partial derivatives with respect to the coordinates and their velocities.
These two forms are a covariant form (3.22) and a contravariant form (3.23). If we
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
92 Exercise 3.1
decompose the forces acting on the particle into conservative and nonconservative
forces, then we can transform (3.1) to (3.2):
d
dt
∂L
∂ ˙ q
i
−
∂L
∂q
i
= F
ncon
· a
i
.
Now suppose that an integrable constraint is imposed on the particle, that this con
straint can be written as q
3
− f (t) = 0, and that the constraint force associated with
this constraint is F
c
= λa
3
. In this case, Lagrange’s equations of motion can be used
to readily provide a set of differential equations for the generalized coordinates q
1
and q
2
:
d
dt
∂
˜
L
∂ ˙ q
α
−
∂
˜
L
∂q
α
= F
ncon
· ˜ a
α
, α = 1, 2. (3.19)
These equations feature the constrained Lagrangian
˜
L that we obtain from L by
imposing the integrable constraint q
3
= f (t), and, most important, do not feature λ.
That is, equations of motion (3.19) are reactionless. This case, in which all the con
straints are integrable and the constraint forces are prescribed by use of Lagrange’s
prescription, is an example of a mechanical system subject to “ideal constraints.”
We also discussed the situation in which nonintegrable constraints were im
posed on the system and outlined how the equations of motion could be obtained in
these circumstances. The imposition of nonintegrable constraints will not affect the
number of generalized coordinates, the conﬁguration manifold, or the kinematical
lineelement.
The summary just presented will be identical for systems of particles, rigid bod
ies, and systems of both particles and rigid bodies. The only major differences are
that the calculation of the kinetic energy becomes signiﬁcantly more complicated
for these systems and that the righthand sides of Lagrange’s equations feature sev
eral forces and moments. Despite these differences, the decoupling of the equations
of motion into a set of reactionless equations governing the generalized coordinates
will hold if Lagrange’s prescription for the constraint forces is used. This is one of
the most remarkable features of Lagrange’s equations for systems subject to inte
grable constraints.
EXERCISES
3.1. Recall from Exercise 1.5 in Chapter 1 that, for a parabolic coordinate system
{u, v, θ},
a
1
=
∂r
∂u
= ve
r
+uE
3
, a
2
=
∂r
∂v
= ue
r
−vE
3
,
a
3
=
∂r
∂θ
= uve
θ
,
and
a
1
=
1
u
2
+v
2
a
1
, a
2
=
1
u
2
+v
2
a
2
, a
3
=
1
uv
e
θ
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
Exercises 3.1–3.2 93
(a) Consider a particle of mass m that is acted on by a force F and is free to
move in E
3
. Show that the equations of motion of the particle are
d
dt
m(u
2
+v
2
) ˙ u
−m( ˙ u
2
+ ˙ v
2
)u −mv
2
u
˙
θ
2
= F · a
1
,
d
dt
m(u
2
+v
2
)˙ v
−m( ˙ u
2
+ ˙ v
2
)v −mu
2
v
˙
θ
2
= F · a
2
,
d
dt
mu
2
v
2
˙
θ
= F · a
3
.
(b) Next, we are interested in a particle that is moving on the parabolic surface
of revolution:
c
2
= −z +
z
2
+r
2
,
where c is a constant. A vertical gravitational force −mgE
3
acts on the par
ticle. Using the results of (a), derive the equations governing the uncon
strained motion of the particle and show that the normal force acting on
the particle is
N = −
m˙ u
2
c +mu
2
c
˙
θ
2
+mgc
a
2
.
Show that the two secondorder differential equations governing the
generalized coordinates can be written as a single secondorder differential
equation:
m(u
2
+c
2
)
¨ u +m˙ u
2
u −
h
2
mu
3
c
2
= −mgu, (3.20)
where h is a constant. (This constant is none other than H
O
· E
3
, which is
an integral of motion). Noting that the units of u and c are meters
1/2
, what
is a dimensionless form of equations of motion (3.20)?
(c) Show that the solutions of (3.20) conserve the energy
E =
m
2
(u
2
+c
2
) ˙ u
2
+
h
2
2mu
2
c
2
+
mg
2
(u
2
−c
2
).
How does one arrive at this expression for E?
3.2. For many mechanical systems a canonical form of Lagrange’s equations can
be established that is suited to numerical integrations. Here, we establish one such
form [see (3.23)].
∗
This problem is adapted from the texts of McConnell [139] and
Synge and Schild [208]. We take this opportunity to note that (3.23) can be found in
an early paper by Ricci and LeviCivita [178].
We start by recalling the covariant component forms of Lagrange’s equations of
motion for a particle that is in motion under the inﬂuence of a resultant external
∗
As will become evident from the developments of later chapters, a related form can be established
for any mechanical system that features scleronomic integrable constraints and constraint forces
and moments that are prescribed by use of Lagrange’s prescription.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
94 Exercise 3.2
force F =
¸
3
i=1
F
i
a
i
=
¸
3
i=1
F
i
a
i
∗
:
d
dt
∂T
∂ ˙ q
k
−
∂T
∂q
k
= F · a
k
,
where
T = T(q
r
, ˙ q
s
) =
m
2
3
¸
i=1
3
¸
k=1
a
ik
˙ q
i
˙ q
k
, (3.21)
and
a
ik
= a
ik
(q
r
) = a
i
· a
k
, a
ik
= a
ik
(q
r
) = a
i
· a
k
.
You should notice that a
ik
= a
ki
and a
ik
= a
ki
.
Here, we wish to show that Lagrange’s equations can be written in two other
equivalent forms. The ﬁrst one is the covariant form:
m
3
¸
i=1
a
ki
¨ q
i
+m
3
¸
i=1
3
¸
s=1
[si, k] ˙ q
i
˙ q
s
=
˙
G· a
k
= F
k
, (3.22)
where a Christoffel symbol of the ﬁrst kind is deﬁned by
[si, k] =
∂a
s
∂q
i
· a
k
.
It is important to note that [si, k] =[is, k]. The second form of Lagrange’s equations
is known as the contravariant form:
m¨ q
k
+m
3
¸
i=1
3
¸
s=1
k
si
˙ q
i
˙ q
s
=
˙
G· a
k
= F
k
, (3.23)
where a Christoffel symbol of the second kind is deﬁned by
k
i j
=
∂a
i
∂q
j
· a
k
.
Notice that
k
i j
=
k
ji
. This form of Lagrange’s equations is used in numerical simu
lations of mechanical systems.
†
(a) Show that the Christoffel symbols have the representations
[si, k] =
1
2
∂a
ki
∂q
s
+
∂a
sk
∂q
i
−
∂a
si
∂q
k
,
and
k
i j
=
3
¸
r=1
a
kr
[i j, r].
∗
The indices i, j, k, r, and s range from 1 to 3.
†
Most numerical integration packages assume that the differential equations to be integrated are of
the form ˙ x = f(x, t). By deﬁning the set of variables (states) x
1
= q
1
, . . . , x
3
= q
3
, x
4
= ˙ q
1
, . . . , x
6
=
˙ q
3
, the contravariant form of Lagrange’s equations can be easily placed in the form ˙ x = f(x, t).
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
Exercises 3.2–3.3 95
(b) Starting from Lagrange’s equations,
d
dt
∂T
∂ ˙ q
k
−
∂T
∂q
k
= F · a
k
,
derive the following representation for the covariant component form
∗
:
m
3
¸
i=1
a
ki
¨ q
i
+m
3
¸
i=1
3
¸
s=1
[si, k] ˙ q
i
˙ q
s
=
˙
G· a
k
= F
k
.
(c) Starting from Lagrange’s equations in the form
m
3
¸
i=1
a
ki
¨ q
i
+m
3
¸
s=1
3
¸
i=1
[si, k] ˙ q
i
˙ q
s
=
˙
G· a
k
= F
k
,
derive the following representation for the contravariant component form
†
:
m¨ q
k
+m
3
¸
s=1
3
¸
i=1
k
si
˙ q
i
˙ q
s
=
˙
G· a
k
= F
k
.
(d) For which coordinate system do the Christoffel symbols vanish?
3.3. Recall that for spherical polar coordinates, {R, φ, θ}, the covariant basis vectors
are
a
1
= e
R
, a
2
= Re
φ
, a
3
= Rsin(φ)e
θ
,
and the contravariant basis vectors are
a
1
= e
R
, a
2
=
1
R
e
φ
, a
3
=
1
Rsin(φ)
e
θ
.
Furthermore, the linear momentum and kinetic energy of a particle of mass m are
G = m
˙
Ra
1
+m
˙
φa
2
+m
˙
θa
3
, T =
m
2
˙
R
2
+R
2
˙
φ
2
+R
2
sin
2
(φ)
˙
θ
2
.
(a) For a particle of mass m that is in motion in E
3
under the inﬂuence of a
resultant force F, establish the three covariant components of Lagrange’s
equations of motion. In your solution, avoid explicitly calculating the 27
Christoffel symbols of the ﬁrst kind.
(b) For a particle of mass m that is in motion in E
3
under the inﬂuence of a re
sultant force F, establish the three contravariant components of Lagrange’s
equations of motion. In your solution, avoid explicitly calculating the 27
Christoffel symbols of the second kind.
∗
Hint : Expand the partial derivatives of Tusing the representation (3.21). Then, take the appropriate
time derivative and reorganize the resulting equation by using the aforementioned symmetries. You
may need to relabel certain indices to obtain the desired results.
†
Hint : Multiply the covariant form by a
sk
and sum over k. After some rearranging and relabeling
of the indices, you should get the ﬁnal desired result. Notice that the covariant component and
contravariant component forms of these equations can be viewed as linear combinations of each
other.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
96 Exercises 3.4–3.5
3.4. Consider a particle that is in motion on a rough surface. A curvilinear coordi
nate system q
1
, q
2
, q
3
is chosen such that the surface can be described by the equa
tion
q
3
= d(t),
where d(t) is a known function of time t.
(a) Suppose that the particle is moving on the rough surface.
(i) Argue that v
rel
= ˙ q
1
˜ a
1
+ ˙ q
2
˜ a
2
.
(ii) Give a prescription for the constraint force acting on the particle.
(b) Suppose that the particle is stationary on the rough surface. In this case, two
equivalent prescriptions for the constraint force are
F
c
= N+F
f
=
3
¸
i=1
λ
i
a
i
,
where the tildes are dropped for convenience.
(i) Show that
⎡
⎢
⎣
λ
1
λ
2
λ
3
⎤
⎥
⎦
=
⎡
⎢
⎣
a
11
a
12
0
a
12
a
22
0
a
13
a
23
1
⎤
⎥
⎦
⎡
⎢
⎣
F
1
f
F
2
f
N
⎤
⎥
⎦
,
where N, and F
1
f
and F
2
f
uniquely deﬁne the normal force N and friction
force F
f
, respectively, and a
ik
= a
i
· a
k
with i, k = 1, 2, 3.
(ii) For which coordinate systems do F
1
f
= λ
1
, F
2
f
= λ
2
, and N = λ
3
? Give
an example to illustrate your answer.
(c) Suppose that a spring force and a gravitational force also act on the particle.
Prove that the total energy of the particle is not conserved, even when the
friction force is static.
3.5. Consider a particle of mass m that is in motion on a helicoid. In terms of cylin
drical polar coordinates r, θ, z, the equation of the right helicoid is
z = αθ,
where α is a constant. A gravitational force −mgE
3
acts on the particle.
(a) Consider the following curvilinear coordinate system for E
3
:
q
1
= θ, q
2
= r, q
3
= ν = z −αθ.
Show that
a
1
= re
θ
+αE
3
, a
2
= e
r
, a
3
= E
3
,
and that
a
1
=
1
r
e
θ
, a
2
= e
r
, a
3
= E
3
−
α
r
e
θ
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
Exercises 3.5–3.6 97
(b) Consider a particle moving on the smooth helicoid:
(i) What is the constraint on the motion of the particle, and what is a pre
scription for the constraint force F
c
enforcing this constraint?
(ii) Show that the equations governing the unconstrained motion of the
particle are
d
dt
m
r
2
+α
2
˙
θ
= −mgα,
d
dt
(m˙ r) −mr
˙
θ
2
= 0. (3.24)
(iii) Prove that the angular momentum H
O
· E
3
is not conserved.
(c) Suppose the nonintegrable constraint
r
˙
θ +h(t) = 0
is imposed on the particle. Establish a secondorder differential equation
for r(t), a differential equation for θ(t), and an equation for the constraint
force enforcing the nonintegrable constraint. Indicate how you would solve
these equations to determine the motion of the particle and the constraint
forces acting on it.
3.6. Consider a particle of mass m that is free to move on the smooth inner surface
of a hemisphere of radius R
0
(cf. Figure 3.6). The particle is under the inﬂuence of a
gravitational force −mgE
3
.
m
O
E
1
E
2
E
3
r
g
Figure 3.6. Schematic of a particle of mass m moving on the inside of a hemisphere of radius
R
0
.
(a) Using a spherical polar coordinate system, what is the constraint on the
motion of the particle? Give a prescription for the constraint force acting
on the particle.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
98 Exercises 3.6–3.7
(b) Using Lagrange’s equations, establish the equations of motion for the par
ticle and an expression for the constraint force.
(c) Prove that the total energy E and the angular momentum H
O
· E
3
of the
particle are conserved.
(d) Show that the normal force acting on the particle can be expressed as a
function of the position of the particle and its initial energy E
0
:
N =
−
2E
0
R
0
+2mg sin(φ) +mg cos (φ)
e
R
.
(e) Numerically integrate the equations of motion of the particle and show that
there are instances for which it will always remain on the surface of the
hemisphere.
3.7. As shown in Figure 3.7, consider a bead of mass m that is free to move on
a smooth semicircular wire of radius R
0
. The wire has a constant angular velocity
0
E
3
and whirls about the conﬁguration shown in the ﬁgure. The particle is also
m
O
E
1
E
2
E
3
r
g
(t)
(t)
Figure 3.7. Schematic of a particle of mass m moving on a semicircular path that is being
whirled about the vertical at a speed (t) =
0
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
Exercises 3.7–3.8 99
under the inﬂuence of a gravitational force −mgE
3
. This is a classical problem that
is discussed in several textbooks (see, for example, [78]).
(a) Using a spherical polar coordinate system, what are the two constraints on
the motion of the particle? Give a prescription for the constraint force F
c
acting on the particle.
(b) Using Lagrange’s equations, establish the equation of motion for the
particle:
¨
φ =
2
0
cos (φ) +
g
R
0
sin(φ) . (3.25)
After nondimensionalizing (3.25), numerically integrate the resulting dif
ferential equation and construct its phase portrait for values of 0.5, 1.0, 1.5
of the parameter
g
R
0
2
0
.
(c) Recall that an equilibrium point x = x
0
of the differential equation ¨ x = f (x)
is such that ˙ x = 0 and f (x
0
) = 0. Show that (3.25) has three equilibria:
φ
0
= 0, φ
0
= π, φ
0
= cos
−1
−
g
R
0
2
0
.
Give physical interpretations for these equilibria and show that the third
one is possible if, and only if,
2
0
is sufﬁciently large. How do these results
correlate to your phase portraits?
(d) Starting from the work–energy theorem
˙
T = F · v, prove that the total
energy of the particle is not conserved:
˙
E = N
θ
R
0
0
sin(φ) ,
where N
θ
is the e
θ
component of the normal force acting on the particle.
3.8. Consider a particle of mass m moving in E
3
. If coordinate system (1.6) is used
to describe its kinematics, then establish expressions for the velocity vector v and
the kinetic energy T of the particle.
(a) Suppose a particle is constrained to move on a rough parabolic surface de
scribed by the equation
x −y
2
= −4.
Give a prescription for the constraint force F
c
acting on the particle, and
establish the equations of motion for the particle.
(b) As illustrated in Figure 3.8, suppose a particle is constrained to move on the
smooth parabola
x −y
2
= −4, z = 0.
Give a prescription for the constraint force F
c
acting on the particle, and
establish the equations of motion for the particle.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
100 Exercise 3.9
m
y
x
6
2
−4
−4
q
2
= −4
Figure 3.8. Schematic of a particle moving
on a parabola in the x −y plane.
3.9. Consider a particle of mass m moving on a smooth ellipsoid:
x
2
a
2
+
y
2
b
2
+
z
2
c
2
= 1.
(a) With the help of a suitable curvilinear coordinate system, establish expres
sions for the constrained kinetic energy of the particle.
∗
(b) Assuming that a gravitational force −mgE
3
acts on the particle, establish
the two secondorder differential equations governing the motion of the
particle.
(c) Numerically integrate the equations you found in (b) and discuss features
of the motion of the particle.
∗
The coordinate systems you need are not discussed in this text but are readily found in the literature.
They are often known as “confocal ellipsoidal coordinates.”
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
PART TWO
DYNAMICS OF A SYSTEM OF
PARTICLES
101
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
102
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
4 The Equations of Motion for a System
of Particles
4.1 Introduction
In this chapter, we establish Lagrange’s equations for a system of particles by start
ing with the balances of linear momentum for each of the particles. Our derivation
is based on the results presented in Chapter 15 of Synge and Grifﬁth [207].
∗
We
supplement their work with a discussion of constraints and potential energies. To
examine the geometry inherent in Lagrange’s equations of motion for the system of
particles, we use the construction of a representative single particle by Casey [27].
All the work presented in this chapter emphasizes the equivalence of Lagrange’s
equations of motion for a system of particles and the balances of linear momenta.
For completeness, a brief discussion of the principle of virtual work, D’Alembert’s
principle, Gauss’ principle of least constraint, and Hamilton’s principle are also pre
sented in Section 4.11. The chapter closes with a discussion of a canonical form of
Lagrange’s equations of motion in which timeindependent integrable constraints
are present.
For many speciﬁc problems, we can obtain Lagrange’s equations by merely
calculating the kinetic and potential energies of the system. This approach is
used in most dynamics textbooks, and neither the construction of a single particle
nor the components of force vectors are mentioned.
†
Indeed, once we establish
Lagrange’s equations we can also ignore the explicit construction of the single
particle. However, for many cases – which are not possible to treat using the
approach adopted in most dynamics textbooks – we ﬁnd that the use of Synge’s and
Grifﬁth’s representation of Lagrange’s equations of motion allows us to tremen
dously increase the range of application of Lagrange’s equations. For instance, as
will be shown in some of the examples in Chapter 5, dynamic Coulomb friction is
accommodated.
∗
Related derivations for systems of particles can be found in several texts. For example, Section 66
of Greenwood [79] and Section 21 of Whittaker [228].
†
These texts use either the principle of virtual velocities or Hamilton’s principle (also known as the
principle of least action) to derive Lagrange’s equations.
103
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
104 The Equations of Motion for a System of Particles
m
i
O
r
i
F
i
E
1
E
2
E
3
Figure 4.1. A single particle of mass m
i
in E
3
. The position
vector of this particle is r
i
, and the resultant external force
acting on the particle is F
i
.
4.2 A System of N Particles
Here, we are interested in establishing the equations of motion for a system of N
particles. The ﬁrst step in this development is to discuss the individual elements in
the system of particles.
We consider a system of N particles, each of which is in motion in three
dimensional Euclidean space E
3
. For the particle of mass m
i
(see Figure 4.1), the
position vector is
r
i
=
3
¸
j=1
x
j
i
E
j
.
We also recall that the kinetic energy of the particle is
T
i
=
1
2
m
i
v
i
· v
i
.
It is also convenient to recall that the linear momentum of the particle of mass m
i
is
G
i
= m
i
˙ r
i
= m
i
v
i
.
The resultant force acting on the particle of mass m
i
has the representation
F
i
=
3
¸
j=1
F
j
i
E
j
.
The balance of linear momentum for the particle of mass m
i
is
F
i
= m
i
˙ v
i
.
As a consequence of the balance of linear momentum for the particle, we have the
angular momentum theorem:
˙
H
O
i
= r
i
× F
i
,
where H
O
i
= r
i
×m
i
v
i
is the angular momentum of the particle relative to the ﬁxed
point O. A second consequence of the linear momentum balance is the work–energy
theorem
˙
T
i
= F
i
· v
i
for the particle of mass m
i
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
4.3 Coordinates 105
The Kinetic Energy
For the system of particles, the combined (total) kinetic energy T is the sum of the
kinetic energies:
T = T
1
+ · · · + T
N
.
With the help of the work–energy theorem for each particle, we can determine the
corresponding result for the system of particles:
˙
T = F
1
· v
1
+ · · · + F
N
· v
N
. (4.1)
This result is used to establish energy conservation (or lack thereof) in a system of
particles.
The Center of Mass
For the system of particles, we can deﬁne a center of mass C. This point, which lies
in E
3
, has the position vector ¯ r, where
¯ r =
1
m
1
+ · · · + m
N
(m
1
r
1
+ · · · + m
N
r
N
) .
It is easy to show from this result that the linear momentum of a system of particles
has the representation G = (m
1
+ · · · +m
N
)
˙
¯r. Next we examine a particle of mass
m moving in E
3N
, and it is important not to confuse this particle with C.
4.3 Coordinates
In many problems, Cartesian coordinates for r
i
are not a convenient choice. Indeed
for many systems of two particles, we use one coordinate system to describe r
1
and
another to describe the relative position vector r
2
− r
1
. For instance, for the system
shown in Figure 4.2, we might use Cartesian coordinates for r
1
and spherical polar
coordinates for r
2
− r
1
:
r
1
= xE
1
+ yE
2
+ zE
3
, r
2
= xE
1
+yE
2
+ zE
3
+R
2
e
R
2
. (4.2)
In this equation, R
2
= L
0
is the length of the rod connecting the particles, and
e
R
2
= sin(φ
2
) (cos(θ
2
)E
1
+ sin(θ
2
)E
2
) + cos(φ
2
)E
3
.
m
1
m
2
O
g
E
1
E
2
E
3
Figure 4.2. A particle of mass m
1
attached
by a rigid rod of length L
0
to a particle of
mass m
2
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
106 The Equations of Motion for a System of Particles
with related deﬁnitions for e
φ
2
and e
θ
2
. Notice that e
R
2
points from m
1
to m
2
. The
coordinate system most suited to a given system of particles will, as in the case of a
single particle, depend on the constraints and applied forces.
In general, for the system of N particles we will use a coordinate system de
noted by q
1
, . . . , q
3N
. As in the case of a single particle, we assume that we can de
termine the unique Cartesian coordinates for a system of particles once q
1
, . . . , q
3N
are known, and vice versa:
q
K
= ˆ q
K
x
1
1
, x
2
1
, x
3
1
, . . . , x
1
N
, x
2
N
, x
3
N
(K = 1, . . . , 3N),
x
j
i
= ˆ x
j
i
(q
1
, . . . , q
3N
) (i = 1, . . . , Nand j = 1, 2, 3).
To calculate Lagrange’s equations of motion, we need to calculate the following 3N
2
vectors:
∂r
i
∂q
K
(K = 1, . . . , 3Nand i = 1, . . . , N).
These vectors play the role of the basis vectors a
k
that we used with the single par
ticle earlier.
As an example, let us return to (4.2). In these equations, we gave examples of
the coordinates for a twoparticle system:
q
1
= x = x
1
1
, q
2
= y = x
2
1
, q
3
= z = x
3
1
, q
4
= R
2
, q
5
= θ
2
, q
6
= φ
2
.
For this coordinate system, and with the help of (4.2),
∂r
1
∂q
j
= E
j
,
∂r
1
∂q
(j+3)
= 0,
and
∂r
2
∂q
j
= E
j
,
∂r
2
∂q
4
= e
R
2
,
∂r
2
∂q
5
= R
2
sin(φ
2
)e
θ
2
,
∂r
2
∂q
6
= R
2
e
φ
2
,
where j = 1, 2, 3 in these two sets of equations. It is left to the student to realize how
coordinate system (4.2) can also be used to describe the kinematics of the particle
system shown in Fig. 4.3.
m
1
m
2
O
linear spring
E
1
E
2
E
3
Figure 4.3. A particle of mass m
1
attached by a linear spring of stiffness K and unstretched
length L
0
to a particle of mass m
2
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
4.4 Constraints and Constraint Forces 107
4.4 Constraints and Constraint Forces
For a single particle, one of the key notions we encountered earlier was kinematic
constraints. We now turn to examining this topic for a system of particles. As we
shall see, the extension to this case is not as difﬁcult as it ﬁrst may seem.
Let us start with a physical system and use it to establish expressions for con
straint forces. We will later show that these constraint forces are compatible with
Lagrange’s prescription. Suppose we have two particles connected by a rigid rod of
length L
0
(see Figure 4.2). Then the constraint imposed by the rod on their motion
is r
1
− r
2
 −L
0
= 0. To enforce this constraint, the particles will be subject to equal
and opposite constraint forces:
F
c1
= −µt,
F
c2
= µt, (4.3)
where µ is the tension force in the rod and the unit vector t points from m
1
to m
2
:
t =
r
2
− r
1
r
2
− r
1

.
If we describe r
2
− r
1
= L
0
e
R
2
, which we did earlier, then t = e
R
2
. Writing the con
straint as
= 0,
where
= r
2
− r
1
− L
0
,
then, as
∂
∂r
1
= −
r
2
− r
1
r
1
− r
2
,
∂
∂r
2
=
r
2
− r
1
r
1
− r
2
,
we observe that (4.3) can be expressed as
F
c1
= µ
∂
∂r
1
, F
c2
= µ
∂
∂r
2
. (4.4)
This result will help motivate Lagrange’s prescription for the constraint forces acting
on a system of particles.
As a second example, consider the threeparticle system shown in Figure 4.4.
This model is a prototypical model for a braking vehicle and is used to explain the
vehicle instability that often occurs when the front wheels lock during braking.
∗
We
assume that the motion of this cart is planar. As a result the motion of the center
of mass C of the system is planar r = xE
1
+ yE
2
. Further, we need to supplement x
and y only by an angle φ to determine the position of any point on the cart. Thus,
∗
For further details on this matter, and references to the vast literature on this prototypical noninte
grably constrained system, see O’Reilly and Tongue [165] and Ruina [186].
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
108 The Equations of Motion for a System of Particles
m
1
m
2
m
3
O
Q
C
φ
Vehicle
E
1
E
1
E
2
e
1
e
1
e
2
Figure 4.4. A model for a braking threewheeled vehicle. The distributed mass of the vehicle
is modeled as three mass particles of masses m
1
, m
2
, and m
3
. The particles of mass m
1
and
m
2
model wheels that are rolling without slipping, and the particle of m
3
models a slipping
wheel.
for this system, we make the following choices for the nine coordinates q
1
, . . . , q
9
:
q
1
= x, q
2
= y, q
3
= φ, q
3+k
= r
2
· E
k
, q
6−k
= r
3
· E
k
.
Here, k = 1, 2, 3 and r = xE
1
+yE
2
. The unit vector e
1
= cos(φ)E
1
+ sin(φ)E
2
in
Figure 4.4 is perpendicular to the line connecting m
1
and m
2
and e
2
is tangent to this
line. The constraint that the velocity of the particle of mass m
1
is always normal to
e
2
can be written in several equivalent forms. For example,
e
2
· v
1
= 0.
This constraint is nonintegrable. By inspection, the force that enforces this con
straint is parallel to e
2
:
F
c1
= µe
2
. (4.5)
This force ensures that m
1
moves only in the e
1
direction.
Lagrange’s Prescription
We now have sufﬁcient motivation to motivate Lagrange’s prescription for the con
straint forces acting on a system of particles. We suppose that the system of particles
is subject to an integrable constraint:
= 0, (4.6)
where
= (r
1
, r
2
, . . . , r
N
, t).
Constraint (4.6) can be differentiated to yield a constraint of the form
N
¸
i=1
f
i
· v
i
+ e = 0,
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
4.4 Constraints and Constraint Forces 109
where
f
i
=
∂
∂r
i
,
e =
∂
∂t
.
When evaluating
∂
∂r
1
, for example, we assume that r
2
, . . . , r
N
are ﬁxed. Lagrange’s
prescription for the constraint force F
ci
acting on the particle of mass m
i
is
F
ci
= µf
i
= µ
∂
∂r
i
.
Result (4.4) discussed earlier is an example of this prescription. As with a single par
ticle in the presence of dynamic friction, Lagrange’s prescription is not universally
applicable. Indeed, we shall encounter some examples later on when this prescrip
tion cannot be used because of the presence of friction.
∗
Clearly Lagrange’s prescription can also be applied to nonintegrable con
straints. Such constraints have the functional form
f
1
· v
1
+ · · · + f
N
· v
N
+ e = 0, (4.7)
where f
i
= f
i
(r
1
, . . . , r
N
, t) and e = e (r
1
, . . . , r
N
, t). Lagrange’s prescription for such
a constraint is
F
ci
= µf
i
(i = 1, . . . , N).
This prescription is equivalent to one we discussed earlier for the example of a
nonintegrably constrained system of particles [see (4.5)]. It should be obvious
how Lagrange’s prescription can be extended to systems of (integrable and
nonintegrable) constraints on the system of particles.
The Power of the Constraint Forces
The constraints we consider can all be written in the form (4.7). Further, if the con
straint forces are consistent with Lagrange’s prescription, then the power expended
by these forces is easily calculated:
P =
N
¸
i=1
F
ci
· v
i
= µ
N
¸
i=1
f
i
· v
i
= −µe.
Thus, when e = 0, the constraint forces do no work. For an integrable constraint this
arises when ψ is not an explict function of time: = (r
1
, . . . , r
N
). In a previous
example, shown in Figure 4.2, we note that depended on r
1
and r
2
. If, in addition,
L
0
were a function of time, then = (r
1
, r
2
, t).
∗
See, for example, Subsection 8.6.2 in Chapter 8.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
110 The Equations of Motion for a System of Particles
Newton’s Third Law
For the constraint r
1
− r
2
 − L
0
= 0 mentioned earlier, Lagrange’s prescription
yields
F
c1
= µ
r
1
− r
2
r
1
− r
2

,
F
c2
= µ
r
2
− r
1
r
1
− r
2

.
Notice that these constraint forces point along the rod – which is what we would
expect from physical grounds. Furthermore, F
c2
= −F
c1
– which is none other than
Newton’s third law.
∗
4.5 Conservative Forces and Potential Energies
Conservative forces in the dynamics of systems of particles commonly occur. For
instance, they arise when two particles are connected by a spring (see Figure 4.3) or
when each particle is attracted to a central body by a gravitational force ﬁeld. In this
section, we discuss how to prescribe the conservative forces F
con1
, . . . , F
conN
acting
on their respective particle of a system of particles. The system is assumed to have a
potential energy
U = U(r
1
, r
2
, . . . , r
N
) . (4.8)
Notice that we are presuming that this is the most general form of the potential
energy of conservative forces in a system of particles.
General form (4.8) encompasses the inverse gravitational law between two par
ticles of mass m
1
and m
2
,
U
n
= −
Gm
1
m
2
r
2
− r
1

,
and the potential energy of a spring force between two particles of mass m
1
and m
2
:
U
s
=
K
2
(r
2
− r
1
 −L
0
)
2
.
Here, G is the universal gravitational constant and K is the spring constant.
To calculate the conservative forces, we equate the time derivative of U to the
negative power of the conservative forces. After some rearraging, we ﬁnd that
N
¸
i=1
F
coni
+
∂U
∂r
i
· v
i
= 0. (4.9)
∗
For more details on this interesting result, see Noll [153], O’Reilly and Srinivasa [163], and Subsec
tion 8.6.4 in this book.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
4.6 Lagrange’s Equations of Motion 111
Assuming that, for each i, F
coni
+
∂U
∂r
i
is independent of v
1
, . . . , v
N
and that (4.9) is
true for all possible v
1
, . . . , v
N
, then we conclude that
F
coni
= −
∂U
∂r
i
(i = 1, . . . , N). (4.10)
This is the prescription for the conservative forces.
For the spring force and gravitational forces associated with U
n
and U
s
that we
deﬁned earlier, prescription (4.10) yields
F
con1
= −
Gm
1
m
2
r
2
− r
1

2
r
1
− r
2
r
2
− r
1

, F
con2
= −
Gm
1
m
2
r
2
− r
1

2
r
2
− r
1
r
2
− r
1

,
and
F
con1
= −
K
2
(r
2
− r
1
 − L
0
)
r
1
− r
2
r
2
− r
1

,
F
con2
= −
K
2
(r
2
− r
1
 − L
0
)
r
2
− r
1
r
2
− r
1

,
respectively. You might again notice that the pairs of forces obey Newton’s third
law identically. This interesting result was ﬁrst pointed out by Lanczos [124] and
Noll [153].
4.6 Lagrange’s Equations of Motion
For a system of unconstrained particles, the equations of motion consist of N
balances of linear momenta:
m
i
¨ r
i
= F
i
(i = 1, . . . , N).
We now show that these equations are equivalent to Lagrange’s equations of
motion:
d
dt
∂T
∂ ˙ q
K
−
∂T
∂q
K
=
K
(K = 1, . . . , 3N) , (4.11)
where
K
=
N
¸
i=1
F
i
·
∂r
i
∂q
K
.
As there are no constraints on the system, all of the coordinates q
1
, . . . , q
3N
are
generalized coordinates and
K
are the generalized forces. Our results here are
adapted from the work of Synge and Grifﬁth. Their derivation is presented in
Section 15.1 of [207]. Shortly, an alternative derivation of (4.11) that is due to Casey
[27] will be presented.
∗
∗
See Equation (4.20) in Section 4.7.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
112 The Equations of Motion for a System of Particles
Lagrange’s equations presume that each individual particle’s position vector is
expressed as a function of the coordinates q
1
, . . . , q
3N
:
r
i
= r
i
q
1
, . . . , q
3N
(i = 1, . . . , N).
With these coordinates, we then compute v
i
and construct T.
Derivation of Lagrange’s Equations of Motion
Before proceeding with the derivation of Lagrange’s equations of motion, we estab
lish an important result:
∂v
i
∂ ˙ q
K
=
∂r
i
∂q
K
. (4.12)
This identity is often (fondly) referred to as “canceling the dots.” To verify the iden
tity, we use the chain rule to calculate v
i
:
v
i
= ˙ r
i
=
3N
¸
J =1
˙ q
J
∂r
i
∂q
J
.
Taking the partial derivative of both sides of this expression with respect to ˙ q
K
then
yields (4.12).
We are now in a position to show that
∂T
∂ ˙ q
K
=
N
¸
i=1
m
i
v
i
·
∂r
i
∂q
K
,
∂T
∂q
K
=
N
¸
i=1
m
i
v
i
·
d
dt
∂r
i
∂q
K
. (4.13)
First,
∂T
∂ ˙ q
K
=
∂
∂ ˙ q
K
N
¸
i=1
m
i
2
v
i
· v
i
=
N
¸
i=1
m
i
v
i
·
∂v
i
∂ ˙ q
K
=
N
¸
i=1
m
i
v
i
·
∂r
i
∂q
K
.
Notice that we used (4.12) in the ﬁnal stages of this calculation. The other result
follows similarily and so we skip some of the intermediate stages:
∂T
∂q
K
=
N
¸
i=1
m
i
v
i
·
∂v
i
∂q
K
=
N
¸
i=1
m
i
v
i
·
∂
∂q
K
3N
¸
J =1
˙ q
J
∂r
i
∂q
J
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
4.7 Construction and Use of a Single Representative Particle 113
=
N
¸
i=1
m
i
v
i
·
3N
¸
J =1
˙ q
J
∂
∂q
J
∂r
i
∂q
K
=
N
¸
i=1
m
i
v
i
·
d
dt
∂r
i
∂q
K
.
In the last stages of this calculation, we used the identities
∂
2
r
i
∂q
K
∂q
J
=
∂
2
r
i
∂q
J
∂q
K
,
˙
f =
3N
¸
J =1
˙ q
J
∂f
∂q
J
,
where f = f
q
1
, . . . , q
3N
.
To derive Lagrange’s equations of motion, we now follow a familiar series of
steps with the help of (4.13):
d
dt
∂T
∂ ˙ q
K
−
∂T
∂q
K
=
d
dt
N
¸
i=1
m
i
v
i
·
∂r
i
∂q
K
−
N
¸
i=1
m
i
v
i
·
d
dt
∂r
i
∂q
K
=
N
¸
i=1
m
i
˙ v
i
·
∂r
i
∂q
K
=
N
¸
i=1
F
i
·
∂r
i
∂q
K
=
K
, (4.14)
where the force
K
is
K
=
N
¸
i=1
F
i
·
∂r
i
∂q
K
.
We have now shown how Lagrange’s equations of motion for a system of particles
can be established by using the balances of linear momenta for each of the particles.
4.7 Construction and Use of a Single Representative Particle
An alternative derivation of Lagrange’s equations was presented recently by
Casey [27]. In this work, he constructs a single representative particle moving in
a 3Ndimensional space subject to a force. With the help of this construction, the
derivation of Lagrange’s equations of motion follows easily, and tremendous insight
can be gained into the geometry of Lagrange’s equations of motion.
In this section, we follow Casey [27], and construct a single particle of mass m
that is moving in E
3N
. The position vector of this particle is r. The kinetic energy
T of this particle is the same as the kinetic energy of the system of particles that it
represents, and the force on the particle is such that
˙
T = · ˙ r, = m¨ r.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
114 The Equations of Motion for a System of Particles
In Casey [27], m is chosen (without loss in generality) to be the sum of the masses:
m = m
1
+ · · · +m
N
. Here, we assume only that m > 0. As we noted earlier in Sec
tion 3.6, the interested reader is also referred to Chapter 1 of Lanczos [124] for a
related discussion of the geometry of a mechanical system.
The Conﬁguration Space
The space E
3N
, which is known as the conﬁguration space, is equipped with a Carte
sian coordinate system. Consequently, for any vector b,
b =
3N
¸
K=1
b
K
¯ e
K
=
N
¸
i=1
3
¸
j=1
b
3i+j−3
¯ e
3i+j−3
.
Here, { ¯ e
1
, . . . , ¯ e
3N
} is a ﬁxed orthonormal basis for E
3N
. We also deﬁne two other
sets of basis vectors:
e
3i+j−3
=
m
i
m
¯ e
3i+j−3
,
e
3i+j−3
=
m
m
i
¯ e
3i+j−3
.
Here i = 1, . . . , N and j = 1, 2, 3. Notice that
e
K
· e
J
= δ
J
K
.
The particle has a position vector r and a force vector . Both of these vectors have
representations similar to that for b.
Prescription for the Position Vector r
As mentioned earlier, the position vector r is deﬁned by the criterion that the kinetic
energy of the particle of mass m is identical to the kinetic energy of the system of
particles:
T =
1
2
mv · v =
1
2
N
¸
i=1
m
i
v
i
· v
i
.
Substituting,
v = ˙ r =
N
¸
i=1
3
¸
j=1
˙ r
3i+j−3
¯ e
3i+j−3
, v
i
= ˙ r
i
=
3
¸
j=1
˙ x
j
i
E
j
,
and equating kinetic energies, we ﬁnd one solution for r:
r =
N
¸
i=1
3
¸
j=1
(r
i
· E
j
)
m
i
m
¯ e
3i+j−3
=
N
¸
i=1
3
¸
j=1
(r
i
· E
j
) e
3i+j−3
. (4.15)
It is a good exercise to show that the other solutions for r can be shown to be equiv
alent to (4.15) modulo a translation of the origin and a relabeling of axes of E
3N
.
Notice how the mass ratios
m
i
m
are subsumed into the basis vectors e
3i+j−3
in (4.15).
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
4.7 Construction and Use of a Single Representative Particle 115
Prescription for the Force Vector
The force vector is prescribed by the requirement that
m˙ v = .
Substituting (4.15) and using the balance of linear momentum for each particle, one
is led to the prescription
=
N
¸
i=1
3
¸
j=1
(F
i
· E
j
)
m
m
i
¯ e
3i+j−3
=
N
¸
i=1
3
¸
j=1
(F
i
· E
j
) e
3i+j−3
. (4.16)
Again its interesting to note how the mass ratios
m
i
m
are subsumed into the basis
vectors e
3i+j−3
in (4.15).
Next, it is often convenient to note that
· e
3i+j−3
= F
i
· E
j
,
which follows from (4.16). It is also an easy exercise to show that the work–energy
theorems for the individual particles,
˙
T
i
= F
i
· v
i
(no sum on i), give a work–energy
theorem for the particle of mass m:
˙
T = · v.
This theorem can be used to establish energy conservation results for the system of
particles.
Curvilinear Coordinates
We then use the curvilinear coordinates q
1
, . . . , q
3N
to deﬁne basis vectors for E
3N
:
a
K
=
∂r
∂q
K
=
∂
∂q
K
⎛
⎝
N
¸
i=1
3
¸
j=1
r
i
· E
j
= ˆ x
j
i
e
3i+j−3
⎞
⎠
and
a
J
= grad(q
J
) =
N
¸
i=1
3
¸
j=1
∂ ˆ q
J
∂x
j
i
e
3i+j−3
.
It can be shown that a
K
· a
J
= δ
J
K
. For most problems, it is not necessary to explicitly
calculate the contravariant basis vectors a
J
.
Equivalences
For future reference, it is important to note that
· a
K
=
N
¸
i=1
F
i
·
∂r
i
∂q
K
. (4.17)
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
116 The Equations of Motion for a System of Particles
To establish this result a series of identities and the deﬁnition ˆ x
p
s
= r
s
· E
p
are
invoked:
· a
K
= ·
∂r
∂q
K
=
⎛
⎝
N
¸
i=1
3
¸
j=1
(F
i
· E
j
) e
3i+j−3
⎞
⎠
·
∂
∂q
K
⎛
⎝
N
¸
s=1
3
¸
p=1
(r
s
· E
p
) e
3s+p−3
⎞
⎠
=
N
¸
i=1
3
¸
j=1
N
¸
s=1
3
¸
p=1
(F
i
· E
j
)
∂ ˆ x
p
s
∂q
K
e
3i+j−3
· e
3s+p−3
=
N
¸
i=1
3
¸
j=1
N
¸
s=1
3
¸
p=1
(F
i
· E
j
)
∂ ˆ x
p
s
∂q
K
δ
i
s
δ
j
p
=
N
¸
i=1
3
¸
j=1
(F
i
· E
j
)
∂ ˆ x
j
i
∂q
K
=
N
¸
i=1
F
i
·
∂
∂q
K
⎛
⎝
3
¸
j=1
ˆ x
j
i
E
j
⎞
⎠
=
N
¸
i=1
F
i
·
∂r
i
∂q
K
.
We conclude that the desired identity has been established.
Identity (4.17) is similar to several other results pertaining to the system of par
ticles and the single particle of mass m. For instance, consider a function
= (r, t) =
¯
(r
1
, . . . , r
N
, t) .
With some minor manipulations, we ﬁnd several representations for the derivative
of with respect to a curvilinear coordinate:
∂
∂q
K
=
∂
∂r
·
∂r
∂q
K
=
N
¸
i=1
∂
¯
∂r
i
·
∂r
i
∂q
K
. (4.18)
These results can be used to establish equivalences between conservative forces and
constraint forces acting on a system of particles and those acting on the single parti
cle of mass m.
Constraints and Constraint Forces
For the single particle, we can express the constraint = 0, where [as in (4.6)]
(r
1
, r
2
, . . . , r
N
, t) = 0,
as a constraint on the motion of the single particle of mass m:
=
¯
(r, t) = 0.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
4.7 Construction and Use of a Single Representative Particle 117
This implies that the motion of the single particle is subject to the constraint and, by
use of Lagrange’s prescription, a constraint force
c
, respectively,
f · v + e = 0,
c
= µf,
where
f =
∂
∂r
, e =
∂
∂t
.
Furthermore, the integrable constraint implies that the particle moves on a conﬁgu
ration manifold Mthat is a (3N − 1)dimensional subset of the conﬁguration space
E
3N
.
It is important to note that the prescriptions of
c
and F
ci
are consistent with
each other. Indeed, the equivalence of the constraint forces F
ci
acting on the system
of particles and the constraint force
c
acting on the single particle of mass m can
be inferred from (4.17) and (4.18).
Conservative Forces and Potential Energies
For the single particle, we can express the potential energy U as a function of the
motion of the single particle of mass m:
U(r
1
, r
2
, . . . , r
N
) =
¯
U(r).
Calculating
˙
¯
Uand equating it to −
con
· v, we ﬁnd that
∗
con
= −
∂
¯
U
∂r
. (4.19)
Again, it is important to note that the prescriptions of
con
and F
coni
[see (4.10)]
are consistent with each other. As with constraint forces, the equivalence can be
inferred from (4.17) and (4.18).
Lagrange’s Equations: An Unconstrained System of Particles
We are now in a position to establish Lagrange’s equations of motion for the single
particle of mass m. Because
⎧
⎪
⎨
⎪
⎩
m
1
˙ v
1
= F
1
. . .
m
N
˙ v
N
= F
N
⎫
⎪
⎬
⎪
⎭
is equivalent tom˙ v = ,
these equations will be none other than Lagrange’s equations for the system of N
particles.
First, for the single particle of mass m, it is easy to see that
v =
3N
¸
i=1
˙ q
K
a
K
.
∗
Essentially, we are solving the equation
con
+
∂
¯
U
∂r
· v = 0 for all possible motions of the system.
For this to hold, the terms in the parentheses, assuming they are independent of v, must vanish, and
we arrive at (4.19).
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
118 The Equations of Motion for a System of Particles
Consequently, we have the two intermediate results
∂T
∂ ˙ q
K
= mv · a
K
,
∂T
∂q
K
= mv · ˙ a
K
.
With the assistance of these results we ﬁnd that
d
dt
∂T
∂ ˙ q
K
−
∂T
∂q
K
=
d
dt
(mv · a
K
) −mv · ˙ a
K
=
d
dt
(mv) · a
K
= · a
K
.
We have just established Lagrange’s equations for a system of particles:
d
dt
∂T
∂ ˙ q
K
−
∂T
∂q
K
= · a
K
. (4.20)
These equations give the motion r = r(t) of the particle of mass m and hence the
motion of the system of particles.
We notice that, because of (4.17), Equations (4.20) are identical to those we
established earlier. In particular,
K
=
N
¸
i=1
F
i
·
∂r
i
∂q
K
= · a
K
.
For many problems it is more convenient to use F
1
, . . . , F
N
instead of . However,
(4.20) provides wonderful insight into the geometry of Lagrange’s equations for a
system of particles.
4.8 The Lagrangian
In many mechanical systems, the sole forces that act on the system are conservative.
In this case, we can use the potential energy to deﬁne a Lagrangian for the system
of particles:
L = T −U.
Using the Lagrangian, and with the help of the identity
∗
∂U
∂q
K
=
N
¸
i=1
∂U
∂r
i
·
∂r
i
∂q
K
,
it is easy to see that Lagrange’s equations for a system of particles is
d
dt
∂L
∂ ˙ q
K
−
∂L
∂q
K
= Q
K
. (4.21)
∗
Those readers who read Section 4.7 will have seen this identity earlier as (4.18).
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
4.9 A Constrained System of Particles 119
Here, Q
K
are the nonconservative forces acting on the system of particles. The non
conservative forces have the equivalent representations
Q
K
=
K
+
∂U
∂q
K
=
N
¸
i=1
F
nconi
·
∂r
i
∂q
K
=
ncon
· a
K
.
Here,
F
nconi
= F
i
+
∂U
∂r
i
,
ncon
= +
∂
¯
U
∂r
.
Notice that if Q
K
= 0 – as it is in many celestial mechanics problems – one can write
the equations of motion for a system of particles without ever having to calculate an
acceleration vector.
When constraints are present, a form of Lagrange’s equations similar to (4.21)
can also be obtained provided the constraints are integrable and the constraint
forces are prescribed by use of Lagrange’s prescription. In this case, the equations
are identical to (4.21) with the kinetic and potential energies being replaced with
their constrained counterparts.
∗
4.9 A Constrained System of Particles
Consider a system comprising particles that are subject to one integrable and one
nonintegrable constraint. We choose the curvilinear coordinates to express these
constraints as
q
3N
−r(t) = 0,
N
¸
i=1
f
i
· v
i
+ e = 0.
These constraints are equivalent to the following constraints on the motion of the
single (representative) particle:
q
3N
− r(t) = 0, f · v +e = 0.
Thus the generalized coordinates for this system are q
1
, . . . , q
3N
. These coordinates
parameterize the (3N − 1)dimensional conﬁguration manifold M. This manifold
lies in the 3Ndimensional conﬁguration space that was discussed in Section 4.7 in
conjunction with Casey’s construction of the representative particle. A representa
tive example is sketched in Figure 4.5.
∗
That is, in the notation subsequently presented, T is replaced with
˜
T and Uis replaced with
˜
U. As a
result, L is replaced with
˜
L =
˜
T −
˜
U.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
120 The Equations of Motion for a System of Particles
Conﬁguration manifold M
m
v
r
c
O
Figure 4.5. The representative particle moving on a conﬁguration manifold. For the example
shown in this ﬁgure, the conﬁguration manifold Mis ﬁxed and
c
is normal to M.
Assuming that the constraint forces associated with these constraints are com
patible with Lagrange’s prescription, the forces acting on the system of particles
have the decompositions
F
i
= −
∂U
∂r
i
+µ
1
∂q
3N
∂r
i
+µ
2
f
i
+ P
i
,
where P
i
is the resultant of the nonconservative and nonconstraint forces acting on
the particle of mass m
i
. It is important to note that
K
=
N
¸
i=1
F
i
·
∂r
i
∂q
K
= −
∂U
∂q
K
+µ
1
δ
3N
K
+µ
2
f
K
+
K
, (4.22)
where
K
=
N
¸
i=1
P
i
·
∂r
i
∂q
K
, f
K
=
N
¸
i=1
f
i
·
∂r
i
∂q
K
.
From these results, it is easy to construct =
¸
3N
K=1
K
a
K
.
Approach I
For the ﬁrst approach, we expand the partial derivatives of Tand Uand then impose
the integrable constraint q
3N
= r(t) on the resulting equations. It can be seen that
the resulting Lagrange’s equations of motion decouple into two sets:
¸
d
dt
∂T
∂ ˙ q
S
−
∂T
∂q
S
= −
∂U
∂q
S
+µ
2
f
S
+
S
¸
q
3N
=r, ˙ q
3N
=˙ r
,
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
4.9 A Constrained System of Particles 121
and
¸
d
dt
∂T
∂ ˙ q
3N
−
∂T
∂q
3N
= −
∂U
∂q
3N
+µ
1
+µ
2
f
3N
+
3N
¸
q
3N
=r, ˙ q
3N
=˙ r
,
where S = 1, . . . , 3N − 1, and we emphasize that the constraint q
3N
= r(t) is im
posed after the partial derivatives of T and Uhave been calculated.
Approach II
For comparison, we now use the second approach. In this case we ﬁrst impose the
integrable constraint on the kinetic and potential energies, in addition to the basis
vectors and constraints. For instance,
˜
T =
˜
T(q
1
, . . . , q
3N−1
, ˙ q
1
, . . . , ˙ q
3N−1
, t)
= T(q
1
, . . . , q
3N−1
, q
3N
= r(t), ˙ q
1
, . . . , ˙ q
3N−1
, ˙ q
3N
= ˙ r(t)).
As in the case of a single particle,
˜
T can be used to construct a lineelement for the
conﬁguration manifold M.
We note that, because
∂
˜
T
∂q
3N
= 0,
∂
˜
T
∂ ˙ q
3N
= 0,
we are unable to obtain an expression for µ
1
. In other words, because we have elim
inated the coordinate associated with the integrable constraint, we can obtain only
3N − 1 Lagrange’s equations:
d
dt
∂
˜
T
∂ ˙ q
S
−
∂
˜
T
∂q
S
= −
∂
˜
U
∂q
S
+µ
2
˜
f
S
+
˜
S
.
Further, no information on the constraint force enforcing the integrable constraint
is obtained when this approach is used.
Geometric Considerations
Once the integrable constraint has been imposed, the single particle of mass m
moves on the (3N − 1)dimensional submanifold of the conﬁguration space E
3N
. As
before, this submanifold is known as the conﬁguration manifold M. A measure of
the distance traveled by the single particle on this manifold can be found by use of
˜
T. Indeed, this energy can be decomposed as
˜
T =
˜
T
0
+
˜
T
1
+
˜
T
2
,
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
122 The Equations of Motion for a System of Particles
where
˜
T
0
=
m
2
˜ a
3N3N
˙ r
2
,
˜
T
1
= m
3N−1
¸
K=1
˜ a
K3N
˙ q
K
˙ r,
T
rel
=
˜
T
2
=
m
2
3N−1
¸
K=1
3N−1
¸
J =1
˜ a
KJ
˙ q
K
˙ q
J
. (4.23)
In the event that the single particle of mass m is determined, we also have the
decomposition:
a
JK
= a
K
· a
J
(J, K = 1, . . . , 3N) .
The mass m in (4.23) is usually chosen to be m
1
+ · · · + m
N
, although other
selections such as m = 1 are equally admissible.
The kinematical lineelement ds for Mis
ds =
2T
rel
m
dt.
This quantity may also be written as
ds =
3N−1
¸
K=1
3N−1
¸
J =1
˜ a
KJ
dq
K
dq
J
.
As was the case previously, the imposition of a nonintegrable constraint will not
change Mor ds.
4.10 A Canonical Form of Lagrange’s Equations
We now consider a system of particles in which the constraints are integrable and
time independent and the constraint forces are prescribed by use of Lagrange’s pre
scription. In this case, it sufﬁces to examine those equations associated with the
generalized coordinates in order to ﬁnd the equations of motion. In what follows,
we establish two alternative forms of Lagrange’s equations of motion [see (4.30) and
(4.32)].
Preliminaries
We are interested in a system of N particles subject to a set of C (scleronomic)
integrable constraints of the form
∗
1
(r
1
, . . . , r
N
) = 0, . . . ,
C
(r
1
, . . . , r
N
) = 0. (4.24)
∗
The generalization of our developments to instances in which the constraints are rheonomic can be
found elsewhere, for example, Ginsberg [71].
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
4.10 A Canonical Form of Lagrange’s Equations 123
We assume that the coordinates q
1
, . . . , q
3N
are chosen such that the generalized co
ordinates for the system are q
1
, . . . , q
M
, where M = 3N − C. That is, the constraints
1
= 0, . . . ,
C
= 0 are expressed in terms of the coordinates q
M+1
, . . . , q
3N
.
The constrained kinetic energy for the system has the representation
˜
T =
m
2
M
¸
K=1
M
¸
J =1
˜ a
KJ
˙ q
J
˙ q
K
,
where we have chosen m = m
1
+ · · · +m
N
. We note that ˜ a
KJ
= ˜ a
JK
and assume that
this matrix is invertible. The inverse is ˜ a
JK
. For example, if M = 3, then
⎡
⎢
⎣
˜ a
11
˜ a
12
˜ a
13
˜ a
12
˜ a
22
˜ a
23
˜ a
13
˜ a
23
˜ a
33
⎤
⎥
⎦
⎡
⎢
⎣
˜ a
11
˜ a
12
˜ a
13
˜ a
12
˜ a
22
˜ a
23
˜ a
13
˜ a
23
˜ a
33
⎤
⎥
⎦
=
⎡
⎢
⎣
1 0 0
0 1 0
0 0 1
⎤
⎥
⎦
.
We also deﬁne the Christoffel symbols of the ﬁrst kind as
[SJ, K] =
1
2
∂a
KJ
∂q
S
+
∂a
KS
∂q
J
−
∂a
SJ
∂q
K
(J, K, S = 1, . . . , 3N) . (4.25)
There are (3N)
3
of these symbols, but many of them are not distinct: [SJ, K] =
[JS, K]. The Christoffel symbols of the second kind are
K
IJ
=
3N
¸
R=1
a
KR
[IJ, R] (I, J, K = 1, . . . , 3N) . (4.26)
To gain further insight into these symbols, it is very useful to examine the single
particle of mass m.
The Representative Particle
Casey’s construction of the representative particle is very useful for exploring the
Christoffel symbols. Using the particle and its conﬁguration space, we have a set of
covariant, a
K
, and contravariant, a
J
, basis vectors for E
3N
. These vectors can be used
to deﬁne a
JK
, a
JK
, and the Christoffel symbols:
a
JK
= a
J
· a
K
, a
JK
= a
J
· a
K
,
[SI, K] =
∂a
S
∂q
I
· a
K
,
J
KS
=
∂a
K
∂q
S
· a
J
. (4.27)
Here, all the indices range from 1 to 3N. Using (4.27), the symmetries a
JK
= a
KJ
,
a
JK
= a
KJ
, [SI, K] = [IS, K], and
J
KS
=
J
SK
should be transparent. We also observe
that the Christoffel symbols are none other than the covariant and contravariant
components of
∂a
K
∂q
S
. It is a good exercise to show how (4.25) and (4.26) can be used
to establish (4.27)
3,4
when (4.27)
1,2
apply.
Derivatives of the Kinetic Energy
Preparatory to establishing the covariant and contravariant forms, we ﬁrst record
the derivatives of the kinetic energy. What follows is based entirely on the
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
124 The Equations of Motion for a System of Particles
constrained kinetic energy, and we now drop the tildes ornamenting the various
kinematical quantities.
First, we note that
∂T
∂ ˙ q
R
=
∂
∂ ˙ q
R
m
2
M
¸
K=1
M
¸
J =1
a
KJ
˙ q
J
˙ q
K
=
m
2
M
¸
K=1
M
¸
J =1
a
KJ
⎛
⎜
⎜
⎜
⎝
U
δ
J
R
∂ ˙ q
J
∂ ˙ q
R
˙ q
K
+ ˙ q
J
U
δ
K
R
∂ ˙ q
K
∂ ˙ q
R
⎞
⎟
⎟
⎟
⎠
= m
M
¸
K=1
a
RK
˙ q
K
.
To arrive at this result, we used the symmetries a
RK
= a
KR
and the fact that a
RK
are
independent of ˙ q
S
. Similarly,
∂T
∂q
R
=
∂
∂q
R
m
2
M
¸
K=1
M
¸
J =1
a
KJ
˙ q
J
˙ q
K
=
m
2
M
¸
K=1
M
¸
J =1
∂a
KJ
∂q
R
˙ q
J
˙ q
K
.
We next differentiate m
¸
M
K=1
a
RK
˙ q
K
with respect to time. With some rearranging
and relabeling of the indices,
∗
we ﬁnd
d
dt
∂T
∂ ˙ q
R
= m
M
¸
K=1
a
RK
¨ q
K
+m
M
¸
K=1
M
¸
S=1
∂a
RK
∂q
S
˙ q
S
˙ q
K
= m
M
¸
K=1
a
RK
¨ q
K
+
m
2
M
¸
K=1
M
¸
S=1
∂a
RK
∂q
S
˙ q
S
˙ q
K
+
m
2
M
¸
J =1
M
¸
I=1
∂a
IR
∂q
J
˙ q
I
˙ q
J
= m
M
¸
K=1
a
RK
¨ q
K
+
m
2
M
¸
K=1
M
¸
S=1
∂a
RK
∂q
S
+
∂a
SR
∂q
K
˙ q
S
˙ q
K
.
∗
Although
∂a
RK
∂q
S
=
∂a
RS
∂q
K
, because we are summing over the indices S and K, we always have that
¸
M
K=1
¸
M
S=1
∂a
RK
∂q
S
˙ q
S
˙ q
K
=
¸
M
K=1
¸
M
S=1
∂a
RS
∂q
K
˙ q
S
˙ q
K
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
4.10 A Canonical Form of Lagrange’s Equations 125
Summarizing, the derivatives of T have the representations
∂T
∂q
R
=
m
2
M
¸
K=1
M
¸
J =1
∂a
KJ
∂q
R
˙ q
J
˙ q
K
,
d
dt
∂T
∂ ˙ q
R
= m
M
¸
K=1
a
RK
¨ q
K
+
m
2
M
¸
K=1
M
¸
S=1
∂a
RK
∂q
S
+
∂a
SR
∂q
K
˙ q
S
˙ q
K
. (4.28)
A Covariant Form of Lagrange’s Equations of Motion
A covariant form of Lagrange’s equations of motion was previously established:
d
dt
∂T
∂ ˙ q
R
−
∂T
∂q
R
=
R
(R = 1, . . . , M) . (4.29)
Here
1
, . . . ,
M
are the generalized forces. We now expand the derivatives of T to
establish another form of this equation. With the help of (4.28) and the deﬁnition of
the Christoffel symbol of the ﬁrst kind, it is straightforward to show that Lagrange’s
equations of motion can be written in the form
m
M
¸
K=1
a
RK
¨ q
K
+m
M
¸
K=1
M
¸
S=1
[SK, R] ˙ q
K
˙ q
S
=
R
(R = 1, . . . , M) . (4.30)
As mentioned in the exercises at the end of Chapter 3, this form of Lagrange’s equa
tions appears in several texts on differential geometry,
∗
and, in the case of a single
particle, was an exercise at the end of Chapter 3.
From (4.30), it should be apparent that, if we know a
JK
, then we can immedi
ately write the lefthand side of Lagrange’s equations of motion. Further, for the
system at hand the kinematical lineelement ds is
ds =
M
¸
R=1
M
¸
K=1
a
RK
˙ q
R
˙ q
K
dt.
Knowledge of ds (which also implies knowledge of a
JK
) enables us to write the left
hand side of Lagrange’s equations.
A Contravariant Form of Lagrange’s Equations of Motion
If we multiply both sides of the covariant form of Lagrange’s equations by the in
verse of a
RK
, then we will ﬁnd the contravariant form of Lagrange’s equations of
motion. Because we have imposed the integrable constraints and are interested in
∗
See, for example, McConnell [139] or Synge and Schild [208].
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
126 The Equations of Motion for a System of Particles
only the ﬁrst Mequations of motion, we can ﬁnd only the Mcontravariant equations
from the M covariant equations provided
∗
:
⎡
⎢
⎢
⎣
a
11
· · · a
1M
.
.
.
.
.
.
.
.
.
a
1M
· · · a
MM
⎤
⎥
⎥
⎦
⎡
⎢
⎢
⎣
a
11
· · · a
1M
.
.
.
.
.
.
.
.
.
a
1M
· · · a
MM
⎤
⎥
⎥
⎦
=
⎡
⎢
⎢
⎣
1 · · · 0
.
.
.
.
.
.
.
.
.
0 · · · 1
⎤
⎥
⎥
⎦
. (4.31)
That is, we require several a
KJ
’s to be zero:
⎡
⎢
⎢
⎢
⎢
⎣
a
1(M+1)
a
1(M+2)
· · · a
1(3N)
a
2(M+1)
a
2(M+2)
· · · a
2(3N)
.
.
.
.
.
.
.
.
.
.
.
.
a
3N(M+1)
a
3N(M+2)
· · · a
3N(3N)
⎤
⎥
⎥
⎥
⎥
⎦
=
⎡
⎢
⎢
⎢
⎢
⎢
⎣
0 0 · · · 0
0 0 · · · 0
.
.
.
.
.
.
.
.
.
.
.
.
0 0 · · · 0
⎤
⎥
⎥
⎥
⎥
⎥
⎦
.
When (4.31) holds, the righthand side of Lagrange’s equations of motion trans
forms from the covariant components of to the contravariant components:
J
=
M
¸
R=1
a
JR
R
.
In summary, the contravariant form of Lagrange’s equations of motion is
m¨ q
J
+ m
M
¸
K=1
M
¸
S=1
J
KS
˙ q
K
˙ q
S
=
J
(J = 1, . . . , M) . (4.32)
This form of Lagrange’s equations is very useful in numerical simulations because
we have explicit expressions for ¨ q
1
, . . . , ¨ q
M
.
An Example
A simple example with which to explore Equations (4.30) and (4.32) is to consider
a single particle moving in E
3
under the action of a force F. We describe the motion
of this particle by using a spherical polar coordinate system: q
1
= R, q
2
= φ, and
q
3
= θ. That is, M = 3. We can also consider our discussion here to be a solution to
Exercise 3.3.
It should by now be trivial to establish that
T =
m
2
˙
R
2
+ R
2
sin
2
(φ)
˙
θ
2
+R
2
˙
φ
2
.
∗
In terms of the representative particle, these restrictions are equivalent to a
J
· a
K
= 0 for all J =
1, . . . , M and all K = M+ 1, . . . , 3N.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
4.10 A Canonical Form of Lagrange’s Equations 127
From the expression for T, we can immediately deduce that
⎡
⎢
⎣
a
11
a
12
a
13
a
12
a
22
a
23
a
13
a
23
a
33
⎤
⎥
⎦
=
⎡
⎢
⎣
1 0 0
0 R
2
0
0 0 R
2
sin
2
(φ)
⎤
⎥
⎦
,
⎡
⎢
⎣
a
11
a
12
a
13
a
12
a
22
a
23
a
13
a
23
a
33
⎤
⎥
⎦
=
⎡
⎢
⎣
1 0 0
0 R
−2
0
0 0 R
−2
sin
−2
(φ)
⎤
⎥
⎦
.
We readily deduce the covariant form of Lagrange’s equations of motion from
d
dt
∂T
∂ ˙ q
i
−
∂T
∂q
i
= F
i
by expanding the time derivative:
m
⎡
⎢
⎣
1 0 0
0 R
2
0
0 0 R
2
sin
2
(φ)
⎤
⎥
⎦
⎡
⎢
⎣
¨
R
¨
φ
¨
θ
⎤
⎥
⎦
+
⎡
⎢
⎣
C
1
C
2
C
3
⎤
⎥
⎦
=
⎡
⎢
⎣
F
1
F
2
F
3
⎤
⎥
⎦
, (4.33)
where
⎡
⎢
⎣
C
1
C
2
C
3
⎤
⎥
⎦
=
⎡
⎢
⎢
⎣
−mR
sin
2
(φ)
˙
θ
2
+
˙
φ
2
−mRsin(φ) cos(φ)
˙
θ
2
+ 2mR
˙
R
˙
φ
2mRsin
2
(φ)
˙
R
˙
θ + 2mR
2
sin(φ) cos(φ)
˙
φ
˙
θ
⎤
⎥
⎥
⎦
,
⎡
⎢
⎣
F
1
F
2
F
3
⎤
⎥
⎦
=
⎡
⎢
⎣
F · e
R
F · Re
φ
F · Rsin(φ)e
θ
⎤
⎥
⎦
,
⎡
⎢
⎣
F
1
F
2
F
3
⎤
⎥
⎦
=
⎡
⎢
⎢
⎣
F · e
R
F ·
1
R
e
φ
F ·
1
Rsin(φ)
e
θ
⎤
⎥
⎥
⎦
=
⎡
⎢
⎣
a
11
a
12
a
13
a
12
a
22
a
23
a
13
a
23
a
33
⎤
⎥
⎦
⎡
⎢
⎣
F
1
F
2
F
3
⎤
⎥
⎦
.
From (4.33), we can read off the Christoffel symbols of the ﬁrst kind.
For example, [22, 1] = −R, [33, 1] = −Rsin
2
(φ), [12, 2] = R, and [13, 3] =
Rsin
2
(φ).
We can determine the contravariant form of (4.33) by multiplying both sides of
(4.33) by the inverse of
¸
a
i j
¸
:
m
⎡
⎢
⎣
¨
R
¨
φ
¨
θ
⎤
⎥
⎦
+
⎡
⎢
⎢
⎣
−mR(sin
2
(φ)
˙
θ
2
+
˙
φ
2
)
−
msin(φ) cos(φ)
R
˙
θ
2
+
2m
R
˙
R
˙
φ
2m
R
˙
R
˙
θ + 2mcot(φ)
˙
φ
˙
θ
⎤
⎥
⎥
⎦
=
⎡
⎢
⎣
F
1
F
2
F
3
⎤
⎥
⎦
. (4.34)
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
128 The Equations of Motion for a System of Particles
Again, we can read off the Christoffel symbols of the second kind from the right
hand side of these equations. For example,
1
22
= −R,
2
33
= −sin(φ) cos(φ), and
3
23
= cot(φ).
4.11 Alternative Principles of Mechanics
Newton’s tremendous contributions to mechanics in the 17th century still left many
unanswered questions. Among these questions were the formulation of equations of
motion for rigid bodies and deformable media. To this end, several principles of me
chanics were postulated in the subsequent centuries: among them, Jean Bernoulli’s
principle of virtual work from 1717, D’Alembert’s principle from 1743 [44], Gauss’
principle of least constraint in 1829 [68], and Hamilton’s principle in 1835 [89].
∗
The
purpose of this section is to brieﬂy outline how some of these principles are related
to the balances of linear momenta used to establish the equations of motion for a
system of particles.
4.11.1 Principle of Virtual Work and D’Alembert’s Principle
We ﬁrst consider the principle of virtual work and D’Alembert’s principle applied
to a system of N particles. These principles are often the basis for treatments of
Lagrange’s equations of motion in many texts.
†
We assume that the particles are
subject to a single constraint:
f
1
· v
1
+ · · · + f
N
· v
N
+e = 0. (4.35)
The principle of virtual work and D’Alembert’s principle collectively state that the
motion of the system of particles is such that the following equation is satisﬁed:
(F
a1
− m
1
¨ r
1
) · d
1
+ · · · + (F
aN
−m
N
¨ r
N
) · d
N
= 0 (4.36)
for all possible choices of the vectors d
1
, . . . , d
N
that satisfy the condition
f
1
· d
1
+ · · · + f
N
· d
N
= 0. (4.37)
Notice that the e present in (4.35) is notably absent from (4.37). The vectors d
K
are
known as virtual displacements and are usually denoted by δr
K
. The virtual work
performed by the applied force F
aK
is deﬁned as F
aK
· d
K
; thus (4.36) states that the
combined virtual work of the applied forces F
aK
and the inertial forces −m
K
¨ r
K
is
zero.
Our aim is to obtain the equations of motion of the system of particles from
(4.36). To this end, we can introduce Lagrange multipliers to accommodate the
∗
Further background on these (and many other principles) can be found in Dugas [48], Szab´ o [209],
and Truesdell [216, 217].
†
See, for example, Section 4.9 of Baruh [14], Section 2.1 of Greenwood [78], or Section 46 of Synge
[206]. As noted by many authors, the principle of virtual work pertains to static problems, and its
extension to dynamics requires the application of D’Alembert’s principle; hence the combination
of these principles in this section.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
4.11 Alternative Principles of Mechanics 129
constraint on the vectors d
K
∗
:
N
¸
K=1
(F
aK
−m
K
¨ r
N
) · d
K
+µ
N
¸
K=1
f
K
· d
K
= 0. (4.38)
As a consequence of the Lagrange multiplier µ, the vectors d
K
can be varied inde
pendently. For (4.38) to hold for all such displacements, it is necessary and sufﬁcient
that
m
K
¨ r
K
= F
aK
+µf
K
(K = 1, . . . , N) . (4.39)
However, (4.39) are none other than the balances of linear momenta for a system of
particles subject to constraint (4.35), for which the constraint forces are prescribed
by use of Lagrange’s prescription:
F
cK
= µf
K
(K = 1, . . . , N) . (4.40)
We can now easily proceed to establish Lagrange’s equations of motion for the sys
tem of particles. Thus the principle of virtual work and D’Alembert’s principle com
bined with Lagrange’s prescription for the constraint forces collectively lead to the
balance laws for a constrained system of particles.
When constraint (4.35) is integrable, we can interpret (4.37) as a normality con
dition for the vectors d
K
. Further, it is not too difﬁcult to see that constraint force
prescription (4.40) implies that this force is normal to the conﬁguration manifold M
in this case.
4.11.2 Gauss’ Principle of Least Constraint
Gauss’ principle of least constraint was published in 1829 [68]. It is a remarkable
interpretation of the role played by constraint forces in a mechanical system. Re
stricting our attention to particles, suppose that we have a system of N particles
that are subject to constraint (4.35) and suppose that the constraint forces are pre
scribed by use of Lagrange’s prescription [i.e., (4.40) holds]. Then, in any motion of
the system that satisﬁes the constraints, the constraint forces F
cK
= µf
K
are the least
needed to ensure that the constraint is satisﬁed. That is, Lagrange’s prescription is
in a sense optimal!
Since its introduction in 1829, the principle of least constraint has played a key
role in several seminal developments in mechanics (see, for example, Hertz [92]). A
lucid discussion of the principle can be found in Udwadia and Kalaba [218], and its
extension to cases for which Lagrange’s prescription does not hold can be found in
O’Reilly and Srinivasa [162].
∗
When constraint (4.35) is integrable, this is precisely the approach taken by Lagrange in [121, 122]
(cf. Section 4 in Chapter 11 of Dugas [48]).
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
130 The Equations of Motion for a System of Particles
t
q(t
0
)
q
q(t
1
)
t
0
t
1
Figure 4.6. Some of the possible paths q connecting two conﬁgurations, q(t
0
) and q(t
1
), of a
particle. The actual path is the one that satisﬁes the differential equations F = ma. Hamilton’s
principle states that this path extremizes I =
t
1
t
0
Ldt [see (4.41)]. The portrait of Hamilton is
from the Royal Irish Academy in Dublin, Ireland.
4.11.3 Hamilton’s Principle
The remaining principle of interest was discovered by Sir William R. Hamilton
(1805–1865) and ﬁrst published in [89] for a system of unconstrained particles sub
ject solely to conservative forces. For ease of exposition, we restrict our discussion
to a single particle of mass mand suppose that three coordinates q =
q
1
, q
2
, q
3
are
used to parameterize its position vector. Then Hamilton’s principle states that the
motion of the system between a given initial conﬁguration q(t
0
) and a given ﬁnal
conﬁguration q(t
1
) is such that it extremizes the action integral
∗
:
I =
t
1
t
0
Ldt, (4.41)
where the Lagrangian L = T
q
i
, ˙ q
i
−U
q
i
and t
0
and t
1
are ﬁxed instances of time.
As illustrated in Figure 4.6, there are an inﬁnite number of paths q(t) that can con
nect two possible conﬁgurations, and so ﬁnding the one that extremizes I appears to
be a daunting task.
However, it was known long before (4.41) appeared in print in 1835 that the
necessary conditions for q
1
(t), q
2
(t), q
3
(t) to extremize I was that q
1
(t), q
2
(t), q
3
(t)
satisfy the following differential equations:
d
dt
∂L
∂ ˙ q
k
−
∂L
∂q
k
= 0 (k = 1, 2, 3) . (4.42)
In fact, in the context of extremizing I, (4.42) are known as the Euler–Lagrange
equations.
†
Thus Hamilton’s principle implies that the motion of the system satisﬁes
∗
To extremize is to minimize or maximize.
†
Problems featuring the extremization of I can be solved by use of the calculus of variations. In the
middle of the 18th century, Euler and Lagrange played seminal roles in the development of this
calculus [72]. It was on the subject of this calculus that a 19yearold Lagrange ﬁrst wrote to Euler
in 1755.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
Exercise 4.1 131
(4.42). At this stage, it should be transparent that (4.42) are equivalent to F = ma
for a particle. In conclusion, for an unconstrained particle, Hamilton’s principle is
equivalent to the balance of linear momentum. Thus the motion of the system pro
vided by the solution to F = ma extremizes I!
Subsequent extensions to Hamilton’s principle were made by several authors,
and these included the case in which the system of particles was subject to integrable
constraints. Jacobi [102] in particular used this principle to great effect when exam
ining the motion of particles on smooth surfaces and the shortest distance between
two points on a surface. Although Hamilton’s principle cannot readily be applied to
systems of particles subject to nonintegrable constraints, it has helped form several
pillars of modern physics.
4.12 Closing Remarks
In this chapter, a derivation of Lagrange’s equations of motion was presented for
a system of particles, and it was shown how they can be modiﬁed when constraints
are introduced. Our developments emphasized that these equations are equivalent
to the Newtonian balance laws of linear momenta for each particle. This important
feature enables us to conﬁdently calculate the forces
K
=
¸
N
i=1
F
i
·
∂r
i
∂q
K
that appear
on the righthand side of Lagrange’s equations of motion:
d
dt
∂T
∂ ˙ q
K
−
∂T
∂q
K
=
K
(R = 1, . . . , 3N) .
We are also able to introduce integrable and nonintegrable constraints and the con
straint forces associated with them.
On a deeper level, if we wish to consider the system of particles as a single par
ticle moving on a conﬁguration manifold Mthat is embedded in a 3Ndimensional
Euclidean space, then Casey’s construction of the representative particle enables us
to do this in a straightforward manner. In the next chapter, we turn to examples in
which Lagrange’s equations of motion for several systems of particle are established
and discussed.
EXERCISES
4.1. What are the kinematical lineelements ds, generalized coordinates, and con
ﬁguration manifolds Mof the following systems:
(a) a particle attached to a ﬁxed point by a spring;
(b) a particle attached to a ﬁxed point by a rod of length L(t);
(c) a harmonic oscillator consisting of one particle;
(d) a planar double pendulum;
(e) a spherical double pendulum.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
132 Exercises 4.2–4.5
4.2. Under which circumstances do constraint forces perform no work?
4.3. Many authors assume that the kinetic energy of a system of particles is a
positivedeﬁnite function of the velocities ˙ q
K
. Letting T =
¸
3N
I=1
¸
3N
J =1
m
2
a
IJ
˙ q
I
˙ q
J
is
equivalent to saying that the matrix whose components are
m
2
a
IJ
is positivedeﬁnite.
∗
Using the spherical pendulum as an example, show that certain representations
of T are not always positivedeﬁnite. Speciﬁcally, if one uses spherical polar coordi
nates, then the positivedeﬁniteness breaks down at the singularities of this coordi
nate system.
4.4. In Casey’s construction of the single particle, what are the distinctions among
the bases {e
K
}, {e
K
}, and { ¯ e
K
}? For a given system of two particles, how does one
construct these bases for E
6
?
4.5. Consider the following function that depends on the motion of two particles:
V = V(r
2
− r
1
 , t) .
This function is representative of a potential energy function and an integrable
constraint.
(a) For any vector x, establish the following result:
dx
dt
=
x
x
· ˙ x.
The easiest way to show this result is to represent x by Cartesian coordi
nates.
(b) Show that
˙
V =
∂V
∂r
1
· v
1
+
∂V
∂r
2
· v
2
+
∂V
∂t
,
where
∂V
∂r
1
= −
∂V
∂r
2
= −
∂V
∂x
x
x
and
x = x, x = r
2
− r
1
.
(c) Consider a system of two particles subject to a constraint = 0, where
= (r
2
− r
1
, t) . (4.43)
Using Lagrange’s prescription, show that F
c1
= −F
c2
. Notice that this is
Newton’s third law of motion. Give two examples of a physical constraint
for which has the form (4.43) and for which the constraint forces that
enforce this constraint are equal and opposite.
∗
Recall that a matrix C is positivedeﬁnite if, for all nonzero x, x
T
Cx > 0 and x
T
Cx = 0 only when
x = 0.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
Exercises 4.5–4.6 133
(d) Consider a system of two particles subject to a conservative force that has
a potential energy function:
U = U(r
2
− r
1
) . (4.44)
Using the identity,
˙
U = −F
con1
· v
1
− F
con2
· v
2
,
show that F
con1
= −F
con2
. Notice that this (again) is Newton’s third law of
motion. Give two instances in which Uhas the form (4.44) and demonstrate
that the conservative forces are equal and opposite.
4.6. The twobody problem consists of a system of two particles, one of mass
m
1
= M and the other of mass m
2
= m. The sole forces on the system are the
result of a Newtonian gravitational force ﬁeld that has a potential energy function
U
n
= −
GMm
r
2
−r
1

.
(a) For this system, show that the linear momentum of the system is conserved
and that the center of mass C moves at a constant speed in a straight line.
In addition, show that
r
2
− ¯ r =
M
m+M
(r
2
− r
1
) , r
1
− ¯ r = −
m
m+ M
(r
2
− r
1
) .
Here, ¯ r is the position vector of the center of mass.
(b) Show that the angular momentum of the system of particles relative to C is
conserved.
(c) Show that the total energy of the system of particles is conserved.
(d) Show that the differential equations governing the motions of m
1
and m
2
can be written in the following forms
∗
:
M
¨ r
1
−
¨
¯ r
= −GMm
m
M+m
2
r
1
− ¯ r
r
1
− ¯ r
3
,
m
¨ r
2
−
¨
¯ r
= −GMm
M
M+m
2
r
2
− ¯ r
r
2
− ¯ r
3
. (4.45)
(e) Argue that the results of Section 2.8 can be applied to (4.45) to determine
the orbital motions of the particles about their center of mass C.
∗
In celestial mechanics, expressing the equations of motion in this manner is equivalent to using what
is known as a barycentric coordinate system (see, for example, [220]), that is, a coordinate system
whose origin is at the center of mass.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
5 Dynamics of Systems of Particles
5.1 Introduction
In this chapter, several examples of systems of particles are discussed. We pay
particular attention to how the equations of motion for these systems are estab
lished by use of Lagrange’s equations. The examples we discuss are classical and
range from simple harmonic oscillators to dumbbell satellites and pendula. Our
goals are to illuminate the developments of the previous chapter and to present
representative examples.
Examples that are closely related to the ones we discuss can be found in many
dynamics texts. Most of these texts use alternative formulations of Lagrange’s
equations of motion that do not readily accommodate nonconservative forces.
Here, because we have established an equivalence between Lagrange’s equations
of motion and the balances of linear momenta, we are easily able to incorporate
nonconservative forces such as dynamic Coulomb friction. This chapter closes
with a brief discussion of some recent works on the dynamics of systems of
particles.
5.2 Harmonic Oscillators
We ﬁrst consider simple examples involving a system of two particles. The system
shown in Figure 5.1 is the ﬁrst of several related systems that we discuss in this
section.
Referring to the ﬁgure, we see that a particle of mass m
1
is connected by a spring
of stiffness K
1
and unstretched length L
1
to a ﬁxed support. It is also connected by
a spring of stiffness K
2
and unstretched length L
2
to a particle of mass m
2
. Both
particles are constrained to move in the E
1
direction:
r
1
= (L
1
+x
1
) E
1
+y
1
E
2
+z
1
E
3
,
r
2
= (L
1
+L
2
+x
2
) E
1
+y
2
E
2
+z
2
E
3
.
Notice that x
1
and x
2
measure the displacement of the particles from the unstretched
spring states.
134
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
5.2 Harmonic Oscillators 135
m
1
m
2
O
Smooth surface
g
E
1
E
2
Figure 5.1. The system of two particles moving on a smooth horizontal surface. The particles
are connected by linear springs.
Coordinates
Here we denote the six coordinates by
q
1
= x
1
, q
2
= x
2
, q
3
= y
1
, q
4
= y
2
, q
5
= z
1
, q
6
= z
2
.
For the systems of interest, the particles are constrained to move on a line. Thus the
system is subject to four constraints:
1
= 0,
2
= 0,
3
= 0,
4
= 0,
where
1
= y
1
,
2
= y
2
,
3
= z
1
,
4
= z
2
.
It should be noted that
∂
1
∂r
1
= E
2
,
∂
2
∂r
1
= 0,
∂
3
∂r
1
= E
3
,
∂
4
∂r
1
= 0,
and
∂
1
∂r
2
= 0,
∂
2
∂r
2
= E
2
,
∂
3
∂r
2
= 0,
∂
4
∂r
2
= E
3
.
These expressions will be used later to prescribe constraint forces.
Kinetic and Potential Energies
It is easy to see that the kinetic and potential energies of the system are
T =
m
1
2
˙ x
2
1
+ ˙ y
2
1
+ ˙ z
2
1
+
m
2
2
˙ x
2
2
+ ˙ y
2
2
+ ˙ z
2
2
,
U =
K
1
2
x
2
1
+
K
2
2
(x
2
−x
1
)
2
+m
1
gy
1
+m
2
gy
2
. (5.1)
Imposing the four integrable constraints on these energies, we can determine their
constrained counterparts:
˜
T =
m
1
2
˙ x
2
1
+
m
2
2
˙ x
2
2
,
˜
U =
1
2
K
1
x
2
1
+
1
2
K
2
(x
2
−x
1
)
2
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
136 Dynamics of Systems of Particles
Constraint Forces
If the surface that the particles are moving on is smooth, then Lagrange’s prescrip
tion can be invoked:
F
c1
= µ
1
E
2
+µ
3
E
3
, F
c2
= µ
2
E
2
+µ
4
E
3
.
Here, µ
1
, µ
2
, µ
3
, and µ
4
are normal force components. If the surface is rough, then
these prescriptions must be altered to include the friction forces:
F
c1
= µ
1
E
2
+µ
3
E
3
−µ
d
µ
1
E
2
+µ
3
E
3

˙ x
1
 ˙ x
1

E
1
,
F
c2
= µ
2
E
2
+µ
4
E
3
−µ
d
µ
2
E
2
+µ
4
E
3

˙ x
2
 ˙ x
2

E
1
.
The total force acting on each particle is composed of the constraint force and the
conservative forces that are due to gravity and the springs.
The Representative Particle
It is easy to construct the representative particle of mass m for this system. First,
the position vector of this particle, which is moving in E
6
, is deﬁned with the help
of (4.15):
r = (x
1
+L
1
) e
1
+y
1
e
2
+z
1
e
3
+(x
2
+L
1
+L
2
) e
4
+y
2
e
5
+z
2
e
6
.
Using this expression, we easily calculate the six covariant vectors a
J
and the six
contravariant vectors a
J
:
a
1
= e
1
, a
2
= e
4
, a
3
= e
2
, a
4
= e
5
, a
5
= e
3
, a
6
= e
6
,
and
a
1
= e
1
, a
2
= e
4
, a
3
= e
2
, a
4
= e
5
, a
5
= e
3
, a
6
= e
6
.
If the kinetic energy T =
m
2
˙ r · ˙ r were calculated, it would be identical to (5.1).
The constraints on the motion of the single particle have the following repre
sentations:
1
= r · e
2
,
2
= r · e
5
,
3
= r · e
3
,
4
= r · e
6
.
Using these representations, we easily see that the constraint force
c
that we would
prescribe by using Lagrange’s prescription is
c
= µ
1
e
2
+µ
2
e
5
+µ
3
e
3
+µ
4
e
6
. (5.2)
When friction is present, this prescription is inadequate.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
5.2 Harmonic Oscillators 137
The force vector can be determined by use of prescription (4.16):
=
−K
1
(x
1
) +K
2
(x
2
−x
1
) −µ
d
µ
1
E
2
+µ
3
E
3

˙ x
1
 ˙ x
1

e
1
+(µ
1
−m
1
g) e
2
+µ
3
e
3
+(µ
2
−m
2
g) e
5
+µ
4
e
6
+
−K
2
(x
2
−x
1
) −µ
d
µ
2
E
2
+µ
4
E
3

˙ x
2
 ˙ x
2

e
4
.
Notice that the constraint force contributions to this vector are consistent with our
prescription for
c
in (5.2) when µ
d
= 0. We have now completely speciﬁed all the
ingredients needed to establish Lagrange’s equations of motion by using the repre
sentative particle of mass m.
The Generalized Coordinates and Conﬁguration Manifold
There are four integrable constraints on the system, so the generalized coordinates
are x
1
and x
2
. The conﬁguration manifold M is simply the plane E
2
and the kine
matical lineelement is
ds =
m
1
m
1
+m
2
dx
1
dt
2
+
m
2
m
1
+m
2
dx
2
dt
2
dt.
Notice that this lineelement is easily related to the standard measure of distance
traveled along a curve (x(τ), y(τ)) on a plane:
dx
dτ
2
+
dy
dτ
2
dτ. We shall shortly
introduce a nonintegrable constraint into the system. This constraint will not affect
M or ds.
Equations of Motion for the Oscillator
We ﬁrst consider the case shown in Figure 5.1 in which friction is absent. For this
system, the constraint forces are prescribed by use of Lagrange’s prescription and
the other forces acting on the system are conservative. Thus
1
= −
∂
˜
U
∂q
1
and
2
=
−
∂
˜
U
∂q
2
. In conclusion, the equations of motion can be obtained by use of the following
form of Lagrange’s equations of motion:
d
dt
∂
˜
T
∂ ˙ q
α
−
∂
˜
T
∂q
α
= −
∂
˜
U
∂q
α
, (5.3)
where α = 1, 2. With a small amount of work, we ﬁnd the equations of motion:
m
1
¨ x
1
= −K
1
x
1
−K
2
(x
1
−x
2
) ,
m
2
¨ x
2
= −K
2
(x
2
−x
1
) .
These equations are classical and are easily seen to be equivalent to F
1
· E
1
= m
1
¨ r
1
and F
2
· E
1
= m
2
¨ r
2
, respectively.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
138 Dynamics of Systems of Particles
m
1
m
2
O
Rough surface
g
P
E
1
E
2
Figure 5.2. A system of two particles moving on a rough horizontal surface.
The Inﬂuences of Dynamic Friction and an Applied Force
We now consider the same system for particles, but assume that dynamic Coulomb
friction is present and a forcing P = Pcos(ωt)E
1
is applied to the particle of mass
m
2
(see Figure 5.2). We have already determined all of the needed ingredients to
determine the equations of motion for this system. The major difference is that
Lagrange’s equations of motion in the form (5.3) are inadequate. It is necessary to
use the following form of these equations:
d
dt
∂T
∂ ˙ q
K
−
∂T
∂q
K
=
K
. (5.4)
Here,
K
= F
1
·
∂r
1
∂q
K
+F
2
·
∂r
2
∂q
K
, or equivalently if we use the single representative
particle,
K
= · a
K
. Clearly we need to append P = Pcos(ωt)E
1
to F
2
and
Pcos(ωt)e
4
to .
From (5.4) and the expressions we have established for T and the force, we ﬁnd
the following six equations:
m
1
¨ x
1
= −K
1
x
1
−K
2
(x
1
−x
2
) −µ
d
m
1
g
˙ x
1
 ˙ x
1

,
m
2
¨ x
2
= −K
2
(x
2
−x
1
) −µ
d
m
2
g
˙ x
2
 ˙ x
2

+Pcos(ωt),
0 = N
1y
−m
1
g,
0 = N
2y
−m
2
g,
0 = N
1z
,
0 = N
2z
. (5.5)
The last four equations yield the constraint forces: N
1y
= µ
1
, N
2y
= µ
2
, N
1z
= µ
3
,
and N
2z
= µ
4
.
Stick–Slip Oscillations
One of the most interesting features of the oscillator arises when either or both
of the velocities of the particles vanish. In this case, the number of integrable
constraints increases and the constraint forces must be altered. The resulting
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
5.2 Harmonic Oscillators 139
oscillations are often termed stick–slip: behavior we encountered earlier in our
model for the roller coaster. Because the number of differential equations govern
ing the motion is different for these two types of friction, numerical simulations of
the equations of motion can be challenging.
∗
For instance, suppose m
1
is instantaneously at rest (sticks); then the additional
constraint x
1
= x
10
, where x
10
is a constant, needs to be imposed. The constraint
force F
c1
is now
F
c1
= µ
1
E
2
+µ
3
E
3
+µ
5
E
1
,
where µ
5
E
1
is the static friction force. This force is subject to the static friction
criterion:
µ
5
 ≤ µ
s
µ
2
1
+µ
2
3
,
where µ
s
is the coefﬁcient of static friction. Instead of (5.5), the equations of motion
for the oscillator in this case are
µ
5
= K
1
x
10
+K
2
(x
10
−x
2
) ,
m
2
¨ x
2
= −K
2
(x
2
−x
10
) −µ
d
m
2
g
˙ x
2
 ˙ x
2

+Pcos(ωt), (5.6)
where
K
1
x
10
+K
2
(x
10
−x
2
)
≤ µ
s
m
1
g.
Thus, if there is sufﬁcient friction to match the spring forces, then x
10
will remain
stationary. Otherwise, it will tend to slip in the direction of the resultant spring force
and (5.5) is then used to determine the motion.
Imposing a Nonintegrable Constraint
We now consider the introduction of a nonintegrable constraint in the system con
sidered in Section 5.2. The constraint is
˙ x
1
x
2
− ˙ x
2
= 0.
We can also express this constraint as
(x
2
E
1
) · v
1
+(−E
1
) · v
2
= 0,
or, for the representative particle of mass m, as
x
2
e
1
−e
4
· v = 0.
It is left as an exercise to show that the constraint is nonintegrable.
Using Lagrange’s prescription, we ﬁnd that the additional constraint forces on
the particles are µ
5
x
2
E
1
and −µ
5
E
1
. For the representative particle of mass m,
c
∗
For further details on numerical schemes for stick–slip oscillators, see [23, 50, 112]. Interesting ex
amples of such oscillators and their application to various ﬁelds, including brake squeal, can be
found in [21, 111].
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
140 Dynamics of Systems of Particles
m
1
m
2
r
1
r
2
¯ r
O
C
E
1
E
2
E
3
Figure 5.3. A particle of mass m
1
attached by linear spring of stiffness K and unstretched
length L
0
to a particle of mass m
2
. Each particle is attracted to the ﬁxed point Oby a Newto
nian gravitational force ﬁeld.
needs to be augmented by µ
5
(x
2
e
1
−e
4
). Starting from Lagrange’s equations of mo
tion in the form (5.4), we quickly arrive at the equations of motion:
m
1
¨ x
1
= −K
1
x
1
−K
2
(x
1
−x
2
) −µ
d
m
1
g
˙ x
1
 ˙ x
1

+µ
5
x
2
,
m
2
¨ x
2
= −K
2
(x
2
−x
1
) −µ
d
m
2
g
˙ x
2
 ˙ x
2

−µ
5
.
The last four equations are identical to those recorded in (5.5). The two equations
of motion are supplemented by the constraint ˙ x
1
x
2
− ˙ x
2
= 0.
5.3 A Dumbbell Satellite
In the 1960s, several simple models for deformable satellites orbiting a planet of
mass M appeared. Here we consider one of these models and discuss some features
of the dynamics predicted by it. In particular, we will notice a coupling between the
motion of the center of mass and the rotation of the satellite. This coupling, which
is elegantly explained in Beletskii’s text on satellite dynamics [16], is induced by the
gravitational forces acting on the satellite.
The model we discuss here lumps the mass distribution of the satellite into two
mass particles at the extremeties of a spring of stiffness K and unstretched length
L
0
(see Figure 5.3). The gravitational force exerted on the satellite by the planet of
mass M it is orbiting is modeled as a Newtonian gravitational force ﬁeld exerted on
each of the particles. In what follows, we discuss how the equations of motion for
this model can be determined.
Coordinates
Our ﬁrst task is to choose coordinates for the position vectors r
1
and r
2
of the mass
particles. One reasonable choice would be to pick Cartesian coordinates for both
particles; another would be to choose Cartesian coordinates for r
1
and spherical
polar coordinates for r
2
−r
1
. A third alternative would be to pick Cartesian
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
5.3 A Dumbbell Satellite 141
coordinates for the position vector ¯ r of the center of mass C of the system and a set
of spherical polar coordinates for r
2
− ¯ r and r
1
− ¯ r. We elect the third choice here.
Denoting r
2
−r
1
by Re
R
, where R = r
2
−r
1
,
e
R
= sin(φ) (cos(θ)E
1
+sin(θ)E
2
) +cos(φ)E
3
.
Some elementary algebra shows that
r
1
= ¯ r −
m
2
m
1
+m
2
Re
R
, r
2
= ¯ r +
m
1
m
1
+m
2
Re
R
.
With our choice of coordinates, ¯ r = xE
1
+yE
2
+zE
3
.
For future reference, we label our coordinates
q
1
= x, q
2
= y, q
3
= z, q
4
= R, q
5
= φ, q
6
= θ. (5.7)
We also record the following 12 partial derivatives:
∂r
1
∂x
= E
1
,
∂r
1
∂y
= E
2
,
∂r
1
∂z
= E
3
,
∂r
1
∂R
= −
m
2
m
1
+m
2
e
R
,
∂r
1
∂φ
= −
m
2
m
1
+m
2
Re
φ
,
∂r
1
∂θ
= −
m
2
m
1
+m
2
Rsin(φ)e
θ
,
and
∂r
2
∂x
= E
1
,
∂r
2
∂y
= E
2
,
∂r
2
∂z
= E
3
,
∂r
2
∂R
=
m
1
m
1
+m
2
e
R
,
∂r
2
∂φ
=
m
1
m
1
+m
2
Re
φ
,
∂r
2
∂θ
=
m
1
m
1
+m
2
Rsin(φ)e
θ
.
Because the forces acting on the system of particles are conservative, we can deter
mine Lagrange’s equations of motion without calculating these vectors. Our moti
vation for calculating them here is that they will illuminate a certain relationship in
Section 5.3.
Kinetic and Potential Energies
The kinetic energy of the system is the sum of the kinetic energies of the particles:
T =
m
1
2
v
1
· v
1
+
m
2
2
v
2
· v
2
=
m
1
+m
2
2
˙ x
2
+ ˙ y
2
+ ˙ z
2
+
m
1
m
2
2(m
1
+m
2
)
˙
R
2
+R
2
˙
φ
2
+R
2
sin
2
(φ)
˙
θ
2
.
To arrive at this expression, a substantial amount of algebraic cancellations
occurred.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
142 Dynamics of Systems of Particles
The potential energy of the system is due to Newtonian gravitation and the
spring kinetic energy:
U = −
GMm
1
¯ r −
m
2
m
1
+m
2
Re
R
−
GMm
2
¯ r +
m
1
m
1
+m
2
Re
R
+
K
2
(R−L
0
)
2
.
We could also have included the Newtonian gravitational force that m
1
exerts on m
2
and vice versa but this will not signiﬁcantly add to the discussion.
Lagrange’s Equations of Motion
For this system of particles, there are no constraints and the applied forces are con
servative. As a result, we can use Lagrange’s equations of motion in the form
d
dt
∂L
∂ ˙ q
S
−
∂L
∂q
S
= 0 (S = 1, . . . , 6) .
Equipped with the expressions for T and U recorded in the previous subsection,
we can derive the equations straightforwardly, and the details are not presented
here.
Because the sole forces acting on this system are conservative, the solutions to
the equations of motion should preserve the total energy E = T +U. In addition,
the angular momentum of this system relative to O, H
O
, should also be conserved.
The former momentum has the following representation:
H
O
= r
1
×m
1
v
1
+r
2
×m
2
v
2
= ¯ r ×m¯ v +
m
1
m
2
m
1
+m
2
R
2
˙
φe
θ
−
˙
θ sin(φ)e
φ
.
The conservation of H
O
implies that the linear speeds, ˙ x, ˙ y, and ˙ z, of the center of
mass are coupled to the angular speeds
˙
θ and
˙
φ of the satellite. We shall observe this
coupling later in rigid body models for satellites.
The Generalized Coordinates and Conﬁguration Manifold
There are no constraints on the system, so the generalized coordinates are x, y, z, R,
θ, and φ. The conﬁguration manifold is simply E
6
, and the kinematical lineelement
is ds =
2T
m
1
+m
2
dt.
Comments on the Equations of Motion
The equations of motion found by use of Lagrange’s equations are equivalent to
those that would be obtained directly from F
1
= m
1
¨ r
1
and F
2
= m
2
¨ r
2
. Indeed, if we
return to the derivation of Lagrange’s equations in Section 4.6, then it is easy to
see that these equations are linear combinations of F
1
= m
1
¨ r
1
and F
2
= m
2
¨ r
2
. For
instance, Lagrange’s equation of motion for R,
d
dt
∂L
∂
˙
R
−
∂L
∂R
= 0,
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
5.4 A Pendulum and a Cart 143
is equivalent to
(F
1
= m
1
¨ r
1
) ·
∂r
1
∂R
+(F
2
= m
2
¨ r
2
) ·
∂r
2
∂R
= 0.
If we change our selection for the coordinates used for the system, this changes
the precise linear combinations of the components of F
1
= m
1
¨ r
1
and F
2
= m
2
¨ r
2
that
constitute Lagrange’s equations of motion.
Related Systems
We can modify the model presented in this section in a variety of manners. First, the
spring can be replaced with a rigid bar. For this model, E and H
O
are still conserved.
It is interesting to observe that, even for this simple system, the gravitational force
ﬁeld is equivalent to a force acting at the center of mass Cand a moment relative to
this point.
5.4 A Pendulum and a Cart
Consider the system of two particles shown in Figure 5.4. A particle of mass m
1
is
free to move on a smooth horizontal rail and is connected to a particle of mass m
2
by a spring of stiffness K and unstretched length L
0
. The motion of both particles is
assumed to be planar.
Coordinates and Constraints
As usual, the ﬁrst task is to choose coordinates for the position vectors r
1
and r
2
of
the mass particles. In anticipation of imposing the constraints, we choose Cartesian
coordinates for r
1
and cylindrical polar coordinates for r
2
−r
1
:
r
1
= xE
1
+yE
2
+z
1
E
3
, r
2
= xE
1
+yE
2
+re
r
+(z
1
+z
2
) E
3
.
m
1
m
2
O
Smooth horizontal rail
Linear spring
g
r
θ
E
1
E
2
Figure 5.4. A system of two particles connected by a linear spring of stiffness K and un
stretched length L
0
. The particle of mass m
1
is free to move on a smooth horizontal rail and
the second particle moves on the x −y plane.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
144 Dynamics of Systems of Particles
The angle θ is measured from the E
1
direction and is counterclockwise positive. That
is, when m
2
is directly below m
1
, θ =
3π
2
.
We label the coordinates as follows:
q
1
= x, q
2
= r, q
3
= θ, q
4
= y, q
5
= z
1
, q
6
= z
2
= (r
2
−r
1
) · E
3
. (5.8)
The constraints on this system of particles are all integrable:
1
= 0,
2
= 0,
3
= 0,
where
1
= y = r
1
· E
2
,
2
= z
1
= r
1
· E
3
,
3
= z
2
= (r
2
−r
1
) · E
3
.
At this stage, it is prudent to compute that
∂
1
∂r
1
= E
2
,
∂
1
∂r
2
= 0,
∂
2
∂r
1
= E
3
,
∂
2
∂r
2
= 0,
∂
3
∂r
1
= −E
3
,
∂
3
∂r
2
= E
3
.
We could also have used Cartesian coordinates to parameterize r
2
−r
1
, and this
choice would be a good exercise to pursue.
Kinetic and Potential Energies
The potential energy of the system is due to gravity and the spring potential energy:
U = m
1
gy +m
2
gy +m
2
gr sin(θ) +
K
2
(r
2
−r
1
 −L
0
)
2
.
Imposing the integrable constraints, we ﬁnd that
˜
U = m
2
gr sin(θ) +
K
2
(r −L
0
)
2
.
The kinetic energy of the system is also easy to calculate:
T =
m
1
+m
2
2
˙ x
2
+ ˙ y
2
+ ˙ z
2
1
+
m
2
2
˙ r
2
+r
2
˙
θ
2
+ ˙ z
2
2
+m
2
˙ x ˙ r cos(θ) + ˙ y ˙ r sin(θ) − ˙ xr
˙
θ sin(θ) + ˙ yr
˙
θ cos(θ)
+m
2
˙ z
1
˙ z
2
. (5.9)
This expression simpliﬁes when we impose the following constraints:
˜
T =
m
1
+m
2
2
˙ x
2
+
m
2
2
˙ r
2
+r
2
˙
θ
2
+m
2
˙ x ˙ r cos(θ) − ˙ xr
˙
θ sin(θ)
.
Generalized Coordinates and Constraints
The system has three generalized coordinates: x, r, θ. The conﬁguration manifold M
for this system is a threedimensional manifold in E
6
that is parameterized by these
coordinates. Because r ranges from 0 to ∞ and θ ranges from 0 to 2π, these two
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
5.4 A Pendulum and a Cart 145
coordinates parameterize E
2
in its entirety. Further, as x ranges from −∞to ∞, we
see that this coordinate parameterizes E. We conclude that M is E
3
.
We can ﬁnd the kinematical lineelement for the conﬁguration manifold by us
ing
˜
T:
ds =
2
˜
T
m
1
+m
2
dt,
where we have chosen m = m
1
+m
2
.
Constraint Forces and Resultant Forces
The constraint forces on the system can be computed by use of Lagrange’s prescrip
tion:
F
c1
=
3
¸
i=1
µ
i
∂
i
∂r
1
= µ
1
E
2
+µ
2
E
3
−µ
3
E
3
,
F
c2
=
3
¸
i=1
µ
i
∂
i
∂r
2
= µ
3
E
3
.
Notice that the Lagrange multipliers µ
i
are equivalent to the normal forces: N
1
=
µ
1
E
2
+(µ
2
−µ
3
) E
3
and N
2
= µ
3
E
3
.
For completeness, the total resultant forces F
1
and F
2
acting on the particles are
F
1
= K(r −L
0
) e
r
+(µ
1
−m
1
g) E
2
+(µ
2
−µ
3
) E
3
,
F
2
= −K(r −L
0
) e
r
−m
2
gE
2
+µ
3
E
3
,
respectively. We shall use these expressions later on to determine the force vector .
Lagrange’s Equations of Motion
We can ﬁnd the equations of motion for this system by using any of the variety of
forms for Lagrange’s equations of motion. Arguably, the easiest approach is to use
the form involving Lagrangian (4.21):
d
dt
∂L
∂ ˙ q
S
−
∂L
∂q
S
= Q
S
=
2
¸
i=1
F
nconi
·
∂r
i
∂q
S
(S = 1, . . . , 6) . (5.10)
However, as the constraint forces are compatible with Lagrange’s prescription, for
the ﬁrst three of these equations the righthand side will be zero: Q
1
= 0, Q
2
= 0,
and Q
3
= 0.
To ﬁnd the equations of motion, it sufﬁces to examine the ﬁrst three of
Lagrange’s equations of motion:
d
dt
∂
˜
L
∂ ˙ q
J
−
∂
˜
L
∂q
J
= 0 (J = 1, . . . , 3) .
Here,
˜
L =
˜
T −
˜
U. Evaluating the partial derivatives of
˜
L =
˜
T −
˜
U and then calcu
lating
d
dt
, we ﬁnd with some rearrangement that the equations of motion can be
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
146 Dynamics of Systems of Particles
expressed in the form
⎡
⎢
⎣
m
1
+m
2
m
2
cos(θ) −m
2
r sin(θ)
m
2
cos(θ) m
2
0
−m
2
r sin(θ) 0 m
2
r
2
⎤
⎥
⎦
⎡
⎢
⎣
¨ x
¨ r
¨
θ
⎤
⎥
⎦
= −
⎡
⎢
⎣
f
1
f
2
f
3
⎤
⎥
⎦
, (5.11)
where f
1
, f
2
, and f
3
are quadratic in the velocities:
f
1
= −2m
2
˙ r
˙
θ sin(θ) −m
2
r
˙
θ
2
cos(θ),
f
2
= −m
2
r
˙
θ
2
+m
2
g sin(θ) +K(r −L
0
) ,
f
3
= 2m
2
r ˙ r
˙
θ +m
2
gr cos(θ). (5.12)
The form (5.11) of the equations of motion is easy to implement numerically. It
is also a canonical form for many mechanical systems featuring timeindependent
(scleronomic) integrable constraints. You should also notice that the matrix on the
lefthand side of (5.11) can be obtained by inspection from
˜
T.
Solving for the Constraint Forces
To determine the constraint forces F
c1
and F
c2
, we ﬁrst solve for µ
i
by using three
of Lagrange’s equations of motion:
d
dt
∂L
∂ ˙ q
J
−
∂L
∂q
J
= Q
J
(J = 4, . . . , 6) .
We leave the intermediate steps as an exercise:
F
c1
= (m
1
+m
2
) gE
2
+
d
dt
m
2
˙ r sin(θ) +m
2
r
˙
θ cos(θ)
E
2
,
F
c2
= 0.
It is easy to observe from these expressions the expected result that F
c1
· v
1
+F
c2
·
v
2
= 0.
Conservations
The solutions to equations of motion (5.11) conserve two kinematical quantities.
The ﬁrst is the total energy E of the system. To see this conservation, it sufﬁces to
note that none of the constraint forces do work and all of the remaining forces acting
on the system are conservative. Thus the work–energy theorem easily leads to the
conclusion that
˙
E = 0, where E =
˜
T +
˜
U. The second conservation can be deduced
from (5.11)
1
. That is, the linear momentum of the system in the horizontal direction,
G· E
1
, is conserved.
The Single Representative Particle
We computed Lagrange’s equations of motion for the system without explicitly cal
culating the position vector r of the single particle of mass m moving in E
6
. If we
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
5.5 Two Particles Tethered by an Inextensible String 147
were to use this particle to compute (5.11), then we would ﬁrst need to use (4.15) to
deﬁne
r = xe
1
+ye
2
+z
1
e
3
+(x +r cos(θ)) e
4
+(y +r sin(θ)) e
5
+(z
1
+z
2
) e
6
.
Using this expression, we can easily calculate the six covariant vectors a
J
and the six
contravariant vectors a
J
. For example, a
1
= e
1
+e
4
and a
6
= e
6
. As expected, the
kinetic energy T =
m
2
˙ r · ˙ r will be identical to (5.9).
The force vector can be determined by use of prescription (4.16):
= K(r −L
0
) cos(θ)e
1
+(µ
1
−m
1
g +K(r −L
0
) sin(θ)) e
2
+(µ
2
−µ
3
) e
3
−K(r −L
0
) cos(θ)e
4
+(−m
2
g −K(r −L
0
) sin(θ)) e
5
+µ
3
e
6
.
We used the expressions for F
1
and F
2
recorded earlier to compute this vector. If we
compute · a
S
and compare the results with those we obtained by using F
1
·
∂r
1
∂q
S
+
F
2
·
∂r
2
∂q
S
, then the two sets of expressions for
S
should be identical.
Remarks
The system discussed here is also a good candidate to explore the covariant (4.30)
and contravariant (4.32) forms of Lagrange’s equations of motion that we discussed
earlier. Indeed, by using (5.11), we can readily compute the
¸
a
i j
¸
matrix and the
Christoffel symbols of the ﬁrst kind.
5.5 Two Particles Tethered by an Inextensible String
As shown in Figure 5.5, a particle of mass m
1
is connected by an inextensible string
of length L
0
, which passes through a smooth eyelet at O to a particle of mass m
2
.
The particle of mass m
1
moves on a rough horizontal plane, and the particle of mass
m
2
is free to move in space. The goal of the following analysis is to establish the
equations of motion for the system of particles and then discuss certain conserved
quantities associated with their solutions. To make the problem tractible, we assume
that the string remains taut and that the particle of mass m
2
does not collide with
the underside of the horizontal plane.
The Coordinates and Other Kinematical Quantities
To describe the kinematics of this system of particles, a cylindrical polar coordinate
system {r
1
, θ
1
, z
1
} is used to describe the motion of the particle of mass m
1
and a
spherical polar coordinate system {R
2
, φ
2
, θ
2
} is used to describe the motion of the
particle of mass m
2
:
r
1
= r
1
e
r
1
+z
1
E
3
, r
2
= R
2
e
R
2
.
We deﬁne six coordinates as follows:
q
1
= r
1
, q
2
= θ
1
, q
3
= φ
2
, q
4
= θ
2
, q
5
= x, q
6
= z
1
. (5.13)
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
148 Dynamics of Systems of Particles
m
1
m
2
O
g
Inextensible string
Rough horizontal plane
E
1
E
2
E
3
Figure 5.5. A system of two particles connected by an inextensible string of length L
0
.
Notice that we have introduced a new coordinate x:
R
2
= L
0
−r
1
+x.
Using the cylindrical polar coordinate system and the spherical polar coordinate
systems, we have
r
1
= r
1
e
r
1
+z
1
E
3
, r
2
= (L
0
+x −r
1
) e
R
2
.
Hence,
∂r
1
∂r
1
= e
r
1
,
∂r
1
∂θ
1
= r
1
e
θ
1
,
∂r
1
∂φ
2
= 0,
∂r
1
∂θ
2
= 0,
∂r
1
∂x
= 0,
∂r
1
∂z
1
= E
3
,
∂r
2
∂r
1
= −e
R
2
,
∂r
2
∂θ
1
= 0,
∂r
2
∂φ
2
= (L
0
+x −r
1
) e
φ
2
,
∂r
2
∂θ
2
= (L
0
+x −r
1
) sin(φ
2
)e
θ
2
,
∂r
2
∂x
= e
R
2
,
∂r
2
∂z
1
= 0.
The Potential and Kinetic Energies
The potential energy of the system is due to gravity:
U = m
1
gE
3
· r
1
+m
2
gE
3
· r
2
= m
1
gz
1
+m
2
g (L
0
+x −r
1
) cos (φ
2
) .
To calculate the kinetic energy of the system we ﬁrst need expressions for the veloc
ity vectors. The expression for v
1
is easy to ﬁnd:
v
1
= ˙ r
1
e
r
1
+r
1
˙
θ
1
e
θ
1
+ ˙ z
1
E
3
.
To calculate v
2
, we recall the expression for this vector in spherical polar coordinates
and then substitute for R
2
and
˙
R
2
:
v
2
= ( ˙ x − ˙ r
1
) e
R
2
+(L
0
+x −r
1
) sin(φ
2
)
˙
θ
2
e
θ
2
+(L
0
+x −r
1
)
˙
φ
2
e
φ
2
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
5.5 Two Particles Tethered by an Inextensible String 149
The kinetic energy T of the system of particles is
T =
m
1
2
v
1
· v
1
+
m
2
2
v
2
· v
2
=
m
1
2
˙ r
2
1
+r
2
1
˙
θ
2
1
+ ˙ z
2
1
+
m
2
2
( ˙ x − ˙ r
1
)
2
+(L
0
+x −r
1
)
2
sin
2
(φ
2
)
˙
θ
2
2
+(L
0
+x −r
1
)
2
˙
φ
2
2
.
Notice that T is a function of q
1
, . . . , q
6
and their time derivatives.
The Constraints and Constraint Forces
The constraints on the motion of the system are twofold. First, the particles are
connected by an inextensible string of length L
0
, and, second, the motion of m
1
is
planar. In terms of the coordinates q
1
, . . . , q
6
, the two constraints are
x = 0, z
1
= 0.
The constraint forces associated with these constraints correspond to the tension
force in the string and the friction and normal forces on m
1
:
F
c1
= µ
1
e
r
1
+µ
2
E
3
−µ
d
µ
2
E
3

v
rel
v
rel

,
F
c2
= µ
1
e
R
2
,
where v
rel
= ˙ r
1
e
r
1
+r
1
˙
θ
1
e
θ
1
.
The constraint forces associated with the inextensible string can also be pre
scribed by use of Lagrange’s prescription. Let
1
= r
2
 +r
1
 −L
0
.
Then the inextensibility constraint is
1
= 0. Using Lagrange’s prescription, we ﬁnd
F
c1
= µ
1
r
1
r
1

= µ
1
e
r
1
,
F
c2
= µ
1
e
R
2
.
Notice that we imposed the constraint z
1
= 0 to simplify the expression for F
c1
. The
constraint forces associated with the constraint
2
= 0, where
2
= z
1
, are not pre
scribed by Lagrange’s prescription because it features a dynamic friction force.
The Equations of Motion
Lagrange’s equations will provide four differential equations for the generalized
coordinates and two equations for µ
1
and µ
2
. For ease of exposition, ﬁrst the
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
150 Dynamics of Systems of Particles
differential equations are presented:
(m
1
+m
2
) ¨ r
1
−m
1
r
1
˙
θ
2
1
+m
2
(L
0
−r
1
)
sin
2
(φ
2
)
˙
θ
2
2
+
˙
φ
2
2
= −µ
d
µ
2
E
3

v
rel
· e
r
1
v
rel

+m
2
g cos(φ
2
) ,
d
dt
m
1
r
2
1
˙
θ
1
= −µ
d
r
1
µ
2
E
3

v
rel
· e
θ
1
v
rel

,
d
dt
m
1
(L
0
−r
1
)
2
˙
φ
2
−m
2
(L
0
−r
1
)
2
˙
θ
2
2
sin(φ
2
) cos(φ
2
) = m
2
g (L
0
−r
1
) sin(φ
2
) ,
d
dt
m
2
(L
0
−r
1
)
2
sin
2
(φ
2
)
˙
θ
2
= 0. (5.14)
In addition, the two equations for µ
1
and µ
2
are
µ
1
= m
2
g cos(φ
2
) −m
2
¨ r
1
−m
2
(L
0
−r
1
)
sin
2
(φ
2
)
˙
θ
2
2
+
˙
φ
2
2
,
µ
2
= m
1
g.
To ﬁnd the preceding equations, we started with Lagrange’s equations:
d
dt
∂T
∂ ˙ q
K
−
∂T
∂ ˙ q
= F
1
·
∂r
1
∂q
K
+F
2
·
∂r
2
∂q
K
(K = 1, . . . , 6) .
We then substituted for T, F
1
, and F
2
. After the partial derivatives of T were cal
culated, we imposed the two constraints and performed some rearranging. Some of
the details of these calculations are subsequently provided:
F
1
· e
r
1
−F
2
· e
R
2
= −µ
d
µ
2
E
3

v
rel
· e
r
1
v
rel

+m
2
g cos (φ
2
) ,
F
1
· r
1
e
θ
1
+F
2
· 0 = −µ
d
r
1
µ
2
E
3

v
rel
· e
θ
1
v
rel

,
F
1
· 0 +F
2
· (L
0
−r
1
) e
φ
2
= m
2
g (L
0
−r
1
) sin(φ
2
) ,
F
1
· 0 +F
2
· (L
0
−r
1
) sin(φ
2
) e
θ
2
= 0,
F
1
· 0 +F
2
· e
R
2
= µ
1
−m
2
g cos (φ
2
) ,
F
1
· E
3
+F
2
· 0 = µ
2
−m
1
g.
The Lack of Energy Conservation
To prove that the total energy of the system of particles is not conserved, we start
with the work–energy theorem
˙
T = F
1
· v
1
+F
2
· v
2
and substitute for the applied
and constraint forces:
˙
T = (−m
1
gE
3
+F
c1
) · v
1
+(−m
2
gE
3
+F
c2
) · v
2
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
5.6 Closing Comments 151
Making use of the potential energy function U, we can express this equation as
˙
E = µ
1
(e
r
1
· v
1
+e
R
2
· v
2
) +µ
2
E
3
· v
1
−µ
d
µ
2
E
3

v
rel
v
rel

· v
1
,
where the total energy E = T +U. Thus, as v
1
= v
rel
and is normal to E
3
, and µ
2
=
m
1
g, we surmise that
˙
E = µ
1
(e
r
1
· v
1
+e
R
2
· v
2
) −µ
d
m
2
gv
1
.
However, e
r
1
· v
1
+e
R
2
· v
2
= ˙ r
1
+
˙
R
2
, and this sum is zero because r
1
+R
2
= L
0
. In
conclusion,
˙
E = −µ
d
m
2
gv
1
.
Notice that
˙
E ≤ 0 as expected because of the friction force.
Conservations of Angular Momenta
If friction is absent, then we ﬁnd from (5.14)
2,4
that H
O1
· E
3
= m
1
r
2
1
˙
θ
1
and H
O2
·
E
3
= m
2
(L
0
−r
1
)
2
sin
2
(φ
2
)
˙
θ
2
are conserved. That is, the angular momentum of
each particle relative to Oin the E
3
direction is conserved.
Conﬁguration Manifold and Its Geometry
The conﬁguration manifold for this system is a fourdimensional subspace of E
6
pa
rameterized by r ∈ (0, L
0
), θ
1
∈ [0, 2π), θ
2
∈ [0, 2π), and φ
2
∈ (0, π). The kinematical
lineelement ds for this manifold is given by
ds =
2
˜
T
2
m
1
+m
2
dt,
where we ﬁnd
˜
T
2
from T by imposing the constraints and collecting all those terms
that are quadratic in the generalized velocities:
˜
T
2
=
m
1
2
˙ r
2
1
+r
2
1
˙
θ
2
1
+
m
2
2
˙ r
2
1
+(L
0
−r
1
)
2
sin
2
(φ
2
)
˙
θ
2
2
+(L
0
−r
1
)
2
˙
φ
2
2
.
To visualize the conﬁguration manifold, one would give a twodimensional picture
of a plane with the coordinates r
1
cos (θ
1
)–r
1
sin(θ
1
). This would be supplemented by
a threedimensional image featuring a sphere of radius 1 parameterized by φ
2
and θ
2
.
5.6 Closing Comments
Problems involving systems of particles have played a key role in the development
of dynamics. Speciﬁcally, mention is made here of a model for the celestial system of
the Sun, Earth, and Moon, known as the threebody problem. In this problem, the
three bodies are modeled as particles subject to the mutual interaction that is due to
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
152 Dynamics of Systems of Particles
m
1
m
1
m
1
m
2
m
2
m
2
m
3
m
3
m
3
(a)
(b)
(c)
Figure 5.6. Representative orbits from the threebody problem: (a), (b) examples of La
grange’s equilateral triangle solutions, and (c) the ﬁgureeight solution. For the solutions
shown in this ﬁgure m
1
= m
2
= m
3
.
a Newtonian gravitational force ﬁeld. That is, the potential energy for the system is
[cf. (4.9)]
U
n
= −
Gm
1
m
2
r
2
−r
1

−
Gm
3
m
1
r
1
−r
3

−
Gm
2
m
3
r
3
−r
2

, (5.15)
where r
1
, r
2
, and r
3
are the position vectors of the particles of masses m
1
, m
2
, and
m
3
, respectively.
Famous exact solutions to special cases of the threebody problem range from
the equilateral triangle solution by Lagrange [117] in 1772
∗
to the ﬁgureeight so
lution that was only recently found numerically by Moore [147] and Moore and
Nauenberg [148] and proven to exist by Chenciner and Montgomery [37, 145] (see
Figure 5.6).
†
Apart from its paucity of exact solutions, the threebody problem is
also well known because of the profound analysis of this system by Henri Poincar´ e
(1854–1912) in the late 1880s (see [4, 13, 45]). His analysis is considered to be the
ﬁrst description of chaos in mathematical models for physical systems and formed
one of the cornerstones for the ﬁeld of chaos in dynamical systems that achieved
popular attention some 100 years later in the late 1980s.
The threebody and twobody problems are special cases of the more general
nbody problem. In celestial mechanics, the nbody problem is synonymous with
models for our Solar System and has attracted some of the most celebrated scien
tists in history. It was also the problem that led Hamilton to discover his famous
∗
A discussion of this famous solution can be found in many texts on celestial mechanics, for example,
[93, 150, 220].
†
The interested reader is referred to the online article by Casselman [31], where simulations of
several threebody problems can be found.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
Exercises 5.1–5.3 153
equations of motion in [88] and his variational principle in [89]. Unfortunately, we
do not have the opportunity to explore the threebody and nbody problems in any
detail here; the interested reader is referred to the previously cited texts.
The problems just mentioned do not feature constraints on the motions of
the particles, and they are often formulated without using Lagrange’s equations
of motion. However, problems featuring particles connected by rigid links often
feature in simple models for artiﬁcial satellites orbiting a celestial body and in
various pendulum systems. For these models, Lagrange’s equations of motion are
ideally suited to the task of establishing a set of governing ordinary differential
equations that are free from constraint forces. In the following exercises, problems
featuring systems of particles of this type are emphasized.
EXERCISES
5.1. Consider the systems of particles discussed in Section 5.2. Suppose a time
dependent force P(t)E
1
acted on the particle m
2
.
∗
What are the equations of motion
for each of these systems?
5.2. Again, consider the systems of particles discussed in Section 5.2. Suppose, in
addition to the springs, there are viscous dashpots in these systems.
†
Then, what are
the equations of motion?
5.3. Here, we are interested in establishing a particular representation for the equa
tions governing the motion of two unconstrained particles. In a subsequent exercise,
one can impose constraints to yield the equations of motion of a pendulum system.
Consider the system of particles shown in Figure 5.7. The particles are free to
move in E
3
under the inﬂuences of resultant external forces F
1
and F
2
, respectively.
m
1
m
2
O
g
E
1
E
2
E
3
Figure 5.7. A system of two particles.
(a) To establish the equations of motion for the single particle, we use a
cylindrical polar coordinate system {r
1
, θ
1
, z
1
} for the particle of mass m
1
.
For the second particle, it is convenient to describe its motion with the
assistance of the relative position vector r
21
= r
2
−r
1
. We describe this
∗
This force is not conservative.
†
The forces from these dashpots are not conservative.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
154 Exercise 5.3
vector by using a spherical polar coordinate system {R
2
, φ
2
, θ
2
}. Show that
the position vector of the single particle is
r = r
1
cos(θ
1
)e
1
+r
1
sin(θ
1
)e
2
+z
1
e
3
+(r
1
cos(θ
1
) +R
2
sin(φ
2
) cos(θ
2
))e
4
+(r
1
sin(θ
1
) +R
2
sin(φ
2
) sin(θ
2
))e
5
+(z
1
+R
2
cos(φ
2
))e
6
.
(b) Using r and the curvilinear coordinate system it induces on E
6
,
q
1
= r
1
, q
2
= θ
1
, q
3
= z
1
, q
4
= R
2
, q
5
= φ
2
, q
6
= θ
2
,
show that the six covariant basis vectors a
J
=
∂r
∂q
J
are
a
1
=
∂r
∂r
1
= cos(θ
1
)e
1
+sin(θ
1
)e
2
+cos(θ
1
)e
4
+sin(θ
1
)e
5
,
a
2
=
∂r
∂θ
1
= −r
1
sin(θ
1
)e
1
+r
1
cos(θ
1
)e
2
−r
1
sin(θ
1
)e
4
+r
1
cos(θ
1
)e
5
,
a
3
=
∂r
∂z
1
= e
3
+e
6
,
a
4
=
∂r
∂R
2
= sin(φ
2
) cos(θ
2
)e
4
+sin(φ
2
) sin(θ
2
)e
5
+cos(φ
2
)e
6
,
a
5
=
∂r
∂φ
2
= R
2
cos(φ
2
) cos(θ
2
)e
4
+R
2
cos(φ
2
) sin(θ
2
)e
5
−R
2
sin(φ
2
)e
6
,
a
6
=
∂r
∂θ
2
= −R
2
sin(φ
2
) sin(θ
2
)e
4
+R
2
sin(φ
2
) cos(θ
2
)e
5
.
(c) Show that the six contravariant basis vectors have the following represen
tations:
a
1
= cos(θ
1
)e
1
+sin(θ
1
)e
2
,
a
2
= −
sin(θ
1
)
r
1
e
1
+
cos(θ
1
)
r
1
e
2
,
a
3
= e
3
,
a
4
= sin(φ
2
)(cos(θ
2
)(e
4
−e
1
) +sin(θ
2
)(e
5
−e
2
)) +cos(φ
2
)(e
6
−e
3
),
a
5
=
cos(φ
2
)
R
2
cos(θ
2
)(e
4
−e
1
) +sin(θ
2
)(e
5
−e
2
)
−
sin(φ
2
)
R
2
(e
6
−e
3
),
a
6
= −
sin(θ
2
)
R
2
sin(φ
2
)
(e
4
−e
1
) +
cos(θ
2
)
R
2
sin(φ
2
)
(e
5
−e
2
).
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
Exercises 5.3–5.4 155
(d) Show that the kinetic energy T of the particle of mass m = m
1
+ m
2
is
T =
m
1
+m
2
2
˙ r
2
1
+r
2
1
˙
θ
2
1
+ ˙ z
2
1
+
m
2
2
˙
R
2
2
+R
2
2
˙
φ
2
2
+R
2
2
sin
2
(φ
2
)
˙
θ
2
2
+m
2
cos(φ
2
)
˙
R
2
˙ z
1
+
˙
φ
2
˙ r
1
R
2
cos(θ
21
) +
˙
φ
2
˙
θ
1
r
1
R
2
sin(θ
21
)
−m
2
sin(φ
2
)
R
2
˙
φ
2
˙ z
1
−r
1
R
2
˙
θ
1
˙
θ
2
cos(θ
21
) − ˙ r
1
˙
R
2
cos(θ
21
)
−m
2
sin(φ
2
)
−r
1
˙
R
2
˙
θ
1
sin(θ
21
) + ˙ r
1
˙
θ
2
R
2
sin(θ
21
)
,
where we have used the abbreviation θ
21
= θ
2
−θ
1
. This expression for the
kinetic energy follows from the deﬁnition
T =
m
2
v · v =
m
1
2
v
1
· v
1
+
m
2
2
v
2
· v
2
.
(e) If the forces acting on the particles are F
1
= −m
1
gE
3
and F
2
= −m
2
gE
3
,
then what are the force and potential energy Uassociated with this force?
(f) What are the six Lagrange’s equations governing the motion of the particle
of mass m?
∗
5.4. As shown in Figure 5.8, two particles of mass m
1
and m
2
are connected by a
rigid massless rod of length L
2
. The rod is connected to m
1
by a ballandsocket
joint. In addition, the particle of mass m
1
is connected by a rigid massless rod of
length L
1
to a ﬁxed point O. The connection between the rod and the point O is
through a pin joint and is such that the motion of m
1
is in the E
1
−E
2
plane.
m
1
m
2
O
g
E
1
E
2
E
3
Figure 5.8. A planar double pendulum.
(a) What are the three constraints on the motion of the particle of mass m?
(b) Using Lagrange’s prescription, what is the constraint force
c
acting on the
particle of mass m? You should also, if possible, verify that the components
of this force are physically realistic.
∗
You should refrain if possible from expanding the time derivative here – it will entail a considerable
amount of algebra.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
156 Exercise 5.4–5.5
(c) Starting from the ﬁnal results of Exercise 5.3, establish Lagrange’s
equations of motion for the pendulum system. In your solution, clearly
distinguish the equations governing the motion of the particle and the
equations giving the components of
c
.
(d) Let us now establish some of the equations of (c) by using an equivalent
approach. Impose the constraints on T to determine the constrained kinetic
energy
˜
T. In addition, determine the constrained potential energy
˜
U. Verify
that the following equations correspond to those you obtained from (c)
∗
:
d
dt
∂
˜
T
∂
˙
θ
1
−
∂
˜
T
∂θ
1
= · a
2
= −
∂
˜
U
∂θ
1
,
d
dt
∂
˜
T
∂
˙
φ
2
−
∂
˜
T
∂φ
2
= · a
5
= −
∂
˜
U
∂φ
2
,
d
dt
∂
˜
T
∂
˙
θ
2
−
∂
˜
T
∂θ
2
= · a
6
= −
∂
˜
U
∂θ
2
.
(e) Suppose a nonintegrable constraint is imposed on the pendulum system
discussed in (d):
f
1
· v
1
+f
2
· v
2
+e = 0.
Show that this constraint can be expressed as
f · v +e = 0.
In addition, what are the equations governing the motion of the non
integrably constrained system? Illustrate your solution with an non
integrable constraint of your choice.
5.5. As shown in Figure 5.9, a model for an artiﬁcial satellite consists of two
particles of mass m
1
and m
2
connected by a rigid massless rod of length L
0
. A third
particle of mass m
3
is assumed to be stationary at the ﬁxed point O. In addition to
the constraint force in the rod, the system is subject to conservative forces whose
potential energy function is given by (5.15).
(a) What are the four constraints on the motion of the system of particles?
(b) Using Lagrange’s prescription, what are the constraint forces acting on the
particles of mass m
1
and m
2
? You should also, if possible, verify that the
components of these forces are physically realistic.
(c) Using a set of Cartesian coordinates to describe the location of the center
of mass C of the satellite of mass m
1
+m
2
and a set of spherical polar
coordinates to parameterize the position of m
2
relative to C, establish an
expression for the kinetic energy of the system.
(d) Establish the equations of motion for the system.
∗
It is crucial to note that
∂
˜
T
∂ ˙ r
1
=
∂
˜
T
∂
˙
R
2
=
∂
˜
T
∂ ˙ z
1
=
∂
˜
T
∂r
1
=
∂
˜
T
∂R
2
=
∂
˜
T
∂z
1
= 0.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
Exercises 5.5–5.6 157
m
1
m
2
m
3
O
E
1
E
2
E
3
Figure 5.9. Schematic of a model for a satellite
orbiting a ﬁxed body of mass m
3
.
(e) Show that the solutions to the equations of motion for the system conserve
the total energy of the system and the angular momentum of the system
relative to O.
(f) Show that it is possible for C to execute a steady circular motion about O.
What are the possible orientations of the rigid massless rod of length L
0
during such motions?
5.6. As shown in Figure 5.10, a particle of mass m
1
is connected by a linear spring
of stiffness K
1
and unstretched length L
0
to a ﬁxed point O. A second particle of
mass m
2
is attached by a rod of length L
2
to the particle of mass m
1
with a pin
joint. For this system, which is a variation on the classical system of a planar double
pendulum, we assume that the motions of m
1
and m
2
are constrained to move on
the E
1
−E
2
plane.
m
1
m
2
O
E
1
E
2
g
linear spring
rigid massless rod
Figure 5.10. A system of two parti
cles connected by a rigid massless
rod of length L
2
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
158 Exercise 5.6
To describe the kinematics of this system, a cylindrical polar coordinate system
{r
1
, θ
1
, z
1
} is used to parameterize the motion of the particle of mass m
1
and another
cylindrical polar coordinate system{r
2
, θ
2
, z
2
} is used to parameterize the motion of
the particle of mass m
2
relative to m
1
:
r
1
= r
1
e
r
1
+z
1
E
3
, r
2
= r
1
+r
2
e
r
2
+z
2
E
3
.
We deﬁne six coordinates as follows:
q
1
= θ
1
, q
2
= θ
2
, q
3
= r
1
, q
4
= r
2
, q
5
= z
1
, q
6
= z
2
. (5.16)
(a) With the help of (5.16), what are the 12 vectors
∂r
1
∂q
K
and
∂r
2
∂q
K
? Here,
K = 1, . . . , 6.
(b) What are the three constraints on the motion of the system of particles?
Argue that the constraint forces F
c1
and F
c2
acting on the individual
particles have the prescriptions
F
c1
= µ
1
e
r
2
+µ
2
E
3
, F
c2
= −µ
1
e
r
2
+µ
3
E
3
. (5.17)
Compute the following six components:
cK
= F
c1
·
∂r
1
∂q
K
+F
c2
·
∂r
2
∂q
K
.
Comment on the values of the ﬁrst three components.
(c) In terms of the coordinates q
1
, . . . , q
3
and their time derivatives, what are
the kinetic energy
˜
T and potential energy
˜
U of the constrained system of
particles?
(d) What are Lagrange’s equations of motion for the generalized coordinates
of this system of particles?
(e) Suppose a nonintegrable constraint
∗
r
1
˙
θ
1
+L
2
˙
θ
2
= 0 (5.18)
is imposed on the system of particles. After expressing this constraint in
the form f
1
· v
1
+f
2
· v
2
= 0, argue that
F
c1
= µ
1
e
r
2
+µ
2
E
3
+µ
4
(e
θ
1
−e
θ
2
) , F
c2
= −µ
1
e
r
2
+µ
3
E
3
+µ
4
e
θ
2
.
With the help of your results from (d), determine the equations of motion
for the system of particles.
(f) Starting from the work–energy theorem
˙
T = F
1
· v
1
+F
2
· v
2
, show that the
total energy E is conserved.
(g) Suppose that the spring is replaced with a rigid rod of length L
1
and
nonintegrable constraint (5.18) is removed. In this case, which is the classic
∗
Referring to the discussion of constraint (1.16) in Chapter 1, this constraint is arguably the simplest
nonintegrable constraint that we can impose on this system.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
Exercises 5.6–5.7 159
planar double pendulum, show that the equations governing the motion of
the system are
(1 +α)
d
2
θ
1
dτ
2
+αβcos (θ
2
−θ
1
)
d
2
θ
2
dτ
2
− αβ
dθ
2
dτ
2
sin(θ
2
−θ
1
)
= −(1 +α) cos (θ
1
) ,
d
2
θ
2
dτ
2
+
1
β
cos (θ
2
−θ
1
)
d
2
θ
1
dτ
2
+
1
β
dθ
1
dτ
2
sin(θ
2
−θ
1
)
= −
1
β
cos (θ
2
) . (5.19)
In writing (5.19), we have used the following dimensionless parameters and
time variable:
α =
m
2
m
1
, β =
L
2
L
1
, τ =
g
L
1
t.
The conﬁguration manifold M for this system is a torus. What is the
kinematical lineelement for M?
(h) Numerically integrate (5.19) for a variety of initial conditions and illustrate
your solutions on the conﬁguration manifold for the planar double pendu
lum. You should verify that your solutions conserve the total energy of the
system.
5.7. This problem is adapted from Section 156 of Whittaker [228] and the introduc
tion to [30]. Consider a system of N particles, and, following Lecture 4 from Jacobi
[102], deﬁne the following function:
J =
1
2
N
¸
k=1
m
k
r
k

2
.
The quantity 2J is often known as the moment of inertia of the system of particles.
(a) Assuming that the center of mass C of the system is stationary and located
at the origin, show that J has the equivalent representation:
J =
1
4
N
¸
k=1
N
¸
j=1
m
k
m
j
M
r
k
−r
j
2
, (5.20)
where M = m
1
+· · · +m
N
.
(b) As in (a), assuming that the center of mass C of the system is station
ary, show that the kinetic energy of the system of particles has the
representation
T =
1
4
N
¸
k=1
N
¸
j=1
m
k
m
j
M
v
k
−v
j
2
. (5.21)
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
160 Exercise 5.7
(c) Now suppose that the system of particles is in motion subject to a
conservative Newtonian force ﬁeld:
U
n
= −
1
2
N
¸
j=1
N
¸
k=1,k=j
Gm
k
m
j
r
k
−r
j
. (5.22)
The presence of the
1
2
in the expression for U
n
should be noted: It is needed
to ensure that the summations on k and j yield the correct expression for
U
n
. With the help of (5.20)–(5.22) and balances of linear momenta for each
particle, establish Jacobi’s equation:
¨
J = 2T +U
n
. (5.23)
This equation is also known as the Lagrange–Jacobi equation (see, for
example, [220]).
(d) For the orbits of the threebody problem shown in Figure 5.6(b), show that
T = −
1
2
U
n
.
(e) Show that J is a measure of the distance squared from the origin of the con
ﬁguration space E
3N
for the representative particle discussed in Section 4.7.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
PART THREE
DYNAMICS OF A SINGLE RIGID
BODY
161
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
162
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
6 Rotation Tensors
6.1 Introduction
One of the key features of the rigid body dynamics problems that we will shortly
examine is the presence of a variable axis of rotation. This is one of the reasons
for the richness of phenomena in rigid body dynamics. It is also a reason why this
subject is intimidating. To quote the mechanician Louis Poinsot (1777–1859), from
[172], “. . . if we have to consider the motion of a body of sensible shape, it must be
allowed that the idea which we form of it is very obscure.” In this chapter, several
representations of rotations are discussed that will enable us to establish both a clear
picture of rigid body motions and straightforward proofs of several major results. To
this end, many results on two key kinematical quantities for rigid bodies, rotation
tensors and their associated angular velocity vectors, are discussed in considerable
detail.
The subject of rotations in rigid body dynamics has a wonderful history, a wide
range of interesting results, and an impressive list of contributors. Here, however,
space limits the presentation of only the handful of results that are most relevant to
our purposes. From a historical perspective, much of what is presented was estab
lished by Leonhard Euler (1707–1783) in his great works on rigid body dynamics
that started to appear in the 1750s. The foundations Euler established were built
upon by such notables as Cayley, Gauss, Hamilton, and Rodrigues in the early part
of the 19th century. Despite the wealth of their results, the subject of rotations
remains an active area of discovery (and some unfortunate reinvention) to this
day.
Our exposition makes extensive use of tensors, and the background required
for these quantities is provided in the Appendix. Following related developments
in continuum mechanics, tensors are an invaluable tool for providing concise expla
nations of many important results. They are not universally invoked in dynamics
texts, and the interested reader is invited to compare our treatment with those in
the textbooks cited in the bibliography.
We start this chapter with a discussion of a rotation in which the axis of rotation
is ﬁxed, and this example is used to introduce several of the key concepts in this
163
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
164 Rotation Tensors
chapter. The example also serves to illuminate many of the generalizations that are
required for characterizing more elaborate rotations. To this end, our exposition
then turns to properorthogonal tensors in Section 6.3, and many of their interesting
features are shown there. Then Euler’s representation of a rotation tensor is
presented, and we examine his important theorem that any properorthogonal
tensor is a rotation tensor. This result has many consequences for rigid body
dynamics. One of the intriguing aspects of rotation tensors is the vast range of
representations. Here the Euler angle representation is emphasized, but sufﬁcient
material is present for the interested reader to examine many of the other represen
tations such as the Euler–Rodrigues symmetric parameters (often synonymous with
quaternions).
Our principal reference for this chapter is the authoritative review by Shuster
[196]. This material is supplemented with the elegant relative angular velocity vector
that was proposed by Casey and Lam [29], the dual Euler basis [160], and a proof
of Euler’s theorem by Guo [83]. In the interests of presenting a uniﬁed treatment,
our notation differs from these references in several places, but the differences are
easily deciphered once the material in this chapter has been comprehended.
6.2 The Simplest Rotation
To motivate many of our later developments, we start with the simplest case of a
rotation about a ﬁxed axis p
3
through an angle θ = θ(t). This example should be fa
miliar to you from many different venues. To describe this rotation, we consider
the action of this rotation on this set of orthonormal righthanded basis vectors
¸
p
1
, p
2
, p
3
¸
. As shown in Figure 6.1, we suppose that these vectors are transformed
to the set {t
1
, t
2
, t
3
} by the rotation.
Using a matrix notation, we can represent the transformation from the set of
basis vectors
¸
p
1
, p
2
, p
3
¸
to {t
1
, t
2
, t
3
} as
⎡
⎢
⎣
t
1
t
2
t
3
⎤
⎥
⎦
= R
⎡
⎢
⎣
p
1
p
2
p
3
⎤
⎥
⎦
, (6.1)
θ
θ
θ
p
1
p
2
t
1
t
2
p
3
= t
3
Figure 6.1. The transformations of various basis vectors in
duced by a rotation about p
3
through an angle θ.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
6.2 The Simplest Rotation 165
where
R =
⎡
⎢
⎣
cos(θ) sin(θ) 0
−sin(θ) cos(θ) 0
0 0 1
⎤
⎥
⎦
.
It is easy to see that the matrix R in this equation has a determinant of +1 and that
its inverse is its transpose: R
−1
= R
T
. That is, the matrix R is properorthogonal.
By differentiating (6.1) with respect to time, we ﬁnd that
⎡
⎢
⎣
˙
t
1
˙
t
2
˙
t
3
⎤
⎥
⎦
=
˙
θ
⎡
⎢
⎣
−sin(θ) cos(θ) 0
−cos(θ) −sin(θ) 0
0 0 0
⎤
⎥
⎦
⎡
⎢
⎣
p
1
p
2
p
3
⎤
⎥
⎦
.
Using the result R
−1
= R
T
, we can easily replace p
i
in this equation with t
i
:
⎡
⎢
⎣
˙
t
1
˙
t
2
˙
t
3
⎤
⎥
⎦
=
˙
θ
⎡
⎢
⎣
−sin(θ) cos(θ) 0
−cos(θ) −sin(θ) 0
0 0 0
⎤
⎥
⎦
⎡
⎢
⎣
cos(θ) −sin(θ) 0
sin(θ) cos(θ) 0
0 0 1
⎤
⎥
⎦
⎡
⎢
⎣
t
1
t
2
t
3
⎤
⎥
⎦
=
˙
θ
⎡
⎢
⎣
0 1 0
−1 0 0
0 0 0
⎤
⎥
⎦
. .. .
˙
RR
T
⎡
⎢
⎣
t
1
t
2
t
3
⎤
⎥
⎦
. (6.2)
Notice that this is equivalent to the familiar results
˙
t
1
=
˙
θ t
2
and
˙
t
2
= −
˙
θ t
1
. It should
also be clear from (6.2) that
˙
RR
T
is a skewsymmetric matrix. A vector
˙
θ p
3
=
˙
θt
3
can
be introduced that has the useful property that
˙
t
k
=
˙
θ p
3
×t
k
(k = 1, 2, 3) . (6.3)
You should notice how the vector
˙
θ p
3
can be inferred from the components of
˙
RR
T
.
It is convenient to use a tensor notation to describe the rotation we have been
discussing. In particular, we can write (6.1) in the form
∗
t
k
= Rp
k
(k = 1, 2, 3) , (6.4)
where R is the tensor
R = cos(θ) (p
1
⊗p
1
+p
2
⊗p
2
) −sin(θ) (p
1
⊗p
2
−p
2
⊗p
1
) +p
3
⊗p
3
.
It is left as an exercise to verify that this representation of the rotation is equivalent
to (6.1).
†
Indeed, because
I −p
3
⊗p
3
= p
1
⊗p
1
+p
2
⊗p
2
, p
3
= p
1
⊗p
2
−p
2
⊗p
1
,
∗
Background on tensor notation can be found the Appendix.
†
The deﬁnition of the tensor product needed to establish this equivalence can be found in (A.1) in
the Appendix. The alternating tensor is deﬁned in Section A.7 [see, in particular, (A.11).]
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
166 Rotation Tensors
we can express the tensor R entirely in terms of the axis of rotation p
3
and the angle
of rotation θ:
R = cos(θ) (I −p
3
⊗p
3
) −sin(θ)p
3
+p
3
⊗p
3
. (6.5)
As we shall see later, this representation naturally leads to the general form of a
tensor that represents a rotation about an arbitrary axis through an arbitrary angle
of rotation. It is left as another exercise to show that R
T
= R
−1
and that det(R) = 1.
Differentiating (6.4), we ﬁnd that
˙
t
k
=
˙
Rp
k
+
˙
R
0
˙ p
k
=
˙
RR
T
t
k
....
=p
k
=
˙
RR
T
t
k
.
Some straightforward computations are used to show that
˙
R = −
˙
θ sin(θ) (p
1
⊗p
1
+p
2
⊗p
2
) −
˙
θ cos(θ) (p
1
⊗p
2
−p
2
⊗p
1
) ,
˙
RR
T
=
˙
θ (−p
1
⊗p
2
+p
2
⊗p
1
) .
With the help of these results, we conclude that
˙
t
k
=
˙
RR
T
t
k
=
˙
θ p
3
×t
k
(k = 1, 2, 3) .
As expected, this result is in agreement with (6.3). You might have already noticed
that
˙
θ p
3
=
˙
θ t
3
is the axial vector of
˙
RR
T
.
We have demonstrated how several results for a familiar rotation can be ex
pressed by using a tensor notation. This notation will prove to be very useful
next when we wish to examine more complex rotations. To this end, we need to
answer several questions. The ﬁrst among them is this: What is the representa
tion for a rotation about an arbitrary axis? Once this is known, the question of
whether its time derivative leads to a vector such as
˙
θp
3
then presents itself. The
answers to these questions were ﬁrst formulated by Euler in the 1750s. However,
numerous alternative representations for his solutions have appeared since then,
and we shall leverage several of these representations in the remainder of this
chapter.
6.3 ProperOrthogonal Tensors
To proceed with our treatment of rotations, we appear to regress somewhat and
discuss properorthogonal tensors. This discussion will lay the foundations for the
possibility of threeparameter representations of rotations and subsequently the ex
istence of angular velocity vectors. It is also a starting point for several investigations
on experimental measurements of rotations. We recall that a properorthogonal
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
6.3 ProperOrthogonal Tensors 167
secondorder tensor R is a tensor that has a unit determinant and whose inverse
is its transpose:
R
T
R = RR
T
= I, det(R) = 1. (6.6)
The ﬁrst of these equations implies that there are six restrictions on the nine com
ponents of R. Consequently, only three components of R are independent. In other
words, any properorthogonal tensor can be parameterized by three independent
parameters.
The tensors of interest here are secondorder tensors, any one of which has the
representation
R =
3
¸
i=1
3
¸
k=1
R
ik
p
i
⊗p
k
.
Let us now consider the transformation induced by R on the basis vectors p
1
, p
2
,
and p
3
. We deﬁne
t
1
= Rp
1
=
3
¸
i=1
R
i1
p
i
,
t
2
= Rp
2
=
3
¸
i=1
R
i2
p
i
,
t
3
= Rp
3
=
3
¸
i=1
R
i3
p
i
.
You should notice that, by using the vectors t
i
, R has the representation
R = t
1
⊗p
1
+t
2
⊗p
2
+t
3
⊗p
3
.
We now wish to show that {t
1
, t
2
, t
3
} is a righthanded orthonormal basis. First, let
us verify the orthonormality:
t
i
· t
k
= Rp
i
· Rp
k
= R
T
Rp
i
· p
k
= p
i
· p
k
= δ
ik
.
Hence the vectors t
i
are orthonormal. To establish righthandedness, we use the
deﬁnition of the determinant [see (A.6)]:
[t
1
, t
2
, t
3
] = [Rp
1
, Rp
2
, Rp
3
]
= det(R)[p
1
, p
2
, p
3
]
= (1)(1) = 1.
Hence {t
1
, t
2
, t
3
} is a righthanded orthonormal basis.
∗
∗
It is left as an exercise to verify that this result holds for the example of a simple rotation presented
in (6.1).
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
168 Rotation Tensors
A properorthogonal tensor also has a rather unusual representation. To arrive
at it, we embark on a series of manipulations:
R = RRR
T
= R
3
¸
i=1
3
¸
k=1
R
ik
p
i
⊗p
k
R
T
=
3
¸
i=1
3
¸
k=1
R
ik
R(p
i
⊗p
k
) R
T
=
3
¸
i=1
3
¸
k=1
R
ik
(Rp
i
⊗Rp
k
)
=
3
¸
i=1
3
¸
k=1
R
ik
t
i
⊗t
k
.
In summary, we have the following representations for R:
R =
3
¸
i=1
3
¸
k=1
R
ik
p
i
⊗p
k
=
3
¸
i=1
3
¸
k=1
R
ik
t
i
⊗t
k
=
3
¸
i=1
t
i
⊗p
i
.
Notice that the components of R for the ﬁrst two representations are identical, and
the handedness of {p
1
, p
2
, p
3
} is transferred without change by R to {t
1
, t
2
, t
3
}.
We observed that the components R
ik
of R are equal to t
k
· p
i
. As this product
is equal to the cosine of the angle between t
k
and p
i
, each R
ik
is often referred to as
a direction cosine. Consequently the matrix [R
ik
] is often known as the direction co
sine matrix. Clearly, the nine angles whose cosines are t
k
· p
i
are not all independent,
for if they were then [R
ik
] would have nine independent components and this would
contradict the requirement R
T
R = I. Indeed, as we shall shortly see, it is possible to
arrive at three independent angles to parameterize R, but these angles are not all
easily related to the angles between p
i
and t
k
.
6.4 Derivatives of a ProperOrthogonal Tensor
Here we consider a properorthogonal tensor R that is a function of time: R = R(t).
Consider the derivative of RR
T
:
d
dt
RR
T
=
˙
RR
T
+R
˙
R
T
.
However,
˙
I = O, so the righthand side of the preceding equation is zero. Hence,
˙
RR
T
= −R
˙
R
T
= −
˙
RR
T
T
.
In other words,
˙
RR
T
is a skewsymmetric secondorder tensor. We deﬁne
R
=
˙
RR
T
,
in part because this tensor appears in numerous places later on. The tensor
R
is
known as the angular velocity tensor (of R).
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
6.4 Derivatives of a ProperOrthogonal Tensor 169
The skewsymmetry of
˙
RR
T
allows us to deﬁne the angular velocity vector ω
R
:
ω
R
= −
1
2
¸
˙
RR
T
¸
.
As mentioned in the Appendix, ω
R
×a = (
˙
RR
T
)a for all vectors a.
∗
The most com
mon example of the calculation of an axial vector arises when we consider the mo
tion of a rigid body rotating about the E
3
direction. In this case, we will see later
that the skewsymmetric tensor
R
= (E
2
⊗E
1
−E
1
⊗E
2
) .
Consequently, we can compute that
[
R
] = −2E
3
.
We conclude that the axial vector of
R
is the angular velocity vector E
3
. It also
useful to check that
((E
2
⊗E
1
−E
1
⊗E
2
)) a = E
3
×a
for all vectors a.
In a similar manner, we can also show that R
T
˙
R is a skewsymmetric tensor and
deﬁne an angular velocity tensor
0
R
and another angular vector ω
0
R
:
0
R
= R
T
˙
R, ω
0
R
= −
1
2
[R
T
˙
R]. (6.7)
At a later stage, the reader should be able to show that Rω
0
R
= ω and R
0
R
R
T
=
R
. A key to establishing one of these results is the identity, which holds for all
orthogonal Q,
¸
QBQ
T
¸
= det (Q) Q( [B]) .
Notice how this identity simpliﬁes when Q is a properorthogonal tensor.
Corotational Derivatives
Recall that, for a properorthogonal tensor R, we have the representation
R =
3
¸
i=1
t
i
⊗p
i
.
Now suppose that ˙ p
i
= 0. Then,
R
=
˙
RR
T
=
3
¸
i=1
˙
t
i
⊗p
i
+
3
¸
i=1
t
i
⊗
U
0
˙ p
i
R
T
=
3
¸
i=1
˙
t
i
⊗p
i
R
T
=
3
¸
i=1
˙
t
i
⊗t
i
.
∗
In Section A.7 of the Appendix, several examples featuring the calculation of the axial vector ω
R
of
a given skewsymmetric tensor
R
are presented.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
170 Rotation Tensors
If we now consider
R
t
k
, we ﬁnd a familiar result:
˙
t
i
=
R
t
i
= ω
R
×t
i
.
It is left as an exercise to show the less familiar result
˙
t
i
= R(ω
0
R
×p
i
).
It is important to note that if t
i
are deﬁned by use of a properorthogonal tensor
R and a ﬁxed basis p
i
, then their time derivatives can be expressed in terms of the
angular velocity vector of the rotation tensor and the basis vectors t
i
.
Given any secondorder tensor A and any vector a, we have the following rep
resentations:
a =
3
¸
i=1
a
i
t
i
, A =
3
¸
i=1
3
¸
k=1
A
ik
t
i
⊗t
k
.
If we assume that a is a function of time, then
˙ a =
3
¸
i=1
˙ a
i
t
i
+a
i
˙
t
i
=
3
¸
i=1
˙ a
i
t
i
+a
i
(ω
R
×t
i
)
=
3
¸
i=1
˙ a
i
t
i
+ω
R
×a.
Similarly, if we assume that A is a function of time, then
˙
A =
3
¸
i=1
3
¸
k=1
˙
A
ik
t
i
⊗t
k
+
3
¸
i=1
3
¸
k=1
A
ik
˙
t
i
⊗t
k
+
3
¸
i=1
3
¸
k=1
A
ik
t
i
⊗
˙
t
k
=
3
¸
i=1
3
¸
k=1
˙
A
ik
t
i
⊗t
k
+
3
¸
i=1
3
¸
k=1
A
ik
(
R
t
i
) ⊗t
k
+
3
¸
i=1
3
¸
k=1
A
ik
t
i
⊗(
R
t
k
)
=
3
¸
i=1
3
¸
k=1
˙
A
ik
t
i
⊗t
k
+
R
A−A
R
.
The derivatives
o
A
and
o
a are known as the corotational derivatives (with respect to
R) of A and a, respectively. They are the respective derivatives of A and a if the
vectors t
i
are constant:
o
A
=
3
¸
i=1
3
¸
k=1
˙
A
ik
t
i
⊗t
k
,
o
a=
3
¸
i=1
˙ a
i
t
i
.
Using the recently established expression for
o
a, we observe that
˙ a =
o
a +ω
R
×a.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
6.5 Euler’s Representation of a Rotation Tensor 171
Similarly, for any secondorder tensor A, we have
˙
A =
o
A
+
R
A−A
R
.
The terms involving the angular velocity vectors and tensors in these expressions
are the result of the orthonormal vectors t
i
changing with time.
We shall subsequently use several distinct corotational derivatives, and we wish
to do so without introducing a laborious notation. To this end, it will be explicitly
stated which rotation tensor the corotational derivative pertains to in situations in
which confusion is possible.
6.5 Euler’s Representation of a Rotation Tensor
Leonhard Euler deﬁned a rotation by using an angle of rotation φ and an axis of
rotation r.
∗
Using notation introduced over a century later by Gibbs,
†
Euler’s rep
resentation for a tensor that produces this rotation can be written as
R = L(φ, r) = cos(φ)(I −r ⊗r) −sin(φ)(εr) +r ⊗r, (6.8)
where r is a unit vector and φ is a counterclockwise angle of rotation. We refer
to (6.8) as Euler’s representation of a rotation tensor and use the function L to
prescribe the rotation tensor associated with an axis and angle of rotation. The three
independent parameters of R are the angle of rotation and the two independent
components of the unit vector r. The reason r is known as the axis of rotation lies in
the fact that it is invariant under the action of R: Rr = r. We shall shortly examine
the role of φ.
It is interesting to examine some of the features of representation (6.8). To this
end, we deﬁne an orthonormal basis {p
1
, p
2
, p
3
} with p
3
= r. Using this basis, we
have
I = p
1
⊗p
1
+p
2
⊗p
2
+p
3
⊗p
3
,
r ⊗r = p
3
⊗p
3
,
I −r ⊗r = p
1
⊗p
1
+p
2
⊗p
2
,
−(εr) = p
2
⊗p
1
−p
1
⊗p
2
.
Consequently we can write
R = L(φ, r = p
3
) = cos(φ)(p
1
⊗p
1
+p
2
⊗p
2
)
+sin(φ)(p
2
⊗p
1
−p
1
⊗p
2
) +p
3
⊗p
3
.
∗
This representation can be seen in Section 49 in one of Euler’s great papers on rigid body dynamics
from 1775 [56]. There, he provides expressions for the components of the tensor R in terms of an
angle of rotation φ and the direction cosines p, q, r of the axis of rotation. In our notation, r =
pE
1
+qE
2
+rE
3
.
†
See Section 129 of [231].
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
172 Rotation Tensors
Using the identity (a ⊗b)
T
= b ⊗a, we ﬁnd
R
T
= L(φ, r = p
3
) = cos(φ)(p
1
⊗p
1
+p
2
⊗p
2
)
+sin(φ)(p
1
⊗p
2
−p
2
⊗p
1
) +p
3
⊗p
3
.
It is now easy to check that RR
T
is the identity tensor, and the details are left to the
reader. Next, we examine the determinant of R:
det(R) = det
⎡
⎢
⎣
cos(φ) −sin(φ) 0
sin(φ) cos(φ) 0
0 0 1
⎤
⎥
⎦
= 1.
In conclusion, R
T
R = I and det(R) = 1. Thus R is a properorthogonal tensor. We
shall examine the converse of this statement shortly.
Composition of Rotations
Suppose we have a rotation through an angle φ
1
about an axis r
1
and we follow this
by a rotation through an angle φ
2
about an axis r
2
; then the tensor
R
c
= L(φ
2
, r
2
) L(φ
1
, r
1
)
represents the composite rotation. It is left as an exercise to show that R
c
is a proper
orthogonal tensor. As can be shown later by Euler’s theorem, this implies that R
c
is
a rotation tensor. Thus the composition of two rotations is also a rotation.
As tensor multiplication is noncommutative, the order in which we consider the
composition is important. In general,
L(φ
2
, r
2
) L(φ
1
, r
1
) = L(φ
1
, r
1
) L(φ
2
, r
2
) .
The exceptions arise when r
1
r
2
or one of the angles of rotation is 0. The fact that
rotations are sequence dependent will play a major role later on in the deﬁnition of
Euler angle sequences.
Euler’s Formula
We now examine the action of R on a vector a. As shown in Figure 6.2, the part of
a that is parallel to r is unaltered by the transformation, whereas the part of a that
is perpendicular to r is rotated through an angle φ counterclockwise about r. To see
this, it is convenient to decompose a:
a = a
⊥
+a
,
where
a
⊥
= a −(a · r)r = (I −r ⊗r)a, a
= (a · r)r = (r ⊗r)a.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
6.5 Euler’s Representation of a Rotation Tensor 173
Ra
r
a
φ
a
a
⊥
Ra
⊥
Figure 6.2. The transformation of a vector a by the rotation tensor
R = L(φ, r).
With the assistance of this decomposition, we now compute that
Ra = L(φ, r)a
= (cos(φ)(I −r ⊗r) −sin(φ)(εr) +r ⊗r) a
= cos(φ)(I −r ⊗r)a −sin(φ)(εr)a +(r ⊗r)a
= cos(φ)(I −r ⊗r)a +sin(φ)r ×a +(r · a)r,
= cos(φ)a
⊥
+sin(φ)r ×a
⊥
+a
. (6.9)
Noting that cos(φ)a
⊥
+sin(φ)r ×a
⊥
is a rotation of a
⊥
about r, we obtain the desired
conclusion. The ﬁnal expression for Ra in (6.9) is known as Euler’s formula.
Remarks on Euler’s Representation
Euler’s representation (6.8) is unusual in several respects. First, you should notice
that
L(φ, r) = L(−φ, −r).
This implies that there are two different representations for the same rotation ten
sor. Second, as R is a rotation tensor, R
−1
= R
T
, and this leads to two easy repre
sentations for these tensors:
R
−1
= R
T
= L(−φ, r) = L(φ, −r).
In words, the inverse can be calculated by either reversing the angle of rotation or
inverting the axis of rotation. Another peculiarity is that L(φ = 0, r) = I holds for
all vectors r. Finally, we note that Euler’s representation is used to deﬁne other
representations of rotation tensors in this book.
6.5.1 Calculating the Axis and Angle of Rotation
Given a rotation tensor R, it is a standard exercise to calculate the axis of rotation
r and the angle of rotation θ associated with this tensor. First, one is normally
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
174 Rotation Tensors
presented with the matrix components of R with respect to a basis, say {p
1
, p
2
, p
3
}:
R =
3
¸
i=1
3
¸
k=1
R
ik
p
i
⊗p
k
.
If we compare this representation with (6.8), we ﬁnd that
R
ik
= ((cos(θ)(I −r ⊗r) −sin(θ)(εr) +r ⊗r) p
k
) · p
i
= cos(θ) (δ
ik
−r
i
r
k
) +r
i
r
k
−
3
¸
j=1
sin(θ)
jik
r
j
,
where
r =
3
¸
i=1
r
i
p
i
.
Expanding the expressions for the components of R, we ﬁnd the matrix represen
tation
⎡
⎢
⎢
⎣
R
11
R
12
R
13
R
21
R
22
R
23
R
31
R
32
R
33
⎤
⎥
⎥
⎦
= cos (θ)
⎡
⎢
⎢
⎣
1 −r
2
1
−r
1
r
2
−r
1
r
3
−r
1
r
2
1 −r
2
2
−r
2
r
3
−r
1
r
3
−r
2
r
3
1 −r
2
3
⎤
⎥
⎥
⎦
+
⎡
⎢
⎢
⎣
r
2
1
r
1
r
2
r
1
r
3
r
1
r
2
r
2
2
r
2
r
3
r
1
r
3
r
2
r
3
r
2
3
⎤
⎥
⎥
⎦
+sin(θ)
⎡
⎢
⎣
0 −r
3
r
2
r
3
0 −r
1
−r
2
r
1
0
⎤
⎥
⎦
. (6.10)
Notice that this matrix can be expressed as the sum of a symmetric and skew
symmetric matrix. The skewsymmetric part is the only part that changes when we
transpose the tensor.
To determine the angle of rotation, we calculate the trace of R:
cos(θ) =
1
2
(tr(R) −1) =
1
2
(R
11
+R
22
+R
33
−1) . (6.11)
Looking at the skewsymmetric part of R, we ﬁnd that
⎡
⎢
⎣
r
1
r
2
r
3
⎤
⎥
⎦
= −
1
2 sin(θ)
⎡
⎢
⎣
R
23
−R
32
R
31
−R
13
R
12
−R
21
⎤
⎥
⎦
. (6.12)
To verify the calculated value of r, we could examine the eigenvectors of R: The
eigenvector corresponding to the unit eigenvalue should be parallel to the axis of
rotation r.
It is interesting to notice that, if R =
¸
3
i=1
t
i
⊗p
i
, then r has the same compo
nents with respect to both sets of basis vectors:
r =
3
¸
i=1
r
i
p
i
=
3
¸
i=1
r
i
t
i
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
6.5 Euler’s Representation of a Rotation Tensor 175
The proof of this result is based on the observation that R has the same components
with respect to the bases p
i
⊗p
k
and t
i
⊗t
k
. That is, R =
¸
3
i=1
¸
3
k=1
R
ik
p
i
⊗p
k
=
¸
3
i=1
¸
3
k=1
R
ik
t
i
⊗t
k
.
AN EXAMPLE. As an example, suppose that the components of a rotation tensor R
are
⎡
⎢
⎢
⎣
R
11
R
12
R
13
R
21
R
22
R
23
R
31
R
32
R
33
⎤
⎥
⎥
⎦
=
⎡
⎢
⎣
0.835959 −0.283542 −0.469869
0.271321 0.957764 −0.0952472
0.47703 −0.0478627 0.877583
⎤
⎥
⎦
.
We can compute the angle of rotation θ of this tensor by using (6.11):
cos(θ) =
1
2
(0.835959 +0.877583 +0.957764 −1) .
That is, θ = 33.3161
◦
. We can calculate the axis of rotation r by using (6.12):
r = 0.043135p
1
−0.861981p
2
+0.505103p
3
= 0.043135t
1
−0.861981t
2
+0.505103t
3
.
In writing this result, we are emphasizing that the components of r in the basis
¸
p
1
, p
2
, p
3
¸
and {t
1
, t
2
, t
3
} are identical.
The Associated Angular Velocity Vector
Given Euler’s representation (6.8) we assume that R = R(t). This implies, in gen
eral, that φ = φ(t) and r = r(t). We now seek to establish representations for ω
R
.
As a preliminary result, we note that, because r is a unit vector, r · ˙ r = 0. In
addition, ˙ = 0 and
˙
I = 0. Now, starting from (6.8),
R = L(φ, r) = cos(φ)(I −r ⊗r) −sin(φ)(εr) +r ⊗r,
we differentiate to ﬁnd
˙
R = −
˙
φsin(φ)(I −r ⊗r) −
˙
φcos(φ)(εr) +(1 −cos(φ))(˙ r ⊗r +r ⊗ ˙ r) −sin(φ)(ε˙ r).
To proceed further, we deﬁne a righthanded orthonormal basis {t
1
, t
2
, t
3
}, such that
t
3
= r at a given instant in time. Then, at the same instant in time,
r = (t
1
⊗t
2
−t
2
⊗t
1
),
˙ r = at
1
+bt
2
,
r × ˙ r = at
2
−bt
1
,
˙ r = a(t
2
⊗t
3
−t
3
⊗t
2
) +b(t
3
⊗t
1
−t
1
⊗t
3
). (6.13)
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
176 Rotation Tensors
In these expressions, a and b are the scalar components of ˙ r. Using (6.13) along with
some manipulations, we ﬁnd that
˙
RR
T
= −
˙
φ(t
1
⊗t
2
−t
2
⊗t
1
) +(a(1 −cos(φ))
+bsin(φ))(t
1
⊗t
3
−t
3
⊗t
1
) +(−b(1 −cos(φ))
+a sin(φ))(t
3
⊗t
2
−t
2
⊗t
3
)
= −
˙
φt
3
−(a(1 −cos(φ)) +bsin(φ))t
2
−(−b(1 −cos(φ))
+a sin(φ))t
1
.
With the assistance of (6.13), we can now write the desired ﬁnal result:
R
=
˙
RR
T
= −
˙
φr −(1 −cos(φ))(r × ˙ r) −sin(φ)˙ r.
The associated angular velocity vector is
ω
R
=
˙
φr +sin(φ)˙ r +(1 −cos(φ))r × ˙ r. (6.14)
If r is constant, then the expression for the angular velocity vector simpliﬁes con
siderably. It is interesting to note that a constant ω
R
does not necessarily imply a
constant r. However, as shown in [161], it is usually possible to choose a ﬁxed ba
sis
¸
p
1
, p
2
, p
3
¸
where R =
¸
3
k=1
t
k
⊗p
k
so that a constant angular velocity implies a
constant
˙
φ and a constant r.
6.6 Euler’s Theorem: Rotation Tensors and ProperOrthogonal Tensors
A tensor Q is properorthogonal if, and only if,
Q
T
Q = I, det(Q) = 1.
From our discussion of rotation tensors, it follows that a rotation tensor is a proper
orthogonal tensor. However, is the converse true? In other words, is every proper
orthogonal tensor a rotation tensor? The afﬁrmative answer to this question is
known as Euler’s theorem.
Our approach to presenting the proof of Euler’s theorem is adapted from a
wonderful paper by Zhongheng Guo [83]. One of the key results needed is to re
call that we can deﬁne two of the invariants, I
A
= tr(A) and III
A
= det(A), of any
secondorder tensor A by using the scalar triple product:
[Aa, b, c] +[a, Ab, c] +[a, b, Ac] = I
A
[a, b, c],
[Aa, Ab, Ac] = III
A
[a, b, c],
where a, b, and c are any three vectors (see Section A.6).
For a properorthogonal tensor, the fact that det(Q) = 1 implies that this tensor
has an eigenvalue λ = 1. There are several ways to see this, but ﬁrst consider
Q
T
(Q−I) = I −Q
T
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
6.6 Euler’s Theorem: Rotation Tensors and ProperOrthogonal Tensors 177
Taking the determinant of both sides and using the fact that det(Q
T
) = 1, one ﬁnds
that
det (Q−I) = (−1)
3
det (Q−I) .
It follows that
det (Q−I) = 0, (6.15)
and thus Q has an eigenvalue of 1.
As Q has a unit eigenvalue, it has an eigenvector u such that
Qu = u.
When this equation is multiplied by Q
T
, it follows that
u = Qu = Q
T
u.
Thus u is a unit eigenvector of both Q and Q
T
.
Now consider a vector v ⊥ u. Some manipulations show that
Qv · Qu = Qv · u,
Qv · Qu = Q
T
Qv · u = v · u = 0.
Consequently,
v ⊥ u if, and only if, Qv ⊥ u.
We henceforth assume that u and v have unit magnitudes and deﬁne a vector w
such that [u, v, w] = 1. Let us now calculate I
Q
:
I
Q
= [Qu, v, w] +[u, Qv, w] +[u, v, Qw]
= [u, v, w] +[u, Qv, w] +[u, v, Qw]
= 1 +v · Qv +w · Qw.
= 1 +v · Qv +(u ×v) · (Qu ×Qv) .
= 1 +v · Qv +(u · u) (v · Qv) −(u · Qv) (v · u) .
= 1 +2v · Qv.
We deﬁne an angle ν such that
cos(ν) = v · Qv.
It is important to notice that this angle is an invariant of Q and that
Qv = cos(ν)v +sin(ν)w
= cos(ν)v +sin(ν) (u ×v) . (6.16)
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
178 Rotation Tensors
As mentioned previously, any vector a can be decomposed into two compo
nents: a
and a
⊥
, where a
⊥
· u = 0 and a = a
+a
⊥
. Now,
Qa
= a
,
and, from (6.16),
Qa
⊥
= cos(ν)a
⊥
+sin(ν) (u ×a
⊥
) .
It follows that
Qa = (a · u) u +cos(ν)(I −u ⊗u)a +sin(ν)u ×a.
Comparing this equation with (6.9), we conclude that Q has the form (6.8) of a
rotation tensor. Further, as the preceding equation holds for any properorthogonal
Q, we conclude that every properorthogonal tensor is a rotation tensor. Thus we
use these terms interchangeably.
The result that every properorthogonal tensor is a rotation tensor is credited
to Euler and dates to 1775 [55]:
Every properorthogonal tensor is a rotation tensor.
We shall shortly revisit this result and see why it is also known as Euler’s theorem
on the motion of a rigid body. To verify that a tensor A is a rotation tensor, we
invoke Euler’s theorem and simply show that A is properorthogonal: A
T
A = I and
det (A) = 1.
6.7 Relative Angular Velocity Vectors
Consider two rotation tensors: R
1
= R
1
(t) and R
2
= R
2
(t). It is straightforward to
show that the product R = R
2
R
1
of these two tensors is also a rotation tensor. We
now wish to calculate its angular velocity tensor and vector. To do this, we use the
work of Casey and Lam [29], who deﬁned a very useful and intuitive relative angular
velocity vector.
To discuss the relative angular velocity vector, it is convenient to deﬁne three
sets of righthanded orthonormal vectors: {
1
t
1
,
1
t
2
,
1
t
3
}, {
2
t
1
,
2
t
2
,
2
t
3
}, and {p
1
, p
2
, p
3
}.
In this section, we assume that ˙ p
i
= 0. For a given R
1
and R
2
, two of these sets can
be deﬁned by use of the representations
R
1
=
3
¸
i=1
1
t
i
⊗p
i
, R
2
=
3
¸
i=1
2
t
i
⊗
1
t
i
.
Notice that
R =
3
¸
i=1
2
t
i
⊗p
i
.
In words, R transforms the vector p
i
into the vector
2
t
i
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
6.7 Relative Angular Velocity Vectors 179
Following Casey and Lam [29], let us consider the following relative angular
velocity tensor:
ˆ
R
2
=
R
−
R
1
.
Using the deﬁnition of the angular velocity tensors and the fact that
˙
R =
˙
R
2
R
1
+
R
2
˙
R
1
, we ﬁnd that
ˆ
R
2
=
R
−
R
1
=
˙
RR
T
−
˙
R
1
R
T
1
=
˙
R
2
R
1
R
T
1
R
T
2
+R
2
˙
R
1
R
T
1
R
T
2
−
˙
R
1
R
T
1
=
˙
R
2
R
T
2
+R
2
R
1
R
T
2
−
R
1
=
˙
R
2
R
T
2
+R
2
R
1
R
T
2
+
T
R
1
=
˙
R
2
+R
2
R
1
+
T
R
1
R
2
R
T
2
.
However,
o
R2
=
˙
R
2
+R
2
R
1
+
T
R
1
R
2
,
where the corotational derivative
o
R2
is deﬁned to be
o
R2
=
3
¸
i=1
3
¸
k=1
˙
R
2ik1
t
i
⊗
1
t
k
,
where R
2ik
= ((R
2
)
1
t
k
) ·
1
t
i
. In words,
o
R2
is the derivative of the tensor R
2
assuming
that
1
t
i
are constant.
In conclusion, the relative angular velocity tensor
ˆ
R
2
is
ˆ
R
2
=
R
−
R
1
=
o
R2
R
T
2
.
If we denote the axis of rotation of R
2
by r
2
and its angle of rotation by φ
2
, then we
can parallel the derivation of (6.14) to ﬁnd that the relative angular velocity vector
has the representation
ˆ ω
R
2
= ω
R
−ω
R
1
=
˙
φ
2
r
2
+sin(φ
2
)
o
r
2
+(1 −cos (φ
2
)) r
2
×
o
r
2
. (6.17)
In this equation,
ˆ ω
R
2
= −
1
2
¸
o
R2
R
T
2
¸
, (6.18)
and the corotational derivative of r
2
=
¸
3
i=1
r
i
(
1
t
i
) is
o
r
2
= ˙ r
1
(
1
t
1
) + ˙ r
2
(
1
t
2
) + ˙ r
3
(
1
t
3
) .
Formula (6.17) will prove to be exceedingly useful when calculating the angular ve
locity vector associated with various representations of a rotation tensor. In partic
ular, for the Euler angle representation, we decompose R into the product of three
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
180 Rotation Tensors
rotation tensors, and we shall invoke (6.17) twice to get a representation for the an
gular velocity vector corresponding to R. The manner in which we do this is similar
to the example now presented.
6.7.1 An Example
To illustrate the convenience of relative angular velocity result (6.17), let us consider
an example. Suppose we have two rotations R
1
and R, where
R
1
= cos(ψ)(I −E
3
⊗E
3
) −sin(ψ)E
3
+E
3
⊗E
3
,
R = R
2
R
1
.
Here, the relative rotation tensor R
2
is chosen to correspond to a rotation through
an angle θ about t
2
= cos(ψ)E
2
−sin(ψ)E
1
:
R
2
= cos(θ)(I −t
2
⊗t
2
) −sin(θ)t
2
+t
2
⊗t
2
.
The tensor R
1
deﬁnes a transformation consisting of a rotation through an angle ψ
about E
3
. This rotation transforms E
i
to t
i
, where
t
1
= cos(ψ)E
1
+sin(ψ)E
1
, t
2
= −sin(ψ)E
1
+cos(ψ)E
2
, t
3
= E
3
.
In addition, R consists of the rotation R
1
followed by a rotation through an angle θ
about t
2
.
∗
To calculate ω
R
1
we can appeal to (6.14) to ﬁnd that
ω
R
1
=
˙
ψE
3
+sin(ψ)
˙
E
3
+(1 −cos(ψ))E
3
×
˙
E
3
=
˙
ψE
3
.
To calculate ω
R
we cannot appeal directly to (6.14) because we do not know the axis
and angle of rotation of R.
†
Instead, we use the relative angular velocity vector. To
do this, we ﬁrst need to write R
2
with respect to an appropriate basis. As
R
1
= t
1
⊗E
1
+t
2
⊗E
2
+t
3
⊗E
3
,
the appropriate basis is t
i
⊗t
k
:
R
2
= cos(θ)(t
3
⊗t
3
+t
1
⊗t
1
) −sin(θ) (t
3
⊗t
1
−t
1
⊗t
3
) +t
2
⊗t
2
.
Calculating the corotational rate of this tensor, we take its derivative, keeping t
i
ﬁxed:
o
R2
= −
˙
θ sin(θ)(t
3
⊗t
3
+t
1
⊗t
1
) −
˙
θ cos(θ) (t
3
⊗t
1
−t
1
⊗t
3
) .
Hence,
ˆ
R
2
=
˙
θ (t
1
⊗t
3
−t
3
⊗t
1
) ,
∗
In the notation of the previous section, p
i
is replaced with E
i
and
1
t
i
is replaced with t
i
.
†
The interested reader might wish to use the famed Rodrigues formula [(6.49)] that is discussed in
the exercises to compute these quantities and then one could use (6.14).
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
6.8 Euler Angles 181
and, with the help of (6.18), we conclude that
ˆ ω
R
2
=
˙
θ t
2
. (6.19)
As an alternative method of calculating (6.19), we can use (6.17) directly.
Thus we replace r
2
and φ
2
in (6.17) with t
2
and θ, respectively. We then appeal
to the fact that
o
t
2
= 0. In summary,
ˆ ω
R
2
=
˙
θ t
2
+sin(θ)
o
t
2
+(1 −cos(θ))t
2
×
o
t
2
=
˙
θ t
2
.
This alternative method is clearly equivalent to, but more attractive than, the
method that involved calculating
o
R2
.
Combining the expressions for ω
R
1
and ˆ ω
R
2
, we arrive at an expression for the
angular velocity vector for R:
ω
R
=
˙
θ t
2
+
˙
ψE
3
.
The intuitive nature of this result is often surprising. It is very easy to use (6.14) to
see that ω
R
2
= ˆ ω
R
2
.
6.8 Euler Angles
The most popular representation of a rotation tensor is based on the use of three
Euler angles. As discussed in [38, 230], this representation dates to works by Euler
[54, 57] that he ﬁrst presented in 1751.
∗
In these papers, he shows how three angles
can be used to parameterize a rotation, and he also establishes expressions for the
corotational components of the angular velocity vector.
One interpretation of the Euler angles involves a decomposition of the rotation
tensor into a product of three fairly simple rotations:
R =
`
R(γ
1
, γ
2
, γ
3
) = L(γ
3
, g
3
)L(γ
2
, g
2
)L(γ
1
, g
1
). (6.20)
Here, {γ
i
} are the Euler angles and the set of unit vectors {g
i
} is known as the Euler
basis. The function L(θ, b) is deﬁned by use of the Euler representation:
L(θ, b) = cos(θ)(I −b ⊗b) −sin(θ)(εb) +b ⊗b,
∗
For discussions of these papers, and several other interesting historical facts on the development
of representations for rotations, see Blanc’s introduction to parts of Euler’s collected works in
[59, 60], Cheng and Gupta [38], and Wilson [230]. Although [57] dates to the 18th century, it was
ﬁrst published posthumously in 1862.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
182 Rotation Tensors
ψ
ψ
θ
θ
φ
φ
φ
p
1
p
1
p
1
p
2
p
3
= g
1
t
1
t
2
t
3
= g
3
g
2
Figure 6.3. Schematic of the 3–2–3 set of Euler angles and the individual rotations these angles
represent. In this ﬁgure, the three Euler angles are denoted by ψ = γ
1
, θ = γ
2
, and φ = γ
3
,
respectively, and the rotation tensor R that they parameterize transforms p
i
to t
i
. The image
on the lefthand side is a portrait of Leonhard Euler.
where b is a unit vector and θ is the counterclockwise angle of rotation. In general,
g
3
is a function of γ
2
and γ
1
and g
2
is a function of γ
1
. As we shall shortly see,
there are 12 possible choices of the Euler angles. For example, Figure 6.3 illustrates
these angles for a set of 3–2–3 Euler angles. Because there are three Euler angles,
the parameterization of a rotation tensor by use of these angles is an example of a
threeparameter representation.
If we assume that g
1
is constant, then the angular velocity vector associated
with the Euler angle representation can be established by use of the relative
angular velocity vector. In this case, there are two relative angular velocity vectors
[cf. (6.17)]. For the ﬁrst rotation, the angular velocity vector is ˙ γ
1
g
1
[cf. (6.14)].
The angular velocity of the second rotation relative to the ﬁrst rotation is ˙ γ
2
g
2
,
and the angular velocity of the third rotation relative to the second rotation is
˙ γ
3
g
3
.
∗
Combining the two relative angular velocity vectors with ˙ γ
1
g
1
, we conclude
that
ω
R
= ˙ γ
3
g
3
+ ˙ γ
2
g
2
+ ˙ γ
1
g
1
. (6.21)
If the rotation tensor R transforms the vectors p
i
into the set t
i
, then it is possible to
express the Euler basis in terms of either set of vectors.
∗
The calculation of these angular velocity vectors is similar to the example discussed in Subsection
6.7.1.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
6.8 Euler Angles 183
φ
φ
ψ
ψ
θ
θ
p
1
p
2 t
1
t
1
t
2
t
3
t
1
t
2
t
3
t
3
t
2
t
3
Figure 6.4. The transformations of various basis vectors induced by the individual angles in a
set of 3–2–1 Euler angles.
Alternative approaches to establishing (6.21) can be found in textbooks. In
one such approach, all three Euler angles are considered to be inﬁnitesimal. A
good example of this approach can be found in Section 2.9 of Lurie [131]. Another
approach, which can be found in [109, 185] and dates to Euler, features spherical
geometry. Finally, a third (lengthy) approach involves directly differentiating (6.20)
and then computing
R
and its axial vector.
For the Euler angles to effectively parameterize all rotations, we need to
assume that we can ﬁnd γ
k
and ˙ γ
k
such that, for any given ω
R
, (6.21) holds. For
this to happen, it is necessary and sufﬁcient that the vectors g
1
, g
2
, and g
3
span E
3
.
When these vectors are not linearly independent, we say that the Euler angles have
a singularity. This singularity is unavoidable for Euler angles and is often known
as a “gimbal lock” (cf. [196]). We shall ﬁnd that this singularity occurs for certain
values of γ
2
, and we shall restrict this angle to avoid these singularities.
For future purposes, it is also convenient to deﬁne the dual Euler basis {g
j
}:
g
j
· g
i
= δ
j
i
, where δ
j
i
is the Kronecker delta.
∗
Notice that
ω
R
· g
i
= ˙ γ
i
. (6.22)
To determine this basis, one expresses g
i
in terms of a righthanded basis, say {t
i
}.
Then, to determine g
2
, say, we write g
2
= at
1
+bt
2
+ct
3
and solve the three equa
tions, g
1
· g
2
= 0, g
2
· g
2
= 1, and g
3
· g
2
= 0 for the three unknowns a, b, and c. When
the Euler angles have singularities, one will ﬁnd that the two dual Euler basis vec
tors g
1
and g
3
cannot be deﬁned. Another point of interest is the observation that
g
2
= g
2
for all possible sets of Euler angles.
We now turn to examining two different choices of the Euler angles: the 3–2–1
set and the 3–1–3 set. Both of these sets are popular in different communities. For
instance, the aircraft and vehicle dynamics community favors the 3–2–1 Euler angles
to parameterize yaw–pitch–roll behavior, whereas problems involving spinning
rigid bodies in dynamics often lend themselves to the 3–1–3 set. We shall subse
quently discuss examples of both sets. One of the exercises at the end of this chapter
∗
The dual Euler basis vectors are analogous to the contravariant basis vectors and have similar uses
in dynamics. We shall use the dual Euler basis later in various contexts, and a rapid summary of
these uses can be found in O’Reilly [160].
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
184 Rotation Tensors
involves a comprehensive investigation into the set of 3–2–3 Euler angles. It is highly
recommended that you complete this exercise after reading this section of the book.
6.8.1 321 Euler Angles
To elaborate further on the Euler angles, we now consider the 3–2–1 set of Euler
angles (see Figure 6.4). These are arguably among the most popular sets of Euler
angles.
∗
In several communities, they are known as examples of the Tait and/or
Bryan angles [after Peter G. Tait (1831–1901) and George H. Bryan (1864–1928)]
or the Euler–Cardan angles [after Euler and Girolamo Cardano (1501–1576)].
First, suppose that the rotation tensor has the representation
R =
3
¸
i=1
t
i
⊗p
i
,
where {p
i
} is a ﬁxed Cartesian basis. The ﬁrst rotation is about p
3
through an angle
ψ. This rotation transforms p
i
to t
i
. The second rotation is about the t
2
axis through
an angle θ. This rotation transforms t
i
to t
i
. The third and last rotation is through an
angle φ about the axis t
1
= t
1
. Thus
R =
`
R(γ
1
= ψ, γ
2
= θ, γ
3
= φ) = L(φ, t
1
)L(θ, t
2
) L(ψ, p
3
).
Here,
t
i
= L(φ, t
1
)t
i
, t
i
= L(θ, t
2
= t
2
)t
i
, t
i
= L(ψ, t
3
= p
3
)p
i
.
It is not difﬁcult to express the various basis vectors as linear combinations of each
other:
⎡
⎢
⎣
t
1
t
2
t
3
⎤
⎥
⎦
=
⎡
⎢
⎣
cos(ψ) sin(ψ) 0
−sin(ψ) cos(ψ) 0
0 0 1
⎤
⎥
⎦
⎡
⎢
⎣
p
1
p
2
p
3
⎤
⎥
⎦
,
⎡
⎢
⎣
t
1
t
2
t
3
⎤
⎥
⎦
=
⎡
⎢
⎣
cos(θ) 0 −sin(θ)
0 1 0
sin(θ) 0 cos(θ)
⎤
⎥
⎦
⎡
⎢
⎣
t
1
t
2
t
3
⎤
⎥
⎦
,
⎡
⎢
⎣
t
1
t
2
t
3
⎤
⎥
⎦
=
⎡
⎢
⎣
1 0 0
0 cos(φ) sin(φ)
0 −sin(φ) cos(φ)
⎤
⎥
⎦
⎡
⎢
⎣
t
1
t
2
t
3
⎤
⎥
⎦
. (6.23)
The inverses of these relationships are easy to obtain once you realize that each of
the three matrices in (6.23) is orthogonal. As the inverse of an orthogonal matrix is
∗
For example, they are used in Greenwood’s text [79], Rao’s text [176], and numerous texts on vehi
cle and aircraft dynamics.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
6.8 Euler Angles 185
its transpose, we quickly arrive at the soughtafter results:
⎡
⎢
⎣
p
1
p
2
p
3
⎤
⎥
⎦
=
⎡
⎢
⎣
cos(ψ) −sin(ψ) 0
sin(ψ) cos(ψ) 0
0 0 1
⎤
⎥
⎦
⎡
⎢
⎣
t
1
t
2
t
3
⎤
⎥
⎦
,
⎡
⎢
⎣
t
1
t
2
t
3
⎤
⎥
⎦
=
⎡
⎢
⎣
cos(θ) 0 sin(θ)
0 1 0
−sin(θ) 0 cos(θ)
⎤
⎥
⎦
⎡
⎢
⎣
t
1
t
2
t
3
⎤
⎥
⎦
,
⎡
⎢
⎣
t
1
t
2
t
3
⎤
⎥
⎦
=
⎡
⎢
⎣
1 0 0
0 cos(φ) −sin(φ)
0 sin(φ) cos(φ)
⎤
⎥
⎦
⎡
⎢
⎣
t
1
t
2
t
3
⎤
⎥
⎦
. (6.24)
Relationships (6.23) and (6.24) can be combined to express t
i
in terms of p
k
and vice
versa. Later, they will also be used to obtain representations for the Euler basis in
terms of p
k
and t
i
.
By using (6.24) we can ﬁnd a representation for the components R
i j
= (Rp
j
) · p
i
.
The components are easily expressed by a matrix representation:
⎡
⎢
⎢
⎣
R
11
R
12
R
13
R
21
R
22
R
23
R
31
R
32
R
33
⎤
⎥
⎥
⎦
=
⎡
⎢
⎣
cos(ψ) −sin(ψ) 0
sin(ψ) cos(ψ) 0
0 0 1
⎤
⎥
⎦
⎡
⎢
⎣
cos(θ) 0 sin(θ)
0 1 0
−sin(θ) 0 cos(θ)
⎤
⎥
⎦
×
⎡
⎢
⎣
1 0 0
0 cos(φ) −sin(φ)
0 sin(φ) cos(φ)
⎤
⎥
⎦
. (6.25)
Representation (6.25) is the transpose of what one might naively expect. However,
recalling that R
ik
= p
i
· t
k
will hopefully resolve this initial surprise. Indeed, it is use
ful to note that
⎡
⎢
⎣
p
1
p
2
p
3
⎤
⎥
⎦
=
⎡
⎢
⎢
⎣
R
11
R
12
R
13
R
21
R
22
R
23
R
31
R
32
R
33
⎤
⎥
⎥
⎦
⎡
⎢
⎣
t
1
t
2
t
3
⎤
⎥
⎦
,
⎡
⎢
⎣
t
1
t
2
t
3
⎤
⎥
⎦
=
⎡
⎢
⎢
⎣
R
11
R
21
R
31
R
12
R
22
R
32
R
13
R
23
R
33
⎤
⎥
⎥
⎦
⎡
⎢
⎣
p
1
p
2
p
3
⎤
⎥
⎦
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
186 Rotation Tensors
THE EULER BASIS. As mentioned earlier, by examining the individual rotations
[cf. (6.23)], we can show that the Euler basis vectors have the representations
⎡
⎢
⎣
g
1
g
2
g
3
⎤
⎥
⎦
=
⎡
⎢
⎣
p
3
t
2
t
1
⎤
⎥
⎦
=
⎡
⎢
⎣
−sin(θ) sin(φ) cos(θ) cos(φ) cos(θ)
0 cos(φ) −sin(φ)
1 0 0
⎤
⎥
⎦
⎡
⎢
⎣
t
1
t
2
t
3
⎤
⎥
⎦
. (6.26)
Alternatively, we can also express the Euler basis in terms of the basis vectors
¸
p
1
, p
2
, p
3
¸
:
⎡
⎢
⎣
g
1
g
2
g
3
⎤
⎥
⎦
=
⎡
⎢
⎣
p
3
t
2
t
1
⎤
⎥
⎦
=
⎡
⎢
⎣
0 0 1
−sin(ψ) cos(ψ) 0
cos(θ) cos(ψ) cos(θ) sin(ψ) −sin(θ)
⎤
⎥
⎦
⎡
⎢
⎣
p
1
p
2
p
3
⎤
⎥
⎦
. (6.27)
This representation is useful for establishing the components ω
R
· p
k
.
THE DUAL EULER BASIS. We can now determine the dual Euler basis vectors g
k
. We
recall the remarks following (6.22) and express each of the dual Euler basis vectors
in terms of their components relative to the basis {t
1
, t
2
, t
3
}. That is,
g
k
= g
k1
t
1
+ g
k2
t
2
+ g
k3
t
3
.
Combining these results for the dual Euler basis vectors, we observe that
⎡
⎢
⎣
g
1
g
2
g
3
⎤
⎥
⎦
=
⎡
⎢
⎣
g
11
g
12
g
13
g
21
g
22
g
23
g
31
g
32
g
33
⎤
⎥
⎦
⎡
⎢
⎣
t
1
t
2
t
3
⎤
⎥
⎦
. (6.28)
With some manipulation, the relations g
i
· g
k
= δ
k
i
can be expressed as nine equa
tions for the nine unknowns g
ik
:
⎡
⎢
⎣
g
11
g
21
g
31
g
12
g
22
g
32
g
13
g
23
g
33
⎤
⎥
⎦
⎡
⎢
⎣
−sin(θ) sin(φ) cos(θ) cos(φ) cos(θ)
0 cos(φ) −sin(φ)
1 0 0
⎤
⎥
⎦
=
⎡
⎢
⎣
1 0 0
0 1 0
0 0 1
⎤
⎥
⎦
.
Isolating the matrix
¸
g
ik
¸
on the lefthand side of this equation, we ﬁnd that
⎡
⎢
⎣
g
11
g
12
g
13
g
21
g
22
g
23
g
31
g
32
g
33
⎤
⎥
⎦
=
⎡
⎢
⎣
0 sin(φ) sec(θ) cos(φ) sec(θ)
0 cos(φ) −sin(φ)
1 sin(φ) tan(θ) cos(φ) tan(θ)
⎤
⎥
⎦
.
That is, the dual Euler basis vectors have the representations
⎡
⎢
⎣
g
1
g
2
g
3
⎤
⎥
⎦
=
⎡
⎢
⎣
0 sin(φ) sec(θ) cos(φ) sec(θ)
0 cos(φ) −sin(φ)
1 sin(φ) tan(θ) cos(φ) tan(θ)
⎤
⎥
⎦
⎡
⎢
⎣
t
1
t
2
t
3
⎤
⎥
⎦
. (6.29)
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
6.8 Euler Angles 187
(a)
(b)
ψ
ψ
φ
θ
θ θ
p
1
p
1
p
2
p
2
g
1
= p
3
g
1
= p
3
t
1
t
1
t
3
t
3
g
1
g
2
= t
2
g
2
= t
2
g
3
= t
1
g
3
Figure 6.5. (a) The dual Euler basis vectors g
k
for the 3–2–1 set of Euler angles: g
1
t
3
, g
2
= t
2
,
and g
3
t
1
. (b) The Euler angles ψ and θ serve as coordinates for the Euler basis vector g
3
= t
1
in a manner that is similar to the role that spherical polar coordinates play in parameterizing
e
R
and e
φ
.
If we had used (6.27) in place of (6.26) to calculate the dual Euler basis vectors,
then we would have found the following representations:
⎡
⎢
⎣
g
1
g
2
g
3
⎤
⎥
⎦
=
⎡
⎢
⎣
cos(ψ) tan(θ) sin(ψ) tan(θ) 1
−sin(ψ) cos(ψ) 0
cos(ψ) sec(θ) sin(ψ) sec(θ) 0
⎤
⎥
⎦
⎡
⎢
⎣
p
1
p
2
p
3
⎤
⎥
⎦
. (6.30)
Expressions for the dual Euler basis vectors in terms of t
k
are easily inferred from
(6.30) and are shown in Figure 6.5(a).
For completeness, we note that, when θ = ±
π
2
, one can also express the Euler
basis vectors in terms of the dual Euler basis vectors:
⎡
⎢
⎣
g
1
g
2
g
3
⎤
⎥
⎦
=
⎡
⎢
⎣
1 0 −sin(θ)
0 1 0
−sin(θ) 0 1
⎤
⎥
⎦
⎡
⎢
⎣
g
1
g
2
g
3
⎤
⎥
⎦
. (6.31)
The simplicity of this relationship (related versions of which hold for the other 11
sets of Euler angles) is surprising.
SINGULARITIES. If we examine (6.26), we see that the Euler basis fails to be a ba
sis for E
3
when θ = ±
π
2
. One of the easiest ways to see this fact is to consider
ψ and θ +
π
2
to be spherical polar coordinates for t
1
[see Figure 6.5(b)]. When
θ = ±
π
2
, g
1
= p
3
= ±g
3
, and the Euler basis does not span E
3
. To avoid the afore
mentioned singularity, it is necessary to place restrictions on the second Euler angle:
θ ∈
−
π
2
,
π
2
. The other two angles are free to range from 0 to 2π.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
188 Rotation Tensors
ANGULAR VELOCITY VECTORS. The angular velocity vector ω
R
associated with the
3–2–1 Euler angles has several representations:
ω
R
= −
1
2
¸
˙
RR
T
¸
=
3
¸
i=1
˙ γ
i
g
i
=
˙
φt
1
+
˙
θ t
2
+
˙
ψp
3
= (−
˙
ψ sin(θ) +
˙
φ)t
1
+(
˙
ψ sin(φ) cos(θ) +
˙
θ cos(φ))t
2
+(
˙
ψ cos(φ) cos(θ) −
˙
θ sin(φ))t
3
.
To arrive at the primitive representation ω =
˙
φt
1
+
˙
θ t
2
+
˙
ψp
3
, we needed to
compute two relative angular velocity vectors. We calculated the ﬁrst of these,
˙
θ t
2
,
assuming that t
i
were ﬁxed. The explicit details of this calculation are easily inferred
from our earlier example in Subsection 6.7.1. We calculated the second relative
angular velocity vector,
˙
φt
1
, by using the relative rotation tensor L
φ, t
1
, assuming
that t
i
were ﬁxed.
It is also interesting to note that the angular velocity vector ω
0R
has the repre
sentations
ω
0R
= −
1
2
¸
R
T
˙
R
¸
= R
T
ω
R
=
˙
φR
T
t
1
+
˙
θ R
T
t
2
+
˙
ψR
T
p
3
= (−
˙
ψ sin(θ) +
˙
φ)p
1
+(
˙
ψ sin(φ) cos(θ) +
˙
θ cos(φ))p
2
+(
˙
ψ cos(φ) cos(θ) −
˙
θ sin(φ))p
3
.
In establishing this result, we used the fact that R
T
t
i
= p
i
.
6.8.2 3–1–3 Euler Angles
We can parallel the developments of the previous section for another popular set
of Euler angles: the 3–1–3 Euler angles (see Figure 6.6). These are the set of Eu
ler angles that Lagrange used,
∗
and they are also used in Arnol’d [9], Landau and
Lifshitz [125], and Thomson [214], among many others. For motions of a spinning
top, the Euler angles are identiﬁed with precession, nutation, and spin, respectively.
A closely related set of Euler angles, the 3–2–3 set, are discussed in Exercise 6.2 at
the end of this chapter.
Paralleling the developments for the 3–2–1 Euler angles:
R = L
φ, t
3
= t
3
L
θ, t
1
L(ψ, p
3
),
where {p
3
, t
1
, t
3
} is the Euler basis and
t
i
= L(ψ, p
3
) p
i
, t
i
= L
θ, t
1
t
i
, t
i
= L
φ, t
3
t
i
.
∗
See [118] and Section IX of the Second Part of [121]: His φ, ω, ψ correspond to our φ, θ, ψ, respec
tively.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
6.8 Euler Angles 189
φ
φ
ψ
ψ
θ
θ
p
1
p
2
t
1
t
2
t
2
t
3
t
1
t
2
t
2
t
3
t
1
t
2
Figure 6.6. The transformations of various basis vectors induced by the individual angles in a
set of 3–1–3 Euler angles.
Harking back to many of the celestial mechanics applications for this set of Euler
angles, the line passing through the origin that is parallel to t
1
is often known as the
line of nodes [214]. The angular velocity vector has the representations
ω
R
= −
1
2
¸
˙
RR
T
¸
=
3
¸
i=1
˙ γ
i
g
i
=
˙
φt
3
+
˙
θ t
1
+
˙
ψp
3
. (6.32)
We are using the same notation for the three Euler angles as we did for the 3–2–1
set. However, it should be clear that θ and φ represent different angles of rotation
for these two set of Euler angles.
EULER BASIS. It is not difﬁcult to show that the Euler basis {g
i
} has the representa
tions
⎡
⎢
⎣
g
1
g
2
g
3
⎤
⎥
⎦
=
⎡
⎢
⎣
p
3
t
1
t
3
⎤
⎥
⎦
=
⎡
⎢
⎣
sin(φ) sin(θ) cos(φ) sin(θ) cos(θ)
cos(φ) −sin(φ) 0
0 0 1
⎤
⎥
⎦
⎡
⎢
⎣
t
1
t
2
t
3
⎤
⎥
⎦
=
⎡
⎢
⎣
0 0 1
cos(ψ) sin(ψ) 0
sin(θ) sin(ψ) −sin(θ) cos(ψ) cos(θ)
⎤
⎥
⎦
⎡
⎢
⎣
p
1
p
2
p
3
⎤
⎥
⎦
. (6.33)
With these results and (6.32), two other representations for ω
R
can be obtained, but
this is left as an exercise.
SINGULARITIES. As with all sets of Euler angles, the 3–1–3 Euler angles are subject
to restrictions. For the 3–1–3 set, we can ﬁnd the restrictions by examining when
the Euler basis fails to be a basis. Fortunately, it is easy to see from Figure 6.7(a)
that this is the case when t
3
= ±p
3
. As a result, restrictions are placed on the second
angle:
φ ∈ [0, 2π), θ ∈ (0, π), ψ ∈ [0, 2π).
The fact that the restriction needs to be placed on the second Euler angle is consis
tent with corresponding results for the other 11 sets of Euler angles.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
190 Rotation Tensors
(a)
(b)
ψ ψ
φ φ
θ
θ
p
1
p
1
p
2
p
2
g
1
= p
3
g
1
= p
3
t
2
g
1
g
2
= t
1
g
2
= t
1
g
3
= t
3
g
3
= t
3
g
3
Figure 6.7. (a) The Euler basis and (b) dual Euler basis vectors for the 3–1–3 set of Euler
angles. For this set of Euler angles, g
1
t
2
, g
2
= t
1
, and g
3
t
2
.
DUAL EULER BASIS. By following the procedure that led to (6.29), we ﬁnd that the
dual Euler basis
¸
g
i
¸
has the representation
∗
⎡
⎢
⎣
g
1
g
2
g
3
⎤
⎥
⎦
=
⎡
⎢
⎣
sin(φ)cosec(θ) cos(φ)cosec(θ) 0
cos(φ) −sin(φ) 0
−sin(φ) cot(θ) −cos(φ) cot(θ) 1
⎤
⎥
⎦
⎡
⎢
⎣
t
1
t
2
t
3
⎤
⎥
⎦
. (6.34)
Similarly,
⎡
⎢
⎣
g
1
g
2
g
3
⎤
⎥
⎦
=
⎡
⎢
⎣
−sin(ψ)cot(θ) cos(ψ)cot(θ) 1
cos(ψ) sin(ψ) 0
sin(ψ)cosec(θ) −cos(ψ)cosec(θ) 0
⎤
⎥
⎦
⎡
⎢
⎣
p
1
p
2
p
3
⎤
⎥
⎦
.
The dual Euler basis vectors are shown in Figure 6.7. It is important to observe that
they are not deﬁned when this set of Euler angles has its singularities at θ = 0 and
θ = π.
Using (6.34), we can show that
ω
R
· g
1
=
˙
ψ, ω
R
· g
2
=
˙
θ, ω
R
· g
3
=
˙
φ.
We can also establish the following results:
⎡
⎢
⎣
g
1
g
2
g
3
⎤
⎥
⎦
=
⎡
⎢
⎣
1 0 cos(θ)
0 1 0
cos(θ) 0 1
⎤
⎥
⎦
⎡
⎢
⎣
g
1
g
2
g
3
⎤
⎥
⎦
. (6.35)
It is a valuable exercise to compare the results just presented for the 3–1–3 Euler
angles with those presented earlier for the 3–2–1 set. One issue that will arise in this
comparison is the different ranges that the second Euler angle θ possesses for the
two sets.
∗
That is, one calculates the inverse of the transpose of the 3 × 3 matrix in (6.33) that describes g
i
in
terms of t
k
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
6.9 Further Representations of a Rotation Tensor 191
Other Sets of Euler Angles
For the Euler basis, one has three choices for g
1
and, because g
1
= g
2
, two choices
for g
2
. Finally, there are two choices of g
3
. Consequently, there are 2 ×2 ×3 = 12
choices of the vectors for the Euler basis. The easiest method to see which set of
Euler angles is being used is to specify the angular velocity vector. Here, expres
sions are given for each of the 12 sets of Euler angles for a rotation tensor R =
¸
3
i=1
t
i
⊗p
i
:
1–2–3 Set: ω
R
=
˙
ψp
1
+
˙
θ t
2
+
˙
φt
3
,
3–2–3 Set: ω
R
=
˙
ψp
3
+
˙
θ t
2
+
˙
φt
3
,
1–2–1 Set: ω
R
=
˙
ψp
1
+
˙
θ t
2
+
˙
φt
1
,
1–3–1 Set: ω
R
=
˙
ψp
1
+
˙
θ t
3
+
˙
φt
1
,
1–3–2 Set: ω
R
=
˙
ψp
1
+
˙
θ t
3
+
˙
φt
2
,
2–3–1 Set: ω
R
=
˙
ψp
2
+
˙
θ t
3
+
˙
φt
1
,
2–3–2 Set: ω
R
=
˙
ψp
2
+
˙
θ t
3
+
˙
φt
2
,
2–1–2 Set: ω
R
=
˙
ψp
2
+
˙
θ t
1
+
˙
φt
2
,
2–1–3 Set: ω
R
=
˙
ψp
2
+
˙
θ t
1
+
˙
φt
3
,
3–1–3 Set: ω
R
=
˙
ψp
3
+
˙
θ t
1
+
˙
φt
3
,
2–3–1 Set: ω
R
=
˙
ψp
2
+
˙
θ t
3
+
˙
φt
1
,
3–1–2 Set: ω
R
=
˙
ψp
3
+
˙
θ t
1
+
˙
φt
2
.
The sets of Euler angles, 121, 131, 232, 212, 313, and 323, are known as the symmetric
sets, whereas the other six sets are known as asymmetric sets. The latter sets are also
known as the Cardan angles, Tait angles, or Bryan angles. Tait’s original discussion
(of what we would refer to as 1–2–3 Euler angles) can be seen in Section 12 of his
1868 paper [210]. In his seminal text [24] on aircraft stability that was published in
1911, Bryan introduced what we would refer to as a 2–3–1 set of Euler angles (see
Figure 6.8). It is interesting to recall that the Wright brothers ﬁrst successful ﬂight
was in 1903.
For all sets of Euler angles, a singularity is present for certain values of the
second angle θ. At these values g
1
= ±g
3
, and the Euler basis fails to be a ba
sis for E
3
. To avoid these singularities it is often necessary to use two different
sets of Euler angles and to switch from one set to the other as a singularity is
approached.
6.9 Further Representations of a Rotation Tensor
Apart from Euler’s representation and the Euler angle representation, there are
several other representations of a rotation tensor. Most of them are discussed in
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
192 Rotation Tensors
φ
φ
φ
θ
θ
G
x
z
Figure 6.8. Reproduction of Figure 3 in Bryan’s seminal text [24] on the stability of aircraft.
The angles θ and φ in this ﬁgure are the pitch and roll angles, respectively, of the aircraft. The
point G is the center of mass of the aircraft, and x and z label the corotational bases for the
aircraft.
Shuster’s review article [196], and we now discuss two of them. These are repre
sentations that are due to Olinde Rodrigues [181] in 1840 and the Euler parameter
representation.
The Rodrigues Vector
The Rodrigues representation is based on the vector:
λ = tan
φ
2
r.
This vector is sometimes called the Gibbs vector. Clearly the Rodrigues vector λ is
not a unit vector. Indeed, when φ = 0, λ = 0, and when φ = π, λ is undeﬁned. Con
sequently, if φ varies through π, then we cannot use the Rodrigues representation
subsequently discussed.
With the assistance of the identities
sin(φ) =
2 tan(
φ
2
)
1 +tan
2
(
φ
2
)
, cos(φ) =
1 −tan
2
(
φ
2
)
1 +tan
2
(
φ
2
)
,
you should be able to verify that
cos(φ) =
1 −λ · λ
1 +λ · λ
, sin(φ) =
2λ · r
1 +λ · λ
,
Substituting for r and φ in (6.8), we ﬁnd the Rodrigues representation:
R =
˜
R(λ) =
1
1 +λ · λ
((1 −λ · λ)I +2λ ⊗λ −2(ελ)) .
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
6.9 Further Representations of a Rotation Tensor 193
The angular velocity vector associated with this representation is
ω
R
=
2
1 +λ · λ
˙
λ −
˙
λ ×λ
.
This vector can be calculated by directly substituting into the earlier result associ
ated with Euler representation (6.14).
The Euler–Rodrigues Symmetric Parameters
One of the most popular fourparameter representations uses the four Euler–
Rodrigues symmetric parameters e
0
and e.
∗
These parameters are often known as
the Euler parameters and have very interesting historical connections.
†
They can be
deﬁned as
e
0
= cos
φ
2
, e = sin
φ
2
r.
As a consequence of their deﬁnition, the parameters are subject to what is known
as the Euler parameter constraint
‡
:
e
2
0
+e · e = 1.
We also note that
λ =
e
e
0
.
It is possible to express r and φ in terms of the parameters e
0
and e, but this is left as
an exercise.
By substituting for r and φ in Euler representation (6.8), we ﬁnd the Euler–
Rodrigues symmetric parameter representation:
R =
¯
R(e
0
, e) =
e
2
0
−e · e
I +2e ⊗e −2e
0
(εe).
∗
The four parameters, e
0
and the three components of e, are often considered to be the four com
ponents of a quaternion q = e
0
+e
1
i +e
2
j +e
3
k, where e
i
are the components of e relative to a
righthanded orthonormal basis, and i, j, and k are bases vectors for the quaternion. Consequently,
Euler–Rodrigues symmetric parameters are sometimes referred to as (unit) quaternions.
†
Excellent discussions can be found in Altmann [2, 3] and Gray [77].
‡
The relaxation of this constraint is discussed in O’Reilly and Varadi [166] who show, among other
matters, how it can be visualized by using Hoberman’s sphere. This topic is also intimately related
to Gauss’ mutation of space [69].
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
194 Rotation Tensors
Suppose R =
¸
3
i=1
t
i
⊗p
i
; then it is easy to show that e · t
k
= e · p
k
. If one writes out
expressions for the components of R
ik
= t
k
· p
i
,
∗
⎡
⎢
⎢
⎣
R
11
R
12
R
13
R
21
R
22
R
23
R
31
R
32
R
33
⎤
⎥
⎥
⎦
=
e
2
0
−e
2
1
−e
2
2
−e
2
3
⎡
⎢
⎢
⎣
1 0 0
0 1 0
0 0 1
⎤
⎥
⎥
⎦
+
⎡
⎢
⎢
⎣
2e
2
1
2e
1
e
2
2e
1
e
3
2e
1
e
2
2e
2
2
2e
2
e
3
2e
1
e
3
2e
2
e
3
2e
2
3
⎤
⎥
⎥
⎦
+
⎡
⎢
⎣
0 −2e
0
e
3
2e
0
e
2
2e
0
e
3
0 −2e
0
e
1
−2e
0
e
2
2e
0
e
1
0
⎤
⎥
⎦
, (6.36)
then it is easy to see that the rotation tensor is a quadratic function of the four pa
rameters e
0
, e
1
, e
2
, e
3
. As a result, this representation has several computational ad
vantages over other representations. You may also notice how easy it is to establish
an expression for the components of R
T
from (6.36).
The angular velocity vector associated with this representation is
ω
R
= 2 (e
0
˙ e − ˙ e
0
e +e × ˙ e) .
Again, we can calculate this vector by directly substituting into the earlier result
associated with Euler representation (6.14) for ω
R
. Notice that the angular velocity
vector is a relatively simple function of the Euler–Rodrigues symmetric parameters
and their derivatives.
In one of the exercises at the end of this chapter, further results pertaining to
the Euler–Rodrigues parameters for the composition of two rotation tensors are
presented. These results, which date to Olinde Rodrigues (1794–1851) in 1840, are
remarkably elegant. Modern applications of these parameters arise in the estimation
of the rotation tensor (attitude) of spacecraft and in computer vision and robotics.
In these areas, one considers measurements of two sets of vectors that are related
by an unknown rotation tensor R and a translation d. The estimation of R and d can
be rendered as a leastsquares estimation problem that is known as the orthogonal
Procrustes problem [95, 192] and in the satellite dynamics community as the Wahba
problemafter Grace Wahba [222]. That is, given the N ≥ 2 measurements a
1
, . . . , a
N
and b
1
, . . . , b
N
, which are related by a rotation R and a translation d, determine the
optimal R and d such that the following function is minimized:
W =
1
2
N
¸
K=1
α
K
b
K
−Ra
K
−d
2
, (6.37)
where α
K
is a scalar weight for the Kth measurement. When d = 0, Davenport
showed that, by parameterizing R in (6.37) by e
0
and e, it is possible to ﬁnd a
∗
The corresponding matrix representation for Euler’s representation was established earlier; see
(6.10).
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
6.10 Derivatives of Scalar Functions of Rotation Tensors 195
very elegant solution to the Wahba problem.
∗
The corresponding elegant solution
to the Procrustes problem ( i.e., when d = 0) was established later by Horn [95].
†
Summary of the Angular Velocity Vectors
It is useful to summarize the representations we have discussed for the angular ve
locity vector corresponding to a rotation through an angle φ about an axis r:
ω
R
=
3
¸
i=1
˙ γ
i
g
i
=
˙
φr +sin(φ)˙ r +(1 −cos(φ))r × ˙ r
= 2 (e
0
˙ e − ˙ e
0
e +e × ˙ e)
=
2
1 +λ · λ
˙
λ −
˙
λ ×λ
. (6.38)
With some minor manipulations of these equations, one can also obtain expressions
for ˙ r, ˙ e, and
˙
λ.
6.10 Derivatives of Scalar Functions of Rotation Tensors
Consider a function U = U(R). We wish to calculate the time derivative of this func
tion. One of the complications is that R has several representations, and for each of
them a different representation of the derivative will be found. It will be subse
quently revealed that a simple expression for the derivative of U can be found in
terms of a vector u
R
and the dual Euler basis. This result is presented in Equation
(6.41).
To start, we use the simplest representation for R:
R =
3
¸
i=1
3
¸
k=1
R
ik
p
i
⊗p
k
,
where ˙ p
k
= 0. Then, as U = U(R) = U(R
ik
, p
j
), we ﬁnd by using the chain rule that
˙
U =
3
¸
i=1
3
¸
k=1
∂U
∂R
ik
˙
R
ik
.
We can express this result by using the trace operator:
˙
U = tr
∂U
∂R
˙
R
T
,
∗
Discussions of Davenport’s solution and those of others, along with extensions to Wahba’s original
formulation, can be found in [136, 137, 195, 197].
†
Additional references to other solutions to this problem can be found in [46, 49, 200].
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
196 Rotation Tensors
where
∂U
∂R
=
3
¸
i=1
3
¸
k=1
∂U
∂R
ik
p
i
⊗p
k
.
Noting that
R
=
˙
RR
T
, and, after introducing the angular velocity tensor, we can
then write
˙
U = tr
∂U
∂R
R
T
T
R
.
However, as
R
is skewsymmetric, tr
A
T
R
= 0 for all A = A
T
. Consequently,
only the skewsymmetric part of
∂U
∂R
R
T
contributes to
˙
U:
˙
U = tr
∂U
∂R
R
T
T
R
= tr
U
R
T
R
,
where we have introduced the skewsymmetric operator:
U
R
=
1
2
∂U
∂R
R
T
−R
∂U
∂R
T
.
As U
R
is a skewsymmetric tensor, we can calculate a vector that corresponds to
twice the axial vector of U
R
∗
:
u
R
= −ε [U
R
] . (6.39)
The existence of this vector allows us to establish the following representations for
˙
U:
˙
U = tr
∂U
∂R
˙
R
T
= tr
U
R
T
= u
R
· ω
R
. (6.40)
Because we have established numerous representations for ω
R
, we next invoke the
ﬁnal form of the representation for
˙
Uto determine representations for u
R
.
As an example, suppose Uis parameterized by use of Euler angles: U =
ˆ
U
γ
k
.
Then, invoking (6.40)
2
,
˙
U =
3
¸
k=1
∂
ˆ
U
∂γ
k
˙ γ
k
= u
R
·
3
¸
j=1
˙ γ
j
g
j
.
∗
The reason for the absence of the factor
1
2
in (6.39) can be inferred from identity (A.12), which is
discussed in the Exercises at the end of the Appendix.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
6.10 Derivatives of Scalar Functions of Rotation Tensors 197
With the help of the dual Euler basis, we conclude that
u
R
=
3
¸
i=1
∂
ˆ
U
∂γ
i
g
i
. (6.41)
This is the simplest, and most useful, representation that we know of for u
R
. It ﬁrst
appeared in [163].
We now assume that R is parameterized by use of one of the other three meth
ods mentioned previously. By evaluating
˙
U, using identity (6.40)
2
and representa
tions (6.38), and following the procedure that led to (6.41), we can ﬁnd three other
representations for u
R
. With details omitted, a summary of the representations is
now presented:
u
R
=
3
¸
i=1
∂
ˆ
U
∂γ
i
g
i
=
∂
˜
U
∂φ
r +
1
2
cot
φ
2
(I −r ⊗r) −εr
∂
˜
U
∂r
=
1
2
(e
0
I −εe)
∂
¯
U
∂e
−
∂
¯
U
∂e
0
e
=
1
2
(I +λ ⊗λ −ελ)
∂
˘
U
∂λ
, (6.42)
where U(R) =
¯
U(e
0
, e) =
ˆ
U(γ
1
, γ
2
, γ
3
) =
˜
U(φ, r) =
˘
U(λ). Representation (6.42)
4
was ﬁrst established by Simmonds [199]. We also note that a representation that
is closely related to (6.42)
2
is discussed in Antman [5].
Several of the partial derivatives in (6.42) need to be carefully evaluated. For
example, because r is a unit vector, the derivative
∂
˜
U
∂r
must be evaluated on the sur
face r · r = 1. Related remarks pertain to
∂
¯
U
∂e
,
∂
¯
U
∂e
0
, and
∂U
∂R
. In other words, these are
tangential or surface derivatives. One method of evaluating them is to parameterize
R by the Euler angles, and then transform from the parameters of interest to the
Euler angles. Indeed, (6.42), the chain rule, and the identity εg
i
= −
∂
ˆ
R
∂γ
i
R
T
can be
used to show that
∂
¯
U
∂e
0
= u
R
· 2e,
∂
¯
U
∂e
= 2(e
0
I +εe)u
R
,
∂U
∂R
= −(εu
R
)R,
∂
˜
U
∂φ
= u
R
· r,
∂
˜
U
∂r
=
sin(φ)(I −r ⊗r) +2 sin
2
φ
2
εr
u
R
.
As discussed in O’Reilly [160] and Simmonds [199], results (6.42) can be used to
establish moment potentials associated with conservative moments. We shall use
(6.42)
1
extensively.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
198 Exercises 6.1–6.2
EXERCISES
6.1. Consider the tensor R:
R = cos(θ)E
1
⊗E
1
+sin(θ)E
2
⊗E
1
+cos(θ)E
2
⊗E
2
−sin(θ)E
1
⊗E
2
+E
3
⊗E
3
.
(a) Show that R also has the representations
R = e
r
⊗E
1
+e
θ
⊗E
2
+E
3
⊗E
3
= cos(θ)e
r
⊗e
r
+sin(θ)e
θ
⊗e
r
+cos(θ)e
θ
⊗e
θ
−sin(θ)e
r
⊗e
θ
+E
3
⊗E
3
.
(b) Show that
R
=
˙
θ (E
2
⊗E
1
−E
1
⊗E
2
)
=
˙
θ (e
θ
⊗e
r
−e
r
⊗e
θ
) ,
ω
R
=
˙
θE
3
.
6.2. Recall that three Euler angles can be used to parameterize a rotation tensor R.
In this exercise, we consider the 3–2–3 set of Euler angles:
R = L
φ, t
3
= t
3
L
θ, t
2
L(ψ, E
3
) .
This set of Euler angles is used in several texts, for example, Section 4.2 of Ginsberg
[71], Kelvin and Tait [109], Routh [184, 185], and Whittaker [228],
∗
and is illustrated
in Figure 6.3.
(a) Draw ﬁgures illustrating the relationships between (i) E
i
and t
i
, (ii) t
i
and
t
i
, and (iii) t
i
and t
i
.
(b) Explain why the second angle of rotation θ is restricted to lie between 0 and
π
(c) For this set of Euler angles, show that the Euler basis has the representa
tions
⎡
⎢
⎣
g
1
g
2
g
3
⎤
⎥
⎦
=
⎡
⎢
⎣
−sin(θ) cos(φ) sin(φ) sin(θ) cos(θ)
sin(φ) cos(φ) 0
0 0 1
⎤
⎥
⎦
⎡
⎢
⎣
t
1
t
2
t
3
⎤
⎥
⎦
=
⎡
⎢
⎣
0 0 1
−sin(ψ) cos(ψ) 0
cos(ψ) sin(θ) sin(ψ) sin(θ) cos(θ)
⎤
⎥
⎦
⎡
⎢
⎣
E
1
E
2
E
3
⎤
⎥
⎦
.
∗
Whittaker’s notation is similar to ours except his ψ corresponds to our φ and vice versa. If you are
comparing his expression for the components of ω
R
with ours, you will see that there is a typograph
ical error in his expression for ω
2
in Section 16 of [228]. Routh’s notation is similar to ours, but his
coordinate axes are lefthanded.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
Exercise 6.2 199
In addition, show that the Euler angles are such that
⎡
⎢
⎣
t
1
t
2
t
3
⎤
⎥
⎦
=
⎡
⎢
⎣
cos(φ) sin(φ) 0
−sin(φ) cos(φ) 0
0 0 1
⎤
⎥
⎦
⎡
⎢
⎣
cos(θ) 0 −sin(θ)
0 1 0
sin(θ) 0 cos(θ)
⎤
⎥
⎦
×
⎡
⎢
⎣
cos(ψ) sin(ψ) 0
−sin(ψ) cos(ψ) 0
0 0 1
⎤
⎥
⎦
⎡
⎢
⎣
E
1
E
2
E
3
⎤
⎥
⎦
. (6.43)
(d) Using (6.43), derive expressions for the components R
ik
of R. These com
ponents have a variety of representations
R
ik
= (Rt
k
) · t
i
= (RE
k
) · E
i
= t
k
· E
i
.
(e) Recall that every rotation tensor R has an axis of rotation r and angle of
rotation. With the help of the results presented in Subsection 6.5.1, select
four different values of the set (φ, θ, ψ) and determine the corresponding
axis of rotation and the angle of rotation. Give physical interpretations for
the four sets of values of the Euler angles that you have selected.
(f) For this set of Euler angles, show that the dual Euler basis has the repre
sentation
⎡
⎢
⎣
g
1
g
2
g
3
⎤
⎥
⎦
=
⎡
⎢
⎣
−cos(φ)cosec(θ) sin(φ)cosec(θ) 0
sin(φ) cos(φ) 0
cos(φ) cot(θ) −sin(φ) cot(θ) 1
⎤
⎥
⎦
⎡
⎢
⎣
t
1
t
2
t
3
⎤
⎥
⎦
.
With the help of these results, verify the following expressions for the dual
Euler basis in terms of the bases {E
1
, E
2
, E
3
} and
¸
g
1
, g
2
, g
3
¸
:
⎡
⎢
⎣
g
1
g
2
g
3
⎤
⎥
⎦
=
⎡
⎢
⎣
−cos(ψ)cot(θ) −sin(ψ)cot(θ) 1
−sin(ψ) cos(ψ) 0
cos(ψ)cosec(θ) sin(ψ)cosec(θ) 0
⎤
⎥
⎦
⎡
⎢
⎣
E
1
E
2
E
3
⎤
⎥
⎦
=
⎡
⎢
⎣
cosec
2
(θ) 0 −cot(θ)cosec(θ)
0 1 0
−cot(θ)cosec(θ) 0 cosec
2
(θ)
⎤
⎥
⎦
⎡
⎢
⎣
g
1
g
2
g
3
⎤
⎥
⎦
.
(g) For this set of angles, show that the angular velocity vector has the repre
sentation
ω
R
=
˙
φt
3
+
˙
θ t
2
+
˙
ψE
3
. (6.44)
You will need to use two distinct corotational derivatives to ﬁnd this repre
sentation. The ﬁrst of these ﬁxes t
i
whereas the second ﬁxes t
i
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
200 Exercises 6.2–6.3
(h) For this set of angles, show that the components of ω
R
relative to the basis
{E
1
, E
2
, E
3
} have the representations
1
=
˙
φsin(θ) cos (ψ) −
˙
θ sin(ψ) ,
2
=
˙
φsin(θ) sin(ψ) +
˙
θ cos (ψ) ,
3
=
˙
φcos (θ) +
˙
ψ,
where
i
= ω
R
· E
i
. These representations can be found in Section 257 of
Routh [185].
(i) Suppose that ω
k
(t) = ω
R
· t
k
are known functions. Show that
⎡
⎢
⎣
˙
ψ
˙
θ
˙
φ
⎤
⎥
⎦
=
⎡
⎢
⎣
−cos(φ) cosec(θ) sin(φ) cosec(θ) 0
sin(φ) cos(φ) 0
cos(φ) cot(θ) −sin(φ) cot(θ) 1
⎤
⎥
⎦
⎡
⎢
⎣
ω
1
ω
2
ω
3
⎤
⎥
⎦
. (6.45)
(j) This problem involves the numerical integration of (6.45). Given
ω
1
(t) = 0.2 sin(0.5t), ω
2
(t) = 0.2 sin(0.05t), ω
3
(t) = 10ω
1
(t),
and initial values for the Euler angles of your choice, determine φ(t), θ(t),
and ψ(t). How can these results be used to determine t
k
(t)?
6.3. Recall that a rotation tensor L representing a counterclockwise rotation about
an axis p though an angle ν has the representation
L = L(ν, p) = cos(ν)(I −p ⊗p) −sin(ν)p +p ⊗p,
and its associated angular velocity vector has the representation
ω
L
= ˙ νp +sin(ν) ˙ p +(1 −cos(ν))p × ˙ p.
Consider two rotation tensors:
Q
1
= L(θ, E
3
), Q
2
= L(φ, e
1
)L(θ, E
3
),
where
e
1
= cos(θ)E
1
+sin(θ)E
2
, e
2
= cos(θ)E
2
−sin(θ)E
1
, e
3
= E
3
.
(a) Show that Q
1
has the representation
Q
1
= e
1
⊗E
1
+e
2
⊗E
2
+e
3
⊗E
3
.
(b) Give an example of a system of two rigid bodies for which the rotation
tensor of one body is Q
1
and the rotation tensor of the second body is Q
2
.
(c) Given that the relative rotation tensor R
2
= Q
2
Q
T
1
, show that
ω
R
2
=
˙
φe
1
+sin(φ)
˙
θ e
2
+(1 −cos(φ))
˙
θ e
3
.
(d) Explain why
ω
R
2
= ˆ ω
R
2
= ω
Q
2
−ω
Q
1
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
Exercise 6.4 201
6.4. The parameterization of the rotation tensor by use of Euler parameters (unit
quaternions or symmetric Euler–Rodriques parameters) has the beautiful conse
quence that the formula for the composition of two rotations is very elegant. Indeed,
the same results for two tensors described by Euler angles are very unwieldy. In this
problem, we use Euler parameters to explore some results pertaining to rotation
tensors.
Consider two rotation tensors A and B:
A = (e
2
0
−e · e)I +2e ⊗e −2e
0
(εe),
B = (f
2
0
−f · f)I +2f ⊗f −2f
0
(εf). (6.46)
Here, {e
0
, e} and {f
0
, f} represent two sets of Euler parameters:
e
0
= cos
φ
2
, e = sin
φ
2
a, f
0
= cos
θ
2
, f = sin
θ
2
b, (6.47)
where a is the axis of rotation of A and b is the axis of rotation of B, φ is the angle of
rotation for A, and θ is the angle of rotation for B. That is, the tensor A corresponds
to a counterclockwise rotation of φ about a.
(a) Recall the representation for a rotation tensor A in terms of the angle of
rotation φ and the axis of rotation a:
A(φ, a) = cos(φ)(I −a ⊗a) −sin(φ)a +a ⊗a.
Verify that A has the representation (6.46)
1
.
(b) Letting e =
¸
3
i=1
e
i
E
i
, what are A
ik
= (AE
k
) · E
i
?
(c) Show that
C = BA = (g
2
0
−g · g)I +2g ⊗g −2g
0
(εg), (6.48)
where
g
0
= e
0
f
0
−e · f, g = e
0
f + f
0
e +f ×e. (6.49)
This result was ﬁrst established by Rodrigues [181] in 1840. The earliest
English commentary on it is by Cayley [32] in 1845.
(d) In terms of e
0
, f
0
, f, and e, what are the Euler parameters of the rotation
tensors AB and B
T
A
T
?
(e) Using the results of (d), show that the compositions of rotations is not, in
general, commutative: i.e., AB = BA.
(f) Recall that the angular velocity vector associated with A has the represen
tation
ω
A
= 2 (e
0
˙ e − ˙ e
0
e +e × ˙ e) ,
where ω
A
= −
1
2
¸
˙
AA
T
¸
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
202 Exercises 6.4–6.5
(i) What are ω
B
and ω
C
?
(ii) Give an explanation for the following result:
ω
C
= ω
B
+ω
A
.
(g) If e
0
= f
0
=
1
√
2
, a = E
3
, and b = E
2
, then what does the rotation tensor C
represent? Illustrate your solution by showing how C transforms the basis
{E
1
, E
2
, E
3
}.
6.5. The latitude (λ) and longitude (θ) of a point on the Earth’s surface are illus
trated schematically in Figure 6.9. In navigation systems, one uses these angles to
deﬁne the downward direction e
z
, the northerly direction e
x
, and the easterly direc
tion e
y
:
⎡
⎢
⎣
e
x
e
y
e
z
⎤
⎥
⎦
=
⎡
⎢
⎣
−cos(θ) sin(λ) −sin(θ) sin(λ) cos(λ)
−sin(θ) cos(θ) 0
−cos(θ) cos(λ) −sin(θ) cos(λ) −sin(λ)
⎤
⎥
⎦
⎡
⎢
⎣
E
1
E
2
E
3
⎤
⎥
⎦
. (6.50)
Here, the triad {E
1
, E
2
, E
3
} is a set of ﬁxed righthanded Cartesian basis vectors.
λ
θ
E
1
E
2
E
3
e
x
e
x
e
y
e
y
e
z
−e
z
North
East
Down
Figure 6.9. The angles of longitude θ and latitude λ.
(a) Suppose that R = e
x
⊗E
1
+e
y
⊗E
2
+e
z
⊗E
3
. Verify that this rotation
tensor can be composed of two rotation tensors R
1
and R
2
, where R
1
cor
responds to a rotation about E
3
through an angle θ and R
2
corresponds to
a rotation about e
y
through an angle −
π
2
−λ.
(b) Given a vector x,
x = x
x
e
x
+x
y
e
y
+x
z
e
z
= X
1
E
1
+X
2
E
2
+X
3
E
3
.
Show that
x
x
=
3
¸
i=1
R
i1
X
i
, x
y
=
3
¸
i=1
R
i2
X
i
, x
z
=
3
¸
i=1
R
i3
X
i
,
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
Exercise 6.12 203
where
R =
3
¸
i=1
3
¸
k=1
R
ik
E
i
⊗E
k
.
How are R
ik
’s related to the matrix in Equation (6.50)?
6.6. Consider two rotation tensors A and B, where
A =
3
¸
i=1
t
i
⊗E
i
, B =
3
¸
i=1
e
i
⊗t
i
,
and {E
1
, E
2
, E
3
} is a ﬁxed, righthanded orthonormal basis for E
3
.
(a) Show that
˙
B =
o
B
+
A
B −B
A
,
where
o
B
is the corotational derivative of B assuming that t
i
are ﬁxed, and
A
=
˙
AA
T
.
(b) Consider the rotation tensor C = BA. Using the results of (a), show that
ˆ
B
=
o
B
B
T
,
where the relative angular velocity tensor
ˆ
B
and the angular velocity ten
sor
C
are, respectively,
ˆ
B
=
C
−
A
,
C
=
˙
CC
T
.
Why is
ˆ
B
skewsymmetric, and what does this imply for the product
ˆ
B
b,
where b is any vector?
(c) Consider the following examples of two tensors:
A = cos(ψ)(I −E
3
⊗E
3
) −sin(ψ)E
3
+E
3
⊗E
3
,
B = cos(θ)(I −t
1
⊗t
1
) −sin(θ)t
1
+t
1
⊗t
1
,
where
t
1
= cos(ψ)E
1
+sin(ψ)E
2
.
(i) What are
˙
A and
o
B
?
(ii) Using Equation (6.14), what are ω
A
and ˆ ω
B
?
(iii) With the help of (6.14), verify that
ω
B
=
˙
θ t
1
−
˙
ψ (cos(θ)E
3
−sin(θ)t
2
) +
˙
ψE
3
.
Here, ω
B
is the angular velocity vector associated with B.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
204 Exercises 6.7–6.11
6.7. Show that the rotation tensor L(
π
2
, E
3
)L(
π
2
, E
1
) is equivalent to a rotation ten
sor L(
2π
3
, p), where
p =
1
√
3
(E
1
+E
2
+E
3
) .
This result is very useful in analyzing material symmetry groups of crystals.
6.8. Show that
∗
L
π
2
, E
2
L
π
2
, E
3
L
−
π
2
, E
1
= L
π
2
, E
3
.
This result has a interpretation that is sometimes used to show that successive rota
tions about three perpendicular axes can be reproduced by a single rotation about
one of the axes. It is also the source for one explanation of a phenomenon in biome
chanics that is known as Codman’s paradox [40].
†
6.9. Examine Sections 1 and 24–30 of Euler [54], in which he introduces the Euler
angles (p, q, r). Show that a set of 1–3–1 Euler angles is being used, where
ψ = p, θ = q, φ = π −r.
You may wish to examine his expression for the components R = e
1
· ω, P = e
2
· ω,
and Q = e
3
· ω on page 205 of [54] to help with this. Related results, but with a
different notation, can be found in [57]. In both of these papers you will ﬁnd, among
other matters, Euler’s discussion of the components of Euler and inertia tensors.
6.10. If p
1
, p
2
and p
3
are any righthanded set of orthonormal basis vectors, then
establish the Rodrigues–Hamilton theorem:
L(π, p
3
) L(π, p
2
) L(π, p
1
) = I. (6.51)
This result is discussed in Section 3 of Whittaker [228], and he credits it to
Rodrigues [181] and Hamilton [90]. Whittaker presents a proof that involves purely
geometric arguments, and it is a good exercise to compare your proof with his. The
Rodrigues–Hamilton theorem will be used later to show that a constant moment is
not conservative.
6.11. Suppose a set of 3–2–1 Euler angles is used to parameterize a rotation tensor
R. If we denote the values of these angles by α
1
, α
2
, and α
3
, respectively, then what
are the corresponding values of the 3–2–1 Euler angles for the inverse R
T
of this
rotation?
∗
For the rotation tensors discussed in this problem, you will need to compute E
k
. If you need assis
tance with this, then please see (A.11) in the Appendix.
†
As quoted in Politti et al. [174], Codman’s paradox occurs if you ﬁrst place your right arm hang
ing down along your side with your thumb pointing forward and your ﬁngers pointing toward the
ground. Now elevate your arm horizontally so that your ﬁngers point to the right, then rotate your
arm in the horizontal plane so that your ﬁngers now point forward, and ﬁnally rotate your arm
downward so that your ﬁngers eventually point toward the ground. After these three rotations, you
will notice that your thumb points to the left. That is, your arm has rotated by 90
◦
. The fact that
you get this rotation without having performed a rotation about the longitudinal axis of your arm is
known as Codman’s paradox.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
Exercise 6.12 205
6.12. In Section 1.7 of Kane et al. [106] two distinct sets of Euler angles are
deﬁned. The ﬁrst corresponds to “bodyangles” and the second corresponds to
“spaceangles.” Thus they deﬁne 24 sets of Euler angles. Verify that the 3–2–1
Euler angles discussed in Subsection 6.8.1 are equivalent to the “bodythree: 3–2–1”
angles in [106]. In addition, show how a set of “spacethree: 3–2–1” angles can be
used to parameterize R. Denoting the “spacethree: 3–2–1” angles by β
1
, β
2
, and β
3
,
respectively, then, in your solution to this problem, you will ﬁnd the representation
⎡
⎢
⎢
⎣
R
11
R
12
R
13
R
21
R
22
R
23
R
31
R
32
R
33
⎤
⎥
⎥
⎦
=
⎡
⎢
⎣
1 0 0
0 cos (β
3
) −sin(β
3
)
0 sin(β
3
) cos (β
3
)
⎤
⎥
⎦
×
⎡
⎢
⎣
cos (β
2
) 0 sin(β
2
)
0 1 0
−sin(β
2
) 0 cos (β
2
)
⎤
⎥
⎦
×
⎡
⎢
⎣
cos (β
1
) −sin (β
1
) 0
sin(β
1
) cos (β
1
) 0
0 0 1
⎤
⎥
⎦
.
This result should be compared with corresponding representation (6.25) for the
“bodythree: 3–2–1” angles. For assistance with this problem, the discussion of
“bodyﬁxed” and “spaceﬁxed” rotations in Section 3.2 of Ginsberg [71] and a
related discussion in Section 7.14 of Pars [170] might be helpful.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
7 Kinematics of Rigid Bodies
7.1 Introduction
This chapter contains results on the threedimensional kinematics of rigid bodies.
We discuss several useful classical representations for the velocity and acceleration
vectors of any material point of a rigid body. We also discuss the angular velocity
vector ω, the linear momentum G, angular momenta H, H
O
, and H
A
, and kinetic en
ergy T of rigid bodies and the inertias that accompany them. The chapter concludes
with a discussion of the conﬁguration manifold for a rigid body.
The origin of most of the material in this chapter can be traced to Euler’s sem
inal work on rigid body dynamics in the 1750s. Since that time, his theory has been
used to develop models for a wide range of mechanical systems and various treat
ments of his work have appeared. Recently, a tensorbased notation has been used
by Beatty [15], Casey [26, 28], Fox [65], Greenwood [80], and Gurtin [84]. In this
book, we follow their work as it leads to transparent developments particularly with
regards to inertias and angular velocities.
7.2 The Motion of a Rigid Body
To discuss the kinematics of rigid bodies, it is convenient to follow some develop
ments in continuum mechanics and deﬁne the reference and present conﬁguration
of a rigid body. First, a body B is considered to be a collection of material points
(mass particles or particles). We denote a material point of B by X. The position
of the material point X, relative to a ﬁxed origin, at time t is denoted by x (see
Figure 7.1). The present (or current) conﬁguration κ
t
of the body is a smooth, one
toone, onto function that has a continuous inverse. It maps material points X of B
to points in threedimensional Euclidean space: x = κ
t
(X). As the location x of the
particle Xchanges with time, this function depends on time, hence the subscript t. It
is important to note that κ
t
deﬁnes the state of the body at time t.
We also deﬁne a ﬁxed reference conﬁguration κ
0
of the body. This conﬁgura
tion is deﬁned by the invertible function X = κ
0
(X). Using the invertibility of this
function, we can use the position vector X of a material point X in the reference
conﬁguration to uniquely deﬁne the material point of interest. Later, we will use the
206
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
7.2 The Motion of a Rigid Body 207
κ
0
κ
t
O
X
X
Y
Y
¯
X
¯
X
X
Y
¯
X
x
y
¯ x
π
Figure 7.1. The reference κ
0
and present κ
t
conﬁgurations of a body B.
reference conﬁguration to determine many of the properties of a body, such as its
mass m and inertia tensor J
0
. Using the reference conﬁguration, we can deﬁne the
motion of the body as a function of X and t:
x = χ(X, t).
Notice that the motion of a material point of B depends on the instant of time and
the material point of interest.
Euler’s Theorem
For rigid bodies, the nature of the function χ(X, t) can be simpliﬁed dramatically.
First, for rigid bodies the distance between any two mass particles, say X
1
and X
2
,
remains constant for all motions. Mathematically, this is equivalent to saying that
x
1
−x
2
 = X
1
−X
2
 . (7.1)
Second, the motion of the rigid body preserves orientations. In 1775, Euler [55, 56]
showed that the motion of a body that satisﬁes (7.1) is such that
x
1
−x
2
= Q(X
1
−X
2
) , (7.2)
where Q is a rotation tensor. As you may recall, this rotation tensor has an associ
ated axis and angle of rotation.
We can use (7.2) to tell if the rotation of a rigid body is nontrivial. That is, if
Q = I, we can pick two points X
1
and X
2
of the rigid body and examine how the
relative position vector x
1
−x
2
varies as a function of time. If, for all choices of X
1
and X
2
, the relative position vector is unaltered in direction, then Q = I; otherwise
the rigid body is rotating. This is, perhaps, the most useful interpretation of Q: the
transformation that takes vectors between two points in the body and transforms
them into their present state.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
208 Kinematics of Rigid Bodies
If we assume that one point of the body is ﬁxed then we can simplify (7.2) by
choosing the ﬁxed point to be the origin:
x(t) = Q(t)X. (7.3)
We can then infer Euler’s theorem on the motion of a rigid body:
Every motion of a rigid body about a ﬁxed point is a rotation about an axis through the
ﬁxed point.
The axis here is the axis of rotation of Q(t). Because the motion of the body in ques
tion is from the conﬁguration κ
0
to κ
t
, this axis depends on the choice of reference
conﬁguration.
We can arrive at an alternative, and more common, interpretation of Euler’s
theorem that does not feature the reference conﬁguration κ
0
. To do this, we again
consider the motion of the body with a ﬁxed point during the interval t ∈ [t
0
, t]. We
ﬁnd from (7.3) that
x (t) = Q(t) Q
T
(t
0
) x (t
0
) .
Thus the motion of the body at the end of the time interval is characterized by
the rotation tensor Q(t) Q
T
(t
0
).
∗
Invoking Euler’s theorem, the axis of rotation of
Q(t) Q
T
(t
0
) is the axis of rotation for the motion of the rigid body. It is emphasized
that, during the motion in question, the body’s present conﬁguration changes from
κ
t
0
to κ
t
.
Representations for the General Motion
A general motion of a rigid body is one in which the body may not have a ﬁxed
point. In this case, it is easy to argue that a uniform translation of the reference con
ﬁguration can be imposed on the rigid body so that an arbitrary one of its material
points X
P
is placed at its location x
P
(t) in the present conﬁguration. The rigid body
is then rotated about X
P
so as to occupy its present conﬁguration κ
t
:
x(t) −x
P
(t) = Q(t) (X−X
P
) .
That is,
x = Q(t)X+d(t), (7.4)
where d(t) = x
P
(t) −Q(t)X
P
. In words, (7.4) states that the most general motion of
a rigid body is a translation and a rotation.
†
It is one of the general representations
of the rigid body motion that we will often use in our subsequent discussions.
∗
In general, the tensor Q(t) Q
T
(t
0
) has an axis of rotation and an angle of rotation that differ from
those associated with Q(t).
†
It should not come as a surprise that this result was also known to Euler. See his remarks in the
introductory sections to [52, 54].
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
7.2 The Motion of a Rigid Body 209
κ
0
κ
t
κ
t
0
O
X
X
X
X
x (t)
x (t
0
)
Q(t)
Q(t
0
)
Q(t) Q
T
(t
0
)
Figure 7.2. Schematic of the conﬁgurations κ
t
0
and κ
t
of a rigid body. The reference conﬁgu
ration κ
0
and illustrations of the roles played by several rotation tensors are also shown.
To discuss an alternative to (7.4) that does not feature the reference conﬁgura
tion, we consider the motion of the body during a time interval [t
0
, t] (see Figure 7.2).
With the help of (7.4), we ﬁnd that
x (t
0
) = Q(t
0
) X+d(t
0
) , x (t) = Q(t) X+d(t) .
Combining these equations, we arrive at an alternative representation of (7.4):
x (t) = Q(t) Q
T
(t
0
) x (t
0
) +z (t) , (7.5)
where z (t) = d(t) −Q
T
(t
0
) d(t
0
).
∗
Result (7.5) is a convenient departure point to discuss a third alternative repre
sentation of rigid body motion. This representation of rigid body motion is synony
mous with a famous theorem, credited to Michel Chasles (1793–1880), on this topic
(see Section 5 of [228] and references therein). Here, the motion of a rigid body
is decomposed into a screw motion (see Figure 7.3). That is, the motion is consid
ered to be a rotation through an angle θ about an axis s(t) followed by a translation
σ(t)s(t) along that axis. The screw axis s(t) is the axis of rotation of Q(t) Q
T
(t
0
), and
φ(t) is this tensor’s angle of rotation. The translational component σs(t) and the lo
cation ρ of the intercept of the screw axis can in principle be determined from the
three components of z(t) by use of three equations:
(I −Q(t)) ρ(t) +σ(t)s(t) = z(t). (7.6)
∗
You may notice that we can choose the reference conﬁguration κ
0
to be identical to κ
t
0
. Then,
Q(t
0
) = I, and representations (7.4) and (7.5) will be identical.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
210 Kinematics of Rigid Bodies
κ
t
κ
t
0
φ(t)
s(t)
O
X
X
x (t)
x (t
0
)
ρ
1
ρ
2
Q(t) Q
T
(t
0
)
Figure 7.3. Schematic of the conﬁgurations κ
t
0
and κ
t
of a rigid body showing the screw axis
s(t) and angle of rotation φ(t). Here, the rigid body is translated along the screw axis by an
amount σs and rotated about s(t) through an angle φ. Two possible choices of ρ(t) are also
shown. For the ﬁrst, ρ(t) = ρ
1
= ρ
1
E
1
+ρ
2
E
2
is chosen to be the intercept of the screw axis
with the E
1
−E
2
plane, and, for the second, ρ(t) = ρ
2
, where ρ
2
is the vector from the origin
that intersects the screw axis at a right angle: ρ
2
· s = 0.
However, (I −Q(t)) is noninvertible,
∗
and so ρ(t) is not uniquely deﬁned. As a
result, several choices of ρ can be found in the literature (see, for example, the two
choices shown in Figure 7.3). Choosing ρ(t) to be normal to s leads to the following
solutions for σ(t) and ρ(t):
ρ(t) =
1
2
z
⊥
(t) +cot
φ(t)
2
s(t) ×z(t)
, σ(t) = s(t) · z(t), (7.7)
where
z
⊥
(t) = z(t) −(z(t) · s(t)) s(t).
It is left as an exercise for the reader to verify (7.7). You should notice that a special
case of the solution occurs when Q = I and ρ is indeterminate.
∗
There are many ways to see this result. The ﬁrst is to refer the reader to the derivation of (6.15).
Alternatively, one can use Euler representation (6.8) for a rotation tensor and observe that s is the
zero eigenvalue of (I −Q(t)).
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
7.3 The Angular Velocity and Angular Acceleration Vectors 211
7.3 The Angular Velocity and Angular Acceleration Vectors
The motion of a rigid body is deﬁned by (7.4) or, equivalently, by (7.5). Because Q is
a rotation tensor, we can deﬁne an angular velocity tensor and an angular velocity
vector ω
∗
:
=
˙
QQ
T
, ω = −
1
2
¸
˙
QQ
T
¸
.
The vector ω is the angular velocity vector of the rigid body, and is the angular ve
locity tensor of the rigid body. As we know, the rotation tensor Qcan be represented
in a variety of manners, for instance, Euler angles or the Euler representation, and
so too can its angular velocity vector. However, here it is convenient to omit explicit
mention of these representations.
By differentiating the angular velocity vector, we ﬁnd the angular acceleration
vector of the rigid body:
α = ˙ ω.
You should notice that
α = −
1
2
¸
¨
QQ
T
+
˙
Q
˙
Q
T
¸
= −
1
2
[
˙
],
where we used the fact that ˙ = O.
We can use result (7.2) to determine the relative velocity and acceleration vec
tors of any two points X
1
and X
2
of the rigid body:
v
1
−v
2
= ˙ x
1
− ˙ x
2
=
˙
Q(X
1
−X
2
)
=
˙
QQ
T
Q(X
1
−X
2
)
= Q(X
1
−X
2
)
= ω ×(x
1
−x
2
) .
A further differentiation and some manipulations give the relative acceleration
vectors:
a
1
−a
2
= ˙ v
1
− ˙ v
2
= ˙ ω ×(x
1
−x
2
) +ω ×( ˙ x
1
− ˙ x
2
)
= α ×(x
1
−x
2
) +ω ×(v
1
−v
2
) .
The ﬁnal forms of the relative velocity and acceleration vectors are expressed as
functions of t, x
1
, and x
2
. They can also be expressed as functions of t, X
1
, and X
2
.
∗
As
˙
Q
T
(t
0
) = 0, ω is also the angular velocity vector associated with the rotation tensor Q(t)Q
T
(t
0
).
Thus representations (7.4) and (7.5) have the same angular velocity vectors.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
212 Kinematics of Rigid Bodies
κ
0
κ
t
O
X
1
X
1
X
2
X
2
X
3
X
3
E
1
E
2
E
3
e
1
e
2
e
3
Figure 7.4. The corotational basis {e
1
, e
2
, e
3
} and the ﬁxed Cartesian basis {E
1
, E
2
, E
3
}.
7.4 A Corotational Basis
It is convenient, when discussing the dynamics of rigid bodies, to introduce another
basis {e
1
, e
2
, e
3
}, which is known as a corotational basis.
∗
Here, we deﬁne such a
basis and point out some features of its use. Our discussion of the corotational basis
follows Casey [26] with some minor changes.
As is well known in the kinematics of rigid bodies, knowledge of the position
vectors of three material points sufﬁces to determine the motion of the rigid body.
Indeed, this is the premise for many navigation schemes and is the motivation
for our construction of a corotational basis. Referring to Figure 7.4, we start by
picking three material points X
1
, X
2
, and X
3
of the body. These points are chosen
such that the orthonormal vectors E
1
and E
2
point from X
3
toward X
1
and X
2
,
respectively:
E
1
X
1
−X
3
, E
2
X
2
−X
3
.
We then complete the (ﬁxed) righthanded Cartesian basis by deﬁning
E
3
= E
1
×E
2
.
Now consider the present locations of the three material points. Because Q
preserves lengths and orientations, the two vectors x
1
−x
3
, x
2
−x
3
will retain their
relative orientation. As a result, using (7.2), we deﬁne two orthonormal members of
a corotational basis by choosing themto point fromx
3
toward x
1
and x
2
, respectively:
e
1
x
1
−x
3
, e
2
x
2
−x
3
.
We then deﬁne e
3
= e
1
×e
2
. As mentioned earlier, the basis {e
1
, e
2
, e
3
} is known as
the corotational basis.
It should be transparent that
Q = e
1
⊗E
1
+e
2
⊗E
2
+e
3
⊗E
3
.
∗
This basis is often referred to as a bodyﬁxed frame or an embedded frame.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
7.5 Three Distinct Axes of Rotation 213
This result follows from (7.2) and our previous discussions on representations of
rotation tensors. Because the corotational basis moves with the body, we can use
our previous results for relative velocities and accelerations to see that
˙ e
i
= ω ×e
i
, ¨ e
i
= α ×e
i
+ω ×(ω ×e
i
) ,
where i = 1, 2, 3.
As the corotational basis is a basis for E
3
, for any vector r we have the represen
tation
r =
3
¸
i=1
r
i
e
i
.
When the components r
i
are constant, then the vector is known as a corotational
vector. The most trivial examples of corotational vectors are e
1
, e
2
, and e
3
. The time
derivative ˙ r of r has the representations
v = ˙ r
=
3
¸
i=1
˙ r
i
e
i
+
3
¸
i=1
r
i
˙ e
i
=
o
r +ω ×r,
where
o
r is the corotational derivative (with respect to Q) of the vector r. A related
expression can be obtained for ¨ r:
a = ˙ v
=
3
¸
i=1
¨ r
i
e
i
+2
3
¸
i=1
˙ r
i
˙ e
i
+
3
¸
i=1
r
i
¨ e
i
=
3
¸
i=1
¨ r
i
e
i
+2ω ×
o
r +ω ×(ω ×r) +α ×r.
The presence of the Coriolis acceleration 2ω ×
o
r in the expression for a arises
because we have chosen to express r in a basis that is not ﬁxed. You should also
observe that, if r is a corotational vector, then
o
r = 0, and the Coriolis acceleration
vanishes.
7.5 Three Distinct Axes of Rotation
It is possible to deﬁne three distinct axes of rotation for a rigid body. These axes are
commonly used in mechanics and navigation, and to discuss them it is convenient to
recall the representations of rigid body motion (7.4) and (7.5). The rotation tensor
Q(t) associated with the former has an axis of rotation q(t) and an angle of rotation
θ(t), and the screw axis s(t) and angle φ(t) are the axis and angle of rotation, respec
tively, of Q(t) Q
T
(t
0
). A third axis of rotation, which is known as the instantaneous
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
214 Kinematics of Rigid Bodies
ω
r
q
r
r
⊥
Figure 7.5. An example of r(t) for a motion of a rigid body for which ω
is constant and r is a corotational vector. In this ﬁgure, r
⊥
· ω = 0 and
r
⊥
+r
= r. The evolution of a possible axis of rotation q(t) for Q(t) is
also shown.
axis of rotation i, can also be deﬁned. This axis is a unit vector parallel to the angular
velocity vector ω
∗
:
i(t) =
ω
ω
.
Except in the simple case in which ω is constant, i does not have an associated angle
of rotation. The terminology “instantaneous axis” can be appreciated from the ob
servation that ˙ r = ω ×r for any corotational vector. Thus, if ω is constant, then, as
shown in Figure 7.5, r(t) will appear to rotate about i.
Our deﬁnition of the instantaneous axis of rotation is identical to that used in
classical works on rigid body dynamics, for example, Poinsot [172] and Sections 405–
406 of Poisson [173]. It is not universally adapted. For instance, in the literature on
kinematics of anatomical joints the terminology instantaneous axis of rotation often
refers to s and not i.
†
In general, the axes q, s, and i are not identical. However, (6.14) can be used to
relate the axes and two angles of rotation:
ω = ω i
=
˙
φs +sin(φ)˙ s +(1 −cos(φ))s × ˙ s
=
˙
θq +sin(θ) ˙ q +(1 −cos(θ))q × ˙ q. (7.8)
In addition, Rodrigues formula (6.49) can be used to relate q, s, φ, and θ, but this is
left as an exercise.
In the course of examining the rotation tensors from various problems in rigid
body dynamics, you can numerically compute s, q, and i. There you will easily ﬁnd
examples in which these axes are distinct. It is, however, also of interest to con
sider examples in which some of these axes are equal. Two such examples are now
presented. First, suppose that the body’s rotation tensor describes a steady rotation
∗
You may wish to recall that the angular velocity vector associated with Q(t) and Q(t) Q
T
(t
0
) are
identical.
†
The interested reader is referred to Woltring [232] and Woltring et al. [233] for further discussion
on this matter.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
7.6 The Center of Mass and Linear Momentum 215
0
0
1
π
2π
t
T
θ(t)
ν(t) −ν
0
i = s = E
3
E
1
E
2
e
1
e
2
e
3
q
Figure 7.6. Plots of the loci of the extremities of the corotational vectors e
i
(t) and the axes
of rotation s, q, and i for the rotation tensor (7.9). A plot of ν(t) is also shown in this ﬁgure
where T =
2π
˙ ν
0
.
about E
3
through an angle ν, where ˙ ν = ˙ ν
0
is constant: Q(t) = L( ˙ ν
0
t, E
3
). In this
case, it is easy to compute that s(t) = i(t) = q(t) = E
3
and ω = ˙ ν
0
E
3
.
Now consider an example in which Q describes a rotation about a timevarying
axis of rotation at constant speed
∗
:
Q(t) = 2q(t) ⊗q(t) −I, (7.9)
where the axis of rotation of Q(t) is
q(t) = cos
ν
2
E
1
+sin
ν
2
E
2
, ν(t) = ˙ ν
0
(t −t
0
) +ν
0
,
and ˙ ν
0
and ν
0
are constants. You should notice that, with the possible help of (6.14),
ω = ˙ ν
0
E
3
, and that the angle of rotation θ for this rotation tensor is π. A standard
calculation also reveals that
Q(t)Q
T
(t
0
) = L(−ν(t) +ν
0
, e
3
) .
That is, the rotation tensor Q(t)Q
T
(t
0
) corresponds to the familiar rotation about
E
3
through an angle ν(t) −ν
0
. It now follows that s = q. Indeed, (7.9) is the simplest
example of a rotation where ω is constant but the axis q is not parallel to ω that we
know of. The temporal behavior of e
i
(t) = QE
i
, the axes of rotation, and the angles
of rotation θ(t) and ν(t) are shown in Figure 7.6.
7.6 The Center of Mass and Linear Momentum
It is convenient (and traditional) to deﬁne a special material point of the rigid body,
which we refer to as the center of mass
¯
X. We then use the velocity vector ¯ v of this
point in the present conﬁguration to deﬁne a very useful expression for the linear
momentum G of the rigid body.
∗
This example is adapted from [161]. Other examples of rotations with constant angular velocity
vectors but distinct axes s and q can be found in [161].
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
216 Kinematics of Rigid Bodies
The Center of Mass
The position vectors of the center of mass of the body in its reference and present
conﬁgurations are deﬁned by
¯
X =
R
0
Xρ
0
dV
R
0
ρ
0
dV
, ¯ x =
R
xρdv
R
ρdv
, (7.10)
where ρ
0
= ρ
0
(X) and ρ = ρ(x, t) are the mass densities per unit volume of the body
in κ
0
and κ
t
, respectively. The regions R
0
and R denote the regions of E
3
occupied
by the body in κ
0
and κ
t
, respectively. If a body is homogeneous, then ρ
0
is a constant
that is independent of X.
The principle of mass conservation states that the mass of the body is conserved.
That is,
dm = ρ
0
dV = ρdv,
or, equivalently,
m =
R
0
ρ
0
dV =
R
ρdv.
It follows immediately from (7.10) that
m
¯
X =
R
0
Xρ
0
dV, m¯ x =
R
xρdv.
We can also ﬁnd from these results that
0 =
R
0
(X−
¯
X)ρ
0
dV, 0 =
R
(x − ¯ x)ρdv.
These identities play key roles in establishing expressions for momenta and energies
of a rigid body.
A special feature of rigid bodies is that the center of mass behaves as if it
were a material point, which we denote by
¯
X. To see this we follow [26] and
consider
m¯ x =
R
xρdv
=
R
(QX+d) ρdv
=
R
0
(QX+d) ρ
0
dV
= Q
R
0
Xρ
0
dV
+
R
0
ρ
0
dV
d
= mQ
¯
X+md.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
7.6 The Center of Mass and Linear Momentum 217
Consequently,
¯ x = Q
¯
X+d. (7.11)
Recalling (7.4), this implies that the center of mass of the rigid body behaves as if it
were a material point of the rigid body. For many bodies, such as a rigid homoge
neous sphere, the center of mass corresponds to the geometric center of the sphere,
whereas for others, such as a rigid circular ring, it does not correspond to a material
point.
It is convenient in many treatments of rigid bodies to deﬁne a reference frame
consisting of the center of mass
¯
X and the corotational basis {e
1
, e
2
, e
3
}. Such a
frame is known as the corotational reference frame. In the treatment presented in
this book, we often express the relative position vectors of particles x − ¯ x in this
frame.
Linear Momentum
By deﬁnition, the linear momentum G of a rigid body is
G =
R
vρdv.
That is, the linear momentum of a rigid body is the sum of the linear momenta of
its constituents. Using the center of mass, we can establish an alternative expression
for G with the help of the deﬁnitions of m and ¯ x:
G = m¯ v,
where ¯ v =
˙
¯ x is the velocity vector of the center of mass. You may recall that a related
result holds for a (ﬁnite) system of particles.
Relative Position Vectors
For a material point Xof a rigid body, it is convenient to deﬁne the relative position
vectors π and :
π = x − ¯ x, = X−
¯
X.
Representative examples of these vectors are displayed in Figure 7.1. With the as
sistance of (7.2) and (7.11), we see that
π = Q. (7.12)
Using the corotational basis, we also easily see that
π · e
i
= · E
i
.
This implies that the relative position vectors have the representations
=
3
¸
i=1
i
E
i
, π =
3
¸
i=1
i
e
i
. (7.13)
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
218 Kinematics of Rigid Bodies
Furthermore, the corotational derivative (with respect to Q) of π is zero: ˙ π = ω ×π.
That is, π is a corotational vector.
7.7 Angular Momenta
Angular momenta of a rigid body are its most important distinctive feature when
compared with a particle. In particular, the angular momentum relative to two
points, the center of mass
¯
X and a ﬁxed point O, is of considerable importance. For
convenience, we assume that the ﬁxed point Ois also the origin (see Figure 7.1).
By deﬁnition, the angular momenta of a rigid body relative to its center of mass
¯
X, H, a ﬁxed point O, H
O
, and a point A, H
A
, are
H =
R
(x − ¯ x) ×vρdv,
H
O
=
R
x ×vρdv,
H
A
=
R
(x −x
A
) ×vρdv.
The position vectors in these expressions are relative to the ﬁxed point O, and x
A
is the position vector of the point A. You should notice that the velocity vector in
these expressions is the absolute velocity vector.
The aforementioned angular momenta are related by simple and important for
mulae. To ﬁnd one of these formulae, we perform some manipulations on H
O
:
H
O
=
R
x ×vρdv
=
R
(π + ¯ x) ×vρdv
=
R
π ×vρdv +
R
¯ x ×vρdv
= H+ ¯ x ×
R
vρdv.
That is,
H
O
= H+ ¯ x ×G. (7.14)
This relation states that the angular momentum of a rigid body relative to a ﬁxed
point O is the sum of the angular momentum of the rigid body about its center of
mass and the angular momentum of its center of mass relative to O. Paralleling the
establishment of (7.14), it can also be shown that
H
A
= H+(¯ x −x
A
) ×G, H
O
= H
A
+x
A
×G. (7.15)
These results have obvious similarities to (7.14).
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
7.8 Euler Tensors and Inertia Tensors 219
7.8 Euler Tensors and Inertia Tensors
To use the balance laws for a rigid body it is convenient to consider some further
developments of the angular momentum H. These developments are considerably
aided by use of the Euler tensors E
0
and E and the inertia tensors J
0
and J.
Euler Tensors
We next deﬁne the Euler tensors (relative to the center of mass of the rigid body):
E
0
=
R
0
⊗ρ
0
dV, E =
R
π ⊗πρdv.
You should notice that E and E
0
are symmetric.
Using mass conservation, (7.12), and the identity (Aa) ⊗(Bb) = A(a ⊗b)B
T
,
we easily see that
E = QE
0
Q
T
.
Furthermore,
(Ee
i
) · e
k
= (E
0
E
i
) · E
k
.
This implies that E has the representation
E =
3
¸
i=1
3
¸
k=1
E
ik
e
i
⊗e
k
,
where E
ik
= (E
0
E
k
) · E
i
are the constant components of E
0
. In other words, al
though E is a function of time, its components, relative to the corotational basis,
are constant. Furthermore, these constants are identical to the components of the
constant Euler tensor E
0
.
Inertia Tensors
The inertia tensors J
0
and J can be deﬁned by use of the Euler tensors:
J
0
= tr(E
0
)I −E
0
, J = tr(E)I −E.
Using the deﬁnitions of the Euler tensors and the identity tr(a ⊗b) = a · b, we can
restate these deﬁnitions as
J
0
=
R
0
((· )I −⊗) ρ
0
dV, J =
R
((π · π)I −π ⊗π) ρdv. (7.16)
It is easy to see that
J = QJ
0
Q
T
, J
T
0
= J
0
, J
T
= J.
The ﬁrst of these results implies that
J =
3
¸
i=1
3
¸
k=1
J
ik
e
i
⊗e
k
,
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
220 Kinematics of Rigid Bodies
where J
ki
= (J
0
E
i
) · E
k
are the constant components of J
0
.
∗
The symmetry of the
inertia tensors also implies that J
ki
= J
ik
.
Additional Relationships
Another set of interesting results follows by inverting the relationships between the
Euler and inertia tensors. To do this, we use the deﬁnition of the inertia tensor in
terms of the Euler tensor and the fact that tr(I) = 3:
E
0
=
1
2
tr(J
0
)I −J
0
, E =
1
2
tr(J)I −J. (7.17)
These results are useful for obtaining the Euler tensors from tabulations of the in
ertia tensor that are found in numerous undergraduate textbooks on dynamics.
It is a good exercise to substitute (7.13) into (7.16) to see that the components of
J
0
relative to the basis {E
1
, E
2
, E
3
} are volume integrals involving quadratic powers
of the components of . For example,
J
011
= (J
0
E
1
) · E
1
=
R
0
(y
2
+z
2
)ρ
0
dV,
J
012
= (J
0
E
2
) · E
1
= −
R
0
xyρ
0
dV,
J
013
= (J
0
E
3
) · E
1
= −
R
0
xzρ
0
dV, (7.18)
where
x = π · e
1
= · E
1
, y = π · e
2
= · E
2
, z = π · e
3
= · E
3
.
A similar exercise with the components of E
0
shows that
E
011
= (E
0
E
1
) · E
1
=
R
0
x
2
ρ
0
dV,
E
012
= (E
0
E
2
) · E
1
=
R
0
xyρ
0
dV,
E
013
= (E
0
E
3
) · E
1
=
R
0
xzρ
0
dV.
You should notice the simple relationship between the offdiagonal components of
E
0
and J
0
. Again, one can then use tables of inertias found in textbooks to determine
the components of J
0
and J.
Both inertia tensors J
0
and J are symmetric. It can also be shown that they are
positivedeﬁnite. This allows us to choose {E
1
, E
2
, E
3
} such that these vectors are
the eigenvectors of J
0
, and, consequently,
J
0
= λ
1
E
1
⊗E
1
+λ
2
E
2
⊗E
2
+λ
3
E
3
⊗E
3
.
∗
Not surprisingly, this is similar to the situation we encountered earlier with the Euler tensors E and
E
0
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
7.8 Euler Tensors and Inertia Tensors 221
Here, λ
i
are known as the principal moments of inertia. As J = QJ
0
Q
T
and e
i
=
QE
i
, we also have
J = λ
1
e
1
⊗e
1
+λ
2
e
2
⊗e
2
+λ
3
e
3
⊗e
3
.
It is common to refer to e
i
as the principal axes of the rigid body. Now because of
the deﬁnition of the components J
0ik
of the inertia tensor, the principal moments
need to satisfy certain inequalities:
λ
1
+λ
2
> λ
3
, λ
2
+λ
3
> λ
1
, λ
3
+λ
1
> λ
2
.
These inequalities are easy to establish by use of deﬁnitions (7.18) and are useful in
selecting representative examples of inertias. For instance, (7.19) imply that a rigid
body with λ
1
= 1, λ
2
= 2, and λ
3
= 3 is not physically realizable.
It can also be shown that the eigenvectors of J
0
are the eigenvectors of E
0
.
Consequently, if we choose {E
1
, E
2
, E
3
} to be the eigenvectors of J
0
, then
E
0
= e
1
E
1
⊗E
1
+e
2
E
2
⊗E
2
+e
3
E
3
⊗E
3
.
By use of identity (7.17)
1
, the constants e
i
can be related to the principal moments
of inertia:
e
1
=
1
2
(λ
2
+λ
3
−λ
1
) , e
2
=
1
2
(λ
1
+λ
3
−λ
2
) ,
e
3
=
1
2
(λ
1
+λ
2
−λ
3
) . (7.19)
As E = QE
0
Q
T
and e
i
= QE
i
, we also have
E = e
1
e
1
⊗e
1
+e
2
e
2
⊗e
2
+e
3
e
3
⊗e
3
.
Notice that the principal axis e
j
corresponding to the maximum value of λ
i
corre
sponds to the minimum value of e
i
.
Some Examples
In many studies of satellites, it is convenient to model the satellite as a set of con
nected dumbbells. For such systems, the integrals in the deﬁnition of the Euler and
inertia tensors degenerate into summations over a discrete number of particles. For
instance, an example of such an arrangement is shown in Figure 7.7. For the body
shown in this ﬁgure, it is easy to calculate E
0
:
E
0
=
3
¸
k=1
2m
k
L
2
k
E
k
⊗E
k
.
Consequently, the inertia tensor is
J
0
= 2
m
2
L
2
2
+m
3
L
2
3
E
1
⊗E
1
+2
m
1
L
2
1
+m
3
L
2
3
E
2
⊗E
2
+2
m
1
L
2
1
+m
2
L
2
2
E
3
⊗E
3
.
It is left as an exercise to write E and J.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
222 Kinematics of Rigid Bodies
O
¯
X
m
1
m
1
m
2
m
2
m
3
m
3
L
1
L
2
L
3
E
1
E
2
E
3
Figure 7.7. The reference conﬁguration of a dumbell satellite consisting of six masses joined
to a center by three massless, rigid rods of lengths 2L
1
, 2L
2
, 2L
3
.
The simplest inertia tensor arises when the body is a homogeneous sphere of
radius R or a homogeneous cube of length a:
J = J
0
=
2mR
2
5
I, J = J
0
=
ma
2
6
I,
respectively. For these bodies, any three mutually perpendicular unit vectors are
principal axes.
The next class of bodies is for those with an axis of symmetry. For instance,
a homogeneous circular rod of length L and radius R has a moment of inertia
tensor
J
0
=
mR
2
2
E
3
⊗E
3
+
mR
2
4
+
mL
2
12
(I −E
3
⊗E
3
) ,
where E
3
is the axis of symmetry of the circular rod in its reference conﬁguration.
Most bodies, however, do not have an axis of symmetry. Consider the homo
geneous ellipsoid shown in Figure 7.8. The equation for the lateral surface of the
ellipsoid is
x
2
a
2
+
y
2
b
2
+
z
2
c
2
= 1.
¯
X
Ellipsoid
E
1
E
2
E
3
Figure 7.8. An ellipsoid of mass m.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
7.9 Angular Momentum and an Inertia Tensor 223
The inertia tensor of the ellipsoid is
J
0
=
m
5
b
2
+c
2
E
1
⊗E
1
+
m
5
a
2
+c
2
E
2
⊗E
2
+
m
5
a
2
+b
2
E
3
⊗E
3
.
It is left as an exercise to write the Euler tensor E
0
for the ellipsoid.
∗
It is crucial to note that for all of the preceding examples we have used the
property that any body has three principal axes. Writing the inertia and Euler ten
sors with respect to these axes provides their simplest possible representations.
7.9 Angular Momentum and an Inertia Tensor
The result we now wish to establish is that H = Jω. This is arguably one of the most
important results in rigid body dynamics. In particular, with the assistance of J
0
, for
a particular rigid body it allows us to write a tractable expression for H.
We now reconsider the angular momentum H,
H =
R
(x − ¯ x) ×vρdv
=
R
π ×vρdv
=
R
π ×(¯ v +ω ×π)ρdv
=
R
π × ¯ vρdv +
R
π ×(ω ×π)ρdv.
However, because
¯
X is the center of mass and the velocity vector ¯ v of
¯
X is indepen
dent of the region of integration, we can take ¯ v outside the integral:
H =
R
π × ¯ vρdv +
R
π ×(ω ×π)ρdv
=
R
πρdv × ¯ v +
R
π ×(ω ×π)ρdv
= 0 × ¯ v +
R
π ×(ω ×π)ρdv
=
R
π ×(ω ×π)ρdv.
Notice that we also used the identity
R
πρdv = 0 in the nexttolast step of this
calculation. Summarizing, we have
H =
R
π ×(ω ×π)ρdv =
R
((π · π)ω −(π · ω)π) ρdv. (7.20)
∗
For assistance, see (7.19).
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
224 Kinematics of Rigid Bodies
In writing this equation, we used the identity a ×(b ×c) = (a · c)b −(a · b)c. By use
of the deﬁnition of the inertia tensor, J, it should now be apparent that
H = Jω.
As mentioned earlier, this is one of the most important results in the kinematics of
rigid bodies.
You should notice that H = Jω implies that there is a linear transformation be
tween angular velocity and angular momentum. Furthermore, unless ω is an eigen
vector of J, H and ω will not be parallel.
7.10 Kinetic Energy
The kinetic energy of a rigid body has a very convenient representation. This repre
sentation, which was ﬁrst established by a contemporary of Euler, Johann S. Koenig
(1712–1757), is known as the Koenig decomposition:
T =
1
2
m¯ v · ¯ v +
1
2
(Jω) · ω. (7.21)
Here, a derivation of this result is given.
The kinetic energy T of a rigid body is deﬁned to be
T =
1
2
R
v · vρdv.
We can simplify this expression for the energy by expressing the velocity vector v as
v = ¯ v +ω ×π.
Substituting this expression into T, we have
T =
1
2
R
v · vρdv
=
1
2
R
¯ v · ¯ vρdv +
R
(ω ×π) · ¯ vρdv +
1
2
R
(ω ×π) · (ω ×π)ρdv.
However, we have the following identities
R
¯ v · ¯ vρdv = ¯ v · ¯ v
R
ρdv = m¯ v · ¯ v,
R
(ω ×π) · ¯ vρdv =
ω ×
R
πρdv
· ¯ v = (ω ×0) · ¯ v = 0,
R
(ω ×π) · (ω ×π)ρdv =
R
(ω · ω)(π · π) −(ω · π)
2
ρdv
= ω ·
R
(π · π)I −π ⊗πρdv
ω
= ω · (Jω) .
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
7.10 Kinetic Energy 225
Substituting these results into the previous expression for T, we ﬁnd the desired
result:
T =
1
2
m¯ v · ¯ v +
1
2
(Jω) · ω.
This result is known as the Koenig decomposition of the kinetic energy of a rigid
body. In words, the kinetic energy of a rigid body is equal to the sum of the kinetic
energy of its center of mass and the rotational kinetic energy T
rot
=
1
2
ω · (Jω) of the
rigid body.
Comments on the Rotational Kinetic Energy
The previous development of the Koenig decomposition showed that 2T
rot
= ω · Jω.
This representation is used in the vast majority of works on rigid body dynamics. An
equivalent, complementary representation using J
0
can also be found,
∗
and we now
discuss this representation.
Consider the angular velocity vector ω
0
:
ω
0
= Q
T
ω.
If ω =
¸
3
i=1
ω
i
e
i
, then it follows that
ω
0
=
3
¸
i=1
ω
i
E
i
.
In addition, it can also be shown that ω
0
is the axial vector of
˙
Q
T
Q:
ω
0
= −
1
2
[Q
T
˙
Q] = −
1
2
[Q
T
Q].
It is a good exercise to see what the representation of ω
0
is when the 3–2–1 or 3–1–3
Euler angles are used.
We now use the relationship between J and J
0
to see that
2T
rot
= ω · (Jω)
= Qω
0
· (JQω
0
)
= ω
0
·
Q
T
JQω
0
= ω
0
· (J
0
ω
0
) .
This is the ﬁnal desired result:
T
rot
=
1
2
ω · Jω =
1
2
ω
0
· J
0
ω
0
.
The advantage of the representation involving ω
0
is that J
0
is a constant. Thus, when
taking derivatives of T
rot
with respect to the parameters used for Q, we need to
consider only the derivatives of ω
0
.
∗
See, for example, [28] and Chapter 15 of [138].
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
226 Exercise 7.1
7.11 Concluding Remarks
We have now assembled most of the kinematical quantities required for character
izing the motion of a rigid body. The precise representations for these quantities
that are used generally depend on the problem at hand. For instance, we shall later
use a set of Euler angles for a spinningtop problem that differs from a set used for a
problem featuring a satellite. Another example arises when we examine the dynam
ics of a rolling sphere. There, we will choose to use a ﬁxed basis representation of
its angular velocity vector, ω =
¸
3
i=1
i
E
i
, whereas we will use the Euler basis for
the satellite problem. These choices are guided by experience and are often not in
tuitively obvious. It is hoped that the examples in the exercises and chapters ahead
will help you gain this needed experience. You might also have noticed that we have
yet to discuss constraints on the motions of rigid bodies. The next chapter is devoted
to this topic.
EXERCISES
7.1. Recall that the rotation tensor Q of a rigid body has the representations
Q =
3
¸
i=1
e
i
⊗E
i
=
3
¸
i=1
3
¸
k=1
Q
ik
e
i
⊗e
k
=
3
¸
i=1
3
¸
k=1
Q
ik
E
i
⊗E
k
.
We recall that the corotational rates of a vector a and a tensor A relative to the
rotation tensor Q are deﬁned as
o
a = ˙ a −ω ×a,
o
A
=
˙
A−A+A.
In these expressions, =
˙
QQ
T
and ω = −
1
2
[
˙
QQ
T
].
(a) If a
i
= a · e
i
and A
ik
= (Ae
k
) · e
i
, then show that
o
a =
3
¸
i=1
˙ a
i
e
i
,
o
A
=
3
¸
i=1
3
¸
k=1
˙
A
ik
e
i
⊗e
k
.
Give physical interpretations of these results.
(b) A vector z is said to be corotational if z = QZ, where Z is constant. Give
examples of such vectors from rigid body dynamics and show that the coro
tational rate relative to Q of a corotational vector is 0.
(c) Construct a deﬁnition of a corotational tensor and give two prominent ex
amples of such tensors from rigid body dynamics.
(d) Establish the following identities:
˙ ω =
o
ω,
˙
Q = Q
T
o
QQ.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
Exercises 7.1–7.2 227
(e) Suppose a rigid body is in motion and the corotational rate of ω is 0. What
is the angular acceleration of the rigid body and what is the rotation tensor
of the rigid body?
∗
7.2. As shown in Figure 7.9, a robotic arm can be used to move a payload. The robot
consists of
1. a drive shaft that rotates about the E
3
axis through an angle ψ,
2. an axle Athat rotates about g
2
through an angle θ relative to the drive shaft,
3. an arm that rotates about g
3
through an angle φ relative to axle A.
The payload is rigidly attached to the robotic arm. In this question, the drive shaft,
the axle A, the robotic arm, and the payload are assumed to be rigid.
P
O
Axle
Drive shaft
Robotic arm
Payload
ψ
θ
φ
E
1
E
2
E
3
g
2
g
3
Figure 7.9. Schematic of a robot consisting
of a robotic arm, drive shaft, and axle. The
motors used to actuate the robot are not
shown.
(a) What are the angular velocity vectors of the drive shaft, the axle A, the
robotic arm, and the payload?
(b) Which set of Euler angles is being used to parameterize the rotation tensor
Q of the payload?
(c) If the position vector r of a point X on the payload is
r = HE
3
+Lg
3
,
then establish an expression for the velocity vector v of the point X.
(d) Suppose after a time interval t
1
−t
0
the point X of has returned to its origi
nal location in space:
r(t
1
) = r(t
0
).
Show that the payload will have rotated through an angle φ(t
1
) −φ(t
0
)
about the axis g
3
(t
0
) during this interval of time. In other words, the
∗
A solution to this problem can be found in O’Reilly and Payen [161].
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
228 Exercises 7.2–7.3
rotation tensor Q(t
1
)Q
T
(t
0
) corresponds to a rotation of φ(t
1
) −φ(t
0
) about
g
3
(t
0
).
7.3. Consider the circular disk shown in Figure 7.10. The motion of the disk is given
by the position vector y of an arbitrary material point Yof the disk and the rotation
tensor Q of the rigid disk.
g
P
O
¯
X
ψ
φ
E
1
E
2
E
3
e
1
e
1
e
2
e
1
e
2
Figure 7.10. The present conﬁguration of a circular disk moving with one point in contact with
a ﬁxed horizontal plane.
(a) Starting from the results that the motions of any points X and Yof the disk
have the representations
x = QX+d, y = QY+d,
show that their relative velocity vector and relative acceleration vector
satisfy
˙ x − ˙ y = ω ×(x −y) ,
¨ x − ¨ y = ˙ ω ×(x −y) +ω ×(ω ×(x −y)) .
(b) To parameterize the rotation tensor of the disk, a set of 3–1–3 Euler angles
is used. With the assistance of Figure 7.10, prescribe a reference conﬁgura
tion for the disk. For which orientations of the disk in its present conﬁgura
tion do the singularities of the 3–1–3 Euler angles occur?
(c) A sensor is mounted onto the disk and is aligned with the e
3
axis so that it
measures ω · e
3
= ω
3
(t). Show that, in general,
t
1
t
0
ω
3
(t)dt = φ(t
1
) −φ(t
0
). (7.22)
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
Exercises 7.3–7.5 229
Give a physical interpretation of the case in which the integral of ω
3
(t) does
yield the angle φ.
∗
7.4. Consider a rigid body of mass m whose inertia tensors are deﬁned as
J = QJ
0
Q
T
, J
0
=
R
0
(· )I −⊗ ρ
o
dV.
(a) Explain why J
0
and J are symmetric.
(b) Show that E
i
· (J
0
E
k
) = e
i
· (Je
k
), where Q =
¸
3
i=1
e
i
⊗E
i
.
(c) Show that J has the representation
J =
3
¸
i=1
λ
i
e
i
⊗e
i
, (7.23)
where λ
i
are the principal values of J
0
and E
i
= Q
T
e
i
are the principal di
rections of J
0
.
(d) Establish the following identities:
˙
J = J −J,
˙
H = J ˙ ω +ω ×(Jω),
˙
Jω · ω = 2
˙
H· ω.
How do these results simplify if J
0
= µI, where µ is a constant? Note that,
in the ﬁrst of these results, you are showing that the corotational rate of J
relative to Q is zero:
o
J
= 0.
(e) Suppose one used the representation
J =
3
¸
p=1
3
¸
n=1
J
pn
E
p
⊗E
n
.
Starting from (7.23), show that
J
pn
=
3
¸
i=1
Q
pi
λ
i
Q
ni
,
where Q
pi
= (QE
i
) · E
p
. Why are the components J
pn
of J not constant?
7.5. Recall that a tensor that is intimately related to the familiar inertia tensor is the
Euler tensor:
E = QE
0
Q
T
, E
0
=
R
0
⊗ ρ
o
dV.
(a) Show that
J
0
= tr(E
0
)I −E
0
, J = tr(E)I −E.
(b) Verify that E
i
· (E
0
E
k
) = e
i
· (Ee
k
), where Q =
¸
3
i=1
e
i
⊗E
i
.
(c) Establish the following results:
o
E
= 0,
˙
E = E −E,
where
o
E
denotes the corotational derivative of E with respect to Q.
∗
An application of (7.22) to the navigation of motorcycles can be found in Coaplen et al. [39].
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
230 Exercises 7.5–7.7
(d) What are the Euler tensors for a sphere of mass m and radius R, and a
cylinder of mass m, radius R, and length L? You might ﬁnd it convenient
to use the relationships of the form E
0
=
1
2
tr (J
0
) I −J
0
that were discussed
earlier in this chapter in Section 7.8.
7.6. Recall that the angular momentum H and rotational kinetic energy T
rot
of a
rigid body have the representations
H = Jω, T
rot
=
1
2
ω · Jω.
Here, J is the moment of inertia tensor of the rigid body relative to its center of
mass. Choosing {E
i
} to be the principal directions of J
0
, then
J
0
=
3
¸
i=1
λ
i
E
i
⊗E
i
, J = QJ
0
Q
T
=
3
¸
i=1
λ
i
e
i
⊗e
i
,
where Q =
¸
3
i=1
e
i
⊗E
i
.
(a) Using a set of 3–2–3 Euler angles to parameterize Q,
∗
show that
ω = (
˙
θ sin(φ) −
˙
ψsin(θ) cos(φ))e
1
+(
˙
θ cos(φ) +
˙
ψsin(φ) sin(θ))e
2
+(
˙
ψcos(θ) +
˙
φ)e
3
.
(b) Using a set of 3–2–3 Euler angles, establish an expression for H and T
rot
as
functions of the Euler angles and their time derivatives.
7.7. Recall the deﬁnitions of the angular momenta of a rigid body relative to its
center of mass and a point A:
H =
R
π ×vρdv, H
A
=
R
π
A
×vρdv. (7.24)
Here, π = x − ¯ x and π
A
= x −x
A
, where x
A
is the position vector of the point Aand
¯ x is the position vector of the center of mass.
(a) Starting from (7.24)
2
, and with the assistance of (7.24)
1
, show that
H
A
= H+(¯ x −x
A
) ×G,
where G is the linear momentum of the rigid body.
(b) Suppose that Ais a point of the rigid body. Show that
π
A
= Q
A
,
where
A
= X−X
A
and X
A
is the position vector of A in the reference
conﬁguration. If v
A
= 0, show in addition that
v = ω ×(x −x
A
).
∗
This set of Euler angles is discussed in Exercise 6.2. In particular, the student is referred to (6.44) in
Exercise 6.2(g).
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
Exercises 7.7–7.8 231
(c) Again supposing that A is a point of the rigid body and that v
A
= 0, show
that
H
A
= J
A
ω,
where J
A
is the inertia tensor of the rigid body relative to A:
J
A
=
R
(π
A
· π
A
)I −π
A
⊗π
A
ρdv.
Verify that
J
A
= QJ
A
0
Q
T
,
where
J
A
0
=
R
0
((
A
·
A
)I −
A
⊗
A
) ρ
0
dV.
(d) Using the previous results, prove the parallel axis theorem
∗
:
J
A
0
= J
0
+m((
¯
X−X
A
) · (
¯
X−X
A
)I −(
¯
X−X
A
) ⊗(
¯
X−X
A
)). (7.25)
You should be able to see from this result how a parallel axis theorem re
lating two Euler tensors could be established.
(e) Suppose that the rigid body is a circular cylinder of mass m, length L, and
radius R, where
¯
X = 0, X
A
= −xE
1
−zE
3
,
J
0
=
mL
2
12
(I −E
3
⊗E
3
) +
mR
2
4
(I +E
3
⊗E
3
) .
Then, what is J
A
0
?
7.8. Consider the rectangular parallelepiped of mass m and dimensions a, b, and c
shown in Figure 7.11. Relative to the principal axes {A
1
, A
2
, A
3
}, the inertia tensor
of this rigid body has the representation
J
O
=
3
¸
i=1
λ
i
A
i
⊗A
i
,
where
λ
1
=
m
12
(b
2
+c
2
), λ
2
=
m
12
(a
2
+c
2
), λ
3
=
m
12
(a
2
+b
2
).
The righthanded set of basis vectors {E
1
, E
2
, E
3
} is oriented so that E
3
passes di
agonally through the parallelepiped. That is, E
3
is parallel to the line connecting
¯
X
and the point P:
E
3
=
1
√
a
2
+b
2
+c
2
(aA
1
−bA
2
+cA
3
) .
∗
This representation of the parallel axis theorem is due to Fox [65].
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
232 Exercise 7.8
In addition,
E
1
= cos(α)A
1
−sin(α)A
3
,
where
α = tan
−1
a
c
.
P
a
b
c ¯
X
E
1
E
2
E
3
A
1
A
2
A
3
Figure 7.11. Schematic of a parallelepiped.
(a) After calculating E
2
, verify that the transformation from the basis
{A
1
, A
2
, A
3
} to the basis {E
1
, E
2
, E
3
} can be written in the form
¸
¸
¸
E
1
E
2
E
3
= R
¸
¸
¸
A
1
A
2
A
3
, (7.26)
where
R =
¸
¸
¸
R
11
R
12
R
13
R
21
R
22
R
23
R
31
R
32
R
33
=
¸
¸
¸
1 0 0
0 cos(β) sin(β)
0 −sin(β) cos(β)
¸
¸
¸
cos(α) 0 −sin(α)
0 1 0
sin(α) 0 cos(α)
,
and
cos(β) =
√
a
2
+c
2
√
a
2
+b
2
+c
2
, sin(β) =
b
√
a
2
+b
2
+c
2
.
(b) Show that the components J
0ik
= (J
0
E
k
) · E
i
are
J
0ik
=
3
¸
r=1
R
ir
λ
r
R
kr
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
Exercises 7.8–7.9 233
Using a matrix notation, these equations are equivalent to
¸
¸
¸
J
011
J
012
J
013
J
012
J
022
J
023
J
013
J
023
J
033
= R
¸
¸
¸
λ
1
0 0
0 λ
2
0
0 0 λ
3
R
T
.
(c) If the body is given an angular velocity ω = ωe
3
, where e
i
= QE
i
, then es
tablish expressions for the angular momentum H of the body relative to its
center of mass and the rotational kinetic energy T
rot
of the body. How do
these expressions simplify if a = b = c?
7.9. In this question, you will establish results for a set of Euler angles: the 3–1–2
set. This set of angles is similar to the 3–2–1 set discussed in Subsection 6.8.1. We
then apply these results to solve a navigation problem.
We shall decompose the rotation tensor Q into the product of three rotation ten
sors. The ﬁrst rotation is about E
3
through an angle ψ: L(ψ, E
3
). The second ro
tation is an angle θ about e
1
= cos(ψ)E
1
+sin(ψ)E
2
: L(θ, e
1
). The third rotation is
through an angle φ about
e
2
= e
2
= cos(θ)e
2
+sin(θ)e
3
.
(a) Show that
¸
¸
¸
e
1
e
2
e
3
=
¸
¸
¸
¸
Q
11
Q
21
Q
31
Q
12
Q
22
Q
32
Q
13
Q
23
Q
33
¸
¸
¸
E
1
E
2
E
3
,
where Q
i j
= (QE
j
) · E
i
.
(b) Show that the Euler basis for the 3–1–2 set of Euler angles has the repre
sentation
¸
¸
¸
g
1
g
2
g
3
=
¸
¸
¸
¸
0 0 1
cos(ψ) sin(ψ) 0
−cos(θ) sin(ψ) cos(θ) cos(ψ) sin(θ)
¸
¸
¸
E
1
E
2
E
3
.
Using the relations g
k
· g
i
= δ
k
i
, show that the dual Euler basis for the 3–1–2
set of Euler angles has the representation
¸
¸
¸
g
1
g
2
g
3
=
¸
¸
¸
¸
sin(ψ) tan(θ) −cos(ψ) tan(θ) 1
cos(ψ) sin(ψ) 0
−sec(θ) sin(ψ) sec(θ) cos(ψ) 0
¸
¸
¸
E
1
E
2
E
3
.
You should notice that the second Euler angle needs to be restricted to
θ ∈
−
π
2
,
π
2
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
234 Exercise 7.9
(c) By examining the three rotations that compose Q and with the help of the
results in (a), show that
¸
¸
¸
¸
Q
11
Q
12
Q
13
Q
21
Q
22
Q
23
Q
31
Q
32
Q
33
= CBA, (7.27)
where
A =
¸
¸
¸
cos(φ) 0 sin(φ)
0 1 0
−sin(φ) 0 cos(φ)
,
B =
¸
¸
¸
1 0 0
0 cos(θ) −sin(θ)
0 sin(θ) cos(θ)
,
C =
¸
¸
¸
cos(ψ) −sin(ψ) 0
sin(ψ) cos(ψ) 0
0 0 1
.
(d) Select four different values of the set (φ, θ, ψ) and determine the axis of
rotation q and the angle of rotation of Q.
(e) The angular velocity vector associated with the 3–1–2 Euler angles has the
representations
ω =
˙
ψE
3
+
˙
θe
1
+
˙
φe
2
= ω
1
e
1
+ω
2
e
2
+ω
3
e
3
.
Show that
¸
¸
¸
ω
1
ω
2
ω
3
=
¸
¸
¸
−sin(φ) cos(θ) 0 cos(φ)
sin(θ) 1 0
cos(φ) cos(θ) 0 sin(φ)
¸
¸
¸
˙
ψ
˙
φ
˙
θ
,
and
¸
¸
¸
˙
ψ
˙
φ
˙
θ
=
¸
¸
¸
−sec(θ) sin(φ) 0 sec(θ) cos(φ)
tan(θ) sin(φ) 1 −tan(θ) cos(φ)
cos(φ) 0 sin(φ)
¸
¸
¸
ω
1
ω
2
ω
3
. (7.28)
These sets of differential equations are very important in strapdown navi
gation systems.
(f) Suppose that a set of 3–1–2 Euler angles is used to parameterize the rota
tional motion of a vehicle: Q =
¸
3
i=1
e
i
⊗E
i
. Here, e
1
is taken to point in
the forward direction along the vehicle and E
3
is taken to point vertically
downward. The ﬁrst Euler angle ψ is known as the yaw angle, the second
angle θ is the roll angle, and the third angle φ is called the pitch angle.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
Exercises 7.9–7.10 235
(i) A set of three gyroscopes is mounted on the vehicle and provides three
signals ω
i
(t) (see, for example, the signals shown in Figure 7.12).
∗
Argue
that, with knowledge of the initial orientation of the vehicle, integrating
(7.28), the orientation of the vehicle can be determined.
(ii) Suppose the vehicle is initially level. As time progresses the gyroscopes
register the following signals:
ω
1
(t) = 0.02 sin(0.5t),
ω
2
(t) = 0.02 sin(0.05t),
ω
3
(t) = 2.0 sin(0.5t).
Determine e
k
(t) for the vehicle.
(iii) How does the axis of rotation q of the vehicle change and is this axis
parallel to the instantaneous axis of rotation i =
ω
ω
?
0
100
100
−100
time (s)
o
/s
ω
1
+40
ω
2
−40
ω
3
Figure 7.12. Time traces of ω
3
(t), ω
2
(t) −40, and ω
1
(t) +40 from a set of three gyroscopes
mounted on a motorcycle. The motorcycle is moving in a rectangular path and the offsets ±40
are imposed so that the three signals can be distinguished. It is also interesting to observe the
amount of noise in the signals for ω
i
(t).
7.10. As shown in Figure 7.13, a Poisson top is an axisymmetric body with a sharp
apex that is free to move on a ﬂat surface. The material point of the top in con
tact with the surface is labeled P. The position vectors of this point relative to the
ﬁxed point Oin the present and reference conﬁgurations are denoted by x
P
and X
P
,
respectively. The material point corresponding to the center of mass of the top is de
noted by
¯
X, and the position vectors of
¯
Xrelative to the ﬁxed point Oin the present
and reference conﬁgurations are denoted by ¯ x and
¯
X, respectively.
(a) A set of 3–2–1 Euler angles is used to describe the orientation of the top.
Denoting the ﬁrst angle by ψ, the second by θ, and the third by φ, give
sketches showing how these angles describe the orientation of the top.
∗
The principle of operation of a class of gyroscopes is discussed in Section 11.4.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
236 Exercise 7.10
g
O
¯
X
¯
X
P
P
E
1
E
1
E
2
E
2
E
3
E
3
e
1
e
2
e
3
κ
0
κ
t
Figure 7.13. The ﬁxed reference κ
0
and present κ
t
conﬁgurations of a Poisson top. The surface
that the top moves on is taken to be the E
1
−E
2
plane.
(b) For which orientations of the top does the set of 3–2–1 Euler angles have
singularities? Now, suppose a set of 3–1–3 Euler angles were used to pa
rameterize the rotation tensor of the top. For which orientations of the top
does the 3–1–3 set have singularities?
(c) Starting from the result that for any point X on a rigid body,
x = QX+d,
show that
x = Q(X−X
P
) +x
P
, ˙ x = ω ×(x −x
P
) + ˙ x
P
.
(d) The moment of inertia tensor of the top in its reference conﬁguration has
the representation
J
O
= λ
a
E
3
⊗E
3
+λ
t
(I −E
3
⊗E
3
) .
If
P
= X
P
−
¯
X = −L
3
E
3
, ω = ω
1
e
1
+ω
2
e
2
+ω
3
e
3
,
then showthat the angular momentumvector of the top relative to the point
P is
H
P
= λ
a
ω
3
e
3
+
λ
t
+mL
2
3
(ω
1
e
1
+ω
2
e
2
) +L
3
e
3
×m˙ x
P
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
8 Constraints on and Potentials
for Rigid Bodies
8.1 Introduction
In this chapter, constraints and the forces and moments that enforce them in
the dynamics of rigid bodies are discussed. In particular, the constraints associ
ated with interconnected rigid bodies, rolling rigid bodies, and sliding rigid bod
ies along with prescriptions for the associated constraint forces and moments are
presented. We also discuss Lagrange’s prescription for constraint forces and con
straint moments and outline its limitations. It also proves convenient to discuss
potential energies and their associated conservative forces and moments. Our dis
cussion includes as examples springs and central gravitational ﬁelds. It has ob
vious parallels to the treatment of constraints and their associated forces and
moments.
8.2 Constraints
Constraints in the motions of rigid bodies usually arise in two manners. First, the
rigid body is connected to another rigid body in such a way that its relative motion
is constrained. The connections in question are usually in the form of joints. The
second manner in which constraints arise occurs when one rigid body is rolling or
sliding on the other. As in particles, the constraints we discuss can be classiﬁed as
integrable or nonintegrable, and this classiﬁcation is important in dynamics because
it may lead to considerable simpliﬁcation in the formulation of the equations gov
erning the motion.
In our discussion, we consider two rigid bodies, B
1
and B
2
(see Figure 8.1). We
can consider the case of a rigid body interacting with the ground by prescribing the
velocity vector of the center of mass and the angular velocity vector of either one
of the bodies as functions of time. If more than two rigid bodies are involved in
the constraint, then it is not too difﬁcult to generalize the discussion subsequently
presented. It should also be noted that the discussion presented here of the various
types of joints and contacts is not exhaustive.
237
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
238 Constraints on and Potentials for Rigid Bodies
O
¯
X
1
¯
X
2
¯ x
1
¯ x
2
Figure 8.1. The present conﬁgurations and centers of mass of two rigid bodies B
1
and B
2
.
Connected Rigid Bodies
As shown in Figure 8.2, consider two rigid bodies B
1
and B
2
that are connected at
the point P by a joint. The position vector of P on B
α
relative to
¯
X
α
is denoted by
π
Pα
. Because the point P occupies the same location for both bodies, we have the
following three constraints
∗
:
¯ x
1
+π
P1
= ¯ x
2
+π
P2
. (8.1)
These constraints also imply that
¯ v
1
+ω
1
×π
P1
= ¯ v
2
+ω
2
×π
P2
. (8.2)
Notice that by integrating (8.2) with respect to time and setting the constants of in
tegration to 0, we will arrive at (8.1). Consequently, constraints (8.2) are integrable.
The joint at P may have the ability to restrict the rotation of B
2
relative to B
1
.
There are two types of joints to consider: the pin (or revolute) joint and the ball
andsocket joint. The pin joint arises in gyroscopes where it serves to connect the
inner and outer gimbals. On the other hand, ballandsocket joints do not place any
restriction on the relative angular velocity vector ω
2
−ω
1
.
Consider the case of a pin joint. Let s
3
be a unit vector that is parallel to the
axis of relative rotation permitted by the joint, and let {s
1
, s
2
, s
3
} be an orthonormal
basis. The pin joint ensures that the relative rotation Q
2
Q
T
1
is a rotation about s
3
.
These restrictions are most easily expressed in terms of the relative angular velocity
vector. Speciﬁcally, a pin joint imposes the constraints
(ω
1
−ω
2
) · s
1
= 0, (ω
1
−ω
2
) · s
2
= 0.
It should be clear that these two (integrable) constraints are supplemented by (8.2).
In addition, it is sufﬁcient to use a single angle to parameterize Q
2
Q
T
1
.
∗
By taking the components of this vector equation with respect to a basis, one arrives at three inde
pendent scalar equations. Hence the vector equation represents three constraints.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
8.2 Constraints 239
O
P
¯
X
1
¯
X
2
¯ x
1
¯ x
2
π
P1
π
P2
Figure 8.2. The present conﬁgurations of two rigid
bodies B
1
and B
2
that are connected by a joint at P.
Clearly each of the individual constraints mentioned to this point can be written
in the form
π = 0, (8.3)
where
π = f
1
· ¯ v
1
+f
2
· ¯ v
2
+h
1
· ω
1
+h
2
· ω
2
+e.
Here, f
1
, f
2
, h
1
, h
2
, and e are functions of ¯ x
1
, ¯ x
2
, Q
1
, Q
2
, and t. Furthermore, (8.3) can
be integrated with respect to time to yield a function :
= (¯ x
1
, ¯ x
2
, Q
1
, Q
2
, t) .
Here,
˙
= π.
In other words, for this case, the constraint π = 0 is said to be integrable or holo
nomic. In general, the constraints associated with connections are usually integrable.
This is in marked contrast to the next set of situations we consider.
Rolling and Sliding Rigid Bodies
Consider the situation shown in Figure 8.3. Here, two rigid bodies are instanta
neously in contact at the point P. It is assumed that there is a welldeﬁned unit
normal n to the surfaces of both bodies at the point P. In addition, we use n to de
ﬁne an orthonormal basis {s
1
, s
2
, n}, where s
1
and s
2
are tangent to the surfaces of
both bodies at the point of contact P.
As the point of contact P coincides with a material point of each body, we have
x
p1
= ¯ x
1
+π
p1
, v
p1
= ¯ v
1
+ω
1
×π
p1
, (8.4)
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
240 Constraints on and Potentials for Rigid Bodies
O
P
¯
X
1
¯
X
2
¯ x
1
¯ x
2
π
P1
π
P2
n
s
1
s
2
Figure 8.3. The present conﬁgurations of two rigid bodies B
1
and B
2
that are in contact at the
point P.
and
x
p2
= ¯ x
2
+π
p2
, v
p2
= ¯ v
2
+ω
2
×π
p2
. (8.5)
It is important to note that it is not generally possible to differentiate (8.4)
1
to arrive
at (8.4)
2
, nor (8.5)
1
to arrive at (8.5)
2
. The reason for this is that π
pα
does not identify
the same material of B
α
at each instant of time. In other words, P
α
corresponds to a
different material point of B
α
at each instant of time.
The problems in rigid body dynamics involving rolling or sliding rigid bodies
that have received the most attention involve rigid bodies for which the vectors π
pα
have relatively simple forms, speciﬁcally, rolling/sliding spheres, the sliding top, and
the rolling/sliding disks. The most famous example for which π
pα
is very difﬁcult
to express in a tractable form is the wobblestone or celt. As ﬁrst reported by G. T.
Walker in 1896 [223], this rigid body has the unique feature of reversing its direction
of spin in a counterintuitive manner.
∗
8.2.1 The Sliding Condition
For two bodies in contact, it is assumed that the contact persists and that the two
bodies do not interpenetrate. These assumptions lead to the sliding condition:
v
p1
· n = v
p2
· n. (8.6)
We can express this condition in another form:
(¯ v
1
− ¯ v
2
) · n = (ω
2
×π
p2
−ω
1
×π
p1
) · n.
∗
For further details on this interesting mechanical system, and more recent research on it, see [17]
and references therein.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
8.3 A Canonical Function 241
The sliding condition implies that a certain relative velocity has components in only
the tangential directions:
v
s
= v
p1
−v
p2
= v
s1
s
1
+v
s2
s
2
. (8.7)
The velocity v
s
is often known as the sliding velocity and is important for specifying
the friction forces at the point P.
8.2.2 The Rolling Condition
Rolling occurs when the velocity vectors of the point of contact for each body are
identical. In this case, we have the rolling condition:
v
p1
= v
p2
.
Again, we can express this equation in another form:
¯ v
1
+ω
1
×π
p1
= ¯ v
2
+ω
2
×π
p2
. (8.8)
These three equations are equivalent to sliding condition (8.6) and the condition
that the sliding velocity v
s
is zero.
Rolling condition (8.8) is equivalent to three scalar equations [cf. (8.3)]:
π
1
= 0, π
2
= 0, π
3
= 0,
where, for i = 1, 2, 3,
π
i
= f
i1
· ¯ v
1
+f
i2
· ¯ v
2
+h
i1
· ω
1
+h
i2
· ω
2
+e
i
.
However, for two of these equations, say π
2
= 0 and π
3
= 0, it is not possible to
ﬁnd functions
2
and
3
such that
˙
2
= π
2
and
˙
3
= π
3
. In other words, two of
constraints (8.8) are nonintegrable or nonholonomic. The one constraint of (8.8)
that is integrable corresponds to the n component of (8.8). We shall shortly discuss
some examples that illustrate this point.
8.3 A Canonical Function
For future purposes, it is convenient to consider two rigid bodies B
1
and B
2
and
construct the general functional form of a possible integrable constraint or potential
energy function that features the motions of both bodies. This function is presented
in (8.9) and, for future purposes, we also calculate its derivative. Our treatment here
follows [163].
As shown in Figure 8.1, the position vector of the center of mass
¯
X
α
of the body
B
α
is denoted by ¯ x
α
, where α = 1 or 2. Similarly, the rotation tensor of B
α
is denoted
by Q
α
. For each rigid body, we can deﬁne corotational bases:
Q
1
=
3
¸
i=1
1
e
i
⊗E
i
, Q
2
=
3
¸
i=1
2
e
i
⊗E
i
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
242 Constraints on and Potentials for Rigid Bodies
It should be noted that we are using the same ﬁxed basis {E
1
, E
2
, E
3
} to deﬁne the
corotational bases. The angular velocity vectors of the bodies are
ω
1
= −
1
2
¸
˙
Q
1
Q
T
1
¸
, ω
2
= −
1
2
¸
˙
Q
2
Q
T
2
¸
.
It should also be noted that the rotation tensor of B
2
relative to B
1
is
Q
2
Q
T
1
=
3
¸
i=1
2
e
i
⊗
1
e
i
.
This tensor represents the rotation of B
2
relative to an observer who is stationary
on B
1
.
A ScalarValued Function of the Motions
We now consider a scalar function :
= (¯ x
1
, ¯ x
2
, Q
1
, Q
2
, t) .
Clearly this function depends on the motions of both rigid bodies and time. One
observes functions of this type when representing integrable constraints on the mo
tions of rigid bodies and potential energies of rigid bodies.
As in the case of a particle, it is of interest to calculate
˙
. To calculate this time
derivative, we invoke the chain rule:
˙
=
∂
∂ ¯ x
1
· ¯ v
1
+
∂
∂ ¯ x
2
· ¯ v
2
+tr
∂
∂Q
1
˙
Q
T
1
+tr
∂
∂Q
2
˙
Q
T
2
+
∂
∂t
.
Recalling our earlier discussion in Section 6.10 of derivatives of scalar functions of
rotation tensors, it is convenient to deﬁne the skewsymmetric tensor
Q
α
=
1
2
∂
∂Q
α
Q
T
α
−Q
α
∂
∂Q
α
T
and an associated vector
ψ
Q
α
= − [
Q
α
] .
Representations for these tensors and vectors, based on the parameterization of Q
α
,
were discussed earlier. In particular, if the Euler angles ν
i
are used to parameterize
Q
1
, say, then
ψ
Q
1
=
3
¸
i=1
∂
∂ν
i
g
i
, ψ
Q
1
· ω
1
=
3
¸
i=1
∂
∂ν
i
˙ ν
i
, (8.9)
where g
i
are the dual Euler basis vectors associated with the Euler angles.
We can use the aforementioned skewsymmetric tensors to rewrite
˙
:
˙
=
∂
∂ ¯ x
1
· ¯ v
1
+
∂
∂ ¯ x
2
· ¯ v
2
+tr
Q
1
T
1
+tr
Q
2
T
2
+
∂
∂t
=
∂
∂ ¯ x
1
· ¯ v
1
+
∂
∂ ¯ x
2
· ¯ v
2
+ψ
Q
1
· ω
1
+ψ
Q
2
· ω
2
+
∂
∂t
,
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
8.4 Integrability Criteria 243
where
α
=
˙
Q
α
Q
T
α
, ω
α
= −
1
2
[
α
], (α = 1, 2) ,
are the angular velocity tensors and vectors of the rigid bodies.
It is more efﬁcient next to use the form of
˙
that involves vectors:
˙
=
∂
∂ ¯ x
1
· ¯ v
1
+
∂
∂ ¯ x
2
· ¯ v
2
+ψ
Q
1
· ω
1
+ψ
Q
2
· ω
2
+
∂
∂t
. (8.10)
With some minor rearrangements, we can eliminate ¯ v
2
and ω
2
in favor of the rel
ative velocity vectors ¯ v
2
− ¯ v
1
and ω
2
−ω
1
in the expression for
˙
. Equation (8.10),
which ﬁrst appeared in [163], will play a key role in examining potential forces and
moments.
Result (8.10) is rarely apparent in treatments of rigid body dynamics. This is
partially because speciﬁc parameterizations of ¯ x
α
and Q
α
are used. To elaborate on
this point, let
¯ x
1
=
3
¸
i=1
x
i
E
i
, ¯ x
2
=
3
¸
i=1
y
i
E
i
,
and suppose that Q
1
is parameterized by the Euler angles ν
i
and the relative rotation
tensor Q
2
Q
T
1
is parameterized by the Euler angles γ
i
. That is,
ω
1
=
3
¸
i=1
˙ ν
i
g
i
, ω
2
−ω
1
=
3
¸
i=1
˙ γ
i
n
i
,
where g
i
and n
i
are the Euler basis vectors associated with ν
i
and γ
i
, respectively.
Then,
= (¯ x
1
, ¯ x
2
, Q
1
, Q
2
, t) =
˜
x
i
, y
j
, ν
i
, γ
k
, t
.
Furthermore,
˙
=
˙
˜
=
3
¸
i=1
∂
˜
∂x
i
˙ x
i
+
∂
˜
∂y
i
˙ y
i
+
∂
˜
∂ν
i
˙ ν
i
+
∂
˜
∂γ
i
˙ γ
i
+
∂
˜
∂t
.
It is good exercise to compare this expression with (8.10) and identify the corre
sponding terms in both expressions.
8.4 Integrability Criteria
Earlier, in Section 1.8, we examined integrability criteria for constraints of the form
f · v +e = 0 on the motion of a single particle. Here, we examine the corresponding
criteria for a constraint on the motions of two rigid bodies. In the dynamics of rigid
bodies, situations involving systems of constraints are often inescapable: One only
has to think of the case of a rolling rigid body. It is also possible that, when a sufﬁ
cient number of constraints are imposed on a rolling rigid body, the resulting system
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
244 Constraints on and Potentials for Rigid Bodies
of constraints becomes integrable. Fortunately, there is a theorem, which is due to
Frobenius, that provides a criterion for the integrability of a system of constraints.
His criterion will be presented shortly. Before doing so, we discuss the situation of
a single constraint on a pair of rigid bodies.
A Single Constraint
For rigid bodies, what is needed is a generalization of criterion (1.21) for a single
particle to constraints of the form
π = 0,
where
π = f
1
· ¯ v
1
+f
2
· ¯ v
2
+h
1
· ω
1
+h
2
· ω
2
+e.
As mentioned earlier, this constraint is integrable if we can ﬁnd an integrating fac
tor k and a function (¯ x
1
, ¯ x
2
, Q
1
, Q
2
, t) such that k
˙
= π. Although the integrability
criterion is daunting in the number of algebraic calculations needed, for many prob
lems most of these calculations are trivial (but tedious).
The necessary and sufﬁcient conditions for π = 0 to be integrable involve a set
of up to 66 independent conditions. To establish these conditions, which are similar
to those we discussed earlier for a single particle, we assume that ¯ x
1
is parameterized
by use of a coordinate system{q
1
, q
2
, q
3
}, ¯ x
2
is parameterized by use of a coordinate
system{q
4
, q
5
, q
6
}, Q
1
is parameterized by use of {ν
1
, ν
2
, ν
3
}, and Q
2
is parameterized
by use of {ν
4
, ν
5
, ν
6
}. Thus the function π can be expressed as
π =
3
¸
i=1
f
1i
˙ q
i
+ f
2i
˙ q
i+3
+h
1i
˙ ν
i
+h
2i
˙ ν
i+3
+e.
We also deﬁne the intermediate functions
W
i
= f
1i
, W
i+3
= f
2i
, W
i+6
= h
1i
, W
i+9
= h
2i
, W
13
= e,
and variables
U
i
= q
i
, U
i+3
= q
i+3
, U
i+6
= ν
i
, U
i+9
= ν
i+3
, U
13
= t,
where i = 1, 2, 3. It remains to deﬁne
I
JKL
= W
J
S
LK
+W
K
S
JL
+W
L
S
KJ
, (8.11)
where S
LK
are the components of a skewsymmetric matrix:
S
LK
=
∂W
L
∂U
K
−
∂W
K
∂U
L
. (8.12)
In (8.11) and (8.12), the integer indices J , K, and L range from 1 to 13.
With all the preliminaries taken care of, we are now in a position to state a
theorem that is due to Frobenius:
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
8.4 Integrability Criteria 245
A necessary and sufﬁcient condition for the constraint π = 0 to be integrable is that the
following
13
6
(13 −1)(13 −2) equations hold for all U
1
, . . . , U
13
:
I
JKL
= 0, for all J, K, L ∈ {1, . . . , 13}, J = K = L, J = L. (8.13)
A proof of the theorem can be found in texts on differential equations (for
example, Section 161 of Forsyth [64]). We also note that only 66 of the 286 equa
tions I
JKL
= 0 are independent. Criterion (8.13) can be specialized to the case of a
constraint f · ¯ v +h · ω +e = 0 on a single rigid body. There, the indices J, K, L will
range from 1 to 7, and the number of independent conditions I
JKL
= 0 to verify will
be 10.
Systems of Constraints
When rigid bodies are subject to several constraints, a criterion is available by which
the possible integrability of the systemof constraints can be evaluated. This criterion
is known as Frobenius’ theorem. To simplify our exposition, we frame this criterion
in terms of a single rigid body. Using the developments of the previous section, we
can easily extend this to the case of two rigid bodies.
Suppose we have a system of R ≤ 6 constraints on the motion of a single rigid
body:
π
1
= 0, . . . , π
R
= 0, (8.14)
where
π
A
= f
A
· ¯ v +h
A
· ω +e
A
(A= 1, . . . , R).
Again, we assume that coordinates
¸
q
1
, q
2
, q
3
¸
have been chosen for the position
vector of the center of mass and parameters
¸
ν
1
, ν
2
, ν
3
¸
have been selected for the
rotation tensor of the rigid body. Thus each of the π
A
’s can be rewritten as
π
A
=
3
¸
i=1
f
Ai
˙ q
i
+h
Ai
˙ ν
i
+e
A
.
Next, we deﬁne the following functions and variables:
W
Ai
= f
Ai
, W
A(i+3)
= h
Ai
, W
A7
= e
A
,
S
A
LK
=
∂W
AL
∂U
K
−
∂W
AK
∂U
L
, (8.15)
where we have used the following notation for the variables
U
i
= q
i
, U
i+3
= ν
i
, U
7
= t. (8.16)
In (8.15) and (8.16), i = 1, 2, 3 and L, K = 1, . . . , 7. The functions W
AL
and S
A
LK
can
be used to construct two matrices, for example,
W =
⎡
⎢
⎢
⎣
W
11
· · · W
17
.
.
.
.
.
.
.
.
.
W
R1
· · · W
R7
⎤
⎥
⎥
⎦
. (8.17)
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
246 Constraints on and Potentials for Rigid Bodies
With the help of (8.15), it is easy to see that S
1
, . . . , S
R
are skewsymmetric matrices:
S
A
KL
= −S
A
LK
. We also deﬁne the following sevendimensional vectors:
a =
⎡
⎢
⎢
⎣
a
1
.
.
.
a
7
⎤
⎥
⎥
⎦
, b =
⎡
⎢
⎢
⎣
b
1
.
.
.
b
7
⎤
⎥
⎥
⎦
, x =
⎡
⎢
⎢
⎣
x
1
.
.
.
x
7
⎤
⎥
⎥
⎦
. (8.18)
We are now in a position to state the Frobenius integrability theorem:
For a single rigid body, the necessary and sufﬁcient conditions for a system of R < 7
constraints π
1
= 0, . . . , π
R
= 0 to be integrable are for the following equations to hold
a
T
·
S
A
b
= 0 (A= 1, . . . , R),
for all values of the variables U
1
, . . . , U
7
, and for all distinct solutions a and b to the
equation
Wx = 0,
that is, the sevendimensional vectors a and b lie in the null space of W.
REMARKS ON FROBENIUS’ THEOREM. The statement of Frobenius’ remarkable theo
rem is adapted from Forsyth [63], and the reader is referred to this text for a clas
sical proof and to a paper by Hawkins [91] for additional historical perspectives.
More modern proofs, featuring differential forms, are readily found; see, for exam
ple, [34, 63, 101, 189].
As S
A
are skewsymmetric, a
T
·
S
A
a
= 0 for all possible a. Often this result is
a key to using the theorem: For if W has a onedimensional null space, then there is
only one vector, say b, satisfying Wb = 0, and then the theorem is trivially satisﬁed.
EXAMPLES. An example of the use of the theorem is discussed in the exercises at
the end of this chapter. There it is used to show that a rolling disk is generally sub
ject to nonintegrable constraints. However, when the disk is further constrained to
be vertical and its center of mass moves in a straight line, then the constraints on
the rolling disk become integrable. As discussed in Exercise 8.9, this is the familiar
situation from introductory dynamics classes.
A second example of the use of the theorem was mentioned earlier in conjunc
tion with two scleronomic constraints (1.23) on the motion of a single particle:
f
11
˙ x + f
12
˙ y + f
13
˙ z = 0, f
21
˙ x + f
22
˙ y + f
23
˙ z = 0,
where f
1
=
¸
3
k=1
f
1k
E
k
and f
2
=
¸
3
k=1
f
2k
E
k
are functions of the position vector of
the particle, and f
2
×f
1
= 0. To apply Frobenius’ theorem to this case, some trivial
modiﬁcations to (8.15)–(8.18) are needed. First,
U
1
= x, U
2
= y, U
3
= z, U
4
= t.
For this case, R = 2, and the indices Land Krange from1 to 4. It is left as an exercise
to establish that the 2 ×4 matrix Whas a twodimensional null space that is spanned
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
8.5 Forces and Moments Acting on a Rigid Body 247
by the vectors
a = [0, 0, 0, 1]
T
, b = [g
1
, g
2
, g
3
, 0]
T
,
where g
k
= (f
1
×f
2
) · E
k
. Now, as
S
A
=
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
0
∂f
A1
∂x
2
−
∂f
A2
∂x
1
∂f
A1
∂x
3
−
∂f
A3
∂x
1
0
∂f
A2
∂x
1
−
∂f
A1
∂x
2
0
∂f
A2
∂x
3
−
∂f
A3
∂x
2
0
∂f
A3
∂x
1
−
∂f
A1
∂x
3
∂f
A3
∂x
2
−
∂f
A2
∂x
3
0 0
0 0 0 0
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
,
it follows that a
T
·
S
1
b
= 0 and a
T
·
S
2
b
= 0. Thus the hypotheses of Frobenius’
integrability theorem are satisﬁed, and we conclude that system of constraints (1.23)
is integrable.
8.5 Forces and Moments Acting on a Rigid Body
Before discussing constraint forces and moments, we dispense with some prelimi
naries. The resultant force F acting on a rigid body is the sum of all the forces acting
on a rigid body. Similarly, the resultant moment relative to a ﬁxed point O, M
O
,
is the resultant external moment relative to O of all of the moments acting on the
rigid body. We also denote the resultant moment relative to the center of mass
¯
Xby
M. These moments may be decomposed into two additive parts, the moment that
is due to the individual external forces acting on the rigid body and the applied ex
ternal moments that are not due to external forces. We refer to the latter as “pure”
moments.
As an example, consider a system of forces and moments acting on a rigid body.
Here, a set of Lforces F
K
(K = 1, . . . , L) acts on the rigid body. The force F
K
acts at
the material point X
K
, which has a position vector x
K
. In addition, a pure moment
M
p
, which is not due to the moment of an applied force, also acts on the rigid body
(see Figure 8.4). For this system of applied forces and moments, the resultants are
F =
L
¸
K=1
F
K
,
M
O
= M
p
+
L
¸
K=1
x
K
×F
K
,
M = M
p
+
L
¸
K=1
(x
K
− ¯ x) ×F
K
.
You should notice how the pure moment M
p
features in these expressions.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
248 Constraints on and Potentials for Rigid Bodies
O
κ
t
X
K
¯
X
¯ x
x
K
π
K
M
p
F
K
Figure 8.4. A force F
K
and a moment M
p
acting on a
rigid body.
The mechanical power P of a force F
K
acting at a material point X
K
is deﬁned
to be
P = F
K
· ˙ x
K
.
Using the center of mass, we can obtain a different representation of this power.
Speciﬁcally, we use the identity, ˙ x
K
= ¯ v +ω ×π
K
, where π
K
= x
K
− ¯ x. It then fol
lows that
P = F
K
· ˙ x
K
= F
K
· ¯ v +(π
K
×F
K
) · ω.
In words, the power of a force is identical to the combined power of the same force
acting at the center of mass, and its moment, relative to the center of mass. The
mechanical power of a pure moment M
p
is deﬁned to be
P = M
p
· ω.
You should notice how this expression is consistent with the previous expression for
the mechanical power of a force F
K
.
Using the results for the mechanical power of a force F
K
and a pure moment
M
p
, we ﬁnd that the resultant mechanical power of the system of L forces and a
pure moment discussed previously is
P =
L
¸
K=1
F
K
· ˙ x
K
+M
p
· ω = F · ¯ v +M· ω.
These results will play a key role in our future discussion of constraint forces and
moments and potential energies.
8.6 Constraint Forces and Constraint Moments
Given a system of constraints on the motion of one or more rigid bodies, a system
of forces and moments is required for ensuring that the constraints are enforced for
all possible motions of the bodies that are compatible with the constraints. At issue
here is the prescription of these forces and moments.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
8.6 Constraint Forces and Constraint Moments 249
One of the major points of this chapter is the form of Lagrange’s prescription
associated with constraints of the form [cf. (8.3)]
π = 0,
where
π = f
1
· ¯ v
1
+f
2
· ¯ v
2
+h
1
· ω
1
+h
2
· ω
2
+e.
We showed earlier how most of the commonly used constraints in rigid body dy
namics could be written in this form.
Pedagogically, it is easiest to present some examples and then give Lagrange’s
prescription for the constraint forces and constraint moments. However, this pre
scription is not universally applicable. For instance, as in particle dynamics, it does
not give physically realistic constraint forces and constraint moments when dynamic
friction is present.
8.6.1 A Rigid Body Rotating About a Fixed Point
Consider the rigid body shown in Figure 8.5. The body is pinjointed to the ﬁxed
point O. As a consequence of the pin joint, the body performs a ﬁxedaxis rotation
about E
3
= e
3
. Further, if the rotation of the body is known, then the motion of the
center of mass is also known.
As a result of the pin joint, the angular velocity vector ω of the body is subject
to two constraints. These constraints can be expressed in a variety of manners. For
instance,
ω · E
1
= 0, ω · E
2
= 0, (8.19)
or, equivalently,
ω · e
1
= 0, ω · e
2
= 0.
Now because O is ﬁxed, there are three additional constraints on the motion of the
body:
¯ v −ω × ¯ x = 0. (8.20)
O
¯
X
¯ x
E
1
E
2
e
1
e
2
Figure 8.5. A rigid body B that is performing a ﬁxedaxis rotation
about the point O.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
250 Constraints on and Potentials for Rigid Bodies
In other words, the pin joint couples the rotational motion of the body with the
motion of its center of mass.
To enforce constraints (8.19), we assume that two constraint moments are ex
erted by the pin joint on the body. These moments have components perpendicular
to E
3
:
M
c1
= µ
4
E
1
, M
c2
= µ
5
E
2
,
where µ
4
and µ
5
are functions of time that are determined from the balance laws.
These constraint moments are examples of pure moments. Our prescription of the
constraint moments assumes that the pin joint is frictionless. If friction were present,
then they would have components in the E
3
direction. Because the pin joint imposes
the constraints that Ois ﬁxed, it also supplies a reaction force R acting at O:
R =
3
¸
i=1
µ
i
E
i
,
where µ
1
, µ
2
, and µ
3
are functions of times that are determined from the balance
laws.
The reaction force and constraint moments are equipollent to a constraint force
F
c
acting at the center of mass of the rigid body and a constraint moment M
c
relative
to the center of mass of the rigid body:
F
c
=
3
¸
i=1
µ
i
E
i
, M
c
= µ
4
E
1
+µ
5
E
2
− ¯ x ×
3
¸
i=1
µ
i
E
i
.
Notice that there are ﬁve constraints on the rigid body that were imposed by the
pin joint and ﬁve unknown functions µ
1
, . . . , µ
5
in the expressions for the constraint
forces and moments.
8.6.2 A Sphere Rolling or Sliding on an Inclined Plane
The problem of a sphere rolling or sliding on a plane has a long history, in part
because it is the basis for pool (billiards) and bowling. The main contributors to this
problem in the 19th century were Gaspard G. de Coriolis (1792–1843) (see [41]) and
Edward J. Routh (1831–1907) (see [184]). Studies on the dynamics of rolling spheres
on surfaces of revolution are often known as Routh’s problem.
∗
Consider the rigid body shown in Figure 8.6. The body in this case is moving
on a ﬁxed inclined plane. In the ﬁgure, the body is assumed to be a sphere. For the
sphere, it is easy to see that π
P
= −RE
3
, where R is the radius of the sphere. In
addition, you should notice that {s
1
, s
2
, n} = {E
1
, E
2
, E
3
}.
∗
The interested reader is referred to the informative account of Routh’s inﬂuence as a tutor and
teacher in [226].
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
8.6 Constraint Forces and Constraint Moments 251
P
g O
¯
X
β
Sphere of mass m and radius R
Inclined plane
E
1
E
3
Figure 8.6. A rigid sphere moving on an inclined plane.
Let us ﬁrst consider the case in which the sphere is assumed to be sliding. In this
case, sliding condition (8.6), v
P
· n = 0, is simply
¯ v · E
3
= 0.
In addition, the slip velocity v
s
is
v
s
= v
s1
E
1
+v
s2
E
2
= ¯ v +ω ×(−RE
3
) .
To enforce the sliding condition, a force F
c
acts at the point P:
F
c
= µ
3
E
3
−µ
d
µ
3
E
3

v
s
v
s

,
where µ
d
is the coefﬁcient of dynamic friction. As in the previous example, µ
3
is a
function of time that is determined by the balance laws.
For the rolling sphere, there are three constraints:
v
P
· E
1
= 0, v
P
· E
2
= 0, v
P
· E
3
= 0.
These constraints can be expressed in an alternative form:
¯ v · E
i
−R(ω ×E
3
) · E
i
= 0,
where i = 1, 2, or 3. To enforce the three constraints, we assume that a normal
force and a static friction force act at P. The sum of these two forces is a reaction
force R:
R =
3
¸
i=1
µ
i
E
i
,
where µ
i
are functions of time that are determined from the balance laws. You
should notice that the reaction force R is equipollent to a force F
c
= R acting at
the center of mass and a moment M
c
= −RE
3
×R relative to the center of mass. A
related comment pertains to the sliding rigid body.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
252 Constraints on and Potentials for Rigid Bodies
g
P
O
¯
X
ψ
φ
E
1
E
2
E
3
e
1
e
1
e
2
e
1
e
2
Figure 8.7. A thin circular disk moving with one point in contact with a rough horizontal
surface.
8.6.3 The Rolling Disk
As shown in Figure 8.7, a thin rigid circular disk of mass m and radius R rolls (with
out slipping) on a rough horizontal plane. The rotation tensor of the disk will be
described by use of a 3–1–3 set of Euler angles (see Subsection 6.8.2). This set has
the advantage of having the singularities of the Euler angles coincide with the disk
lying ﬂat on the horizontal plane (i.e., θ = 0, π).
For the disk, the position vector of the instantaneous point of contact P relative
to the center of mass
¯
X has the representations
π
P
= −Re
2
= −Rsin(φ)e
1
−Rcos(φ)e
2
= −R(−cos(θ) sin(ψ)E
1
+cos(θ) cos(ψ)E
2
+sin(θ)E
3
) .
Taking the Cartesian components of the rolling condition, v
P
= ¯ v +ω ×π
P
, we ﬁnd
that the motion of the disk is subject to three constraints:
π
1
= 0, π
2
= 0, π
3
= 0, (8.21)
where
π
1
= ˙ x
1
+R
˙
φcos(ψ) +R
˙
ψ cos(θ) cos(ψ) −R
˙
θ sin(θ) sin(ψ),
π
2
= ˙ x
2
+R
˙
φsin(ψ) +R
˙
ψ cos(θ) sin(ψ) +R
˙
θ sin(θ) cos(ψ),
π
3
= ˙ x
3
−R
˙
θ cos(θ),
and ¯ x =
¸
3
i=1
x
i
E
i
. You should notice that the last of these three constraints inte
grates to x
3
= Rsin(θ) +constant. However, as is shown in one of the exercises at
the end of this chapter, system of constraints (8.21) is not integrable.
The rolling of the disk is possible because it is assumed that a static friction force
F
f
acts at P:
F
f
= µ
1
E
1
+µ
2
E
2
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
8.6 Constraint Forces and Constraint Moments 253
Furthermore, this force and the normal force N = µ
3
E
3
are equipollent to
F
c
= F
f
+N =
3
¸
i=1
µ
i
E
i
, M
c
= π
P
×F
c
.
Here, µ
1
, µ
2
, and µ
3
are determined from the balance laws.
8.6.4 Lagrange’s Prescription
For simplicity, we ﬁrst discuss Lagrange’s prescription for the constraint forces and
constraint moments acting on a single rigid body.
∗
Subsequently, the case of mul
tiple constraints in a system of two rigid bodies is discussed. We close this section
with an illustration of Lagrange’s prescription by using two pinjointed rigid bodies.
There we also ﬁnd a generalization of Newton’s third law. Further illustration of the
prescription applied to systems of rigid bodies can be found in Sections 11.3 and 11.4
as well as in Exercise 11.4, which features two rolling rigid bodies and the Dynabee.
A SINGLE RIGID BODY. From the examples discussed in the previous sections, we as
sume that any constraint on a rigid body can be written as [cf. (8.3)]
π = 0,
where
π = f · ¯ v +h · ω +e.
Here, f, h, and e are functions of t, ¯ x, and Q. Lagrange’s prescription states that the
constraint force F
c
and constraint moment M
c
associated with this constraint is
F
c
= µf, M
c
= µh,
where µ is indeterminate. This form of the prescription is an obvious generalization
of the corresponding prescription for a single particle and system of particles.
For a system of R constraints on the motion of a rigid body, we have R con
straints of the form
π
K
= 0,
where K = 1, . . . , R, and
π
K
= f
K
· ¯ v +h
K
· ω +e
K
.
For this system of constraints, Lagrange’s prescription states that
F
c
= µ
1
f
1
+· · · +µ
R
f
R
, M
c
= µ
1
h
1
+· · · +µ
R
h
R
,
where µ
1
, . . . , µ
R
are indeterminate. Notice that the prescription introduces R un
knowns for the R constraints.
∗
The developments presented in this section are based on the work of O’Reilly and Srinivasa [163].
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
254 Constraints on and Potentials for Rigid Bodies
By examining the examples presented in this chapter, you should be able to
verify that, apart from situations in which dynamic Coulomb friction is present, La
grange’s prescription gives physically realistic results.
∗
You should also show that,
unless e is zero for a single constraint, then the constraint force F
c
and the constraint
moment M
c
will collectively do work.
TWO RIGID BODIES. Suppose we have a system of Rconstraints acting on a system of
two rigid bodies:
π
K
= 0,
where K = 1, . . . , R. We assume that the constraints are of the form [cf. (8.3)]:
π
K
= f
K1
· ¯ v
1
+f
K2
· ¯ v
2
+h
K1
· ω
1
+h
K2
· ω
2
+e
K
.
Here, f
K1
, f
K2
, h
K1
, h
K2
, and e
K
are functions of ¯ x
1
, ¯ x
2
, Q
1
, Q
2
, and t. Lagrange’s pre
scription states that the constraint forces and constraint moments that enforce these
R constraints are
F
c1
= µ
1
f
11
+· · · +µ
R
f
R1
,
F
c2
= µ
1
f
12
+· · · +µ
R
f
R2
,
M
c1
= µ
1
h
11
+· · · +µ
R
h
R1
,
M
c2
= µ
1
h
12
+· · · +µ
R
h
R2
. (8.22)
Here, F
cα
is the resultant constraint force acting on B
α
, M
cα
is the resultant con
straint moment relative to the center of mass of B
α
acting on B
α
, and µ
1
, . . . , µ
R
are
functions of time that are determined from the balance laws for the system of two
rigid bodies.
You should notice that, for each of individual R constraints, Lagrange’s pre
scription introduces a single unknown function of time: µ
K
. Consequently, if there
are 10 constraints on the motions of the two rigid bodies, the prescription will intro
duce the unknowns µ
1
, . . . , µ
10
.
TWO PINJOINTED BODIES AND NEWTON’S THIRD LAW. For example, consider the case
of two rigid bodies connected by a pin joint shown in Figure 8.8. For this situation,
there are ﬁve integrable constraints on the relative motion of the bodies:
1
= 0, . . . ,
5
= 0,
∗
See, for example, the discussion of the constraint forces and moments on a rolling disk in Subsection
8.6.3.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
8.6 Constraint Forces and Constraint Moments 255
O
Pin joint
¯ x
1
¯ x
2
¯
X
1
¯
X
2
π
P1
π
P2
Figure 8.8. The present conﬁgurations of two rigid
bodies B
1
and B
2
that are connected by a pin joint.
The axis of the pin joint is aligned with
2
e
3
=
1
e
3
,
and its position vector relative to the centers of mass
of the bodies are π
P1
and π
P2
, respectively.
where
˙
1
= ¯ v
1
· E
1
+(ω
1
×π
P1
) · E
1
− ¯ v
2
· E
1
−(ω
2
×π
P2
) · E
1
,
˙
2
= ¯ v
1
· E
2
+(ω
1
×π
P1
) · E
2
− ¯ v
2
· E
2
−(ω
2
×π
P2
) · E
2
,
˙
3
= ¯ v
1
· E
3
+(ω
1
×π
P1
) · E
3
− ¯ v
2
· E
3
−(ω
2
×π
P2
) · E
3
,
˙
4
= ω
1
·
1
e
1
−ω
2
·
1
e
1
,
˙
5
= ω
1
·
1
e
2
−ω
2
·
1
e
2
.
We emphasize that the axis of the pin joint is aligned with
1
e
3
=
2
e
3
, and its position
vectors relative to the centers of mass of the bodies are π
P1
and π
P2
.
Using Lagrange’s prescription, (8.22), with the ﬁve constraints
1
=
0, . . . ,
5
= 0, we ﬁnd that
F
c1
=
3
¸
i=1
µ
i
E
i
,
F
c2
= −
3
¸
i=1
µ
i
E
i
,
M
c1
= µ
41
e
1
+µ
51
e
2
+π
P1
×
3
¸
i=1
µ
i
E
i
,
M
c2
= −µ
41
e
1
−µ
51
e
2
−π
P2
×
3
¸
i=1
µ
i
E
i
. (8.23)
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
256 Constraints on and Potentials for Rigid Bodies
F
c1
F
c2
M
r
−M
r
¯
X
1
¯
X
2
π
P1
π
P2
Figure 8.9. Freebody diagrams of the two rigid bodies B
1
and B
2
illustrating the constraint
forces F
c1
and F
c2
and the reaction moment M
r
. The reaction moment has the prescription
M
r
= µ
41
e
1
+µ
51
e
2
, and the constraint forces are equal and opposite [cf. (8.24)].
These expressions can be written in a more revealing fashion:
F
c1
= −F
c2
,
M
c1
= µ
41
e
1
+µ
51
e
2
+π
P1
×F
c1
,
M
c2
= −µ
41
e
1
−µ
51
e
2
+π
P2
×F
c2
. (8.24)
As illustrated in Figure 8.9, F
c1
and F
c2
are the equal and opposite reaction forces
at the pin joint, and M
r
= µ
41
e
1
+µ
51
e
2
and −µ
41
e
1
−µ
51
e
2
are the equal and op
posite reaction moments. Relative to the center of mass of each body, the reaction
moments M
c1
and M
c2
are not equal and opposite unless π
P2
= π
P1
.
∗
A NOTE OF CAUTION. For the cases of a frictionless joint between two bodies, for
two bodies in rolling contact, and for two bodies in frictionless sliding contact,
Lagrange’s prescription provides a very convenient method of specifying the con
straint forces and constraint moments. However, the prescription does not yield
physically reasonable forces and moments for joints or contact where dynamic fric
tion is present.
8.7 Potential Energies and Conservative Forces and Moments
The presence of conservative forces and moments acting on rigid bodies is one of the
key features used to solve many problems. Although the deﬁnition of a conservative
force originally arose in the dynamics of a single particle and is well understood, the
same cannot be said for conservative moments (see, for example, Antman [5] and
O’Reilly and Srinivasa [163]). Indeed, as noted by Ziegler [234], for rigid bodies
∗
This was ﬁrst noted in [163] and can be considered as a generalization of Newton’s third law to rigid
bodies.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
8.7 Potential Energies and Conservative Forces and Moments 257
whose axis of rotation is not ﬁxed, a constant moment is not necessarily conserva
tive. To this end, we start with three wellknown examples of forces and their asso
ciated moments. These examples are followed by a discussion of Ziegler’s example.
After the examples have been presented, a general treatment of conservative forces
and moments is given. Following this treatment, you should return to Ziegler’s ex
ample and convince yourself that a constant moment is not conservative.
Constant Forces
A constant force P acting on a rigid body is conservative. If this force acts at all
material points of the rigid body, then it is equipollent to a single force
R
Pρdv
acting at the center of mass of the rigid body. The potential energy of this force
is −
R
Pρdv · ¯ x. Notice that there is no moment (relative to the center of mass)
associated with this force.
The most ubiquitous example of a constant force is a constant gravitational
force acting at each material point. This force is equipollent to a force mgg acting at
the center of mass. Here g is the direction of the gravitational force. The potential
energy of this force is −mgg · ¯ x
Spring Forces
In many mechanical systems, a spring is used to couple the motions of two bodies.
Consider the system of two rigid bodies shown in Figure 8.10. Here, a linear spring
of stiffness K and unstretched length L
0
is connected to the material point X
s1
of the
body B
1
and the material point X
s2
of the body B
2
.
The spring exerts a force F
s1
at the point X
s1
and an equal and opposite force
F
s2
at the point X
s2
:
F
s1
= −K(x
s1
−x
s2
 −L
0
)
x
s1
−x
s2
x
s1
−x
s2

,
F
s2
= −K(x
s1
−x
s2
 −L
0
)
x
s2
−x
s1
x
s1
−x
s2

.
It is easy to see that each of these forces is equipollent to a moment relative to
the center of mass and a force acting at the center of mass. The potential energy
associated with the spring is
U
s
=
K
2
(x
s1
−x
s2
 −L
0
)
2
.
You should notice that this potential energy depends on the position vectors of both
material points X
s1
and X
s2
.
It is also convenient to note the identities
x
s1
= Q
1
X
s1
−
¯
X
1
+ ¯ x
1
, x
s2
= Q
2
X
s2
−
¯
X
2
+ ¯ x
2
,
where X
sα
is the position vector of X
sα
in the reference conﬁguration of B
α
. Using
these identities, we can see that the forces, moments, and potential energy of the
spring can be expressed as functions of the rotation tensors of both rigid bodies and
the position vectors of their centers of mass.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
258 Constraints on and Potentials for Rigid Bodies
O
Linear spring
¯ x
1
¯ x
2
¯
X
1
¯
X
2
X
s1
X
s2
π
s1
π
s2
Figure 8.10. The present conﬁgurations
of two rigid bodies B
1
and B
2
that are
connected by a spring.
Central Gravitational Fields
In celestial and orbital mechanics, a standard problem is to consider the motion of
a body subject to a central force ﬁeld. Such force ﬁelds date to Newton and are
based on his inversesquare force. You may recall that this conservative force is
the force exerted on a particle of mass m by a particle of mass M: F = −
GMm
r
2
r
r
,
where r is the position vector of m relative to M and Gis the universal gravitational
constant.
For the force ﬁelds of interest, we consider two bodies B
1
and B
2
with mass
densities per unit volume of ρ
1
and ρ
2
, respectively. Every material point of B
2
exerts
an attractive force on each material point of B
1
(see Figure 8.11). If we integrate
these forces over all material points in B
1
and B
2
, we will obtain the resultant force
exerted by B
2
on B
1
. Similar integrations apply to the moment and potential energy
of these forces. In short, the resultant gravitational force F
n
and moment M
n
on B
1
that are due to B
2
are
F
n
= −
R
1
R
2
G
x
1
−x
2

2
x
1
−x
2
x
1
−x
2

ρ
1
dV
1
ρ
2
dV
2
,
M
n
= −
R
1
R
2
(x
1
− ¯ x
1
) ×
G
x
1
−x
2

2
x
1
−x
2
x
1
−x
2

ρ
1
dV
1
ρ
2
dV
2
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
8.7 Potential Energies and Conservative Forces and Moments 259
O
C
Rigid body B
1
Rigid body B
2
¯ x
¯ x
1
¯ x
2
¯
X
1
¯
X
2
Figure 8.11. Two rigid bodies B
1
and B
2
that exert mutual gravitational forces. The point Cin
this ﬁgure denotes the center of mass of this system of rigid bodies.
The moment M
n
is relative to the center of mass
¯
X
1
of B
1
. The potential energy
associated with this force is
U
n
= −
R
1
R
2
G
x
1
−x
2

ρ
1
dV
1
ρ
2
dV
2
.
Assuming that B
2
is spherical and has a mass M, the expressions for the force, mo
ment, and potential energy simplify:
F
n
= −
R
1
GM
x
1
− ¯ x
2

2
x
1
− ¯ x
2
x
1
− ¯ x
2

ρ
1
dV
1
,
M
n
= −
R
1
(x
1
− ¯ x
1
) ×
GM
x
1
− ¯ x
2

2
x
1
− ¯ x
2
x
1
− ¯ x
2

ρ
1
dV
1
,
U
n
= −
R
1
GM
x
1
− ¯ x
2

ρ
1
dV
1
.
It is important to notice that, even in this simpliﬁed case, the gravitational force can
exert a moment on the rigid body B
1
.
To simplify the expressions as much as possible, we now use the fact that B
1
is a
rigid body:
π = x
1
− ¯ x
1
= Q.
In addition, we assume that π is small relative to  ¯ x
1
− ¯ x
2
. These assumptions al
low us to approximate the forces, moments, and potential energy associated with the
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
260 Constraints on and Potentials for Rigid Bodies
central force ﬁeld. After a substantial amount of manipulation, we ﬁnd the results
F
n
≈ mg,
M
n
≈
3GM
R
3
c ×(Jc) = −
3GM
R
3
c ×(Ec),
U
n
≈ −
GMm
R
−
GM
2R
3
tr(J) +
3GM
2R
3
(c · (Jc)) , (8.25)
where J is the inertia tensor of B
1
relative to its center of mass, E is the Euler tensor
of B
1
relative to its center of mass, m is the mass of B
1
,
mg = −
GMm
R
2
c −
3GM
2R
4
(2J +(tr(J) −5c · Jc) I) c,
and
R =  ¯ x
1
− ¯ x
2
 , c =
¯ x
1
− ¯ x
2
 ¯ x
1
− ¯ x
2

.
Notice that c points from the center of mass of the spherically symmetric body B
2
to
the center of mass of B
1
. In addition, because of the presence of the inertia tensor
J, (8.25)
1
does not correspond to Newton’s gravitational force on a particle of mass
m by another particle of mass M. Classical expressions (8.25) are used in the vast
majority of works on satellite dynamics (cf. Beletskii [16], Hughes [97], and Kane et
al. [106]). The expression for the potential energy U
n
in the form shown in (8.25) is
credited to James Mac Cullagh (1809–1847).
∗
In many of the works on the dynamics of a satellite about a ﬁxed point O, it
is common to approximate F
n
with F
n
= −
GMm
R
2
c. Such an approximation effectively
decouples the motion of the center of mass
¯
X
1
of B
1
fromits orientation, and one can
then conclude that
¯
X
1
behaves like a particle in the onebody problem and solve for
the orientation of the satellite separately.
†
However, as pointed out by Barkin [12]
and Wang et al. [224], this approximation violates angular momentum conservation
and energy conservation. Studies on the dynamics of satellites for which F
n
= mg
can be found in [12, 164, 224, 225]. These works show that steady motions of the
satellite are possible where the orbital plane of
¯
X
1
does not contain the origin O.
This is in contrast to the onebody problem where the orbital plane contains O.
Constant Moments That Are Not Conservative
To see that a constant moment is not conservative, we recall Ziegler’s example [234].
He considered a constant moment ME
3
. During a motion of a rigid body consisting
of a rotation about E
3
through −π rad, this moment does work equal to Mπ. How
ever, the same ﬁnal orientation of the body can be achieved by a rotation about E
1
through π rad, followed by a rotation about E
2
through π rad.
‡
Now, however, the
∗
Notes on his derivation of U
n
and F
n
can be found in Propositions 4 and 5 of [1]. Comments by
Allman on the derivation of M
n
can also be found in [1].
†
The onebody problem was discussed earlier in Section 2.8.
‡
This is an example of the application of Rodrigues–Hamilton theorem (6.5) discussed in the exer
cises at the end of Chapter 6.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
8.7 Potential Energies and Conservative Forces and Moments 261
work done by the constant moment is zero! Consequently, the work done by the
moment ME
3
depends on the “path” taken by the body – and hence ME
3
cannot be
conservative.
In many courses on dynamics, the rotation of the rigid body is constrained to be
a ﬁxedaxis rotation and the rotations about E
1
and E
2
are not permitted. For these
cases, a constant moment ME
3
is conservative. An alternative proof that a constant
moment is generally not conservative can be found in [160]. The proof presented
there exploits the dual Euler basis.
General Considerations
It is convenient at this point to give a general treatment of conservative forces and
moments in the dynamics of rigid bodies. Our discussion is in the context of two rigid
bodies, but it is easily simpliﬁed to the case of one rigid body and easily generalized
to the case of N rigid bodies.
We assume that the most general form of the potential energy is
U = U(¯ x
1
, ¯ x
2
, Q
1
, Q
2
) .
Clearly, this function depends on the motions of both rigid bodies and time. We can
calculate the time derivative of this function by using (8.10):
˙
U =
∂U
∂ ¯ x
1
· ¯ v
1
+
∂U
∂ ¯ x
2
· ¯ v
2
+u
Q
1
· ω
1
+u
Q
2
· ω
2
,
where u
Q
α
is a vector representing the derivative of U with respect to Q
α
. As dis
cussed in Section 6.10, this vector has numerous representations, and the easiest to
use arises when Q
1
and Q
2
are parameterized by sets of Euler angles. Denoting
these angles and their dual Euler basis vectors by
¸
γ
1
, γ
2
, γ
3
¸
and
¸
g
1
, g
2
, g
3
¸
, and
¸
ν
1
, ν
2
, ν
3
¸
and
¸
h
1
, h
2
, h
3
¸
, respectively, the vectors u
Q
1
and u
Q
2
have the represen
tations
u
Q
1
=
3
¸
k=1
∂U
∂γ
k
g
k
, u
Q
2
=
3
¸
k=1
∂U
∂ν
k
h
k
.
Consider the conservative forces F
conα
and moments M
conα
(relative to the re
spective centers of mass) associated with this potential. We assume that the work
done by these forces and moments is dependent on the initial and ﬁnal conﬁgura
tions of the rigid bodies but is independent of the motions of the rigid bodies. This
implies that
−
˙
U = F
con1
· ¯ v
1
+F
con2
· ¯ v
2
+M
con1
· ω
1
+M
con2
· ω
2
.
Substituting for
˙
Uand collecting terms, we ﬁnd that
F
con1
+
∂U
∂ ¯ x
1
· ¯ v
1
+
F
con2
+
∂U
∂ ¯ x
2
· ¯ v
2
+(M
con1
+u
Q
1
) · ω
1
+ (M
con2
+u
Q
2
) · ω
2
= 0. (8.26)
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
262 Constraints on and Potentials for Rigid Bodies
This can be interpreted as an equation for F
conα
and M
conα
that must hold for all
motions of the rigid bodies. Assuming that F
conα
and M
conα
are independent of the
linear and angular velocity vectors, we ﬁnd that in order for (8.26) to hold for all ¯ v
α
and ω
α
it is necessary and sufﬁcient that
F
conα
= −
∂U
∂ ¯ x
α
, M
conα
= −u
Q
α
. (8.27)
These are the expressions for the conservative forces and moments associated with
a potential energy. It is left as an exercise for you to verify that these expressions are
consistent with the results presented earlier for the spring and central gravitational
forces.
An Example Featuring a Torsional Spring
We now present a very simple example featuring a torsional spring acting on a single
rigid body. This example illustrates representations (8.27) and makes use of the dual
Euler basis. Even in this simple case, the conservative moment associated with the
spring has a surprising feature.
Consider a single rigid body whose rotation tensor is parameterized by a set of
3–1–3 Euler angles and suppose that a torsional spring acts on the body. The spring
is assumed to have a potential energy U =
K
t
2
ψ
2
, where K
t
is a torsional stiffness.
Then, with the help of (8.27) and representation (8.9) for u
Q
in terms of Euler angles
we ﬁnd that the conservative moment is
M
con
= −u
Q
= −
∂U
∂ψ
g
1
= −K
t
ψg
1
= −K
t
ψ (cot(θ) (cos(ψ)E
2
−sin(ψ)E
1
) +E
3
) .
The conservative force F
con
associated with this potential is 0 because the potential
is independent of ¯ x. It is interesting to notice that the components of M
con
are not
entirely in the direction of E
3
.
8.8 Concluding Comments
The main results in this chapter involved prescriptions for constraint and conserva
tive forces and moments. For the former, we argued that Lagrange’s prescription
provided these expressions in cases in which dynamic Coulomb friction was absent.
For instance, if a single rigid body is subject to a constraint that can be expressed as
f · ¯ v +h · ω +e = 0,
then Lagrange’s prescription states that
F
c
= µf, M
c
= µh.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
Exercises 8.1–8.2 263
If we suppose that a single rigid body has a potential energy function
U = U
x
1
, x
2
, x
3
, γ
1
, γ
2
, γ
3
,
where x
i
are the Cartesian coordinates for ¯ x and γ
i
are Euler angles, then our de
velopments showed that the conservative force F
con
and conservative moment M
con
associated with this potential are
F
con
= −
3
¸
i=1
∂U
∂x
i
E
i
, M
con
= −
3
¸
i=1
∂U
∂γ
i
g
i
,
where g
i
are the dual Euler basis vectors associated with the Euler angles.
EXERCISES
8.1. Suppose the motion of a rigid body is subject to two constraints:
ω · g
3
= 0, ω · g
2
= 0.
Here, g
i
are the dual Euler basis for a set of Euler angles of your choice. Give a phys
ical interpretation of these constraints. What are the angular velocity vector, angular
momentum vector, and rotational kinetic energy of the resulting constrained rigid
body?
8.2. Consider a force P acting at a point P of a rigid body. In the present conﬁgura
tion, the point P has the position vector π
P
relative to the center of mass
¯
X of the
rigid body:
π
P
= x
P
− ¯ x.
In addition,
v
P
= ˙ x
P
= ¯ v +ω ×π
P
.
(a) The mechanical power of P is P · v
P
. Show that this power has the equiva
lent representation
P · v
P
= P · ¯ v +(π
P
×P) · ω.
Using a freebody diagram, give a physical interpretation of this identity.
(b) A force P acting at the point P is said to be conservative if there exists a
potential energy U = U(x
P
) such that
P = −
∂U
∂x
P
.
Show that this deﬁnition implies that
−
˙
U = P · ¯ v +(π
P
×P) · ω.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
264 Exercises 8.2–8.3
(c) Show that the potential energy U = U(x
P
) can also be described as a func
tion of ¯ x, Q,
¯
X and
P
:
U = U(x
P
) =
˜
U
¯ x, Q,
P
,
¯
X
.
Here, Q is the rotation tensor of the rigid body, and π
P
= Q
P
. Using this
equivalence, show that
P = −
∂
˜
U
∂ ¯ x
, π
P
×P = −u
Q
.
(d) Consider the case in which P represents a force that is due to a spring of
stiffness K and unstretched length L. One end of the spring is attached to
a ﬁxed point O. What are the functions U(x
P
) and
˜
U(¯ x, Q,
P
,
¯
X) for this
force P?
8.3. Consider two rigid bodies. The rotation tensor of the ﬁrst rigid body is
Q
1
=
3
¸
i=1
1
e
i
⊗E
i
.
The rotation tensor of the second rigid body is
Q
2
=
3
¸
i=1
2
e
i
⊗E
i
.
Here,
1
e
i
corotate with the ﬁrst rigid body and
2
e
i
corotate with the second rigid
body.
(a) Argue that the rotation tensor of the second body relative to the ﬁrst body
is
R = Q
2
Q
T
1
=
3
¸
i=1
2
e
i
⊗
1
e
i
.
What is the rotation tensor of the ﬁrst body relative to the second
body?
(b) Show that the angular velocity vector of the second body relative to the ﬁrst
body is
ˆ ω = −
1
2
¸
3
¸
i=1
2
o
e
i
⊗
2
e
i
,
where we have used the corotational derivative with respect to Q
1
:
o
a= ˙ a −ω
1
×a.
(c) Suppose a set of 3–2–1 Euler angles is used to parameterize Q
1
and a set
of 1–3–1 Euler angles is used to parameterize R. Show that the angular
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
Exercises 8.3–8.4 265
velocity vectors of both bodies and their relative angular velocity vector
have the representations
ω
1
= ˙ γ
1
E
3
+ ˙ γ
2
1
e
2
+ ˙ γ
3
1
e
1
,
ω
2
= ˙ ν
1
1
e
1
+ ˙ ν
2
2
e
3
+ ˙ ν
3
2
e
1
+ ˙ γ
1
E
3
+ ˙ γ
2
1
e
2
+ ˙ γ
3
1
e
1
,
ˆ ω = ˙ ν
1
1
e
1
+ ˙ ν
2
2
e
3
+ ˙ ν
3
2
e
1
.
(d) Suppose the rotation of the second body relative to the ﬁrst body is con
strained such that
ˆ ω = ˙ ν
2
2
e
3
.
Give a physical interpretation of the type of joint needed to enforce this
constraint. In addition, give prescriptions for the constraint moments acting
on both bodies.
8.4. As shown in Figure 8.12, a tippe top is a body with an axis of symmetry. One
of its lateral surfaces can be approximated as a spherical surface of radius R and
the other is a cylinder of radius r. The top is designed so that the center of mass
¯
X
is located below the center of the sphere, and this feature leads to its ability to ﬂip
over (see Figure 8.13).
∗
g
A
P
R
¯
X
E
1
E
3
e
3
Figure 8.12. A tippe top moving on
a rough horizontal plane.
The instantaneous point of contact of the spherical portion of the top with the
horizontal plane is denoted by P. The position vectors of P relative to
¯
X and the
point Arelative to
¯
X are
x
P
− ¯ x = −RE
3
+le
3
, x
A
− ¯ x = he
3
,
respectively.
(a) Using a set of 3–1–3 Euler angles, establish expressions for ω
i
= ω · e
i
and
i
= ω · E
i
. For which orientations of the tippe top does this set of Euler
angles have singularities?
∗
The dynamics of the tippe top has been the subject of several investigations, and the interested
reader is referred to [20, 154, 156, 177, 202, 219] for discussions and references. Among other issues,
these papers point out the important role played by friction forces acting at the point of contact.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
266 Exercises 8.4–8.5
g
A
A
P
¯
X
¯
X
E
1
E
3
e
3
e
3
ω
0
ω
1
Figure 8.13. The two steady motions of a tippe top: the upright position in which e
3
= E
3
and
the inverted state in which e
3
= −E
3
.
(b) Show that the slip velocities of the point P of the tippe top have the repre
sentations
v
s1
= ˙ x
1
−
2
(R−l cos(θ)) +
3
(l sin(θ) cos(ψ)) ,
v
s2
= ˙ x
2
+
1
(R−l cos(θ)) +
3
(l sin(θ) sin(ψ)) ,
where v
P
= v
s1
E
1
+v
s2
E
2
and ¯ v =
¸
3
k=1
˙ x
k
E
k
.
(c) Suppose that the tippe top is sliding. Give a prescription for the constraint
force F
c
and moment M
c
that enforce the sliding constraint.
(d) Suppose that the tippe top is rolling. Give a prescription for the constraint
force F
c
and moment M
c
that enforce the rolling constraints.
(e) With the help of (8.13), show that two of the three constraints on a rolling
tippe top are nonintegrable.
8.5. Consider the mechanical system shown in Figure 8.14. It consists of a rigid body
of mass m that is free to rotate about a ﬁxed point O. The joint at Odoes not permit
the body to have a spin. A vertical gravitational force mgE
1
acts on the body. The
inertia tensor of the body relative to its center of mass C is
J
0
= λ
1
E
1
⊗E
1
+
λ −mL
2
0
(E
2
⊗E
2
+E
3
⊗E
3
) .
g
O
¯
X
φ
θ
E
1
E
2
E
3
e
1
Figure 8.14. A pendulum problem.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
Exercises 8.5–8.6 267
The position vector of the center of mass
¯
X of the body relative to Ois L
0
e
1
.
To parameterize the rotation tensor of the body, we use a set of 1–3–1 Euler
angles: g
1
= E
1
, g
2
= e
3
and g
3
= e
1
. For these angles
ω =
˙
φE
1
+
˙
θe
3
+
˙
ψe
1
=
˙
ψ +
˙
φcos(θ)
e
1
+
˙
θ sin(ψ) −
˙
φsin(θ) cos(ψ)
e
2
+
˙
θ cos(ψ) +
˙
φsin(θ) sin(ψ)
e
3
.
(a) Which orientations of the rigid body coincide with the singularities of the
Euler angles?
(b) Derive expressions for the unconstrained potential Uand kinetic Tenergies
of the rigid body.
(c) Show that the motion of the rigid body is subject to four constraints:
˙
ψ = ω · g
3
= 0, ¯ v −ω ×(L
0
e
1
) = 0.
(d) Derive expressions for the constrained potential U and kinetic T energies
of the rigid body.
8.6. As shown in Figure 8.15, an axisymmetric rigid body is free to rotate about the
ﬁxed point O. The body, which has a mass m, has an inertia tensor
J = λ
a
e
3
⊗e
3
+λ
t
(I −e
3
⊗e
3
) .
The position vector of the center of mass of this body relative to Ois
¯ x = L
1
e
3
.
O
¯
X
e
3
E
1
E
2
E
3
g
Figure 8.15. Arigid body that is free to rotate about the ﬁxed
point O. A set of 3–1–3 Euler angles is used to describe the
rotation tensor Q of this body.
(a) Using a set of 3–1–3 Euler angles, show that the angular velocity vector ω
also has the representation
ω =
˙
θe
1
+
˙
ψ sin(θ)e
2
+(
˙
φ +
˙
ψ cos(θ))e
3
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
268 Exercises 8.6–8.8
(b) Show that the angular momentum H is
H = λ
t
˙
θe
1
+λ
t
˙
ψ sin(θ)e
2
+λ
a
˙
φ +
˙
ψ cos(θ)
e
3
.
(c) Show that the angular momentum of the center of mass relative to Ois
mL
2
1
(ω
1
e
1
+ω
2
e
2
) ,
where ω
i
= ω · e
i
. What is the angular momentum of the rigid body relative
to O?
(d) Show that the kinetic energy T is
T =
λ
t
+mL
2
1
2
˙
θ
2
+
˙
ψ
2
sin
2
(θ)
+
λ
a
2
˙
φ +
˙
ψ cos(θ)
2
.
(e) A conservative moment that is due to a torsional spring is applied to the
rigid body. As a result, the total potential energy of the rigid body is
U = mg¯ x · E
3
+
K
2
ψ
2
.
Here, K is the torsional spring constant. What are the conservative force
F
con
and moment M
con
acting on the rigid body?
8.7. Recall Ziegler’s example of a constant moment ME
3
that was not conserva
tive. After choosing a set of Euler angles,
¸
γ
1
, γ
2
, γ
3
¸
, show that you cannot ﬁnd a
potential energy function Usuch that ME
3
= −
¸
3
i=1
∂U
∂γ
i
g
i
.
∗
8.8. Suppose that a linear spring of stiffness K and unstretched length L
0
is attached
to a point A on a rigid body and the other end of the spring is attached to a ﬁxed
point O. The position vector of Arelative to the center of mass of the rigid body is
π
A
= Re
3
.
(a) With the assistance of a set of 3–1–3 Euler angles, show that
x
A
= (x
1
+Rsin(θ) sin(ψ)) E
1
+(x
2
−Rsin(θ) cos(ψ)) E
2
+(x
3
+Rcos(θ)) E
3
.
In this equation, ¯ x =
¸
3
k=1
x
k
E
k
.
(b) With the assistance of a set of 3–1–3 Euler angles, show that the potential
energy of the spring is
U = U(¯ x, Q) =
K
2
√
u −L
0
2
,
where
u = (x
1
+Rsin(θ) sin(ψ))
2
+(x
2
−Rsin(θ) cos(ψ))
2
+(x
3
+Rcos(θ))
2
.
∗
One solution to this problem can be found in [160].
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
Exercises 8.8–8.10 269
(c) Describe two equivalent methods to ﬁnd the conservative force F
con
and
conservative moment M
con
associated with the spring.
8.9. Consider the rolling disk discussed in Subsection 8.6.3 and consider constraints
(8.21) on its motion. Here, we wish to show that this family of constraints is non
integrable by using Frobenius’ theorem.
To use the theorem, we ﬁrst deﬁne the following seven variables:
U
1
= x
1
, U
2
= x
2
, U
3
= x
3
,
U
4
= φ, U
5
= ψ, U
6
= θ, U
7
= t.
(a) With the help of (8.15), compute the 3 ×7 matrix W. Determine a
1
, . . .,a
4
that span the null space (kernel) of this matrix. You will ﬁnd that one of
these vectors is
a
1
=
¸
0 0 0 0 0 0 1
¸
T
.
(b) With the help of (8.15), calculate the three 7 ×7 skewsymmetric matrices
S
1
, S
2
, and S
3
corresponding to π
1
, π
2
, and π
3
, respectively. You will ﬁnd
that these matrices have at most two nonzero elements.
(c) Using the results of (a) and (b), show that Frobenius’ theorem implies that
the rolling disk is subject to nonintegrable constraints. You should also indi
cate how your results can be used to conclude that the sliding disk is subject
to an integrable constraint.
(d) Suppose the rolling disk is subject to two additional integrable constraints:
x
2
= 0, ψ = 0.
That is, the disk is constrained to roll vertically in a straight line. Compute
the 5 ×7 matrix W and show that its null space is spanned by a
1
and a
2
,
where
a
1
=
¸
0 0 0 0 0 0 1
¸
T
,
a
2
=
¸
−R 0 0 1 0 0 0
¸
T
.
In addition, show that the skewsymmetric matrices S
4
and S
5
correspond
ing to the additional constraints are both zero. Finally, invoking Frobenius’
theorem, show that the family of constraints π
1
= 0, π
2
= 0, π
3
= 0, ˙ x
2
= 0,
and
˙
ψ = 0 is integrable.
8.10. A schematic of a Grifﬁn grinding machine is shown in Figure 8.16.
∗
The grain
to be milled is placed in the bin, and a roller is designed to roll without slipping on
the inner wall of the grain bin. The normal force generated by the roller is substantial
and crushes the grains. The motion of the roller is achieved by two drive shafts. The
∗
This ﬁgure is adapted from Arnold and Mauder [8] and Webster [227].
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
270 Exercise 8.10
O
P
U
e
3
E
1
E
2
E
3
Universal joint
Grain bin
Roller
Drive shaft I
Drive shaft II
ψ
φ
γ
Figure 8.16. Schematic of a Grifﬁn grinding machine.
ﬁrst shaft (drive shaft I) has an angular velocity vector ω
I
=
˙
ψE
3
. It is coupled by
a universal joint at U to the second shaft. The roller is attached to the second shaft
(drive shaft II) by a joint that allows it to have a rotation relative to the second shaft
in the direction of e
3
. The basis {e
1
, e
2
, e
3
} corotates with the roller.
(a) Using a set of 3–1–3 Euler angles {ψ, π −γ, φ}, show that the angular veloc
ity vector of the roller has the representation
ω =
˙
ψ (sin(φ) sin(π −γ)e
1
+cos(φ) sin(π −γ)e
2
+cos(π −γ)e
3
)
− ˙ γ (cos(φ)e
1
−sin(φ)e
2
) +
˙
φe
3
.
For which orientations of the roller does this set of Euler angles have singu
larities? What is the angular velocity vector ω
II
of drive shaft II, and what is
the angular velocity of the roller relative to drive shaft II?
(b) The center of mass
¯
X of the roller has a position vector relative to the ﬁxed
point Uof
¯ x = He
3
.
Establish an expression for the velocity vector ¯ v of the point
¯
X.
(c) The instantaneous point P of contact of the roller with the grain bin has the
following position vector relative to
¯
X:
π
P
= −r (cos(φ)e
2
+sin(φ)e
1
) .
Using the identity v
P
= ω ×π
P
+ ¯ v, establish expressions for the compo
nents v
P
· e
i
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
Exercise 8.10 271
(d) If v
P
= 0, then show that γ is a constant and the rotational speeds
˙
ψ and
˙
φ
are related:
˙
φ =
cos (γ
0
) −
H
r
sin(γ
0
)
˙
ψ.
In this equation, γ
0
is the constant value of the angle γ.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
9 Kinetics of a Rigid Body
9.1 Introduction
In this chapter, the balance laws F=m
˙
¯ v and M=
˙
H for a rigid body are discussed.
Several component forms of these laws are considered. For instance, the compo
nents of the balance of angular momentum with respect to the corotational basis e
i
,
M· e
i
=
˙
H· e
i
lead to a set of equations [(9.9)] that are known as Euler’s equations.
In a subsequent chapter, we shall show that M· g
i
=
˙
H· g
i
lead to Lagrange’s
equations.
Once the balance laws have been discussed, a brief outline is given of the work–
energy theorem, and it is shown how to establish energy conservation for a rigid
body. Our attention then turns to applications. In particular, we discuss the dynam
ics of a body rotating about a ﬁxed point, momentfree motion of a rigid body,
rolling and sliding spheres, and the dynamics of baseballs and footballs. Several
other applications are discussed in the exercises. Despite all these examples, our
consideration of applications is far from exhaustive. At the conclusion of this chap
ter, some details are provided on some other interesting applications that have been
modeled by use of a single rigid body.
9.2 Balance Laws for a Rigid Body
Euler’s laws for a rigid body can be viewed as extensions to Newton’s second law for
a single particle. There are two laws, or postulates, the balance of linear momentum
and the balance of angular momentum:
˙
G = F,
˙
H
O
= M
O
. (9.1)
Here, H
O
is the angular momentum of the rigid body relative to a ﬁxed point Oand
M
O
is the resultant external moment relative to O. As noted in Truesdell [216, 217],
these laws are discussed in several of Euler’s works on rigid body dynamics and ﬁnd
their deﬁnitive form in Euler’s paper [56] that was published in 1776.
∗
∗
The interested reader is referred to [135, 230], where additional commentaries and perspectives on
these balance laws can be found. For additional background on the problem of the precession of
the equinoxes featured in [230], the text of Hestenes [93] is recommended.
272
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
9.2 Balance Laws for a Rigid Body 273
In many cases, it is convenient to give an alternative description of the balance
of angular momentum. To do this, we start with the identity
H
O
= H+ ¯ x ×G.
Differentiating, and using the balance of linear momentum, we obtain
˙
H
O
=
˙
H+ ¯ v ×G+ ¯ x ×
˙
G
=
˙
H+ ¯ x ×F.
Hence, invoking the balance of angular momentum,
˙
H
O
= M
O
, we ﬁnd that
M
O
=
˙
H
O
=
˙
H+ ¯ x ×F.
However, the resultant moment relative to a ﬁxed point O, M
O
, and the resultant
moment relative to the center of mass
¯
X, M, are related by
∗
M
O
= M+ ¯ x ×F.
It follows that
M =
˙
H,
which is known as the balance of angular momentum relative to the center of mass
¯
X. This form of the balance law is used in many problems for which the rigid body
has no ﬁxed point O.
In summary, the balance laws for a rigid body are known as Euler’s laws. Two
equivalent sets of these laws are used. For the ﬁrst set, the balance of angular
momentum relative to a ﬁxed point Ofeatures
˙
G = F,
˙
H
O
= M
O
. (9.2)
In the second set, the balance of angular momentum is taken relative to the center
of mass
¯
X:
˙
G = F,
˙
H = M. (9.3)
It should be noted that
G = m¯ v, H = Jω, H
O
= Jω + ¯ x ×m¯ v.
Here, Ois the origin of the coordinate system used to deﬁne ¯ x.
To determine the motion of the rigid body, it sufﬁces to know ¯ x(t) and Q(t). To
obtain these results, Equations (9.2) or (9.3) must be supplemented by information
on the coordinate system used to parameterize ¯ x and the parameterization of Q.
∗
This may be seen from our previous discussion of a system of forces and moments acting on a rigid
body in Section 8.5.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
274 Kinetics of a Rigid Body
9.3 Work and Energy Conservation
The work–energy theorem for a rigid body equates the rate of change of kinetic
energy to the mechanical power of the external forces and moments acting on the
rigid body. There are two equivalent forms of this theorem:
˙
T = F · ¯ v +M· ω
=
N
¸
K=1
F
K
· v
K
+M
p
· ω. (9.4)
The second form is the most useful for proving that energy is conserved in a speciﬁc
problem. After proving the theorem, we will close this section with a discussion of
energy conservation.
Proving the Work–Energy Theorem
The difﬁculty in establishing the work–energy theorem lies in dealing with the an
gular momentum H = Jω. To overcome this, it is easiest to ﬁrst show that
ω ·
˙
Jω
= 0. (9.5)
This result is established by use of an earlier result,
˙
J = J −J, and the identity
T
ω = −ω ×ω = 0:
˙
Jω
· ω = ((J −J) ω) · ω
= (Jω) · ω −(J (ω ×ω)) · ω
= Jω ·
T
ω
+(J (0)) · ω
= 0.
Now, as
˙
Jω · ω = 0,
˙
H· ω =
˙
Jω
· ω = (J ˙ ω) · ω = (Jω) · ˙ ω. (9.6)
In the last step, we invoked the symmetry of J. From (9.6), we can conclude that
˙
H· ω = H· ˙ ω, (9.7)
a result that we shall presently invoke.
To prove work–energy theorem (9.4), recall the Koenig decomposition for the
kinetic energy T of a rigid body:
T =
1
2
m¯ v · ¯ v +
1
2
(Jω) · ω.
Differentiating T and using (9.5) and (9.7), we ﬁnd that
˙
T =
˙
G· ¯ v +
1
2
˙
H· ω +(Jω) · ˙ ω
=
˙
G· ¯ v +
˙
H· ω.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
9.3 Work and Energy Conservation 275
Invoking the balances of linear and angular momentum, work–energy theorem
(9.4)
1
is established:
˙
T = F · ¯ v +M· ω.
To establish the alternative form, (9.4)
2
, of the work–energy theorem, we con
sider a system of N forces F
K
acting at N material points X
K
of the rigid body and a
pure moment M
p
acting on the rigid body. For such a system of forces and moments,
we recall that
F =
N
¸
K=1
F
K
, M =
N
¸
K=1
(x
K
− ¯ x) ×F
K
+M
p
.
Noting that v
K
= ¯ v +ω ×(x
K
− ¯ x), we ﬁnd with some rearranging that
F · ¯ v +M· ω = M
p
· ω +
N
¸
K=1
F
K
· v
K
.
This result establishes the alternative form, (9.4)
2
, of the work–energy theorem.
Energy Conservation
In most problems in rigid body dynamics in which there is no dynamic friction, the
total energy Eof the body is conserved. To showthis, let us assume that the resultant
conservative force F
con
and conservative moment (relative to the center of mass)
M
con
acting on the body are associated with a potential energy U = U(¯ x, Q):
F
con
= −
∂U
∂ ¯ x
, M
con
= −u
Q
.
In addition, we assume that the body is subject to R constraints of the form
L
f · ¯ v +
L
h · ω +
L
e = 0 (L = 1, . . . , R).
The constraint force and moment are prescribed by use of Lagrange’s prescription:
F
c
=
R
¸
L=1
µ
L
(
L
f) , M
c
=
R
¸
L=1
µ
L
(
L
h) .
Finally, we assume that the only forces and moments acting on the rigid body are
conservative or constraint.
We now examine work–energy theorem (9.4)
1
for the rigid body of interest:
˙
T = F · ¯ v +M· ω
= F
c
· ¯ v +M
c
· ω +F
con
· ¯ v +M
con
· ω.
However,
F
c
· ¯ v +M
c
· ω =
R
¸
L=1
µ
L
(
L
f · ¯ v +
L
h · ω) = −
R
¸
L=1
µ
L
(
L
e) ,
F
con
· ¯ v +M
con
· ω = −
∂U
∂ ¯ x
· ¯ v −u
Q
· ω = −
˙
U.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
276 Kinetics of a Rigid Body
Consequently,
˙
T = −
˙
U−
R
¸
L=1
µ
L
(
L
e) .
This implies that
˙
E = 0, where E = T +U, if
¸
R
L=1
µ
L
(
L
e) = 0.
In summary, one situation in which the total energy E is conserved arises when
1
e = 0, . . . ,
R
e = 0, and the constraint forces and constraint moments are prescribed
by Lagrange’s prescription. This situation arises in most of the problems in rigid
body dynamics that are solvable analytically, and we shall shortly see several
examples.
9.4 Additional Forms of the Balance of Angular Momentum
The balance of angular momentum
˙
H = M has several component forms and is one
of the most interesting equations in mechanics. In this section, several of these forms
are discussed. First, we show that this equation is equivalent to
J
3
¸
i=1
˙ ω
i
e
i
+ω ×Jω = M.
Next, we show that, if e
i
are the principal vectors of J, then we can ﬁnd Euler’s cele
brated equations (9.9) As an intermediate result, we also indicate the corresponding
component form of these equations when e
i
are not principal vectors of J.
A Direct Form
Here, we wish to show that
˙
H = M can be written as
Jα +ω ×(Jω) = M.
To establish this result, we need to examine ˙ ω and
˙
J.
First, we recall that
H = Jω.
Taking the derivative of this expression, we ﬁnd
˙
H =
˙
Jω +J ˙ ω.
To proceed, we need some identities. Speciﬁcally,
α = ˙ ω =
d
dt
3
¸
i=1
ω
i
e
i
=
3
¸
i=1
˙ ω
i
e
i
+ω ×
3
¸
i=1
ω
i
e
i
=
3
¸
i=1
˙ ω
i
e
i
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
9.4 Additional Forms of the Balance of Angular Momentum 277
and
˙
J = J +J
T
,
˙
Jω = Jω −Jω = ω ×Jω.
Using the identities, we ﬁnd that
˙
H =
˙
Jω +J ˙ ω = ω ×Jω +J ˙ ω
= J
3
¸
i=1
˙ ω
i
e
i
+ω ×Jω.
In summary,
˙
H = M is equivalent to
J
3
¸
i=1
˙ ω
i
e
i
+ω ×Jω = M.
This form of
˙
H = M is very useful when J is a constant tensor – for instance when
dealing with rigid spheres and rigid cubes. It is also used to obtain conservation
results for H.
In passing, we note that the result α =
¸
3
i=1
˙ ω
i
e
i
implies that for ω to be constant
it sufﬁces that ω
i
are constant. This is in spite of the fact that e
i
may not be stationary.
A Component Form
If we choose an arbitrary basis {E
1
, E
2
, E
3
} for E
3
, then the inertia tensors J
0
and J
have the representations
J
0
=
3
¸
i=1
3
¸
k=1
J
ik
E
i
⊗E
k
, J =
3
¸
i=1
3
¸
k=1
J
ik
e
i
⊗e
k
,
where e
i
= QE
i
. Consequently,
H = Jω =
3
¸
i=1
3
¸
k=1
J
ik
ω
k
e
i
.
Differentiating H, we ﬁnd that
˙
H =
˙
Jω =
3
¸
i=1
3
¸
k=1
J
ik
˙ ω
k
e
i
+ω ×
3
¸
i=1
3
¸
k=1
J
ik
ω
k
e
i
. (9.8)
If we equate this expression to M, we can ﬁnd the component forms of
˙
H = M.
However, except when ω has a simple form, it is not convenient to consider this
component form of the equations. Examples of where (9.8) are used include mis
balanced rotors, where ω =
˙
θE
3
and J
13
= 0 and/or J
23
= 0.
∗
For cases not involving
a ﬁxed axis of rotation, it is prudent to choose {E
1
, E
2
, E
3
} to be the principal
directions of J
0
.
∗
An example of such a system can be found in Section 7 of Chapter 9 in [159].
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
278 Kinetics of a Rigid Body
The Principal Axis Case
If we choose E
i
to be the principal directions of J
0
, then this tensor has the familiar
representation
J
0
=
3
¸
i=1
λ
i
E
i
⊗E
i
,
where λ
i
are the principal moments of inertia. The vectors E
i
are also known as the
principal axes of the body in its reference conﬁguration.
Deﬁning e
i
= QE
i
, we ﬁnd that the inertia tensor J = QJ
0
Q
T
has the represen
tation J =
¸
3
i=1
λ
i
e
i
⊗e
i
. Consequently,
H = Jω =
3
¸
i=1
λ
i
ω
i
e
i
.
Evaluating
˙
H, we ﬁnd
˙
H = J ˙ ω +ω ×Jω
= J
3
¸
i=1
˙ ω
i
e
i
+ω ×
3
¸
i=1
λ
i
ω
i
e
i
=
3
¸
i=1
λ
i
˙ ω
i
e
i
+(λ
2
−λ
1
)ω
1
ω
2
e
3
+(λ
1
−λ
3
)ω
1
ω
3
e
2
+(λ
3
−λ
2
)ω
3
ω
2
e
1
.
In conclusion,
˙
H = M has the component form
λ
1
˙ ω
1
+(λ
3
−λ
2
)ω
3
ω
2
= M· e
1
,
λ
2
˙ ω
2
+(λ
1
−λ
3
)ω
3
ω
1
= M· e
2
,
λ
3
˙ ω
3
+(λ
2
−λ
1
)ω
1
ω
2
= M· e
3
. (9.9)
These equations, known as Euler’s equations, represent three ﬁrstorder ordinary
differential equations for ω
i
.
To determine the rotation tensor Q, it is necessary to supplement (9.9) by the
three ﬁrstorder ordinary differential equations relating ω to Q,
ω = −
1
2
¸
˙
QQ
T
¸
.
For example, if a set of 3–2–1 Euler angles were used to parameterize Q, then these
differential equations would be
⎡
⎢
⎣
˙
ψ
˙
θ
˙
φ
⎤
⎥
⎦
=
⎡
⎢
⎣
0 sin(φ) sec(θ) cos(φ) sec(θ)
0 cos(φ) −sin(φ)
1 sin(φ) tan(θ) cos(φ) tan(θ)
⎤
⎥
⎦
⎡
⎢
⎣
ω
1
ω
2
ω
3
⎤
⎥
⎦
. (9.10)
You may wish to recall that the 3–2–1 Euler angles were discussed in Subsec
tion 6.8.1, and differential equations (9.10) can be inferred from the developments
there.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
9.5 MomentFree Motion of a Rigid Body 279
9.5 MomentFree Motion of a Rigid Body
Momentfree motion of a rigid body occurs when M = 0. Determining the motion
is resolved by solutions of the balance laws,
˙
G = F,
˙
H = 0,
for ¯ x and Q. It is common to focus exclusively on the balance of angular momen
tum and determine Q. In addition, although an analytical solution for Q was ﬁrst
found by Carl G. J. Jacobi (1804–1851) in 1849,
∗
it is usual to focus on ω(t). Another
ingenious solution to this problem was presented by Poinsot in 1834 [172].
†
The equations governing the components ω
i
= ω · e
i
of the angular velocity vec
tor are found from the three equations
˙
H· e
i
= 0 [cf. (9.9)]:
λ
1
˙ ω
1
+(λ
3
−λ
2
)ω
3
ω
2
= 0,
λ
2
˙ ω
2
+(λ
1
−λ
3
)ω
3
ω
1
= 0,
λ
3
˙ ω
3
+(λ
2
−λ
1
)ω
1
ω
2
= 0. (9.11)
It is easy to see that the solutions to these equations conserve the rotational kinetic
energy T
rot
=
1
2
H· ω and the angular momentum vector H.
Although there are several cases to consider, it sufﬁces to consider three:
symmetric body: λ
1
= λ
2
= λ
3
;
axisymmetric body: λ
1
= λ
2
= λ
3
;
asymmetric body: λ
1
< λ
2
< λ
3
.
For the axisymmetric body, when λ
1
< λ
3
the body is known as oblate. When
λ
1
= λ
2
> λ
3
, the body is known as prolate. For the asymmetric body previously
discussed, e
1
is known as the minor axis of inertia, e
2
is known as the intermediate
axis of inertia, and e
3
is known as the major axis of inertia. We nowturn to discussing
the three cases and the solutions for ω
i
(t).
The Symmetric Body
For the symmetric rigid body, (9.11) simplify to ˙ ω
i
= 0. In other words, the com
ponents of ω are constant. As ˙ ω =
¸
3
k=1
˙ ω
k
e
k
, this implies that ω is constant. If we
choose Q(t
0
) = I, then the axis of rotation q of the body is constant,
‡
and so we ﬁnd
Q(t) = cos(ν)(I −q ⊗q) −sin(ν)q +q ⊗q,
∗
Jacobi’s solution is discussed at length in Section 69 of Whittaker [228] and Section 37 of Landau
and Lifshitz [125].
†
Discussions of Poinsot’s solution can be found in several texts, for instance, Marsden and Ratiu
[138] and Routh [184].
‡
Other choices of Q(t
0
) are possible; however, these may not guarantee that the axis of rotation q of
Q is constant. For further details on this matter see [161].
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
280 Kinetics of a Rigid Body
where the axis and angle of rotation are
q =
ω(t
0
)
ω(t
0
)
, ν =
ω(t
0
)
(t −t
0
) ,
and ω(t) = ω(t
0
).
For the symmetric body, any axis is a principal axis. Consequently, it is possible
to spin such a body at constant speed about any axis. We shall shortly see that there
are related results for axisymmetric and asymmetric rigid bodies.
The Axisymmetric Body
For an axisymmetric body, it is convenient to deﬁne λ
t
= λ
1
= λ
2
and λ
a
= λ
3
. The
equations governing the components of angular velocity (9.11) simplify for this
case to
˙ ω
1
= k
t
ω
2
,
˙ ω
2
= −k
t
ω
1
,
ω
3
= , (9.12)
where = ω
3
(t
0
) is a constant and
k
t
=
λ
t
−λ
a
λ
t
.
Differential equations (9.12) have a simple analytical solution:
¸
ω
1
(t)
ω
2
(t)
=
¸
cos(k
t
(t −t
0
)) sin(k
t
(t −t
0
))
−sin(k
t
(t −t
0
)) cos(k
t
(t −t
0
))
¸
ω
1
(t
0
)
ω
2
(t
0
)
. (9.13)
In summary, ω
i
(t) have been calculated.
There are some special cases to consider. First, notice that it is possible to rotate
the body at constant speed either about the e
3
direction or about any axis in the
e
1
−e
2
plane. All of these axes are principal axes of the body. Hence it is possible
to spin the body at constant speed about a principal axis.
An interesting feature about the axisymmetric body is that the component of
ω in the direction of the axis of symmetry e
3
is always constant. This occurs even
though e
3
(t) may be quite complicated and is a consequence of the angular momen
tum H· e
3
being conserved. The conservation of ω
3
(t) is one of the key results in
rigid body dynamics and is extensively exploited in designing ﬂywheels.
The Asymmetric Body
When the body is asymmetric, its principal moments of inertia are distinct. If we
reexamine (9.11) for this case, then we ﬁnd that if all but one ω
i
are zero, then
the nonzero ω
i
will remain constant. For instance, if ω
2
= 0 and ω
3
= 0, then it is
possible for ω
1
to have any value and for the equations of motion to preserve this
value. These results imply that it is possible to rotate the body about a principal
axis at constant speed under no applied moment M. Clearly, as with the other
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
9.5 MomentFree Motion of a Rigid Body 281
two types of rigid bodies, it is possible to spin the body at constant speed about a
principal axis.
The Momentum Sphere
To visualize the solutions of (9.11), a graphical technique is often used. This tech
nique dates to the mid19th century. It is based on two facts: The solutions ω
i
(t)
to (9.11) preserve the magnitude of H and the rotational kinetic energy T
rot
. As a
result, the solutions
h
i
(t) = H· e
i
= λ
i
ω
i
(t) (i = 1, 2, 3),
lie on the intersection of the constant surfaces h = h
0
and T
rot
= T
E
. Here,
h
2
= h
2
1
+h
2
2
+h
2
3
,
T
rot
=
h
2
1
2λ
1
+
h
2
2
2λ
2
+
h
2
3
2λ
3
,
and the values of h
0
and T
E
are determined by the initial conditions ω
i
(t
0
).
If we pick a value h
0
of h, the surface h = h
0
in the threedimensional space
h
1
−h
2
−h
3
is a sphere – the momentumsphere. Selecting a value of T
E
, we ﬁnd that
the surface T
rot
= T
E
in the threedimensional space h
1
−h
2
−h
3
is an ellipsoid –
the energy ellipsoid. The intersection of the ellipsoid with the sphere is either a dis
crete set of points or a set of curves.
∗
These intersections are the loci of h
i
(t). For
the axisymmetric body, the intersections are shown in Figure 9.1(a). Corresponding
representative intersections for an asymmetric body are shown in Figures 9.1(b).
†
These ﬁgures are among the most famous in dynamics.
For the case presented in Figure 9.1(a), the energy ellipoid has an axis of revo
lution (in this case the third axis). For a symmetric body, the energy ellipsoid degen
erates further into a sphere. This sphere coincides with the momentum sphere and
so the graphical technique used to determine h
i
(t) [and ω
i
(t)] breaks down. How
ever, for the symmetric case, we found previously that ω
i
(t) = ω
i
(t
0
). Consequently,
each point on the momentum sphere corresponds to a steady rotational motion of
the rigid body.
Stability and Instability of the Steady Rotations
We can use the portrait of the trajectories of λ
i
ω
i
(t) on the momentum sphere
to deduce some conclusions on the nature of the steady rotational motions of a
rigid body. Our discussion is very qualitative, and more rigorous presentations
of this topic can easily be found elsewhere. For example, analyses of the stability
of the steady motions can be found in Hahn [85], Hughes [97], and Marsden and
Ratiu [138].
∗
For a more detailed discussion of these intersections, the text of Synge and Grifﬁth [207] is highly
recommended.
†
These ﬁgures were kindly supplied by Patrick Kessler in the Spring of 2007.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
282 Kinetics of a Rigid Body
λ
1
ω
1
λ
1
ω
1
λ
2
ω
2
λ
2
ω
2
λ
3
ω
3
λ
3
ω
3
e
2
e
2
(a)
(b)
Figure 9.1. Trajectories of the components λ
i
ω
i
on the momentum sphere. The curves and
points on the sphere are the intersection of the momentum sphere with the energy ellipsoid.
For these ﬁgures two distinct rigid bodies are shown: (a) λ
1
= λ
2
= 4 and λ
3
= 5, and (b) λ
1
=
2, λ
2
= 4, and λ
3
= 5. The ﬁgures on the righthand side show a momentfree motion of the
e
2
vector that is corotating with a rectangular box motion that corresponds to one of the
trajectories on the sphere. For the trajectories and simulations shown in this ﬁgure, (9.10)
and (9.11) were numerically integrated.
For the axisymmetric case, the trajectories shown in Figure 9.1(a) can be used
to infer that a steady rotation about the e
3
axis is stable. By stability, we mean that,
if we perturb the body’s rotation slightly from this steady state, then h
i
(t) [or equiv
alently ω
i
(t)] will remain close to the state (h
1
(t), h
2
(t), h
3
(t)) = (0, 0, h
3s
), where
h
3s
is the value of h
3
corresponding to the steady rotation.
∗
On the other hand, the
trajectories in Figure 9.1(a) show that steady rotations about any axis in the e
1
−e
2
plane do not satisfy this condition. Consequently, such steady rotations are unstable.
For the asymmetric case, the trajectories shown in Figure 9.1(b) conﬁrm the
previous statements that six steady rotations are possible. The trajectories in this
ﬁgure also illustrate that the four steady rotations about e
1
and e
3
are stable, whereas
∗
Because h
i
= λ
i
ω
i
, it is trivial to ascribe results pertaining to h
i
to those for ω
i
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
9.5 MomentFree Motion of a Rigid Body 283
the pair of steady rotations about the e
2
axis is unstable. That is, a rotation about
the intermediate axis of inertia is unstable, whereas those about the major (e
3
) and
minor (e
1
) axes are stable.
Attitudes of the Rotational Motions
The information we have thus far gleaned from the momentum sphere does not tell
the full story about the motion of the rigid body. What is missing is information on
the behavior of Q. To ﬁnd this information, one can use Jacobi’s analytical solu
tions discussed earlier. Alternatively, one can choose a parameterization for Q and
numerically integrate the equations relating ω
i
to these parameterizations. For ex
ample, if a set of 3–2–1 Euler angles is used, then, in addition to integrating (9.11),
(9.10) would also be integrated to determine φ(t), θ(t), and ψ(t). With the help of
these results, e
i
(t) can be constructed and the motion of the body visualized.
Results from two distinct examples of the numerical integrations of (9.10) and
(9.11) are shown in Figures 9.2, 9.3, and 9.4. One of these simulations corresponds
to a perturbation of the steady rotation of a rigid body rotating about the principal
axes corresponding to its maximal moment of inertia. The resulting behaviors
of ω
i
(t) are displayed in the trajectory labeled (i) in Figure 9.2(b), whereas the
behaviors of the corotational basis vectors can be seen in Figure 9.3. It should be
clear from the former ﬁgure that the perturbation to the steady motion does not
appreciably alter e
i
(t) from their steady rotation behaviors. The easiest method of
visualizing these results is to toss a book into the air with an initial rotation primarily
about e
3
and observe that, although the book will wobble, its instantaneous axis of
rotation does not wander far from its initial state.
In Figure 9.4, the behaviors of e
k
(t) corresponding to a trajectory of ω
i
(t) that
passes close to the equilibrium (ω
1
, ω
2
, ω
3
) = (0, ω
0
, 0) that is labeled with a “star”
0
−5
−5
5
5
6
(i)
(ii)
(a)
(b)
¯
X
ω
0
ω
1
ω
2
ω
3
E
1
E
2
E
3
Figure 9.2. The momentfree motion of a rigid body: (a) a rigid body showing the principal
axes, (b) the components ω
i
(t) = ω · e
i
corresponding to two different sets of initial condi
tions: (i) ω(0) = 0.5E
2
+5.0E
3
and (ii) ω(0) = 5.0E
2
+0.1E
3
. For these simulations, (9.10)
and (9.11) were numerically integrated with the initial conditions Q(0) = I and the parame
ter values λ
1
= 2, λ
2
= 4, and λ
3
= 5.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
284 Kinetics of a Rigid Body
−1
−1
−1
−1
−1
−1
1
1
1
1
1
1
(a)
(b)
(c)
O
O
O
e
1
e
2
e
3
e
1
· E
1
e
1
· E
2
e
1
· E
3
e
2
· E
1
e
2
· E
2
e
2
· E
3
e
3
· E
1
e
3
· E
2
e
3
· E
3
Figure 9.3. Simulation results indicating the stability of the steady rotation of the rigid body
about the principal axis corresponding to the maximal axis of inertia: (a) the E
i
components
of e
1
(t), (b) the E
i
components of e
2
(t), and (c) the E
i
components of e
3
(t). These results
correspond to the trajectory labeled (i) in Figure 9.2(b).
are shown. It should be clear from the behavior of e
2
(t) shown in Figure 9.4(b) that
e
2
(t), which is initially close to E
2
at time t = 0, makes large excursions from its
initial value. This is in contrast to the situation shown in Figure 9.3 and is indicative
of the instability of the steady momentfree motion of a rigid body about its inter
mediate axis of inertia. The easiest way to see this instability is to take a book and
give it an initial angular velocity about e
2
. One will see a wobbling motion where
ω · e
2
will periodically take positive and negative values. Interestingly, it is possible
to execute this motion and have the book perform a rotation about e
1
or e
3
by 180
◦
.
This twisting motion was only recently noted and analyzed by Ashbaugh et al. [11].
∗
∗
In [11], two distinct sets of Euler angles are used to avoid the singularities inherent in Euler angle
parameterizations of rotation tensors. For the results presented in Figures 9.3 and 9.4, it was not
necessary to introduce a second set.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
9.6 The Baseball and the Football 285
−1
−1
−1
−1
−1
−1
1
1
1
1
1
1
(a)
(b)
(c)
O
O
O
e
1
e
2
e
3
e
1
· E
1
e
1
· E
2
e
1
· E
3
e
2
· E
1
e
2
· E
2
e
2
· E
3
e
3
· E
1
e
3
· E
2
e
3
· E
3
Figure 9.4. Simulation results indicating the instability of the steady rotation of the rigid body
about the principal axis corresponding to the intermediate axis of inertia: (a) the E
i
com
ponents of e
1
(t), (b) the E
i
components of e
2
(t), and (c) the E
i
components of e
3
(t). These
results correspond to the trajectory labeled (ii) in Figure 9.2(b).
9.6 The Baseball and the Football
Consider a sphere of mass m and radius R that is thrown into space with an initial
velocity ¯ v(t
0
), angular velocity ω(t
0
), and orientation Q(t
0
). We wish to determine
the motion ¯ x and Q of the sphere. As discussed by Tait [211], it was known to
Isaac Newton that the rotation of the sphere as it moves through the ambient air
causes a curvature of the path of the center of the sphere. This feature results
in interesting dynamics in a variety of sports ranging from golf to baseball and
soccer. Our interest here is to examine the curving of the path of the sphere. We
do this by following several classical works on this problem: most notably Tait
[211].
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
286 Kinetics of a Rigid Body
¯
X
¯
X
ω ω
F
M
F
M
¯ v ¯ v
Figure 9.5. A rigid sphere whose center of mass is moving to the right with a velocity vector
¯ v. When the ball is rotating clockwise the velocity of the air moving over the top of the ball
is slower than the velocity of the air in contact with the bottom of the ball. From Bernoulli’s
equation, the pressure on the top of the ball is greater than the pressure on the bottom of
the ball and a net downward force F
M
results. The opposite occurs when the ball is rotating
counterclockwise.
The Magnus Force
A key force experienced by the sphere is known as the lift or Magnus force (see
Figure 9.5),
∗
F
M
= mBω × ¯ v,
where B is a positive constant. The sign of B is determined by use of Bernoulli’s
equation.
†
Clearly, this force models the coupling between rotation and linear ve
locity. Recent research on free kicks in soccer has shown that there can be a transi
tion in the ﬂow ﬁeld from turbulent to laminar that causes dramatic changes in the
trajectory (see Carr´ e et al. [25]). According to Ireson [100], for some free kicks in
soccer,  ¯ v = 25 m/s and mB ≈ 0.15716 kg.
Apart fromgravity and the Magnus force, the other important force in this prob
lem is the drag force:
F
D
= −
1
2
ρ
f
AC
d
(¯ v · ¯ v)
¯ v
 ¯ v
.
In this expression, C
d
is the drag coefﬁcient, ρ
f
is the density of the ﬂuid that the
sphere is moving in, and Ais the frontal area of the sphere in contact with the ﬂuid:
A= πR
2
.
Equations of Motion
For the system at hand, M = 0 and H =
2mR
2
5
ω, so we ﬁnd the important result that
ω is constant:
ω(t) = ω(t
0
) .
∗
Credited to the German scientist Heinrich Gustav Magnus (1802–1870) in 1851 (see [133, 134]).
†
Bernoulli’s equation applies to inviscid ﬂuid ﬂow and states that the sum of the pressure p and
1
2
ρ
f
U
2
is a constant. Here, Uis the ﬂuid ﬂow velocity and ρ
f
is the ﬂuid density.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
9.6 The Baseball and the Football 287
In other words, the angular velocity of the sphere does not change. As with the
symmetric body discussed earlier, we can easily solve for the rotation tensor of the
sphere if we assume that Q(t
0
) = I:
Q(t) = cos(ν)(I −q ⊗q) −sin(ν)q +q ⊗q,
where the axis and angle of rotation are
q =
ω(t
0
)
ω(t
0
)
, ν =
ω(t
0
)
(t −t
0
) .
We shall see that solving for the motion of the center of mass in this problem is not
trivial.
The rotation tensor Q =
¸
3
k=1
e
k
⊗E
k
for the rigid body is parameterized by
a set of 3–1–3 Euler angles (see Subsection 6.8.2). The Euler basis vectors for this
parameterization have the representations
⎡
⎢
⎣
g
1
g
2
g
3
⎤
⎥
⎦
=
⎡
⎢
⎣
E
3
e
1
e
3
⎤
⎥
⎦
=
⎡
⎢
⎣
sin(φ) sin(θ) cos(φ) sin(θ) cos(θ)
cos(φ) −sin(φ) 0
0 0 1
⎤
⎥
⎦
⎡
⎢
⎣
e
1
e
2
e
3
⎤
⎥
⎦
=
⎡
⎢
⎣
0 0 1
cos(ψ) sin(ψ) 0
sin(θ) sin(ψ) −sin(θ) cos(ψ) cos(θ)
⎤
⎥
⎦
⎡
⎢
⎣
E
1
E
2
E
3
⎤
⎥
⎦
. (9.14)
Further, the Euler angles are subject to the following restrictions: φ ∈ [0, 2π), θ ∈
(0, π), and ψ ∈ [0, 2π). Using these Euler angles, we obtain
ω =
1
E
1
+
2
E
2
+
3
E
3
=
˙
θ cos(ψ) +
˙
φsin(θ) sin(ψ)
E
1
+
˙
θ sin(ψ) −
˙
φsin(θ) cos(ψ)
E
2
+
˙
ψ +
˙
φcos(θ)
E
3
.
If we use a set of Cartesian coordinates for the position vector of the center of mass,
¯ x · E
i
= x
i
, then we would ﬁnd that
ω × ¯ v = ( ˙ x
3
2
− ˙ x
2
3
) E
1
+( ˙ x
1
3
− ˙ x
3
1
) E
2
+( ˙ x
2
1
− ˙ x
1
2
) E
3
. (9.15)
From the previous solution to the balance of angular momentum, we know that
i
are constant.
The balance of linear momentum for the sphere provides the equation for the
motion of the center of mass. Evaluating F =
˙
G in Cartesian coordinates, we ﬁnd
m¨ x
1
= (F
D
+F
M
) · E
1
,
m¨ x
2
= (F
D
+F
M
) · E
2
,
m¨ x
3
= −mg +(F
D
+F
M
) · E
3
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
288 Kinetics of a Rigid Body
Ignoring the drag force and using (9.15), we ﬁnd three differential equations for
x
i
(t):
m¨ x
1
= mB( ˙ x
3
2
− ˙ x
2
3
) ,
m¨ x
2
= mB( ˙ x
1
3
− ˙ x
3
1
) ,
m¨ x
3
= mB( ˙ x
2
1
− ˙ x
1
2
) −mg. (9.16)
For the general case, these equations can be integrated numerically to determine
¯ x(t).
The Path of the Ball
Turning our attention to a simple case, suppose ω(t
0
) =
10
E
1
. From the previous
analysis, we know that ω is constant for this rigid body. Consequently, (9.16)
simpliﬁes to
m¨ x
1
= 0,
m¨ x
2
= −mB˙ x
3
10
,
m¨ x
3
= mB˙ x
2
10
−mg.
The solution to these differential equations, assuming B
10
= 0, is
⎡
⎢
⎣
x
1
(t) −x
1
(t
0
)
x
2
(t) −x
2
(t
0
)
x
3
(t) −x
3
(t
0
)
⎤
⎥
⎦
= A
⎡
⎢
⎣
˙ x
1
(t
0
)
˙ x
2
(t
0
) −
g
B
10
˙ x
3
(t
0
)
⎤
⎥
⎦
+
⎡
⎢
⎢
⎣
0
g(t−t
0
)
B
10
0
⎤
⎥
⎥
⎦
, (9.17)
where
A =
⎡
⎢
⎢
⎢
⎣
t −t
0
0 0
0
sin(B
10
(t−t
0
))
B
10
−
1−cos(B
10
(t−t
0
))
B
10
0
1−cos(B
10
(t−t
0
))
B
10
sin(B
10
(t−t
0
))
B
10
⎤
⎥
⎥
⎥
⎦
.
From (9.17), the trajectory of the sphere can be determined. Two important features
are present. First, the spin
10
inﬂuences the forward speed ˙ x
2
of the sphere. Second,
it also affects the vertical position and speed. It is also of interest to compare (9.17)
with the corresponding solution when the Magnus force is absent. In this case, the
path of the center of mass is the wellknown parabolic trajectory:
⎡
⎢
⎣
x
1
(t) −x
1
(t
0
)
x
2
(t) −x
2
(t
0
)
x
3
(t) −x
3
(t
0
)
⎤
⎥
⎦
=
⎡
⎢
⎣
(t −t
0
) ˙ x
1
(t
0
)
(t −t
0
) ˙ x
2
(t
0
)
(t −t
0
) ˙ x
3
(t
0
)
⎤
⎥
⎦
−
⎡
⎢
⎢
⎣
0
0
g
2
(t −t
0
)
2
⎤
⎥
⎥
⎦
.
Representative examples of the trajectories of a point launched from the origin are
shown in Figure 9.6. For small values of B
10
, we can see from this ﬁgure how the
trajectory differs from that in which the Magnus force is absent: A spin (
10
) in
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
9.7 Motion of a Rigid Body with a Fixed Point 289
10
15
30
−20
x
3
x
2
(i) (ii) (iii)
(iv)
(v) (vi)
Figure 9.6. The trajectories of a sphere that is launched from the origin with an initial velocity
v (t
0
) = 10 (E
2
+E
3
) and an initial angular velocity ω(t
0
) =
10
E
1
. The trajectories shown
correspond to different values of B
10
: (i) B
10
= −0.5, (ii) B
10
= −0.2, (iii) B
10
= 0.0,
(iv) B
10
= 0.2, (v) B
10
= 0.5, and (vi) B
10
= 1.0. All of the trajectories are displayed for
a period of 4 s and g = 9.81 m/s/s.
one direction will result in the ball “rising,” whereas a “dipping” effect can be ob
served by reversing the initial spin. However, we also observe that, for larger values
of B
10
, the behavior of the trajectories becomes unphysical: either through the
appearance of a cusp or the reversal in the sign of ˙ x
3
.
∗
Thus the prescription of the
Magnus force may have a limited range of physical applicability.
Our formulation of the equations of motion for this problem are simpliﬁed by
the fact that the moment of inertia tensor for the sphere has a simple form. Indeed,
it is interesting to compare the ﬂight of a ball predicted by this model with that of a
Frisbee. The interested reader is referred to [95, 97] where the equations of motion
of a Frisbee are formulated and the lift and drag forces on the Frisbee computed
from experiments.
9.7 Motion of a Rigid Body with a Fixed Point
The problem of a body that is free to rotate about a ﬁxed point O, which is also a
material point of the body, occupies a celebrated place in the history of mechanics.
An example of such a body is shown in Figure 9.7, and related systems can be found
in the pendula in clocks and several types of spinning tops. In these systems, there
are three constraints on the motion of the rigid body and F = m¯ a serves to determine
the constraint (reaction) forces at O that enforce these constraints. The remaining
balance law, M
O
=
˙
H
O
, has an interesting form and is used to determine the rotation
tensor Q of the rigid body.
∗
As can be seen in Figure 4 of [211], motions of this type are also present in Tait’s analysis of this
problem.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
290 Kinetics of a Rigid Body
g
¯
X
X
s
A
O
Ballandsocket joint
Linear spring
e
1
e
3
Figure 9.7. An example of a rigid body that is free to rotate about a ﬁxed point O where the
ﬁxed point is a material point of the body. In this example, ¯ x −x
O
= he
3
where h is a constant,
and the body is also subject to conservative forces from the linear spring and gravity.
Kinematics
For the body of interest, the material point O is ﬁxed at a point that we take to be
the origin: x
O
= 0. We assume that the position vector of the center of mass relative
to Ohas the representation
¯ x −x
O
= L
1
e
1
+L
2
e
2
+L
3
e
3
,
where L
i
are constants. Differentiating this equation with respect to time, we see
that
¯ v = ω ×(L
1
e
1
+L
2
e
2
+L
3
e
3
) . (9.18)
This relation can be used to establish a convenient representation for the angular
momentum H
O
in terms of an inertia tensor J
O
for the body relative to O and the
kinetic energy T of the rigid body.
To establish the representation for H
O
, we start with the relation for this quan
tity in terms of H and G:
H
O
= H+ ¯ x ×G
= Jω +m(L
1
e
1
+L
2
e
2
+L
3
e
3
) ×(ω ×(L
1
e
1
+L
2
e
2
+L
3
e
3
))
= J
O
ω. (9.19)
For the ﬁnal step in this result, we use the identity a ×(b ×c) = (a · c)b −(a · b)c.
∗
The inertia tensor J
O
in (9.19) is
J
O
= J +m
L
2
1
+L
2
2
+L
2
3
I
−m(L
1
e
1
+L
2
e
2
+L
3
e
3
) ⊗(L
1
e
1
+L
2
e
2
+L
3
e
3
) .
∗
As can be seen from (7.20), this identity was used earlier to establish the representation H = Jω.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
9.7 Motion of a Rigid Body with a Fixed Point 291
As expected, this expression for J
O
is in agreement with the one that we would
obtain by using the parallel axis theorem [see (7.25)]. Starting with Koenig decom
position (7.21), a series of standard manipulations provides a convenient expression
for the kinetic energy of the rigid body:
T =
1
2
J
O
ω
· ω. (9.20)
The details are left as an exercise.
We can show that J
O
is a positivedeﬁnite symmetric tensor, and consequently it
will have a set of principal axes. We now choose e
i
to be these axes, and, as a result,
we can write
J
O
= λ
O
1
e
1
⊗e
1
+λ
O
2
e
2
⊗e
2
+λ
O
3
e
3
⊗e
3
.
It is straightforward to show how λ
O
i
are related to the components of J and m, L
1
,
L
2
, and L
3
.
Constraint Forces and Constraint Moments
The motion is subject to three (integrable) constraints:
1
= 0,
2
= 0,
3
= 0.
These constraints arise because the point Ois ﬁxed:
i
= (¯ x −L
1
e
1
−L
2
e
2
−L
3
e
3
) · E
i
. (9.21)
Differentiating the constraints, we ﬁnd Equation (9.18). Assuming that the joint at
Ois frictionless, we can easily use (9.18) with Lagrange’s prescription to show that
F
c
= µ
1
E
1
+µ
2
E
2
+µ
3
E
3
, M
c
= (−L
1
e
1
−L
2
e
2
−L
3
e
3
) ×F
c
.
Here, F
c
and M
c
are equipollent to the force F
c
acting at the joint O. You may wish
to recall that we considered the constraint forces and moments for the situation in
which the joint at Owas a pin joint in Subsection 8.6.1. For this case, M
c
would have
two additional components.
Equations of Motion
For this problem, it is convenient to follow a procedure of using F = m¯ a to solve for
the three unknown components of F
c
and to use M
O
=
˙
H
O
to solve for the motion
of the body.
The balance law M
O
=
˙
H
O
can be written in components relative to the basis e
i
.
Recalling that we are choosing this basis to be parallel to the principal axes of J
O
,
we ﬁnd the component form
λ
O
1
˙ ω
1
+
λ
O
3
−λ
O
2
ω
3
ω
2
= M
O
· e
1
,
λ
O
2
˙ ω
2
+
λ
O
1
−λ
O
3
ω
3
ω
1
= M
O
· e
2
,
λ
O
3
˙ ω
3
+
λ
O
2
−λ
O
1
ω
1
ω
2
= M
O
· e
3
. (9.22)
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
292 Kinetics of a Rigid Body
These equations have obvious parallels to Euler’s equations (9.9). Indeed, if M
O
=
0, then we can use the solutions for momentfree motion that we discussed previ
ously with a small number of modiﬁcations.
Equations (9.22) need to be supplemented by equations relating ω to Q in order
to solve for the rotation tensor of the body.
∗
Once the rotation has been found, we
can then use F = m¯ a, where ¯ a = α × ¯ x +ω ×(ω × ¯ x), to determine R.
Euler–Poisson Equations
An alternative formulation of the equations of motion for this case in which a
gravitational force −mgE
3
is acting on the body is known as the Euler–Poisson
equations.
†
For these equations, instead of using a set of 3–1–2 Euler angles to
parameterize Q and then supplementing (9.22) with (7.28), we work with three of
the nine components of Q.
By way of preliminaries, let us write
˙
Q = Q
0
in terms of components. Relative
to the basis E
i
⊗E
k
, we see that
⎡
⎢
⎣
˙
Q
11
˙
Q
12
˙
Q
13
˙
Q
21
˙
Q
22
˙
Q
23
˙
Q
31
˙
Q
32
˙
Q
33
⎤
⎥
⎦
=
⎡
⎢
⎣
Q
11
Q
12
Q
13
Q
21
Q
22
Q
23
Q
31
Q
32
Q
33
⎤
⎥
⎦
⎡
⎢
⎣
0 −ω
3
ω
2
ω
3
0 −ω
1
−ω
2
ω
1
0
⎤
⎥
⎦
. (9.23)
Relations of this form for the time derivatives of the components of the rotation
tensor were ﬁrst found by Poisson in the early 19th century,
‡
and are known as the
Poisson kinematical relations. It is important to note that
E
3
= Q
31
e
1
+Q
32
e
2
+Q
33
e
3
. (9.24)
Because of the moment that is due to gravity we shall subsequently need the com
ponents E
3
· e
i
.
Note that, by differentiating (9.24) and using the identity ˙ e
i
= ω ×e
i
, we would
discover that
˙
Q
31
e
1
+
˙
Q
32
e
2
+
˙
Q
33
e
3
= −
3
¸
k=1
Q
3k
ω ×e
k
.
These relations constitute three differential equations for Q
3i
– which are equivalent
to those from (9.23). The coefﬁcients Q
3i
= e
3
· E
i
are often known as the direction
cosines of e
3
.
The equations of motion for the rigid body consist of three equations governing
Q
3i
and the balance of angular momentum relative to O. Thus we combine the
∗
An explicit form of the differential equations can be found in an exercise at the end of this chapter:
see (9.36).
†
We shall discuss a third alternative, Lagrange’s equations of motion, in Chapter 10.
‡
See Section 411 of his treatise [173].
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
9.7 Motion of a Rigid Body with a Fixed Point 293
balance of angular momentum that is due to Euler and kinematical relations that
are due to Poisson. In component forms, the Euler–Poisson equations are
˙
Q
31
= Q
32
h
O3
λ
O
3
−Q
33
h
O2
λ
O
2
,
˙
Q
32
= Q
33
h
O1
λ
O
1
−Q
31
h
O3
λ
O
3
,
˙
Q
33
= −Q
32
h
O1
λ
O
1
+Q
31
h
O2
λ
O
2
,
˙
h
O1
=
λ
O
2
−λ
O
3
λ
O
2
λ
O
3
h
O2
h
O3
+(mgE
3
× ¯ x) · e
1
,
˙
h
O2
=
λ
O
3
−λ
O
1
λ
O
1
λ
O
3
h
O1
h
O3
+(mgE
3
× ¯ x) · e
2
,
˙
h
O3
=
λ
O
1
−λ
O
2
λ
O
1
λ
O
2
h
O1
h
O2
+(mgE
3
× ¯ x) · e
3
. (9.25)
Here, we have used the representations
H
O
=
3
¸
i=1
h
Oi
e
i
, h
Ok
= λ
O
k
ω
k
(k = 1, 2, 3).
Additional developments and representations of Euler–Poisson equations (9.25)
can be found, for example, in Beletskii [16] and Sudarshan and Mukunda [204].
Conservations
For the problem of interest, we can use representation (9.20) for T to show the
following forms of the work–energy theorem:
˙
T = M
O
· ω =
N
¸
K=1
F
K
· v
K
+M
p
· ω.
If the applied forces and moments acting on the system are conservative, then, be
cause F
c
acts at a point with zero velocity, it is easy to show that the total energy of
the rigid body is conserved. This situation arises when a gravitational force acts on
the body.
If M
O
= 0, then H
O
is conserved. However, in the most common form of this
problem a gravitational force −mgE
3
acts on the rigid body. In this case, M
O
= 0,
but M
O
has no component in the E
3
direction. It is easy to see for this case that
H
O
· E
3
is conserved. If, in addition, the body has an axis of symmetry and λ
O
1
= λ
O
2
,
then you should be able to show with the help of (9.22) that H
O
· e
3
is conserved.
The problemin which the body has an axis of symmetry and M
O
= ¯ x ×(−mgE
3
)
is often known as the (symmetric) Lagrange top.
∗
The motion of this top conserves
∗
In the context of Lagrange’s equations of motion, we will discuss this problem in Section 10.8.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
294 Kinetics of a Rigid Body
P
g O
¯
X
β
Sphere of mass m and radius R
Inclined plane
E
1
E
3
Figure 9.8. A rigid sphere moving on an inclined plane. The angle of inclination of the plane
is β, and a gravitational force −mg cos(β)E
3
+mg sin(β)E
1
acts on the rigid body.
E, H
O
· e
3
, and H
O
· E
3
, and is one of the most famous mechanical systems. Indeed,
as discovered by Lagrange,
∗
analytical solutions for its equations of motion can be
found. This discovery is remarkable, for if we relax the assumption that the body
is symmetric (i.e., λ
O
1
= λ
O
2
), then analytical solutions are possible in only a handful
of special cases (see [128, 228]). Indeed, it is an interesting exercise to numerically
integrate the equations of motion for the case in which M
O
= ¯ x ×(−mgE
3
) and the
λ
O
i
’s are distinct.
9.8 Motions of Rolling Spheres and Sliding Spheres
The problem of the sphere moving on a ﬂat surface has several applications, bowling
and pool being the most famous. The most famous classical treatments of this prob
lem are due to Coriolis [41] and Routh [184], and generalizations of it occupy the
literature on nonholonomically constrained rigid bodies to date.
†
In our treatment,
we assume that the surface is rough with a coefﬁcient of static Coulomb friction of
µ
s
and kinetic friction of µ
d
. Of particular interest to us will be the transition be
tween rolling and sliding and our discussion is heavily inﬂuenced by Routh [184]
and Synge and Grifﬁth [207].
Consider the sphere moving on the surface shown in Figure 9.8. The radius
of the sphere is R, and the velocity of the point of contact of the sphere with the
incline is
v
P
= ¯ v +ω ×(−RE
3
) .
∗
See Section IX.34 of the Second Part of Lagrange’s M´ ecanique Analytique [121]. This analytical
solution is also discussed by Whittaker [228].
†
See, for example, Borisov and Mamaev [19], Frohlich [66], and Huston et al. [99]. The latter papers
discuss the dynamics of bowling balls.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
9.8 Motions of Rolling Spheres and Sliding Spheres 295
Note the simple expression for π
P
here. Because the point P is the instantaneous
point of contact,
v
P
· E
3
= 0.
Consequently, this velocity ﬁeld has the representations
v
P
= v
s1
E
1
+v
s2
E
2
= uc,
where u =
v
2
s1
+v
2
s2
is the slip velocity and c =
v
P
u
is the slip direction. When the
sphere is rolling, there are two additional constraints on v
P
and, as a result, v
s1
= 0
and v
s2
= 0. For the rolling sphere, the slip direction is not deﬁned.
The resultant force and moment on the sphere are
F = −mg cos(β)E
3
+mg sin(β)E
1
+NE
3
+F
f
, M = −RE
3
×F
f
.
When the sphere is rolling,
F
f
= µ
1
E
1
+µ
2
E
2
,
where µ
1
and µ
2
are unknowns. For the sliding sphere, on the other hand, we have
the classical prescription
F
f
= −µ
d
Nc.
For convenience, we use the same notation as that of the friction forces for the
rolling and sliding spheres, but this should not cause confusion.
Rolling Sphere
We determine the motion of the rolling sphere by using the balance laws and the
constraints v
P
= 0. Using Cartesian coordinates for ¯ x and setting ω =
¸
3
i=1
i
E
i
, we
ﬁnd that these equations are
˙ x
1
= R
2
,
˙ x
2
= −R
1
,
˙ x
3
= 0,
m¨ x
1
= mg sin(β) +µ
1
,
m¨ x
2
= µ
2
,
0 = N −mg cos(β),
2
5
mR
2
˙
1
= Rµ
2
,
2
5
mR
2
˙
2
= −Rµ
1
,
2
5
mR
2
˙
3
= 0.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
296 Kinetics of a Rigid Body
To solve these equations, it is convenient to ﬁrst determine the differential equations
governing
i
. From the nine equations just listed, we can eliminate several variables
to ﬁnd
1 +
2
5
mR
2
˙
1
= 0,
1 +
2
5
mR
2
˙
2
= mgRsin(β),
2
5
mR
2
˙
3
= 0.
These equations are easily solved:
ω(t) = ω(t
0
) +
g sin(β) (t −t
0
)
R
1 +
2
5
−1
E
2
. (9.26)
It is left as an exercise to determine ¯ x(t). When β = 0, you will ﬁnd that the sphere
rolls in a straight line at constant speed.
Sliding Sphere
For the sliding sphere, it is convenient to examine the differential equations for v
P
.
Differentiating this velocity, we ﬁnd that
˙ v
P
= ˙ v
s1
E
1
+ ˙ v
s2
E
2
=
˙
¯ v +α ×(−RE
3
) .
Using the balances of linear and angular momentum, we substitute for
˙
¯ v and α to
ﬁnd
m˙ v
s1
=
1 +
5
2
F
f
· E
1
+mg sin(β), m˙ v
s2
=
1 +
5
2
F
f
· E
2
. (9.27)
After substituting for the friction force, these equations provide two differential
equations for the slip velocities.
∗
It is convenient to express these equations as dif
ferential equations for u and the angle χ, where
c = cos(χ)E
1
+sin(χ)E
2
, cos(χ) =
v
s1
u
, sin(χ) =
v
s2
u
.
When χ is constant, the slip direction c is constant. After some manipulations, we
ﬁnd that (9.27) are equivalent to
†
˙ u = −µ
d
g
1 +
5
2
+ g sin(β) cos(χ), u˙ χ = −g sin(β) sin(χ). (9.28)
These differential equations have analytical solutions for χ(t) and u(t). It is also left
as an exercise to write the ﬁve differential equations governing x
1
, x
2
, and
i
.
∗
It is left as an interesting exercise to nondimensionalize and numerically integrate (9.27) to deter
mine the behavior of the slip velocity components as the ratio of µ
d
to sin(β) is varied.
†
To obtain these equations we differentiated u
2
= v
2
s1
+v
2
s2
and sin(χ) =
v
s2
u
and then used (9.27).
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
9.9 Closing Comments 297
If we consider the simple case in which the incline is horizontal, then β = 0. The
differential equations for the slip velocity simplify considerably to
m˙ v
s1
= −µ
d
mg
1 +
5
2
v
s1
u
, m˙ v
s2
= −µ
d
mg
1 +
5
2
v
s2
u
.
From these equations, we will ﬁnd that v
s1
and v
s2
always tend to zero. To see this,
it is best to look at (9.28) and set β = 0:
˙ u = −µ
d
g
1 +
5
2
, ˙ χ = 0. (9.29)
These equations have the solution
u(t) = u(t
0
) −µ
d
g
1 +
5
2
(t −t
0
) , χ(t) = χ(t
0
) .
As a result, u will reach zero in a ﬁnite time T and the slip direction stays constant:
T =
u(t
0
)
µ
d
g
1 +
5
2
−1
.
It can be shown that the path of the center of the sliding sphere is either ﬁxed,
a straight line, or a parabolic arc. Once u = 0, the sphere starts rolling. Now, as
β = 0, this implies that the sphere will roll at constant speed in a straight line (see
Figure 9.9). It is interesting to note that once the sphere starts rolling it will stay
rolling. The transition between the parabolic path during sliding and the straight
line path during rolling is a key to hook shots in bowling and massee shots in pool.
∗
An example of this transition is shown in Figure 9.9(a).
The factor of
2
5
in the equations of motion for rolling and sliding spheres is also
interesting. It is related to the fact that the height of the “center of oscillation” Q
of a sphere relative to the center of mass is
2R
5
. As discussed in Coriolis [41], Q
is the point one aims for when hitting a cue ball so that it rolls without slipping
immediately after the impact of the tip of the cue with the ball.
9.9 Closing Comments
We have touched on some problems in rigid body dynamics. There are several as
pects that we have not had the opportunity to address, and some of them are dis
cussed in the exercises and others in the references at the end of this book. The
treatises of Appell [7], Papastavridis [169], Routh [184], and Whittaker [228], and
the splendid introductory text by Crabtree [42] are particularly recommended.
It is important to realize that, although rolling spheres and thrown baseballs
have been analyzed for over a century, these problems are very rich. Indeed,
a simple change in their kinematical features can lead to dramatically different
results. One of the most celebrated instances of this change arises in a rolling sphere
∗
As discussed in Frohlich [66], bowling balls feature offset centers of mass and moment of inertia
tensors that are not multiples of I. As a result, some of the intricacies of bowling are not explained
by our simple model of rolling spheres and sliding spheres.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
298 Kinetics of a Rigid Body
(a)
(b)
(c)
0
u(t)
u(t
0
)
t −t
0
O
O
O
E
1
E
1
E
1
E
2
E
2
E
2
¯ x
¯ x
¯ x
Figure 9.9. Plot of slip speed u(t) as a function of time for a sphere that is initially sliding
(and eventually rolls) on a rough horizontal plane. Three representative paths of the center
of mass of the sphere are also shown: (a) the path of the sliding center of mass is a parabolic
arc; (b) the path of the sliding center of mass is a straight line; and (c) the center of mass is
stationary while the sphere is sliding. For the rolling phases, the path of the center of mass is
a straight line. The dashed part of these paths denotes the sliding phase of the motion.
where J =
2
5
mR
2
I and π
P
= −RE
3
. Such a sphere is often known as a Chaplygin
sphere. Partially as a consequence of its asymmetry, the path of the point of contact
of the Chaplygin sphere with the ground can be very intricate (see [19, 191] and
references therein).
Two other celebrated examples of rigid bodies that exhibit interesting behavior
are Euler’s disk and the wobblestone (or celt) [42]. The former consists of a heavy
circular cylinder that rolls and slides on a convex mirror. As the disk becomes
increasingly horizontal, a whirring sound is heard whose pitch increases. Eventually,
the disk comes to a dramatic abrupt halt accompanied (some believe) by an impact
of the disk with the convex mirror. The ﬁrst analysis of this system was performed
by Moffatt [143], and his controversial paper was followed by a series of works
promoting alternative mechanisms for the dramatic motion of Euler’s disk (see
[110] and references therein). As mentioned earlier, the wobblestone is a rigid
body whose curved lateral surface rolls on a horizontal plane. At ﬁrst glance, the
curved surface appears to be symmetric, but this is not the case, and, as a result, the
wobblestone exhibits an unusual reversal of spin directions. The wobblestone has
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
Exercises 9.1–9.6 299
been the subject of several papers (see, for example, [129, 171]) and simulations; the
paper by Blackowiak et al. [17] is particularly recommended for a lucid explanation
of the spin reversal mechanism.
Recent analyses of rigid body dynamics have focused on the stability and bi
furcation of their families of steady motions (see, for example, [128, 157]). Increas
ingly, some of these studies are evolving toward an examination of motions of these
systems that (although not steady) are asymptotic to one or connect two steady
motions of the rigid body. For the interested reader, recent work on the tippe top
[20, 177, 219], the possible jumping behavior of a spinning egg [22, 142, 194], and the
attitude of a thrown tennis racket [11] are mentioned as examples of notable mod
ern analyses of rigid body dynamics. It is hoped that the exposition in this chapter
will enable you to explore and appreciate works of this type.
EXERCISES
9.1. Suppose a rigid body is rolling on a ﬁxed surface under the inﬂuence of a grav
itational force −mgE
3
. Starting from the work–energy theorem for the rigid body,
˙
T = F · ¯ v +M· ω,
prove that the total energy E of the rigid body is conserved. Prove that the total
energy is also conserved if the rigid body is sliding on a smooth surface.
9.2. A rigid body of mass m is moving in space under the inﬂuence of an applied
force F
a
= F
a
e
3
and an applied moment M = 0. Outline how you would determine
the attitude Q and motion of the center of mass of the rigid body.
9.3. The orientation of a rigid body relative to a ﬁxed reference conﬁguration is
deﬁned by a rotation tensor Q. At time t
0
this rotation tensor has the value Q(t
0
),
and at time t
1
this rotation tensor has the value Q(t
1
). Give a physical interpretation
of the rotation tensor Q(t
1
)Q
T
(t
0
). You should make use of the corotational basis in
your answer.
9.4. Consider a rigid body with a ﬁxed point O. What are the three constraints on
the motion of this rigid body? Why is it sufﬁcient to solve M
O
=
˙
H
O
to determine
the motion of this rigid body?
9.5. Solutions for ω
i
(t) have been determined for a rigid body dynamics problem.
How would you determine Q(t) from this solution?
9.6. A rigid body has a potential energy U = U
¯ x, γ
i
, where γ
1
, γ
2
, γ
3
are the Euler
angles used to parameterize Q. If the conservative force F and conservative moment
M are such that
−
˙
U = F ·
˙
¯ x +M· ω, (9.30)
then verify that
F = −
3
¸
i=1
∂U
∂x
i
E
i
, M = −
3
¸
i=1
∂U
∂γ
i
g
i
,
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
300 Exercises 9.6–9.8
where x
i
= ¯ x · E
i
and g
i
are the basis vectors for the dual Euler basis. Show that
the gravitational potential energy U
n
for a rigid body orbiting a ﬁxed spherically
symmetric rigid body has the functional form U = U
¯ x, γ
i
.
9.7. As shown in Figure 9.8, a rigid sphere of mass m and radius R rolls (without
slipping) on an inclined plane. The inertia tensors for the sphere are J = J
0
= µI,
where µ =
2mR
2
5
.
(a) What are the three constraints on the motion of the sphere? Showthat these
constraints imply that
¨ x
1
−R
˙
2
= 0, ¨ x
2
+R
˙
1
= 0, ¨ x
3
= 0,
where x
i
= ¯ x · E
i
, and
i
= ω · E
i
.
(b) With the help of the balance of linear momentum for the sphere, show that
F
c
=
mR
˙
2
−mg sin(β)
E
1
−mR
˙
1
E
2
+mg cos(β)E
3
.
(c) Show that the balance of angular momentum for the sphere and the results
of (b) imply that
7
5
mR
2
˙
1
= 0,
7
5
mR
2
˙
2
= mgRsin(β),
2
5
mR
2
˙
3
= 0.
(d) Starting from the work–energy theorem for a rigid body, prove that the
total energy E of the rolling sphere is constant, where
E =
7mR
2
10
2
1
+
7mR
2
10
2
2
+
mR
2
5
2
3
−mgx
1
sin(β) +mgRcos(β).
(e) Why is the angular momentum H of the sphere in the E
3
direction
conserved?
(f) If, at time t = 0, the sphere is given an initial angular velocity ω(0) =
¸
3
i=1
i0
E
i
, then show that the angular velocity ω(t) is (9.26). What is the
angular acceleration vector α of the sphere?
(g) Suppose the sphere is placed on the inclined plane and released from rest
with Q(0) = I. Verify that the sphere will start rolling and that the resulting
attitude Q of the sphere corresponds to a ﬁxedaxis rotation.
9.8. In a model for a rigid body ﬂying through the air, the four primary forces on
the body are gravity, a lift force, a drag force, and a thruster force:
F = −mgE
3
+mBω × ¯ v + f e
1
−
1
2
ρ
f
AC
d
(¯ v · ¯ v)
¯ v
 ¯ v
,
M = π
t
× f e
1
.
(a) Show that one of the four applied forces is conservative.
(b) Show that the lift force does no work.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
Exercises 9.8–9.9 301
(c) If the thrust force acts at a point whose position vector relative to ¯ x is π
t
,
then, with the assistance of the work–energy theorem, establish an expres
sion for
˙
E.
9.9. This famous problem is discussed in most books on satellite dynamics (see, for
example, Beletskii [16] or Hughes [97]). The 24 solutions subsequently discussed
date to Lagrange [116, 118] in the late 18th century. Among the remarkable features
about Lagrange’s extraordinary work on this topic in [118] is the (early) use of his
celebrated equations of motion in the context of a rigid body and his clear discussion
of a set of (what are now known as) 3–1–3 Euler angles. As shown in Figure 9.10,
consider a rigid body B of mass m that is in motion in a central gravitational force
ﬁeld about a massive ﬁxed body of mass M. The center of this force ﬁeld is assumed
to be located at a ﬁxed point O. The force, moment, and potential energy of the ﬁeld
are given by approximations (8.25).
Rigid body B
Fixed body of mass M
O
¯
X
E
1
E
2
E
3
¯ x
Figure 9.10. Schematic of a rigid body of mass m
that is in orbit about a ﬁxed symmetric body of
mass M.
(a) Verify that M
n
= −¯ x ×F
n
. What is the physical relevance of this result?
(b) Why are the angular momentum H
O
and the total energy E of the satellite
conserved?
(c) Using the balance of linear momentum, show that it is possible for the body
to move in a circular orbit ¯ x = R
0
e
r
about Owith a constant orbital angular
velocity
˙
θ
0
, which is known as the modiﬁed Kepler frequency, ω
Km
:
˙
θ
0
= ω
Km
= ω
K
1 +
3
2R
2
0
m
(tr(J) −3e
r
· Je
r
), (9.31)
where the Kepler frequency was deﬁned previously [see (2.12)]:
ω
2
K
=
GM
r
3
0
.
In (9.31), e
r
= cos(θ)E
1
+sin(θ)E
2
, and this vector is an eigenvector of J.
That is, e
r
is parallel to one of the principal axes of the body.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
302 Exercises 9.9–9.10
(d) Using the results of (c), show that a steady motion of the rigid body, that is,
one in which ˙ ω = 0, is governed by the equation
ω ×(Jω) = 3ω
2
K
e
r
×(Je
r
). (9.32)
(e) Suppose that the body is asymmetric. That is, the principal values of J
0
are
distinct. We seek solutions of (9.32) such that ω · e
r
= 0. Show that there
are six possible solutions for ω that satisfy (9.32) and four possible solutions
for e
r
. Here, you should assume that J is known and as a result Q is known.
As a result, there are 6 ×4 possible solutions of (9.32).
(f) Suppose that the body is such that J = µI, where µ is a constant. Show that
any constant ω satisﬁes (9.32) and consequently, any orientation of the rigid
body is possible in this case.
(g) Using the results of (e), explain why it is possible for an Earthbased ob
server to see the same side of a satellite in a circular orbit above the
Earth.
9.10. Consider the problem of a sphere of mass m, radius R, and inertia tensor J
0
=
2mR
2
5
I moving on a turntable, which is discussed by Gersten et al. [70], Lewis and
Murray [127], and Pars [170], among others. The contact between the sphere and
the turntable is rough. In addition, the center O of the turntable is ﬁxed and the
turntable rotates about the vertical E
3
with an angular speed .
(a) Suppose that the sphere is rolling on the turntable. The position vector of
the point of contact of the sphere with the turntable is π
P
= −RE
3
. Show
that the motion of the sphere is subject to three constraints:
¯ v +ω ×(−RE
3
) = E
3
× ¯ x. (9.33)
Using the representations ω =
¸
3
i=1
i
E
i
and ¯ x =
¸
3
i=1
x
i
E
i
, show that the
three constraints imply that
¨ x
1
−R
˙
2
+˙ x
2
= 0, ¨ x
2
+R
˙
1
−˙ x
1
= 0, ¨ x
3
= 0.
(b) Assuming that a vertical gravitational force acts on the sphere, draw a free
body diagram of the sphere.
(c) Using a balance of linear momentum and with the assistance of the con
straints, show that the constraint force acting on the sphere is
F
c
= mgE
3
−m( ˙ x
2
E
1
− ˙ x
1
E
2
) +mR
˙
2
E
1
−
˙
1
E
2
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
Exercises 9.10–9.11 303
(d) Using a balance of angular momentum and with the assistance of the results
of (a)–(c), show that the equations governing the motion of the sphere are
¨ x
1
= R
˙
2
−˙ x
2
,
¨ x
2
= −R
˙
1
+˙ x
1
,
˙ x
3
= 0,
˙
1
=
mR
µ +mR
2
˙ x
1
,
˙
2
=
mR
µ +mR
2
˙ x
2
,
˙
3
= 0. (9.34)
Here, µ =
2mR
2
5
. Why are (9.34) sufﬁcient to determine the motion
(¯ x(t), Q(t)) of the sphere?
(e) For the special case in which = 0, show that the center of mass of the
sphere will move in a straight line with a constant speed and that the angular
velocity vector ω of the sphere will be constant.
(f) Numerically integrate (9.34) for a variety of initial conditions. Is it possible
for the sphere to fall off a turntable of radius R
0
? In choosing your ini
tial conditions (¯ x(t
0
), ¯ v(t
0
), ω(t
0
)) make sure that they are compatible with
rolling condition (9.33).
9.11. Recall the deﬁnition of the kinetic energy T of a rigid body:
T =
1
2
R
v · vρdv.
(a) Starting from the deﬁnition of T, prove the Koenig decomposition:
T =
1
2
m¯ v · ¯ v +
1
2
H· ω.
(b) Establish the following intermediate results:
˙
J = J −J,
˙
Jω·ω = H· ˙ ω,
˙
Jω · ω = 2
˙
H· ω,
where the angular momentum H = Jω.
(c) Using the intermediate results and the balance laws, prove the work–energy
theorem:
˙
T = F · ¯ v +M· ω.
(d) If
F =
K
¸
i=1
F
i
, M =
K
¸
i=1
(x
i
− ¯ x) ×F
i
+M
P
,
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
304 Exercises 9.11–9.13
then show that
˙
T =
K
¸
i=1
F
i
· v
i
+M
P
· ω.
Give three examples of the use of this result for a single rigid body.
9.12. Establish the following theorem, which can be found in Section 44 of Euler
[54]: “For any given (rigid) body, we can always ﬁnd an axis, which passes through
its center of gravity, about which the body can rotate freely and uniformly.” How
is this result related to Euler’s later result that three distinct axes exist about which
a rigid body can rotate freely and at constant angular velocity? This result can be
found in [53], where these axes were ﬁrst termed principal axes of inertia.
9.13. Consider the tippe top shown in Figure 8.12 and discussed in Exercise 8.4.
(a) Continuing the earlier exercise, we follow Or [156] and suppose that the
point P slides on the horizontal surface and experiences both Coulomb and
viscous friction forces:
F
f
= −(µ
k
+µ
v
v
P
) N s,
where µ
k
and µ
v
are friction coefﬁcients, N is the normal force, and s is the
slip direction. Using a balance of linear momentum, show that the normal
force acting on the tippe top is
N = m
g +l
¨
θ sin(θ) +l
˙
θ
2
cos(θ)
E
3
.
(b) Starting from the work–energy theorem, prove that the total energy of a
sliding tippe top decreases with time whereas that for a rolling tippe top is
constant.
(c) If the inertia tensor of the tippe top is
J = λ
t
(I −e
3
⊗e
3
) +λ
a
e
3
⊗e
3
,
then what are the differential equations governing the motion of the tippe
top?
(d) For both rolling and sliding tippe tops, show that the angular momentum
H· π
P
is conserved. This integral of motion, which is known as the Jellett
integral, was discovered in the 1870s by J. H. Jellett (1817–1888).
∗
(e) When the tippe top is rolling on a horizontal surface, show that, in addi
tion to conservation of total energy, the following kinematical quantity is
conserved:
I
1
= ω
2
3
λ
a
λ
t
+mλ
a
(π
P
· e
1
)
2
+(π
P
· e
2
)
2
+mλ
t
(π
P
· e
3
)
2
. (9.35)
∗
A discussion of the history of this integral (and several others) can be found in Gray and Nickel
[76].
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
Exercises 9.13–9.14 305
Here, π
P
is the position vector of the instantaneous point of contact P (rel
ative to the center of mass
¯
X) of the tippe top with the horizontal surface.
The integral of motion I
1
was ﬁrst established by Routh (see Section 243 of
[184]) and can be described as the Routh integral. It is also known as the
Chaplygin integral [107, 115].
9.14. Consider a body that is free to move about one of its material points O that
is ﬁxed (cf. Figure 9.7). The inertia tensor of the body relative to its center of mass
is J =
¸
3
i=1
λ
i
e
i
⊗e
i
, where {e
1
, e
2
, e
3
} is a corotational basis. The position vector of
the center of mass
¯
X relative to Ois
¯ x −x
O
= he
3
,
where h is a constant. A linear spring of stiffness K and unstretched length L
0
is
attached to the body at the point X
S
and the other end is attached to a ﬁxed point
A:
x
s
−x
O
= s
1
e
1
+s
3
e
3
, x
A
−x
O
= L
A
E
1
.
In addition, a gravitational force −mgE
3
acts on the body.
(a) Show that the velocity and acceleration vectors of the center of mass have
the representations
¯ v = hω
2
e
1
−hω
1
e
2
,
¯ a = h ( ˙ ω
2
+ω
1
ω
3
) e
1
−h ( ˙ ω
1
−ω
2
ω
3
) e
2
−h
ω
2
1
+ω
2
2
e
3
.
Here, ω =
¸
3
k=1
ω
k
e
k
.
(b) Show that the angular momentum of the rigid body relative to Ois
H
O
=
λ
1
+mh
2
ω
1
e
1
+
λ
2
+mh
2
ω
2
e
2
+λ
3
ω
3
e
3
.
Show that the kinetic energy of the rigid body has the representation
T =
1
2
H
O
· ω.
(c) What are the three constraints on the motion of the rigid body? Give pre
scriptions for the constraint force F
c
and constraint moment M
c
that enforce
these constraints.
(d) Draw a freebody diagram of the rigid body.
(e) Using a balance of linear momentum, show that the reaction force at Ois
F
c
= mh ( ˙ ω
2
+ω
1
ω
3
) e
1
−mh ( ˙ ω
1
−ω
2
ω
3
) e
2
−mh
ω
2
1
+ω
2
2
e
3
+mgE
3
−F
s
,
where F
s
is the spring force.
(f) Assuming that a set of 3–1–3 Euler angles is used to parameterize Q =
¸
3
i=1
e
i
⊗E
i
, show that the potential energy U = U(¯ x, Q) of the rigid
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 9, 2008 21:29
306 Exercise 9.14
body in this problem can be expressed as a function of these angles: U =
ˆ
U(ψ, θ, φ).
(g) Verify that the conservative force F
con
and conservative moment M
con
act
ing on the rigid body satisfy the identity
F
con
· ¯ v +M
con
· ω = M
O
con
· ω,
where M
O
con
is the resultant conservative moment acting on the body rela
tive to O. Using this result, explain why
M
O
con
= −
∂
ˆ
U
∂ψ
g
1
−
∂
ˆ
U
∂θ
g
2
−
∂
ˆ
U
∂φ
g
3
.
(h) Show that the rotational motion of the rigid body is governed by the follow
ing equations:
λ
1
+mh
2
˙ ω
1
=
λ
2
+mh
2
−λ
3
ω
2
ω
3
+M
O
· e
1
,
λ
2
+mh
2
˙ ω
2
=
λ
3
−λ
1
−mh
2
ω
1
ω
3
+M
O
· e
2
,
λ
3
˙ ω
3
= (λ
1
−λ
2
) ω
1
ω
2
+M
O
· e
3
,
⎡
⎢
⎣
˙
ψ
˙
θ
˙
φ
⎤
⎥
⎦
=
⎡
⎢
⎣
sin(φ)cosec(θ) cos(φ)cosec(θ) 0
cos(φ) −sin(φ) 0
−sin(φ) cot(θ) −cos(φ) cot(θ) 1
⎤
⎥
⎦
⎡
⎢
⎣
ω
1
ω
2
ω
3
⎤
⎥
⎦
. (9.36)
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
10 Lagrange’s Equations of Motion for a Single
Rigid Body
10.1 Introduction
In a famous paper [118] on the dynamics of the Moon orbiting the Earth, which was
published in 1782, Lagrange used his equations of motion in the form
d
dt
∂T
∂ ˙ q
K
−
∂T
∂q
K
= −
∂V
∂q
K
(K = 1, . . . , 6) , (10.1)
where V is the potential energy of the Moon and T is its kinetic energy. His interest
lay in explaining oscillations (librations) in the attitude of the Moon as seen by an
Earthbased observer. What is interesting is that he does not use Euler’s balance
laws, although these were available to him. One might ask what would have hap
pened had he used M =
˙
H and F = m
˙
¯ v instead of (10.1)? In this chapter, we will
show (among other matters) that his equations are equivalent to Euler’s balance
laws and so he would have arrived at the same conclusions.
We start this chapter by showing that the balance laws F = m
˙
¯ v and M =
˙
H for
a rigid body are equivalent to Lagrange’s equations of motion:
d
dt
∂T
∂ ˙ q
i
−
∂T
∂q
i
= F · a
i
,
d
dt
∂T
∂ ˙ ν
i
−
∂T
∂ν
i
= M· g
i
. (10.2)
We illustrate this form of Lagrange’s equations by using a classical problem of a
satellite orbiting a ﬁxed body. Indeed, the problem we consider is equivalent to that
considered by Lagrange [118].
A second form of Lagrange’s equations of motion is also developed. This form
allows broader classes of coordinate choices and, being the most general, is most
useful in applications. The form of the celebrated equations is
d
dt
∂T
∂ ˙ u
A
−
∂T
∂u
A
=
A
(A= 1, . . . , 6) , (10.3)
307
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
308 Lagrange’s Equations of Motion for a Single Rigid Body
where
A
= F ·
∂ ¯ v
∂ ˙ u
A
+M·
∂ω
∂ ˙ u
A
. (10.4)
Following the development of (10.3), we show how constraints may be incorporated
and illustrate this matter by using examples of rolling disks, sliding disks, and spin
ning tops.
When the system of constraints on the system is integrable, the constraint forces
and moments are prescribed by use of Lagrange’s prescription, and the coordinates
are chosen appropriately, we pleasantly ﬁnd that we can decouple (10.3) into a set
of equations involving the unconstrained motion and a set of equations for the
constraint forces and moments. The former equations are known as reactionless.
We then discuss an approach to Lagrange’s equations that we have referred to as
Approach II.
Much of the material concerning Lagrange’s equations in this chapter is based
on Casey [26, 28] supplemented with material on constraint forces and moments
that is due to O’Reilly and Srinivasa [163].
10.2 Conﬁguration Manifold of an Unconstrained Rigid Body
The motion of a rigid body can be decomposed into a translation ¯ x of the center of
mass
¯
Xfollowed by a rotation Q. Here, the set of all vectors ¯ x is a threedimensional
Euclidean space E
3
, whereas the set of all rotation tensors Q is a threedimensional
space known as SO(3).
∗
Thus the conﬁguration manifold M of a rigid body is the
space of all vectors ¯ x and all rotation tensors Q. This space is the product of E
3
and
SO(3):
M = E
3
⊕SO(3).
The product ⊕ is a topological product. For example, E
3
= E ⊕E
2
= E ⊕E ⊕E. It
should be clear that the dimension of Mis 6. It should also be clear that the conﬁg
uration manifold can be considered as a submanifold of the conﬁguration space S,
which in this case is E
3
⊕E
9
. Here the threedimensional space is the space contain
ing the position vector ¯ x, and the ninedimensional space is the space containing the
secondorder tensor Q.
To parameterize M, we can use any curvilinear coordinate system. In one of
the forms to follow, we use q
1
, q
2
, and q
3
, to parameterize the position vector ¯ x of
the center of mass and any set of Euler angles, ν
1
, ν
2
, and ν
3
, to parameterize the
rotation tensor Q. For the velocity vector of the center of mass, we then have
¯ v =
3
¸
i=1
˙ q
i
a
i
,
∗
The space O(3) is the space of all orthogonal tensors. Hence, the S in SO(3) denotes “special”
because rotation tensors have a determinant of +1.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
10.2 Conﬁguration Manifold of an Unconstrained Rigid Body 309
where the covariant basis vectors
a
i
=
∂ ¯ x
∂q
i
.
You should also recall that {a
i
} is a basis for E
3
and that the Euler angles deﬁne the
Euler basis {g
i
}, which is also a basis for E
3
. In particular, we have
ω =
3
¸
i=1
˙ ν
i
g
i
.
The Euler basis vectors are not linearly independent for certain values of the second
Euler angle. It is interesting to note that
g
i
=
∂ω
∂ ˙ ν
i
, a
i
=
∂ ¯ v
∂ ˙ q
i
.
These two identities are easily established.
We can calculate the kinematical lineelement ds for the conﬁguration mani
fold by using the given choice of Euler angles and curvilinear coordinates from the
kinetic energy T:
ds =
2T
m
dt
=
¯ v · ¯ v +
1
m
ω · (J · ω)
dt.
Shortly an example will be presented of how to calculate the desired representation
for T.
As we are using curvilinear coordinates and Euler angles to parameterize
the motion of the rigid body, for certain points on the conﬁguration manifold
singularities in the parameters will arise. At these points, it is possible for the
body to be in motion, yet the value we will get for T will be zero. This situation
violates one of the chief attributes of T, namely that T is zero if and only if ¯ v = 0
and ω = 0, i.e., the rigid body is instantaneously at rest. The singularities will also
result in errors when the equations of motion are being integrated numerically.
For this reason, many computer codes use two or more sets of curvilinear coor
dinates and two sets of Euler angles for analyzing a given problem in rigid body
dynamics.
A Representation for the Kinetic Energy
We now consider a speciﬁc example. First, we choose a set of spherical polar
coordinates R, , and to parameterize ¯ x and a set of 3–1–3 Euler angles to
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
310 Lagrange’s Equations of Motion for a Single Rigid Body
parameterize Q. For these Euler angles, we recall from Subsection 6.8.2 that the
Euler basis {g
i
} has the representations
¸
¸
¸
g
1
g
2
g
3
=
¸
¸
¸
E
3
e
1
e
3
=
¸
¸
¸
sin(φ) sin(θ) cos(φ) sin(θ) cos(θ)
cos(φ) −sin(φ) 0
0 0 1
¸
¸
¸
e
1
e
2
e
3
.
Using these results, we can readily compute the angular velocity vector:
ω =
˙
ψE
3
+
˙
θe
1
+
˙
φe
3
=
˙
ψsin(φ) sin(θ) +
˙
θ cos(φ)
e
1
+
˙
ψcos(φ) sin(θ) −
˙
θ sin(φ)
e
2
+
˙
ψcos(θ) +
˙
φ
e
3
. (10.5)
For convenience, we choose E
i
to be the principal axes of the body.
∗
Hence,
J
0
= λ
1
E
1
⊗E
1
+λ
2
E
2
⊗E
2
+λ
3
E
3
⊗E
3
,
J = QJ
0
Q
T
= λ
1
e
1
⊗e
1
+λ
2
e
2
⊗e
2
+λ
3
e
3
⊗e
3
.
When these results are combined, the kinetic energy T of the rigid body has the
representations
T =
m
2
¯ v · ¯ v +
1
2
ω · Jω
=
m
2
˙
R
2
+R
2
˙
2
+R
2
sin
2
()
˙
2
+
λ
1
2
˙
ψsin(φ) sin(θ) +
˙
θ cos(φ)
2
+
λ
2
2
˙
ψcos(φ) sin(θ) −
˙
θ sin(φ)
2
+
λ
3
2
˙
ψcos(θ) +
˙
φ
2
.
It is left as an exercise to substitute this expression for T into the expression for
the kinematical lineelement ds in order to ﬁnd a measure of distance that the rigid
body travels along the conﬁguration manifold M.
Singularities in the parameterization of the motion arise when θ = 0, π and =
0, π. To see the effects of these singularities on T, let us consider an instant in which
= 0 and θ = 0. At this instant, the preceding expression for T simpliﬁes to
T =
m
2
˙
R
2
+R
2
˙
2
+0
+
λ
1
2
0 +
˙
θ cos(φ)
2
+
λ
2
2
0 −
˙
θ sin(φ)
2
+
λ
3
2
˙
ψ +
˙
φ
2
.
∗
If we do not make this choice, then we would ﬁnd a more complicated representation for the ro
tational kinetic energy: ω · (Jω) = J
011
ω
2
1
+J
022
ω
2
2
+J
033
ω
2
3
+2J
012
ω
1
ω
2
+2J
023
ω
2
ω
3
+2J
013
ω
1
ω
3
,
where ω
i
= ω · e
i
and J
0ik
= e
i
· (Je
k
) = E
i
· (J
0
E
k
).
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
10.3 Lagrange’s Equations of Motion: A First Form 311
If, in addition,
˙
ψ = −
˙
φ,
˙
θ =
˙
R =
˙
= 0, and
˙
= 0, then the preceding expression
for T = 0, but the actual kinetic energy of the body is nonzero.
10.3 Lagrange’s Equations of Motion: A First Form
In this section, we wish to establish Lagrange’s equations for an unconstrained rigid
body. Our proof is based on Casey [28], but our developments are not as general
as his. Resulting equations (10.6) were ﬁrst shown by him to be equivalent to the
balances of linear and angular momentum for a rigid body.
∗
There is also a strong
suggestion of the equivalence of Lagrange’s equations of motion and the balance of
angular momentum for a rigid body in Sections 82 and 86 of Greenwood [79].
†
We
refer to (10.6) as the ﬁrst form of Lagrange’s equations of motion.
To start, we choose a set of curvilinear coordinates q
i
to parameterize ¯ x: ¯ x =
¯ x(q
1
, q
2
, q
3
). Next, we choose a set of Euler angles ν
i
to parameterize the rotation
tensor Q of the rigid body Q = Q(ν
1
, ν
2
, ν
3
). Then, as will be subsequently shown,
Lagrange’s equations for the rigid body are
d
dt
∂T
∂ ˙ q
i
−
∂T
∂q
i
= F · a
i
,
d
dt
∂T
∂ ˙ ν
i
−
∂T
∂ν
i
= M· g
i
. (10.6)
Here, g
i
are the Euler basis vectors and a
i
are the basis vectors for E
3
that are asso
ciated with the curvilinear coordinate q
i
. For the preceding equations, you may wish
to recall the results
¯ v =
3
¸
i=1
˙ q
i
a
i
,
∂ ¯ v
∂ ˙ q
i
= a
i
, ω =
3
¸
i=1
˙ ν
i
g
i
,
∂ω
∂ ˙ ν
i
= g
i
.
It should be evident from (10.6) that Lagrange’s equations for a rigid body have
similarities to those we encountered earlier with particles. The main difference is
the balance of angular momentum.
If some of the forces and moments acting on the rigid body are conservative,
then, for these conservative forces F
con
and moments M
con
, we have
F
con
= −
3
¸
i=1
∂U
∂q
i
a
i
, M
con
= −
3
¸
i=1
∂U
∂ν
i
g
i
, (10.7)
where the potential energy function Uhas the representations
U = U(¯ x, Q) = U
q
1
, q
2
, q
3
, ν
1
, ν
2
, ν
3
.
Notice that F
con
· a
i
= −
∂U
∂q
i
and M
con
· g
i
= −
∂U
∂ν
i
. Consequently, it is not necessary
to evaluate F
con
and M
con
in Lagrange’s equations; rather it sufﬁces to evaluate the
∗
See, in particular, Theorems 4.2 and 4.4 of Casey [28].
†
Greenwood’s exposition is missing the intermediate result
∂T
∂ ˙ γ
i
= H· ˙ g
i
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
312 Lagrange’s Equations of Motion for a Single Rigid Body
partial derivatives of U. It is remarkable that the situation with potential energies in
rigid bodies is similar to that encountered in systems of particles.
It is also possible to write an alternative form of Lagrange’s equations of motion
by using the Lagrangian L = T −U. Speciﬁcally,
d
dt
∂L
∂ ˙ q
i
−
∂L
∂q
i
= (F −F
con
) · a
i
,
d
dt
∂L
∂ ˙ ν
i
−
∂L
∂ν
i
= (M−M
con
) · g
i
. (10.8)
It is left as an (easy) exercise to show that (10.8) can be established from (10.6).
10.3.1 Proof of Lagrange’s Equations
To prove Lagrange’s equations, we need to exploit the Koenig decomposition and
use the angular velocity vector ω
0
= Q
T
ω. The proof proceeds quickly after some
preliminary results have been addressed.
There are four steps in the proof. The ﬁrst step involves parameterizing ¯ x by
use of a set of curvilinear coordinates and parameterizing Q by use of a set of Euler
angles ν
i
. These parameterizations imply that the kinetic energy T is a function of
these quantities and their time derivatives:
T = T
q
i
, ˙ q
i
, ν
k
, ˙ ν
k
=
1
2
m¯ v · ¯ v +
1
2
J
0
ω
0
· ω
0
.
Here, the angular velocity vector ω
0
= Q
T
ω and the inertia tensor J
0
= Q
T
JQ.
We next consider the partial derivatives of T:
∂T
∂ ˙ q
i
=
∂
∂ ˙ q
i
1
2
m¯ v · ¯ v
= m¯ v ·
∂ ¯ v
∂ ˙ q
i
= m¯ v · a
i
,
∂T
∂q
i
=
∂
∂q
i
1
2
m¯ v · ¯ v
= m¯ v ·
∂ ¯ v
∂q
i
= m¯ v · ˙ a
i
,
∂T
∂ ˙ ν
i
=
∂
∂ ˙ ν
i
1
2
J
0
ω
0
· ω
0
= J
0
ω
0
·
∂ω
0
∂ ˙ ν
i
= J
0
ω
0
·
Q
T
g
i
,
∂T
∂ν
i
=
∂
∂ν
i
1
2
J
0
ω
0
· ω
0
= J
0
ω
0
·
∂ω
0
∂ν
i
= J
0
ω
0
·
Q
T
˙ g
i
.
(10.9)
Here we have used the identities
a
i
=
∂ ¯ v
∂ ˙ q
i
, ˙ a
i
=
∂ ¯ v
∂q
i
, g
i
= Q
∂ω
0
∂ ˙ ν
i
, ˙ g
i
= Q
∂ω
0
∂ν
i
. (10.10)
Shortly a derivation of these results will be given.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
10.3 Lagrange’s Equations of Motion: A First Form 313
For the third step, we examine (10.9)
1,2
and note that G = m¯ v. Hence,
d
dt
∂T
∂ ˙ q
i
−
∂T
∂q
i
=
d
dt
(G· a
i
) −G· ˙ a
i
=
˙
G· a
i
= F · a
i
.
Consequently, the ﬁrst three Lagrange equations of motion have been established.
The fourth step in the derivation is to note that, for any vector b,
H· b = Jω · b = QJ
0
Q
T
Qω
0
· b = J
0
ω
0
· Q
T
b.
Using this result in conjunction with (10.9)
3,4
, we ﬁnd that
d
dt
∂T
∂ ˙ ν
i
−
∂T
∂ν
i
=
d
dt
(H· g
i
) −H· ˙ g
i
=
˙
H· g
i
= M· g
i
.
Thus we have established the last three of Lagrange’s equations of motion.
10.3.2 The Four Identities
The proof of Lagrange’s equations is achieved by use of the four identities, (10.10).
The ﬁrst two of these results are similar to those we established for the single parti
cle. You should notice the presence of Q in (10.10)
3,4
. Unfortunately,
∂ω
∂ν
i
= ˙ g
i
.
We ﬁrst establish the easier results:
∂ ¯ v
∂ ˙ q
i
=
∂
∂ ˙ q
i
3
¸
k=1
˙ q
k
a
k
=
3
¸
k=1
δ
ik
a
k
= a
i
and
∂ ¯ v
∂q
i
=
∂
∂q
i
3
¸
k=1
˙ q
k
a
k
=
3
¸
k=1
˙ q
k
∂a
k
∂q
i
=
3
¸
k=1
˙ q
k
∂
2
¯ x
∂q
i
∂q
k
=
3
¸
k=1
˙ q
k
∂
∂q
k
∂ ¯ x
∂q
i
=
3
¸
k=1
˙ q
k
∂a
i
∂q
k
= ˙ a
i
.
Notice that we used the facts that a
i
are both independent of ˙ q
k
and the derivatives
of ¯ x with respect to q
i
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
314 Lagrange’s Equations of Motion for a Single Rigid Body
We next consider the easier of the two remaining identities:
∂ω
0
∂ ˙ ν
i
=
∂
∂ ˙ ν
i
Q
T
ω
=
∂Q
T
∂ ˙ ν
i
ω +Q
T
∂ω
∂ ˙ ν
i
= 0ω +Q
T
(g
i
)
= Q
T
g
i
.
Some rearranging gives
g
i
= Q
∂ω
0
∂ ˙ ν
i
.
Notice that we used the fact that Q does not depend on ˙ ν
i
to establish this result.
For the ﬁnal result, we again need to be cognizant of the fact that Q depends on
the Euler angles ν
i
but not on ˙ ν
i
. First, we note that
g
i
=
∂ω
∂ ˙ ν
i
=
∂
∂ ˙ ν
i
−
1
2
¸
˙
QQ
T
¸
=
∂
∂ ˙ ν
i
−
1
2
¸
3
¸
k=1
˙ ν
k
∂Q
∂ν
k
Q
T
= −
1
2
¸
∂Q
∂ν
i
Q
T
¸
.
Next, we deduce, by differentiating QQ
T
= I twice with respect to the Euler angles,
that
∂Q
∂ν
i
Q
T
= −Q
∂Q
T
∂ν
i
,
∂Q
∂ν
i
∂Q
T
∂ν
k
= Q
∂Q
T
∂ν
i
∂Q
∂ν
k
Q
T
.
The previous results are now used to show the desired identity:
˙ g
i
=
d
dt
−
1
2
¸
∂Q
∂ν
i
Q
T
¸
=
3
¸
k=1
˙ ν
k
−
1
2
¸
∂
2
Q
∂ν
i
∂ν
k
Q
T
+
∂Q
∂ν
i
∂Q
T
∂ν
k
¸
=
3
¸
k=1
˙ ν
k
−
1
2
¸
Q
Q
T
∂
2
Q
∂ν
i
∂ν
k
+
∂Q
T
∂ν
i
∂Q
∂ν
k
Q
T
¸
=
3
¸
k=1
˙ ν
k
−
1
2
¸
Q
∂
∂ν
i
Q
T
∂Q
∂ν
k
Q
T
¸
= −
1
2
¸
Q
∂
∂ν
i
Q
T
3
¸
k=1
˙ ν
k
∂Q
∂ν
k
Q
T
= −
1
2
¸
Q
∂
0
∂ν
i
Q
T
¸
= Q
−
1
2
¸
∂
0
∂ν
i
¸
= Q
∂ω
0
∂ν
i
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
10.4 A Satellite Problem 315
In the ﬁnal stages of the proof we used the facts that ω
0
is the axial vector of
0
=
Q
T
˙
Q = Q
T
Q. We also used the identity
¸
QBQ
T
¸
= det (Q) Q( [B]) .
This identity is one method relating ω
0
to ω, and it was alluded to earlier when
angular velocity vectors were discussed [see (6.7)].
10.4 A Satellite Problem
As an example of a problem from rigid body dynamics for which there are no con
straints, we consider a satellite of mass m that is in orbit about a spherically sym
metric body of mass M (cf. Figure 10.1). We assume that the spherically symmetric
body is ﬁxed and use its center of mass as the origin of the position vector of the
center of mass of the satellite.
Preliminaries
To parameterize the motion of the center of mass of the rigid body we use a spherical
polar coordinate system:
¯ x = Re
R
= Rcos(φ)E
3
+Rsin(φ)e
r
= Rcos(φ)E
3
+Rsin(φ) (cos(θ)E
1
+sin(θ)E
2
)
and
¯ v =
˙
Re
R
+Rsin(φ)
˙
θe
θ
+R
˙
φe
φ
.
You should be able to see from this equation what the covariant basis vectors a
i
are.
We parameterize the rotation tensor Q by using a set of 1–2–3 Euler angles:
ω = ˙ ν
1
E
1
+ ˙ ν
2
e
2
+ ˙ ν
3
e
3
.
You should notice that e
i
= QE
i
. The angular velocity vector of the body also has
the representation
ω = ω
1
e
1
+ω
2
e
2
+ω
3
e
3
,
where
ω
1
= ˙ ν
2
sin(ν
3
) + ˙ ν
1
cos(ν
2
) cos(ν
3
),
ω
2
= ˙ ν
2
cos(ν
3
) − ˙ ν
1
cos(ν
2
) sin(ν
3
),
ω
3
= ˙ ν
1
sin(ν
2
) + ˙ ν
3
.
You should be able to infer the representations for g
i
from these results.
We shall choose E
i
to be the principal axes of the body in its reference conﬁg
uration. Using this speciﬁcation, it follows that the kinetic energy of the rigid body
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
316 Lagrange’s Equations of Motion for a Single Rigid Body
Rigid body B
Fixed symmetric body of mass M
O
θ
β
¯
X
E
1
E
2
E
3
e
1
e
2
e
3
e
r
¯ x
Figure 10.1. A rigid body orbiting a spherically symmetric body of mass M. The angle of lati
tude β =
π
2
−φ.
has the representation
T =
m
2
˙
R
2
+R
2
˙
φ
2
+R
2
sin
2
(φ)
˙
θ
2
+
λ
1
2
( ˙ ν
2
sin(ν
3
) + ˙ ν
1
cos(ν
2
) cos(ν
3
))
2
+
λ
2
2
( ˙ ν
2
cos(ν
3
) − ˙ ν
1
cos(ν
2
) sin(ν
3
))
2
+
λ
3
2
( ˙ ν
1
sin(ν
2
) + ˙ ν
3
)
2
.
If the body has an axis of symmetry so that λ
1
= λ
2
, then the expression for the
rotational kinetic energy will simplify considerably.
The sole force and moment acting on the rigid body is due to the central grav
itational force exerted on it by the body of mass M. These forces and moments are
conservative and are associated with the potential energy
U = −
GMm
R
−
GM
2R
3
tr(J) +
3GM
2R
3
(Je
R
) · e
R
. (10.11)
To express this potential energy in terms of the Euler angles ν
i
, we need to use the
results
e
R
= cos(φ)E
3
+sin(φ) cos(θ)E
1
+sin(φ) sin(θ)E
2
,
E
1
= cos(ν
2
) cos(ν
3
)e
1
−cos(ν
2
) sin(ν
3
)e
2
+sin(ν
2
)e
3
,
E
2
= −sin(ν
2
) cos(ν
3
)e
1
+sin(ν
2
) sin(ν
3
)e
2
+cos(ν
2
)e
3
,
E
3
= (sin(ν
1
) sin(ν
3
) −cos(ν
1
) sin(ν
2
) cos(ν
3
)) e
1
+ (sin(ν
1
) cos(ν
3
) +cos(ν
1
) sin(ν
2
) sin(ν
3
)) e
2
+cos(ν
1
) cos(ν
2
)e
3
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
10.4 A Satellite Problem 317
Using these results, we can express e
R
in terms of the corotational basis, and then a
useful expression for (Je
R
) · e
R
can be established. We can then derive the expres
sions for the force and moment associated with this potential energy by using repre
sentations (10.7). Alternatively, we can appeal to representations (8.25) described
in Chapter 8.
The Balance Laws
For the rigid body of interest, we have the balance laws
m
˙
¯ v = F
= −
GMm
R
2
e
R
−
3GM
2R
4
(2J +((λ
1
+λ
2
+λ
3
) −5e
R
· Je
R
) I) e
R
,
˙
H = M
=
3GM
R
3
e
R
×(Je
R
).
Here, we used our earlier representations, (8.25), for the conservative force and
conservative moment associated with the gravitational potential. You should also
notice that, if e
R
is an eigenvector of J, then Je
R
is parallel to e
R
. In this case, the
socalled gravitygradient torque M = 0. In addition, if R = R
0
e
r
and
˙
θ = ω
0
, where
R
0
and ω
0
are constant, then the center of mass of the rigid body describes a circular
orbit at a constant orbital speed ω
0
.
Lagrange’s Equations of Motion
Lagrange’s equations of motion for the satellite are the a
i
components of the balance
of linear momentum and the g
i
components of the balance of angular momentum:
d
dt
∂T
∂
˙
R
−
∂T
∂R
= F · e
R
= −
GMm
R
2
−
3GM
2R
4
((λ
1
+λ
2
+λ
3
) −3e
R
· Je
R
) ,
d
dt
∂T
∂
˙
φ
−
∂T
∂φ
= F · Re
φ
= −
3GM
R
3
(e
R
· Je
φ
) ,
d
dt
∂T
∂
˙
θ
−
∂T
∂θ
= F · Rsin(φ)e
θ
= −
3GMsin(φ)
R
3
(e
R
· Je
θ
) ,
d
dt
∂T
∂ ˙ ν
i
−
∂T
∂ν
i
= M· g
i
=
3GM
R
3
e
R
×(Je
R
)
· g
i
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
318 Lagrange’s Equations of Motion for a Single Rigid Body
In the last of these equations, i = 1, . . . , 3. It is left as an exercise to evaluate the
partial derivatives of T with respect to the coordinates and their velocities. Al
ternatively, one could also use (10.8) to write Lagrange’s equations by using the
Lagrangian L = T −U.
Conservations
Because the only forces and moments acting on the rigid body are conservative,
and there are no constraints on the motion of the rigid body, it is easy to see that
the total energy E = T +U is conserved. In addition, the angular momentum H
O
is
conserved. To see this notice that
˙
H
O
= M
O
= Re
R
×F +M.
Substituting for F and M, one ﬁnds that M
O
= 0. Consequently H
O
is conserved.
It is interesting to note that we cannot conclude that H is conserved for an ar
bitrary motion of the rigid body – although it is conserved if the gravitygradient
torque M is zero for a speciﬁc motion.
10.5 Lagrange’s Equations of Motion: A Second Form
Previously, we assumed that a set of coordinates q
1
, . . . , q
6
had been chosen to pa
rameterize ¯ x and Q:
¯ x = ¯ x
q
1
, . . . , q
3
, Q = Q
q
4
= ν
1
, . . . , q
6
= ν
3
. (10.12)
We note that the covariant basis vectors associated with this choice of coordinates
are
a
i
=
∂ ¯ v
∂ ˙ q
i
, g
i
=
∂ω
∂ ˙ q
(i+3)
, i = 1, 2, 3.
With choice (10.12), we showed that Lagrange’s equations of motion have the fol
lowing form:
d
dt
∂T
∂ ˙ q
i
−
∂T
∂q
i
= F ·
∂ ¯ v
∂ ˙ q
i
,
d
dt
∂T
∂ ˙ q
(i+3)
−
∂T
∂q
(i+3)
= M·
∂ω
∂ ˙ q
(i+3)
. (10.13)
It is not too difﬁcult to see that these equations can be written in a more compact
form:
d
dt
∂T
∂ ˙ q
A
−
∂T
∂q
A
= F ·
∂ ¯ v
∂ ˙ q
A
+M·
∂ω
∂ ˙ q
A
(A= 1, . . . , 6) . (10.14)
Form (10.14) of Lagrange’s equations is useful in several cases. Among them,
1. there are no constraints on the motion of the rigid body;
2. the constraints (and the associated constraint forces and moments) on the rigid
body do not couple the rotational and translational degrees of freedom.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
10.5 Lagrange’s Equations of Motion: A Second Form 319
However, we shall need a more general form of Lagrange’s equations for other
applications.
We now consider a new choice of coordinates:
¯ x = ¯ x
u
1
, . . . , u
6
, Q = Q
u
1
, . . . , u
6
. (10.15)
With the choice (10.15), we shall see that Lagrange’s equations of motion are
d
dt
∂T
∂ ˙ u
A
−
∂T
∂u
A
=
A
, (10.16)
where the generalized forces are
A
= F ·
∂ ¯ v
∂ ˙ u
A
+M·
∂ω
∂ ˙ u
A
(A= 1, . . . , 6) .
We shall also shortly discuss examples that use this form.
To establish (10.16), we invoke the following identities
∗
:
d
dt
∂ ¯ v
∂ ˙ u
A
=
∂ ¯ v
∂u
A
,
d
dt
∂ω
∂ ˙ u
A
=
d
dt
Q
∂ω
0
∂ ˙ u
A
= Q
∂ω
0
∂u
A
,
where
ω = Qω
0
.
Using these identities and the decomposition of the kinetic energy,
†
we can use a
straightforward set of manipulations to establish (10.16):
d
dt
∂T
∂ ˙ u
A
−
∂T
∂u
A
=
d
dt
m¯ v ·
∂ ¯ v
∂ ˙ u
A
+Jω ·
Q
∂ω
0
∂ ˙ u
A
−
∂T
∂u
A
= m¯ v ·
∂ ¯ v
∂u
A
+J
0
ω
0
·
∂ω
0
∂u
A
=
d
dt
G·
∂ ¯ v
∂ ˙ u
A
+H·
Q
∂ω
0
∂ ˙ u
A
−
∂T
∂u
A
= G·
∂ ¯ v
∂u
A
+H·
Q
∂ω
0
∂u
A
=
˙
G·
∂ ¯ v
∂ ˙ u
A
+
˙
H·
Q
∂ω
0
∂ ˙ u
A
= F ·
∂ ¯ v
∂ ˙ u
A
+M·
∂ω
∂ ˙ u
A
. (10.17)
This form of Lagrange’s equations is the starting point for most applications and a
discussion of Approach II.
We now turn to issues associated with the second form of Lagrange’s equations
in the presence of constraints.
∗
The proof of these results follows from (10.15) in a manner that is similar to the method by which
the four identities were established in Subsection 10.3.2.
†
Recall that Jω · ω = QJ
0
ω
0
· ω = QJ
0
ω
0
· Qω
0
= J
0
ω
0
· ω
0
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
320 Lagrange’s Equations of Motion for a Single Rigid Body
10.5.1 Changing Coordinates: An Identity
Recall that we are using two coordinate systems: u
1
, . . . , u
6
and q
1
, . . . , q
6
. We have
tacitly assumed that these coordinate systems are related by invertible functions:
q
A
= q
A
u
1
, . . . , u
6
.
Then, as
˙ q
A
=
6
¸
B=1
∂q
A
∂u
B
˙ u
B
,
we conclude that
∂q
A
∂u
B
=
∂ ˙ q
A
∂ ˙ u
B
(A= 1, . . . , 6, B = 1, . . . , 6) . (10.18)
These identities are often known as “cancellation of the dots.”
It is a good exercise to show that (10.18) also hold when the coordinate trans
formations are time dependent:
q
A
= q
A
u
1
, . . . , u
6
, t
.
The resulting identities are used in the literature with Approach II of Lagrange’s
equations of motion.
10.5.2 Constraints and Constraint Forces and Moments
Suppose that an integrable constraint is imposed on the rigid body:
q
1
, . . . , q
6
, t
= 0.
We assume that we are at liberty to choose the coordinates u
1
, . . . , u
6
to write this
constraint in a simpler form, such as
u
6
− f (t) = 0.
The question we wish to ask now is, what are the consequences for the righthand
side of Lagrange’s equations?
The constraint = 0 implies that
˙
= 0. With some rearranging we ﬁnd that
˙
= 0 can be expressed as
f · ¯ v +h · ω +e = 0,
where
f =
3
¸
i=1
∂
∂q
i
a
i
, h =
3
¸
i=1
∂
∂q
(i+3)
g
i
, e =
∂
∂t
.
For the present, it is most convenient to have representations for f and h in terms of
the coordinates for ¯ x and Q, respectively, rather than in terms of u
1
, . . . , u
6
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
10.5 Lagrange’s Equations of Motion: A Second Form 321
To answer the question we just posed, we suppose that Lagrange’s prescription
has been used to prescribe the unknown constraint force and constraint moment:
F
c
= µf = µ
3
¸
i=1
∂
∂q
i
a
i
,
M
c
= µh = µ
3
¸
i=1
∂
∂q
i+3
g
i
.
It is interesting to expand the constraint forces and moments for this case:
F
c
·
∂ ¯ v
∂ ˙ u
A
+M
c
·
∂ω
∂ ˙ u
A
=
3
¸
i=1
µ
∂
∂q
i
a
i
·
∂ ¯ v
∂ ˙ u
A
+
3
¸
i=1
µ
∂
∂q
i+3
g
i
·
∂ω
∂ ˙ u
A
= µ
3
¸
i=1
3
¸
k=1
∂
∂q
i
a
i
·
∂ ˙ q
k
∂ ˙ u
A
a
k
+µ
3
¸
i=1
3
¸
k=1
∂
∂q
i+3
g
i
·
∂ ˙ q
k+3
∂ ˙ u
A
g
k
= µ
3
¸
i=1
3
¸
k=1
∂
∂q
i
∂ ˙ q
k
∂ ˙ u
A
δ
k
i
+µ
3
¸
i=1
3
¸
k=1
∂
∂q
i+3
∂ ˙ q
k+3
∂ ˙ u
A
δ
k
i
= µ
3
¸
i=1
∂
∂q
i
∂ ˙ q
i
∂ ˙ u
A
+µ
3
¸
i=1
∂
∂q
i+3
∂ ˙ q
i+3
∂ ˙ u
A
= µ
3
¸
i=1
∂
∂q
i
∂q
i
∂u
A
+µ
3
¸
i=1
∂
∂q
i+3
∂q
i+3
∂u
A
= µ
∂
ˆ
∂u
A
. (10.19)
Notice that we used (10.18) in the penultimate step, and also expressed as a func
tion of u
1
, . . . , u
6
and t:
=
q
1
, . . . , q
6
, t
=
ˆ
u
1
, . . . , u
6
, t
.
To avoid confusion where any may possibly arise, we ornament with a caret (hat)
when it is expressed as a function of u
1
, . . . , u
6
, and t.
We are now in a position to make some important conclusions. For a constraint
= 0, where =
q
1
, . . . , q
6
, t
=
ˆ
u
1
, . . . , u
6
, t
,
Lagrange’s prescription yields
F
c
·
∂ ¯ v
∂ ˙ u
A
+M
c
·
∂ω
∂ ˙ u
A
= µ
∂
ˆ
∂u
A
. (10.20)
Thus, in answer to our question, the presence of the constraint = 0 and the con
straint forces and moments it requires introduces a term µ
∂
ˆ
∂u
A
on the righthand side
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
322 Lagrange’s Equations of Motion for a Single Rigid Body
of Lagrange’s equations. This event is (reassuringly) similar to what arose previously
in the case of a single particle and a system of particles.
10.5.3 Consequences of Lagrange’s Prescription and Integrable Constraints
Let us now explore some of the consequences of (10.20). Suppose we can express
the integrable constraint as
u
6
− f (t) = 0.
Then,
F
c
·
∂ ¯ v
∂ ˙ u
A
+M
c
·
∂ω
∂ ˙ u
A
= µδ
6
A
.
As a result, the constraint force and moment will contribute to only the Lagrange’s
equation associated with u
6
. The familiar decoupling we found with a single particle
and a system of particles thus holds for the rigid body!
10.5.4 Potential Energy and Conservative Forces and Moments
Suppose that a potential energy is associated with the rigid body:
U = U
q
1
, . . . , q
6
=
ˆ
U
u
1
, . . . , u
6
.
The conservative forces and moments associated with this potential energy are
F
con
= −
3
¸
i=1
∂U
∂q
i
a
i
,
M
con
= −u
Q
= −
3
¸
i=1
∂U
∂q
i+3
g
i
.
Now suppose that we have chosen a new set of coordinates. Then, by setting =
−Uand µ = 1 in (10.19), we can show that
F
con
·
∂ ¯ v
∂ ˙ u
A
+M
con
·
∂ω
∂ ˙ u
A
= −
∂
ˆ
U
∂u
A
.
In summary, the conservative forces and moments appear in a familiar form on the
righthand side of Lagrange’s equations of motion.
10.5.5 Mechanical Power and Energy Conservation
If we have a single integrable constraint on the motion of the rigid body and a
potential energy U, then the combined mechanical power P of these forces and
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
10.5 Lagrange’s Equations of Motion: A Second Form 323
moments is
P = F
c
· ¯ v +M
c
· ω −
˙
U
= µ
−
∂
∂t
−
˙
U, (10.21)
where we have assumed that the constraint force and constraint moment are
prescribed by using Lagrange’s prescription. Consequently, if the integrable con
straint is time independent,
∂
∂t
= 0, and the only applied forces and moments are
conservative, then, with the help of (10.21), we can surmise that
˙
T = P = −
˙
U.
As a result, the total energy E = T +U is conserved when the integrable constraint
is time independent and all the applied forces and moments are conservative.
To arrive at (10.21), a straightforward set of manipulations is needed:
P = (F
con
+F
c
) · ¯ v +(M
con
+M
c
) · ω
= µ
3
¸
i=1
∂
∂q
i
a
i
· ¯ v +
3
¸
i=1
∂
∂q
i+3
g
i
· ω
−
3
¸
i=1
∂U
∂q
i
a
i
· ¯ v +
3
¸
i=1
∂U
∂q
i+3
g
i
· ω
= µ
3
¸
i=1
∂
∂q
i
˙ q
i
+
3
¸
i=1
∂
∂q
i+3
˙ q
i+3
−
3
¸
i=1
∂U
∂q
i
˙ q
i
+
3
¸
i=1
∂U
∂q
i+3
˙ q
i+3
= µ
6
¸
A=1
∂
∂q
A
˙ q
A
−
6
¸
B=1
∂U
∂q
B
˙ q
B
= µ
˙
−
∂
∂t
−
˙
U.
The ﬁnal result, P = µ
˙
−
∂
∂t
−
˙
U, is what was used in writing (10.21).
10.5.6 Summary
Suppose we have a rigid body for which all of the applied forces and moments are
conservative:
F = F
c
−
∂U
∂ ¯ x
, M = M
c
−u
Q
.
In addition, suppose that there are two constraints on the motion of the rigid body:
= u
6
− f (t) = 0, f · ¯ v +h · ω +e = 0.
The ﬁrst of these constraints is integrable, whereas the second is nonintegrable. We
assume that the constraint forces and moments associated with them are prescribed
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
324 Lagrange’s Equations of Motion for a Single Rigid Body
using Lagrange’s prescription:
F
c
= µ
1
3
¸
i=1
∂
∂q
i
a
i
+µ
2
f, M
c
= µ
1
3
¸
i=1
∂
∂q
i+3
g
i
+µ
2
h. (10.22)
Turning to Lagrange’s equations of motion (10.16), we ﬁnd that the equations
would read
d
dt
∂T
∂ ˙ u
A
−
∂T
∂u
A
= −
∂U
∂u
A
+Q
A
. (10.23)
As the only nonconservative forces acting on the body are due to the constraint
forces and constraint moments, Q
A
are simply
Q
A
= F
c
·
∂ ¯ v
∂ ˙ u
A
+M
c
·
∂ω
∂ ˙ u
A
(A= 1, . . . , 6) .
Using a Lagrangian L = T −U and using prescriptions (10.22) for F
c
and M
c
, we
ﬁnd that (10.23) can be expressed in the following form:
d
dt
∂L
∂ ˙ u
B
−
∂L
∂u
B
= µ
2
f ·
∂ ¯ v
∂ ˙ u
B
+h ·
∂ω
∂ ˙ u
B
,
d
dt
∂L
∂ ˙ u
6
−
∂L
∂u
6
= µ
1
+µ
2
f ·
∂ ¯ v
∂ ˙ u
6
+h ·
∂ω
∂ ˙ u
6
.
In these equations, B = 1, . . . , 5. Notice that the righthand sides of these forms of
Lagrange’s equations are similar to those for a system of particles with the added
complication of the moment terms.
10.6 Lagrange’s Equations of Motion: Approach II
We examine Approach II applied to Lagrange’s equations for a rigid body whose
motion is constrained. Speciﬁcally, we parallel the discussion of Subsection 10.5.6
and assume that the body is subject to one integrable constraint and one non
integrable constraint. Our discussion is easily generalized to multiple integrable
and nonintegrable constraints. The results we discuss were ﬁrst established by
Casey [28]. The resulting form of Lagrange’s equations is the one most commonly
used in engineering and physics.
First, we assume that the six coordinates u
1
. . . , u
6
are chosen such that the in
tegrable constraint = 0 can be simply written as
= u
6
− f (t).
Imposing this constraint, we can calculate the constrained kinetic
˜
T and potential
˜
U
energies of the system. The former will be a function of u
1
, . . . , u
5
, ˙ u
1
, . . . , ˙ u
5
, and t,
whereas the latter will be a function of u
1
, . . . , u
5
and t. The nonintegrable constraint
π
2
= 0 can be written in terms of the new coordinates and their velocities:
π
2
=
5
¸
B=1
p
B
˙ u
B
+e.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
10.7 Rolling Disks and Sliding Disks 325
Here p
1
, . . . , p
5
and e are functions of t and u
1
, . . . , u
5
. We will use Lagrange’s pre
scription to specify the associated constraint forces and constraint moments [see
(10.22)].
By following the same arguments used to establish Lagrange’s equations for
a single particle by use of Approach II, we ﬁnd that the equations governing the
motion of the rigid body are
5
¸
B=1
p
B
˙ u
B
+e = 0,
d
dt
∂
˜
T
∂ ˙ u
B
−
∂
˜
T
∂u
B
= F ·
∂ ¯ v
∂ ˙ u
B
+M·
∂ω
∂ ˙ u
B
= −
∂
˜
U
∂u
B
+F
anc
·
∂ ¯ v
∂ ˙ u
B
+M
anc
·
∂ω
∂ ˙ u
B
+µ
2
p
B
.
(10.24)
Here, B = 1, . . . , 5 and the constraint forces and moments associated with the inte
grable constraint are absent. In these equations, the forces and moments acting on
the body have been decomposed:
F = F
anc
+F
c
+F
con
,
M = M
anc
+M
c
+M
con
.
For instance, F
anc
are the applied nonconservative forces. The Magnus force F
M
=
mBω × ¯ v, which is used in studies of the ﬂight path of a baseball, is a good example
of such a force.
10.7 Rolling Disks and Sliding Disks
As examples of constrained rigid bodies, we consider a rigid circular disk of mass
m and radius R that either rolls without slipping on a rough horizontal plane (cf.
Figure 10.2) or slides on a smooth horizontal plane. It shall be assumed that the
inertia tensor J
0
has the representation
J
O
= λ (E
1
⊗E
1
+E
2
⊗E
2
) +λ
3
E
3
⊗E
3
. (10.25)
In other words, we are assuming that E
3
is the axis of symmetry of the disk in its
reference conﬁguration. The principal moments of inertia are λ
3
= 2λ and λ =
mR
2
4
,
but we do not use these substitutions in order to enable an easier tracking of the
algebraic manipulations.
Preliminaries
The location of the center of mass of the disk can be parameterized by use of a set
of Cartesian coordinates:
¯ x =
3
¸
i=1
x
i
E
i
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
326 Lagrange’s Equations of Motion for a Single Rigid Body
g
P
O
¯
X
ψ
φ
E
1
E
2
E
3
e
1
e
1
e
2
e
1
e
2
Figure 10.2. A circular disk moving with one point in contact with a horizontal plane.
The rotation tensor of the disk is described by a 3–1–3 set of Euler angles:
ω =
˙
ψE
3
+
˙
θe
1
+
˙
φe
3
.
For this set of Euler angles, we recall, from (6.33) in Subsection 6.8.2, that the Euler
basis vectors g
i
have the representations
¸
¸
¸
g
1
= E
3
g
2
= e
1
g
3
= e
3
=
¸
¸
¸
sin(φ) sin(θ) cos(φ) sin(θ) cos(θ)
cos(φ) −sin(φ) 0
0 0 1
¸
¸
¸
e
1
e
2
e
3
.
(10.26)
In addition, the Euler angles are subject to the restrictions ψ ∈ [0, 2π), θ ∈ (0, π),
and φ ∈ [0, 2π).
Physically, the angle θ represents the inclination angle of the disk. When this
angle is
π
2
, the disk is vertical. On the other hand, when θ = 0 or π, the disk is hor
izontal. In either of these situations, an entire surface of the disk is in contact with
the ground and the equations of motion that will be presented will not be valid.
Constraints
As discussed previously in Subsection 8.6.3, the motion of the rolling disk is subject
to three constraints because the velocity vector of the point of contact P is zero:
v
P
= ¯ v +ω ×π
P
= 0. (10.27)
Earlier, we showed how these equations lead to the following three constraints [see
(8.21)]:
π
1
= 0, π
2
= 0, π
3
= 0.
We also found that the third of these constraints was integrable:
= x
3
−Rsin(θ) = 0, (10.28)
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
10.7 Rolling Disks and Sliding Disks 327
whereas the remaining constraints were nonintegrable.
∗
It should be clear that we
express all three constraints on the rolling disk in the form f · ¯ v +h · ω = 0, and this
facilitates the use of Lagrange’s prescription for F
c
and M
c
.
Coordinates and Energies
Motivated by the presence of the integrable constraint x
3
= Rsin(θ), we now deﬁne
a set of coordinates:
u
1
= x
1
, u
2
= x
2
, u
3
= ψ,
u
4
= θ, u
5
= φ, u
6
= x
3
−Rsin(θ). (10.29)
Given values of u
1
, . . . , u
6
, we can uniquely invert relations (10.29) to determine x
i
and the Euler angles:
x
1
= u
1
, x
2
= u
2
, x
3
= u
6
+Rsin
u
4
,
ψ = u
3
, θ = u
4
, φ = u
5
. (10.30)
It should be clear that u
1
, . . . , u
5
are the generalized coordinates for both the rolling
disk and the sliding disk. For future reference, we compute that
¯ v = ˙ u
1
E
1
+ ˙ u
2
E
1
+
˙ u
6
+R ˙ u
4
cos
u
4
E
3
,
ω = ˙ u
3
E
3
+ ˙ u
4
e
1
+ ˙ u
5
e
3
. (10.31)
We also note that ω does not depend on u
6
or its time derivative.
We shall need the derivatives of ¯ v and ω with respect to the coordinates
u
1
, . . . , u
5
. With the help of (10.31), these can be obtained in a straightforward
manner:
∂ ¯ v
∂ ˙ u
1
= E
1
,
∂ ¯ v
∂ ˙ u
2
= E
2
,
∂ ¯ v
∂ ˙ u
3
= 0,
∂ ¯ v
∂ ˙ u
4
= Rcos (θ) E
3
,
∂ ¯ v
∂ ˙ u
5
= 0,
∂ ¯ v
∂ ˙ u
6
= E
3
, (10.32)
and
∂ω
∂ ˙ u
1
= 0,
∂ω
∂ ˙ u
2
= 0,
∂ω
∂ ˙ u
3
= E
3
,
∂ω
∂ ˙ u
4
= e
1
,
∂ω
∂ ˙ u
5
= e
3
,
∂ω
∂ ˙ u
6
= 0. (10.33)
These vectors will soon feature in Lagrange’s equations of motion.
Given the inertia tensor (10.25), representations (10.31), and the expressions
for ω · e
k
given in (10.5), an easy calculation shows that the (unconstrained) kinetic
∗
This was established by use of Frobenius’ theorem in one of the exercises at the end of Chapter 8.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
328 Lagrange’s Equations of Motion for a Single Rigid Body
energy of the disk has the representation
T =
m
2
˙ x
2
1
+ ˙ x
2
2
+
˙ u
6
+R
˙
θ cos (θ)
2
+
λ
2
˙
ψ
2
sin
2
(θ) +
˙
θ
2
+
λ
3
2
˙
φ +
˙
ψ cos(θ)
2
.
In addition, the potential energy of the disk is
U = mg
u
6
+Rsin(θ)
.
You should notice how the symmetry of the rigid body of interest simpliﬁes the
rotational kinetic energy.
Constraint Forces and Constraint Moments
The constraint forces and moments acting on the rolling disk can be prescribed by
use of Lagrange’s prescription. With the assistance of the expressions for π
i
, we ﬁnd
F
c
= µ
3
E
3
+µ
1
E
1
+µ
2
E
2
,
M
c
= µ
3
−Rcos(θ)g
2
+µ
1
Rcos(θ) cos(ψ)g
1
−Rsin(θ) sin(ψ)g
2
+Rcos(ψ)g
3
+µ
2
Rcos(θ) sin(ψ)g
1
+Rsin(θ) cos(ψ)g
2
+Rsin(ψ)g
3
.
The expressions for the dual Euler basis we use here can be inferred from (6.34) in
Section 6.8.
If we express the dual Euler basis vectors g
i
in terms of e
i
and expand the ex
pressions for M
c
, we will ﬁnd that M
c
= π
P
×F
c
. This implies, not surprisingly, that
F
c
and M
c
are equipollent to a force R = F
c
acting at the instantaneous point of
contact P. The E
3
component of this force is the normal force, whereas the remain
ing components constitute the friction force. Thus Lagrange’s prescription yields a
physically reasonable set of constraint forces and constraint moments.
Imposing the Integrable Constraint
If we impose the integrable constraint, then u
6
= 0 and the resulting constrained
potential and kinetic energies are
˜
U = mgRsin(θ),
˜
T =
m
2
˙ x
2
1
+ ˙ x
2
2
+
λ
2
˙
ψ
2
sin
2
(θ)
+
1
2
λ +mR
2
cos
2
(θ)
˙
θ
2
+
λ
3
2
˙
φ +
˙
ψ cos(θ)
2
.
It is important to notice that
˜
T =
˜
T(x
1
, x
2
, ˙ x
1
, ˙ x
2
, θ,
˙
φ,
˙
θ,
˙
ψ). In other words, u
6
has
been eliminated.
The Rolling Disk’s Equations of Motion
Lagrange’s equations of motion for the rolling disk can now be easily obtained. We
ﬁrst note that the resultant force F acting on the disk is composed of a constraint
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
10.7 Rolling Disks and Sliding Disks 329
force and a conservative force: F = F
c
−mgE
3
. The resultant moment M consists
entirely of the constraint moment: M = M
c
Using (10.6), and with the help of (10.32) and (10.33), we ﬁnd that
d
dt
∂T
∂ ˙ x
1
−
∂T
∂x
1
=
1
,
d
dt
∂T
∂ ˙ x
2
−
∂T
∂x
2
=
2
,
d
dt
∂T
∂
˙
ψ
−
∂T
∂ψ
=
3
,
d
dt
∂T
∂
˙
θ
−
∂T
∂θ
=
4
,
d
dt
∂T
∂
˙
φ
−
∂T
∂φ
=
5
,
d
dt
∂T
∂ ˙ u
6
−
∂T
∂u
6
=
6
. (10.34)
In these equations,
1
= F · E
1
+M· 0 = µ
1
,
2
= F · E
2
+M· 0 = µ
2
,
3
= F · 0 +M· E
3
= Rcos(θ) (µ
1
cos(ψ) +µ
2
sin(ψ)) ,
4
= F · Rcos(θ)E
3
+M· e
1
= −mgRcos(θ) +Rsin(θ) (µ
2
cos(ψ) −µ
1
sin(ψ)) ,
5
= F · 0 +M· e
3
= µ
1
Rcos(ψ) +µ
2
Rsin(ψ),
6
= F · E
3
+M· 0 = µ
3
−mg.
Evaluating the partial derivatives of T in (10.34) and imposing the integrable con
straint u
6
= 0, we ﬁnd, with some minor rearranging, that
m¨ x
1
= µ
1
,
m¨ x
2
= µ
2
,
d
dt
λ sin
2
(θ)
˙
ψ +λ
3
˙
φ +
˙
ψ cos(θ)
cos(θ)
=
3
,
λ
¨
θ +λ
3
˙
φ +
˙
ψ cos(θ)
˙
ψ sin(θ) −λ
˙
ψ
2
sin(θ) cos(θ) =
4
,
d
dt
λ
3
˙
φ +
˙
ψ cos(θ)
=
5
,
m
d
2
dt
2
(Rsin(θ)) = µ
3
−mg. (10.35)
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
330 Lagrange’s Equations of Motion for a Single Rigid Body
These equations are supplemented by the constraints,
x
3
−Rsin(θ) = 0, π
1
= 0, π
2
= 0, (10.36)
to form a closed system of equations for the nine unknowns x
i
, φ, θ, ψ, and µ
i
.
It is important to note that systems of equations (10.35) and (10.36) are not
readily integrated numerically. In particular, they cannot immediately be written in
the form ˙ y = f(y), which is required for most numerical integrators – such as those
based on a Runge–Kutta scheme.
Conservations for the Rolling Disk
The easiest method to see that the rolling disk’s total energy E is conserved is
to use the alternative form of the work–energy theorem. Two forces act on the
rolling disk: the gravitational force −mgE
3
and the constraint force F
c
that acts at P.
Consequently,
˙
T = F
c
· v
P
−mgE
3
· ¯ v.
However, v
P
= 0, and mgE
3
· ¯ v =
d
dt
(mgx
3
); consequently,
d
dt
(T +mgx
3
) = 0.
Because E = T +mgx
3
, this implies that the total energy of the disk is conserved.
The proof of energy conservation presented here applies to any rolling rigid body
under a gravitational force.
Surprisingly, there are two other conserved quantities associated with the
rolling disk. These were discovered by Appell [6], Chaplygin [36], and Korteweg
[113] in the late 19th century (see [19, 43, 107, 157]). Unfortunately, as with
Routh integral (9.35) for the tippe top, their physical interpretation is still an open
question.
The Sliding Disk’s Equations of Motion
The equations governing the motion of the sliding disk on a smooth horizontal plane
can be obtained from (10.35) and (10.36). Speciﬁcally, one sets µ
1
= µ
2
= 0 and
ignores the constraints π
1
= 0 and π
2
= 0.
It is instructive, however, to use Approach II. For the sliding disk there is no
nonintegrable constraint, and the constraint force and constraint moment are pre
scribed by use of Lagrange’s prescription, so (10.24) simpliﬁes to
d
dt
∂
˜
T
∂ ˙ u
A
−
∂
˜
T
∂u
A
= −
∂
˜
U
∂u
A
+F
c
·
∂ ¯ v
∂ ˙ u
A
+M
c
·
∂ω
∂ ˙ u
A
. (10.37)
Here, A= 1, . . . , 5, F
c
= µ
3
E
3
, and M
c
= −Re
1
×F
c
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
10.8 Lagrange and Poisson Tops 331
Using (10.29) in conjunction with (10.37), we ﬁnd that the resulting governing
equations are
m¨ x
1
= 0,
m¨ x
2
= 0,
d
dt
λ sin
2
(θ)
˙
ψ +λ
3
˙
φ +
˙
ψ cos(θ)
cos(θ)
= 0,
λ
¨
θ +λ
3
˙
φ
˙
ψ sin(θ) +(λ
3
−λ)
˙
ψ
2
sin(2θ)
2
= −mgRcos(θ),
d
dt
λ
3
˙
φ +
˙
ψ cos(θ)
= 0. (10.38)
Notice that we could also use these equations to arrive at the equations of motion
for the rolling disk. First, we need to supplement the constraints and, second, we
need to append the constraint forces and moments associated with the integrable
constraints to the righthand side of (10.38).
The equations governing the motion of the sliding disk are clearly far simpler
than those for the rolling disk. Indeed, these equations can be written in the
form ˙ y = f(y), where y is a column vector with 10 rows. These equations can
then be integrated numerically by use of a standard numerical integrator. Any
numerical simulations of these equations should ensure that the linear momentum
G, angular momenta H· e
3
and H· E
3
, and total energy E of the sliding disk are
conserved.
Conﬁguration Manifold
The conﬁguration manifold Mfor the rolling disk and the sliding disk are identical:
M = E
2
⊕SO(3).
Here, x
1
and x
2
are coordinates for E
2
and the Euler angles are coordinates for
SO(3). That is, both disks have ﬁve degreesoffreedom and ﬁve generalized cordi
nates. The kinematical lineelement ds is
ds =
¸
2
˜
T
m
dt,
where
˜
T is the constrained kinetic energy deﬁned earlier.
You should notice that the nonintegrable constraints on the rolling disk do not
affect the conﬁguration manifold. This is identical to a situation we encountered
earlier with the single particle subject to nonintegrable constraints.
10.8 Lagrange and Poisson Tops
There are two classical problems in rigid body dynamics involving spinning tops. In
the ﬁrst, known as the Lagrange top, the axisymmetric rigid body is free to rotate
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
332 Lagrange’s Equations of Motion for a Single Rigid Body
g
O
¯
X
E
1
E
2
E
3
e
1
e
2
e
3
Figure 10.3. The Lagrange top: an axisymmetric rigid body that is free to rotate about a ﬁxed
point O. The image on the lefthand side is a portrait of JosephLouis Lagrange.
about a ﬁxed point Ounder the action of a gravitational force (see Figure 10.3). This
problem was discussed by Lagrange as an illustration of his equations of motion in
1789.
∗
A few years later, Poisson [173] considered the same problem except now the
apex of the top was free to move on a horizontal surface (see Figure 10.4). Poisson’s
interest in this problem arose in part because a spinning top can in principle be used
to determine the vertical on a ship at sea, and hence was considered to be potentially
useful in navigation.
In this section, we do not pursue the complete exposition of the equations of
motions for the two tops; rather, we focus on choices of coordinates for them and
brieﬂy discuss how their equations of motion can be found.
The Lagrange Top: Coordinates, Constraints, and Energies
For the Lagrange top, we assume that the position vector of its center of mass rela
tive to the point Ois
¯ x = x
0
+L
1
e
1
+L
2
e
2
+L
3
e
3
.
Here, L
k
are constants. Further, we assume that the rotation tensor of the top is
parameterized by a set of 3–1–3 Euler angles.
†
That is, the angular velocity vector
of the top is ω =
˙
ψE
3
+
˙
θe
1
+
˙
φe
3
. You should notice that the singularities of
these Euler angles occur when E
3
= ±e
3
, that is, when θ = 0 or π and the top is
vertical.
∗
See Section IX.34 of the Second Part of Lagrange’s M´ ecanique Analytique [121].
†
This set of Euler angles has been used extensively in the present chapter and is discussed in Subsec
tion 6.8.2.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
10.8 Lagrange and Poisson Tops 333
g
O
¯
X
P
E
1
E
2
E
3
e
1
e
2
e
3
Figure 10.4. The Poisson top: an axisymmetric rigid body that moves with one point P in
contact with a horizontal surface.
Motivated by the presence of three integrable constraints x
O
= 0, we now deﬁne
a set of coordinates:
u
1
= ψ, u
2
= θ, u
3
= φ, u
4
= x
1
−
3
¸
k=1
L
k
e
k
· E
1
,
u
5
= x
2
−
3
¸
k=1
L
k
e
k
· E
2
, u
6
= x
3
−
3
¸
k=1
L
k
e
k
· E
3
. (10.39)
In these equations, e
k
are functions of the Euler angles, but the lengthy expressions
are not recorded here. Given values of u
1
, . . . , u
6
, we can uniquely invert relations
(10.39) to determine x
i
and the Euler angles:
x
1
= u
4
+
3
¸
k=1
L
k
e
k
· E
1
, x
2
= u
5
+
3
¸
k=1
L
k
e
k
· E
2
,
x
3
= u
6
+
3
¸
k=1
L
k
e
k
· E
3
, ψ = u
1
, θ = u
2
, φ = u
3
. (10.40)
Using (10.40), we can easily compute that
¯ v = ˙ u
4
E
1
+ ˙ u
5
E
2
+ ˙ u
6
E
3
+ω ×
3
¸
k=1
L
k
e
k
.
ω = ˙ u
1
E
3
+ ˙ u
2
e
1
+ ˙ u
3
e
3
. (10.41)
You should notice that ¯ v will depend on the Euler angles and their time derivatives.
This is entirely due to our choice of coordinates. In place of ¯ x and the three Euler
angles, we are using the Euler angles and the three components of x
O
as our
coordinates.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
334 Lagrange’s Equations of Motion for a Single Rigid Body
For future reference, we shall need several vectors. To calculate these vectors it
is convenient to recall that
∂ω
∂
˙
ψ
= E
3
,
∂ω
∂
˙
θ
= e
1
, and
∂ω
∂
˙
φ
= e
3
. First, we calculate
∂ ¯ v
∂
˙
ψ
= E
3
×
3
¸
k=1
L
k
e
k
,
∂ ¯ v
∂
˙
θ
= e
1
×
3
¸
k=1
L
k
e
k
,
∂ ¯ v
∂
˙
φ
= e
3
×
3
¸
k=1
L
k
e
k
,
∂ ¯ v
∂ ˙ u
4
= E
1
,
∂ ¯ v
∂ ˙ u
5
= E
2
,
∂ ¯ v
∂ ˙ u
6
= E
3
, (10.42)
and
∂ω
∂ ˙ u
1
= E
3
,
∂ω
∂ ˙ u
2
= e
1
,
∂ω
∂ ˙ u
3
= e
3
,
∂ω
∂ ˙ u
4
= 0,
∂ω
∂ ˙ u
5
= 0,
∂ω
∂ ˙ u
6
= 0. (10.43)
These vectors can be used to compute the righthand sides of Lagrange’s equations
of motion.
We assume that the top has an axis of symmetry, and hence its inertia tensor J
0
has the representation
J
0
= λ
a
E
3
⊗E
3
+λ
t
(I −E
3
⊗E
3
) .
With the help of (10.41), we ﬁnd that the kinetic energy of the top has the represen
tation
T =
m
2
˙ u
4
2
+
˙ u
5
2
+
˙ u
6
2
+m
˙ u
4
E
1
+ ˙ u
5
E
2
+ ˙ u
6
E
3
·
ω ×
3
¸
k=1
L
k
e
k
+
λ
t
+mL
2
2
+mL
2
3
2
ω
2
1
+
λ
t
+mL
2
1
+mL
2
3
2
ω
2
2
+
λ
a
+mL
2
1
+mL
2
2
2
ω
2
3
.
For convenience, we have not substituted for ω
i
= ω · e
i
in terms of the Euler an
gles and their derivatives. The needed expressions for ω
i
are given in (10.5) in Sec
tion 10.2. The potential energy of the top is
U = mg
u
3
+
3
¸
k=1
L
k
e
k
· E
3
.
This energy is due entirely to the gravitational force.
There are three constraints on the motion of Lagrange’s top: u
4
= 0, u
5
= 0, and
u
6
= 0. That is, x
O
= 0. These constraints are integrable and imply that the general
ized coordinates for the Lagrange top are the three Euler angles. Thus the conﬁgura
tion manifold M for the top is SO(3). A straightforward calculation with the three
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
10.8 Lagrange and Poisson Tops 335
constraints v
O
· E
k
= 0, where O is the ﬁxed point of the top, shows that the con
straint force F
c
=
¸
3
k=1
µ
k
E
k
and constraint moment M
c
= −
¸
3
k=1
L
k
e
k
×F
c
. As
expected, this system of forces and moments is equipollent to a force F
c
acting at O.
The constrained kinetic and potential energies are
˜
T =
λ
t
+mL
2
2
+mL
2
3
2
ω
2
1
+
λ
t
+mL
2
1
+mL
2
3
2
ω
2
2
+
λ
a
+mL
2
1
+mL
2
2
2
ω
2
3
,
˜
U = mg
3
¸
k=1
L
k
e
k
· E
3
.
It is not too difﬁcult to see that 2
˜
T = H
O
· ω and the angular momentum H
O
can
be easily found by use of the parallel axis theorem. We have kept L
k
distinct
and nonzero to help elucidate some of our kinematical developments. If we now
impose the standard assumptions that L
1
= 0 = L
2
, then our results for
˜
T and
˜
U = mgLcos(θ) are in accord with those found in many dynamics textbooks.
The Poisson Top: Coordinates, Constraints, and Energies
For the Poisson top, it is convenient to choose a different coordinate system:
u
1
= ψ, u
2
= θ, u
3
= φ,
u
4
= x
1
, u
5
= x
2
, u
6
= x
3
−
3
¸
k=1
L
k
e
k
· E
3
. (10.44)
It is left to the reader to compare (10.44) with (10.39) and to compute
∂ ¯ v
∂ ˙ u
A
and
∂ω
∂ ˙ u
A
.
In contrast to the Lagrange top, the Poisson top is subject to a single integrable
constraint u
6
= 0. A straightforward calculation with the constraint v
P
· E
3
= 0,
where P is the point of contact of the top with the horizontal surface, shows that the
constraint force F
c
= µ
1
E
3
and constraint moment M
c
= −
¸
3
k=1
L
k
e
k
×µ
1
E
3
.
As expected, this force and this moment are equipollent to a force F
c
acting at P.
The development of expressions for T and Ufollow in a similar manner to those
discussed previously for the Lagrange top, and so their constrained expressions are
merely summarized:
˜
T =
λ
t
+mL
2
2
+mL
2
3
2
ω
2
1
+
λ
t
+mL
2
1
+mL
2
3
2
ω
2
2
+
λ
a
+mL
2
1
+mL
2
2
2
ω
2
3
+
m
2
˙ x
2
1
+ ˙ x
2
2
,
˜
U = U = mg
3
¸
k=1
L
k
e
k
· E
3
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
336 Exercise 10.1
You should observe from these expressions that the generalized coordinates for the
Poisson top are ψ, θ, φ, x
1
, and x
2
. Thus this system has a ﬁvedimensional conﬁg
uration manifold M. Indeed, the conﬁguration manifold of this top is identical to
that for the sliding disk.
Comments on the Equations of Motion
We can ﬁnd the equations of motion for the Lagrange and Poisson tops by using
Approach II and without calculating
∂ ¯ v
∂ ˙ u
A
and
∂ω
∂ ˙ u
A
. There are three reasons for this.
First, the constraint forces and moments are compatible with Lagrange’s prescrip
tion. Second, the coordinates have been chosen so that the integrable constraints
are each described in terms of one coordinate, and ﬁnally the applied forces acting
on these tops are conservative. Thus the form of Lagrange’s equations of motion
featuring the Lagrangian
˜
L =
˜
T −
˜
U can be used. In the interest of brevity, explicit
equations of motion are not given here: They will be similar to the earlier results for
the disks.
For both tops, it is easy to see that the total energy E is conserved. The Pois
son top also features linear momenta conservations: G· E
1
and G· E
2
. If we now
impose the conditions, L
1
= L
2
= 0 for both tops, then we ﬁnd that the Lagrange
top features two angular momenta conservations: H
O
· E
3
and H
O
· e
3
, whereas the
Poisson top features the angular momenta conservations H· E
3
and H· e
3
.
10.9 Closing Comments
Lagrange’s equations of motion have been illustrated for a variety of systems featur
ing a single rigid body. When there are no constraints on the rigid body or when the
constraints are “ideal,” this formulation of the equations of motion provides a set of
differential equations that can be integrated to determine the motion. However, for
nonintegrably constrained rigid bodies or for rigid bodies when dynamic friction is
present, Lagrange’s equations of motion are not very attractive, and it is often best
to simply examine the balances of linear and angular momentum. As demonstrated
in the examples discussed in Chapter 9, with some insight and patience, F = m
˙
¯ v and
M =
˙
H can lead to a tractable set of equations from which the motion of the rigid
body can be inferred.
EXERCISES
10.1. Consider the mechanical system shown in Figure 10.5. It consists of a rigid
body of mass m that is free to rotate about a ﬁxed point O. A vertical gravitational
force mgE
1
acts on the body. The inertia tensor of the body relative to its center of
mass C is
J
0
= λ
1
E
1
⊗E
1
+(λ −mL
2
0
)(E
2
⊗E
2
+E
3
⊗E
3
).
The position vector of the center of mass
¯
X of the body relative to Ois L
0
e
1
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
Exercise 10.1 337
g
O
¯
X
ψ
θ
E
1
E
2
E
3
e
1
Figure 10.5. A whirling axisymmetric rigid body.
(a) To parameterize the rotation tensor of the body, we use a set of 1–3–1 Euler
angles: g
1
= E
1
, g
2
= e
3
, and g
3
= e
1
. Show that
ω =
˙
ψE
1
+
˙
θe
3
+
˙
φe
1
=
˙
φ +
˙
ψ cos(θ)
e
1
+
˙
θ sin(φ) −
˙
ψ sin(θ) cos(φ)
e
2
+
˙
θ cos(φ) +
˙
ψ sin(θ) sin(φ)
e
3
.
(b) Because the point Ois ﬁxed, we deﬁne the coordinates u
4
, . . . , u
6
to be the
coordinates of O:
¯ x = u
4
E
1
+u
5
E
2
+u
6
E
3
+L
0
e
1
,
and we choose u
1
= ψ, u
2
= θ, and u
3
= φ. With this choice of the coordi
nates, show that
∂ ¯ v
∂ ˙ u
1
= E
1
×L
0
e
1
,
∂ ¯ v
∂ ˙ u
2
= e
3
×L
0
e
1
,
∂ ¯ v
∂ ˙ u
3
= 0,
∂ ¯ v
∂ ˙ u
4
= E
1
,
∂ ¯ v
∂ ˙ u
5
= E
2
,
∂ ¯ v
∂ ˙ u
6
= E
3
, (10.45)
and
∂ω
∂ ˙ u
1
= E
1
,
∂ω
∂ ˙ u
2
= e
3
,
∂ω
∂ ˙ u
3
= e
1
,
∂ω
∂ ˙ u
4
= 0,
∂ω
∂ ˙ u
5
= 0,
∂ω
∂ ˙ u
6
= 0. (10.46)
These results won’t be needed in the remainder of this problem.
(c) Derive expressions for the unconstrained potential Uand kinetic Tenergies
of the rigid body. These expressions should be functions of the coordinates
u
1
, . . . , u
6
and, where appropriate, their time derivatives.
(d) In what follows, the motion of the rigid body is subject to four constraints:
˙
φ = ω · g
3
= 0, ¯ v −ω ×(L
0
e
1
) = 0.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
338 Exercises 10.1–10.2
Assuming the joint at O is frictionless, give prescriptions for the constraint
force and moment acting on the rigid body.
(e) What are the generalized coordinates for this system, and why is the conﬁg
uration manifold a sphere?
(f) Prove that the total energy of the rigid body is conserved.
(g) Show that the (constrained) kinetic and potential energies of the body are
˜
T =
1
2
λ
1
˙
ψ
2
cos
2
(θ) +λ(
˙
θ
2
+
˙
ψ
2
sin
2
(θ))
,
˜
U = mgL
0
cos(θ). (10.47)
In addition, using Approach II, write Lagrange’s equations of motion for
the rigid body:
d
dt
∂
˜
L
∂
˙
θ
−
∂
˜
L
∂θ
= 0,
d
dt
∂
˜
L
∂
˙
ψ
−
∂
˜
L
∂ψ
= 0, (10.48)
where
˜
L is the Lagrangian.
(h) Argue that the solutions ψ(t) and θ(t) to (10.48) determine the motion of
the rigid body. You should also discuss why it was not necessary to calculate
Lagrange’s equations of motion for all six coordinates.
10.2. Consider the simple model for an automobile shown in Figure 10.6. It consists
of a single rigid body of mass m. The moment of inertia tensor of the rigid body
is J =
¸
3
i=1
λ
i
e
i
⊗e
i
. Here, the inertia tensor J and mass m include the masses and
inertias of the wheels, suspension, engine, and occupants. Interest is restricted to the
case in which the front wheels are sliding while the rear wheels are rolling. In other
words, the front wheels’ brakes are locked. To model the rolling of the rear wheels
in this simple model, it is assumed that
v
Q
· e
2
= 0, (10.49)
where π
Q
= −L
1
e
1
−L
2
e
2
is the position vector of Q relative to the center of mass
¯
Xof the rigid body. Constraint (10.49) is often known as Chaplygin’s constraint (see
[151, 165, 186]).
Q
Rigid body of mass m
O
¯
X
φ
E
1
E
1
E
2
e
1
e
1
e
2
Figure 10.6. A rigid body model for an automobile moving on a horizontal plane.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
Exercises 10.2–10.3 339
(a) Assume that the rigid body is performing a ﬁxedaxis rotation through an
angle φ about E
3
, that the motion of its center of mass is planar, and that
(10.49) holds. Using parameterizations of ¯ x and Q of your choice, establish
expressions for the four constraints on the motion of the rigid body.
(b) Verify that one of the constraints in (a) is nonintegrable.
(c) Using Lagrange’s prescription, what are the constraint forces F
c
and mo
ments M
c
acting on the rigid body?
(d) Show that the motion of the rigid body is governed by the equations
m¨ x
1
= −µ
4
sin(φ), m¨ x
2
= µ
4
cos(φ), λ
3
¨
φ = −µ
4
L
1
,
˙ x
1
sin(φ) − ˙ x
2
cos(φ) = −L
1
˙
φ, (10.50)
where x
i
= ¯ x · E
i
.
(e) Show that (10.50) allow the center of mass of the rigid body to move in a
straight line without the body rotating, and prove that the total energy of
the body is conserved.
10.3. As shown in Figure 10.7, a circular rod of mass m, length L, and radius R
slides on a smooth horizontal plane. We assume that the inertia tensor J
0
has the
representation
J
O
= λ (E
1
⊗E
1
+E
3
⊗E
3
) +λ
2
E
2
⊗E
2
.
The sole external applied force acting on the rod is gravitational: −mgE
3
.
g
O
¯
X
Horizontal plane
Circular cylinder
E
1
E
2
E
3
e
1
e
2
e
3
Figure 10.7. A circular rod moving on a horizontal plane.
(a) Assuming that the rotation tensor of the rod is described by a 3–2–3 set of
Euler angles, which orientations of the rod coincide with the singularities of
this set of Euler angles?
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
340 Exercises 10.3–10.4
(b) For this problem, we use Cartesian coordinates to describe the position of
the center of mass. That is, u
1
= x
1
, u
2
= x
2
, u
3
= x
3
, u
4
= ψ, u
5
= θ, and
u
6
= φ. For this set of coordinates, calculate
∂ ¯ v
∂ ˙ u
A
and
∂ω
∂ ˙ u
A
. To write the equa
tions of motion of this rigid body, under which circumstances can you avoid
using these 12 vectors?
(c) Show that the unconstrained kinetic energy of the rod is
T =
m
2
˙ x
2
1
+ ˙ x
2
2
+ ˙ x
2
3
+
λ
2
˙
θ sin(φ) −
˙
ψ sin(θ) cos(φ)
2
+
λ
2
2
˙
θ cos(φ) +
˙
ψ sin(θ) sin(φ)
2
+
λ
2
˙
φ +
˙
ψ cos(θ)
2
,
where ¯ x = x
1
E
1
+x
2
E
2
+x
3
E
3
.
(d) Show that the two constraints on the motion of the rigid body can be written
in the form
¯ v · E
3
= 0, ω · g
3
= 0.
(e) Using Lagrange’s prescription, what are the constraint force F
c
and con
straint moment M
c
that enforce the two constraints? With the assistance of
a freebody diagram of the sliding rod, give physical interpretations of F
c
and M
c
.
(f) Show that the six Lagrange’s equations of motion for the rod yield the fol
lowing differential equations for the motion of the rod:
m¨ x
1
= 0, m¨ x
2
= 0, λ
2
¨
θ = 0, λ
¨
ψ = 0. (10.51)
(g) Show that the six Lagrange’s equations of motion for the rod also yield
solutions for the constraint force and constraint moment:
F
c
= mgE
3
, M
c
= −λ
2
˙
ψ
˙
θ sin(θ)g
3
.
By expressing g
3
in terms of e
1
, you should also be able to establish a simple
expression for M
c
.
(h) Give physical interpretations of the solutions to (10.51) and show that they
allow the cylinder to have motions where ω is not constant.
10.4. As shown in Figure 9.8, a sphere of mass m and radius R is free to move on a
rough inclined plane. Here, we reexamine this problem, which was discussed earlier
in Section 9.8, by using Lagrange’s equations of motion. The instantaneous point of
contact of the sphere with the plane is denoted by P. You will need to recall that
v
P
= ¯ v +ω ×π
P
, (10.52)
where ¯ v is the velocity vector of the center of mass
¯
X of the sphere, and π
P
is the
position vector of P relative to
¯
X.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
Exercises 10.4–10.5 341
(a) Using a set of 3–1–3 Euler angles, show that the slip velocities of the sphere
are
v
s1
= ˙ x
1
−R
˙
θ sin(ψ) +R
˙
φsin(θ) cos(ψ),
v
s2
= ˙ x
2
+R
˙
θ cos(ψ) +R
˙
φsin(θ) sin(ψ),
where v
P
= v
s1
E
1
+v
s2
E
2
and ¯ v =
¸
3
k=1
˙ x
k
E
k
.
(b) What is the constraint on a sliding sphere? Give prescriptions for the con
straint force F
c
and moment M
c
that enforce this constraint.
(c) Show that at least one of the three constraints on a rolling sphere is
nonintegrable. Give prescriptions for the constraint forces F
c
and moments
M
c
that enforce these constraints.
(d) If the unconstrained kinetic energy of the sphere is
T =
mR
2
5
˙
ψ
2
+
˙
θ
2
+
˙
φ
2
+2
˙
ψ
˙
φcos(θ)
+
m
2
˙ x
2
1
+ ˙ x
2
2
+ ˙ x
2
3
,
then what are Lagrange’s equations of motion for the sliding sphere?
(e) What alterations need to be made to the Lagrange’s equations of motion
for (d) so that they now apply to the rolling sphere?
(f) Starting from (10.52) and using balances of linear and angular momentum,
show that
m˙ v
s1
=
1 +
5
2
F
f
· E
1
+mg sin(β),
m˙ v
s2
=
1 +
5
2
F
f
· E
2
,
where F
f
is the friction force acting on the sphere. If the sphere is rolling,
then what is F
f
?
10.5. Consider a rigid body of mass m that is free to rotate about a ﬁxed point O.
The position vector of the center of mass
¯
Xof the rigid body relative to Ois ¯ x −x
O
=
L
0
e
3
, where L
0
is a constant. The inertia tensor of the body relative to its center of
mass is J =
¸
3
k=1
λ
k
e
k
⊗e
k
. A vertical gravitational force −mgE
3
acts on the rigid
body.
(a) Using a set of 3–1–3 Euler angles, establish an expression for the rotational
kinetic energy T
rot
of the rigid body.
(b) What are the three constraints on the motion of the rigid body?
(c) Using Lagrange’s prescription, give prescriptions for the constraint force F
c
and moment M
c
acting on the rigid body.
(d) In terms of the Euler angles and a set of Cartesian coordinates x
k
= ¯ x · E
k
,
prescribe a set of six coordinates u
1
, . . . , u
6
such that the three integrable
constraints on the motion of the rigid body can be expressed as
i
= 0 (i = 1, 2, 3) ,
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
342 Exercises 10.5–10.6
where
1
= u
4
,
2
= u
5
,
3
= u
6
.
Notice that the generalized coordinates for this rigid body are the Euler
angles.
(e) Calculate the following 12 vectors:
∂ ¯ v
∂ ˙ u
A
,
∂ω
∂ ˙ u
A
(A= 1, . . . , 6) .
What is the potential energy U = U
u
1
, . . . , u
6
of the rigid body?
(f) What are Lagrange’s equations of motion for the generalized coordinates?
In your solution, show that the contributions to the righthand sides of these
equations that are due to F
c
and M
c
sum to zero.
(g) With the assistance of the balance laws, show that
F
c
= mgE
3
+mL
0
¨ e
3
.
What is the corresponding result for M
c
?
(h) Prove that the total energy E of the rigid body is conserved.
10.6. In this exercise, we reconsider the satellite problem discussed earlier in Sec
tion 10.4. Here, we employ a coordinate system that is popular in the satellite dy
namics community (see, for example, [164, 187, 203]).
Referring to Figure 10.1, to parameterize ¯ x, we use a spherical polar coordinate
system, R, θ, and β, where β is the latitude and θ is the longitude. For this coordinate
system, we deﬁne the following unit vectors:
e
r
= cos(θ)E
1
+sin(θ)E
2
, e
R
= cos(β)e
r
+sin(β)E
3
,
e
β
= cos(β)E
3
−sin(β)e
r
, e
θ
= cos(θ)E
2
−sin(θ)E
1
.
The rotation tensor Q of the satellite is a transformation from the ﬁxed basis {E
i
}
to the corotational basis {e
i
}. To parameterize Q, it is standard in satellite dynamics
studies to exploit the rotation θ about E
3
inherent in the deﬁnition of the spherical
polar coordinates. To do this, we use the following decomposition:
Q = Q
2
Q
1
,
where the rotation tensor Q
1
is
Q
1
= cos(θ)(I −E
3
⊗E
3
) −sin(θ)E
3
+E
3
⊗E
3
.
The rotation tensor Q
2
will be parameterized by a set of 1–2–3 Euler angles. The ﬁrst
of the Euler angles corresponds to a counterclockwise rotation through an angle ν
1
about the vector e
r
. Similarly, the second rotation corresponds to a counterclockwise
rotation through an angle ν
2
about cos
ν
1
e
θ
+sin
ν
1
E
3
and the last rotation is
about e
3
through a counterclockwise angle ν
3
.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
Exercises 10.6–10.7 343
(a) Starting from the representation ¯ x = Re
R
, show that the velocity vector ¯ v of
the center of mass of the satellite has the representation
¯ v =
˙
Re
R
+Rcos(β)
˙
θe
θ
+R
˙
βe
β
.
(b) Starting from the representation Q =
¸
3
i=1
e
i
⊗E
i
, show that e
1
= Q
2
e
r
,
e
2
= Q
2
e
θ
, and e
3
= Q
2
E
3
.
(c) With the help of three relative angular velocity vectors, show that the angu
lar velocity vector ω of the satellite has the representation
ω =
˙
θE
3
+ ˙ ν
1
e
r
+ ˙ ν
2
cos
ν
1
e
θ
+sin
ν
1
E
3
+ ˙ ν
3
e
3
. (10.53)
(d) Show that the angular velocity vector of the satellite has the representation
ω = ω
1
e
1
+ω
2
e
2
+ω
3
e
3
,
where
ω
1
=
˙
θ
sin(ν
1
) sin(ν
3
) −cos(ν
1
) sin(ν
2
) cos(ν
3
)
+ ˙ ν
2
sin(ν
3
) + ˙ ν
1
cos(ν
2
) cos(ν
3
),
ω
2
=
˙
θ
sin(ν
1
) cos(ν
3
) +cos(ν
1
) sin(ν
2
) sin(ν
3
)
+ ˙ ν
2
cos(ν
3
) − ˙ ν
1
cos(ν
2
) sin(ν
3
),
ω
3
=
˙
θ cos(ν
1
) cos(ν
2
) + ˙ ν
1
sin(ν
2
) + ˙ ν
3
.
(e) Suppose the coordinates chosen for the satellite are u
1
= R, u
2
= θ, u
3
= β,
u
4
= ν
1
, u
5
= ν
2
, and u
6
= ν
3
. With this choice, give expressions for the 12
vectors
∂ ¯ v
∂ ˙ u
A
,
∂ω
∂ ˙ u
A
(A= 1, . . . , 6) .
(f) The potential energy associated with the gravitational force and moment
can be found in Equation (10.11). With the choice of u
1
, . . . , u
6
, it can be
shown that
∂U
∂θ
= 0. Using this result, show that
∂T
∂
˙
θ
is conserved. Prove that
this conservation corresponds to conservation of H
O
· E
3
.
(g) Using Lagrange’s equations, establish the equations governing the motion
of the satellite.
10.7. Consider a body that is free to move about one of its material points O, which
is ﬁxed (cf. Figure 9.7). This system was discussed previously in Exercises 9.15 and
10.5.
(a) Calculate the constrained Lagrangian
˜
L and the equations of motion [see
(10.8)]:
d
dt
∂
˜
L
∂ ˙ γ
i
−
∂
˜
L
∂γ
i
= 0, (10.54)
where γ
1
= ψ, γ
2
= θ, and γ
3
= φ are the 3–1–3 set of Euler angles used to
parameterize Q.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 12, 2008 5:0
344 Exercises 10.7–10.8
(b) Show that equations of motion (10.54) are equivalent to
˙
H
O
· g
i
= M
O
· g
i
,
where
¸
g
1
, g
2
, g
3
¸
are the Euler basis vectors.
(c) Compare equations of motion (10.54) with (9.36), which were discussed in
Exercise 9.14.
10.8. Recall the deﬁnition of the force
A
given by (10.4):
A
= F ·
∂ ¯ v
∂ ˙ u
A
+M·
∂ω
∂ ˙ u
A
.
Now suppose we have a force P that acts at a point X
P
on a rigid body. If the position
vector of X
P
is x
P
, then show that
∗
P ·
∂v
P
∂ ˙ u
A
= P ·
∂ ¯ v
∂ ˙ u
A
+((x
P
− ¯ x) ×P) ·
∂ω
∂ ˙ u
A
.
How can this identity be used to simplify the calculation of
A
?
∗
This identity is used extensively in many treatments of analytical mechanics (see, for example, Kane
and Levinson [105] and Kane et al. [106]).
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
PART FOUR
SYSTEMS OF RIGID BODIES
345
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
346
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
11 Introduction to Multibody Systems
11.1 Introduction
In this chapter, we examine systems of rigid bodies. Our goal is simply to discover
how our previous developments can be used to obtain the equations of motion for
these systems. As you might imagine, the equations of motion can be very complex,
and judicious component selections from the balance laws are often needed to ex
tract the equations of motion. This is illustrated with the example of a rate
gyroscope.
The presentations here are limited in scope and we do not have the opportunity
to discuss many interesting systems featuring several rigid bodies such as the dual
spin spacecraft, bicycles, gyrocompasses, and the Dynabee in detail. As discussed
in [86], a dualspin satellite has the ability to reorient itself in an environment
where the resultant moment on the satellite is negligible. This ability has been
used in communications satellites and was employed in the Galileo spacecraft. This
spacecraft was launched in 1989 and some 6 years later began its orbits of the planet
Jupiter. These orbits were designed so that the spacecraft could capture data on
some of the largest moons of Jupiter; Galileo’s mission was a remarkable success.
∗
The Dynabee (or Rollerball) was invented in the early 1970s by Archie Mishler
[141] and features spinning a rotor to speeds in excess of 5000 rpm by carefully
rotating an outer casing (housing). This novel gyroscopic device is discussed in an
exercise at the end of this chapter.
11.2 Balance Laws and Lagrange’s Equations of Motion
We are interested in examining the dynamics of a system of N rigid bodies,
B
1
, . . . , B
N
. For the body B
K
, we denote its mass by m
K
, its center of mass by
¯
X
K
,
the position of its center of mass by ¯ x
K
, its inertia tensor relative to
¯
X
K
by J
K
, and
its angular velocity by ω
K
. Using these quantities, we can deﬁne the kinetic energy
T
K
, angular momentum H
K
, and linear momentum G
K
in the usual manner:
G
K
= m
K
˙
¯ x
K
, H
K
= J
K
ω
K
, T
K
=
1
2
G
K
·
˙
¯ x
K
+
1
2
H
K
· ω
K
.
∗
A history of the Galileo spacecraft’s mission was recently written by Meltzer [140].
347
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
348 Introduction to Multibody Systems
Setting K = 1, . . . , N we can obtain expressions for the kinematic quantities associ
ated with each of the N rigid bodies. Constraints on, and potential energies for, the
system of rigid bodies can be inferred from our developments in Chapter 8 and we
shall adopt them here.
For the system of rigid bodies, we can deﬁne the center of mass
¯
X. The position
vector of this point is deﬁned by
¯ x =
1
¸
N
K=1
m
K
N
¸
K=1
m
K
¯ x
K
.
The linear momentum G of the system of rigid bodies is the sum of the individual
linear momenta, and G can be related to the momentum of the center of mass:
G =
N
¸
K=1
m
K
˙
¯ x =
N
¸
K=1
m
K
˙
¯ x
K
.
In many problems featuring systems of rigid bodies, it is necessary to calculate the
angular momentum of the system relative to a point, say A: H
A
. This point is often
the center of mass of the system or a ﬁxed point. To compute H
A
, we use a standard
identity [(7.15)] applied to each individual rigid body:
H
A
=
N
¸
K=1
(H
K
+(¯ x
K
−x
A
) ×m
K
˙
¯ x
K
) .
Finally, the kinetic energy of the system of N rigid bodies is simply the sum of the ki
netic energies of the individual bodies: T = T
1
+· · · +T
N
. The resulting expression
for T can be tremendously complex, and often symbolic manipulation packages,
such as Mathematica or Maple, are used.
We ﬁnd the governing equations for the system of N rigid bodies by using the
balances of linear and angular momenta for each of the rigid bodies:
m
1
¨
¯ x
1
= F
1
, . . . , m
N
¨
¯ x
N
= F
N
,
˙
H
1
= M
1
, . . . ,
˙
H
N
= M
N
. (11.1)
Here, F
K
is the resultant force acting on the Kth rigid body and M
K
is the resultant
moment relative to the center of mass
¯
X
K
of the rigid body. Equations (11.1) yield
differential equations for the motion of the bodies and expressions for the constraint
forces and moments.
As shown in [30], balance laws (11.1) are equivalent to Lagrange’s equations of
motion for the system of rigid bodies. For these equations, we denote by u
1
, . . . , u
6N
the coordinates chosen to parameterize the position vectors ¯ x
K
and rotation tensors
Q
K
. Then Lagrange’s equations of motion are
d
dt
∂T
∂ ˙ u
A
−
∂T
∂u
A
=
A
(A= 1, . . . , 6N) , (11.2)
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
11.3 Two PinJointed Rigid Bodies 349
where ¯ v =
˙
¯ x and
A
=
N
¸
K=1
F
K
·
∂ ¯ v
K
∂ ˙ u
A
+M
K
·
∂ω
K
∂ ˙ u
A
. (11.3)
It is emphasized that using either (11.1) or (11.2) will lead to equivalent equations
of motion for the system of rigid bodies.
As in our discussion of Lagrange’s equations of other systems, it can be shown
that, if the constraint forces and moments are prescribed by Lagrange’s prescription,
the system of constraints on the system is integrable, and the coordinates u
1
, . . . , u
6N
are chosen appropriately, then Lagrange’s equations of motion decouple into two
sets of equations: one from which the constraint forces and moments are easily de
termined, and the other from which motion of the system can be computed. The
proof of this can be found in [30], and the interested reader is referred to this paper
for details. Here, we are primarily content to use the result, but before doing so it
is important to comment on the expressions for the generalized forces that follow
from (11.3). Suppose that the system of rigid bodies is subject to C integrable con
straints and that the coordinates are chosen so that these constraints can be simply
represented as
u
6N−C+1
= f
1
(t), . . . , u
6N
= f
C
(t).
We also suppose that the constraint forces F
cK
and constraint moments M
cK
asso
ciated with these constraints are prescribed by Lagrange’s prescription. It can be
shown that
N
¸
K=1
F
cK
·
∂ ¯ v
K
∂ ˙ u
A
+
N
¸
K=1
M
cK
·
∂ω
K
∂ ˙ u
A
= 0 (A= 1, . . . , 6N −C) . (11.4)
That is, the constraint forces and constraint moments do not contribute to the gen
eralized forces
1
, . . . ,
6N−C
. As a result, our expressions (11.3) for the forces
A
are identical to those found in other treatments of Lagrange’s equations of motion
for systems of rigid bodies that are available.
∗
11.3 Two PinJointed Rigid Bodies
As our ﬁrst example, we return to the case of two bodies, B
1
and B
2
, that are pin
jointed. This system was discussed earlier in Subsection 8.6.4 and is illustrated in
Figure 8.8. Our goal is to outline how the equations of motion for this system can be
established by using Lagrange’s equations of motion.
We start with the kinematics. As discussed earlier, the pin joint introduces
ﬁve constraints on the motions of B
1
and B
2
. Mindful of these constraints, we
use Cartesian coordinates to parameterize ¯ x
1
, a set of Euler angles γ
1
, γ
2
, γ
3
to
parameterize Q
1
, and the angle θ to parameterize the relative rotation tensor Q
2
Q
T
1
.
The angle θ represents the rotation of B
2
relative to B
1
and the axis of rotation
∗
See, for example, Baruh [14], Greenwood [80], and Kane [104].
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
350 Introduction to Multibody Systems
associated with this rotation is e
3
: Q
2
Q
T
1
= L(θ, e
3
). It is prudent to deﬁne the
following seven coordinates:
u
k
= ¯ x
1
· E
k
, u
k+3
= γ
k
, u
7
= θ,
where k = 1, 2, 3. We also assume that
π
P1
= L
1
e
1
, π
P1
= (L
2
)
2
e
1
,
where L
1
and L
2
are constants. We also note that the corotational basis vectors
for B
1
are {e
1
=
1
e
1
, e
2
=
1
e
2
, e
3
=
1
e
3
}, and the basis vectors {
2
e
1
,
2
e
2
,
2
e
3
= e
3
}
corotate with B
2
.
From the coordinates chosen, we have the following representations:
¯ v
1
=
3
¸
i=1
˙ x
i
E
i
,
¯ v
2
= ¯ v
1
+ω
1
×L
1
e
1
+
ω
1
+
˙
θe
3
×(L
2
)
2
e
1
,
ω
1
=
3
¸
i=1
˙ γ
i
g
i
,
ω
2
=
3
¸
i=1
˙ γ
i
g
i
+
˙
θe
3
.
These results can then be used to compute the constrained kinetic energy
˜
T of the
system. This energy is the sumof the kinetic energies of B
1
and B
2
, and in the interest
of brevity, we do not record its full expression here.
Expressions for the constraint forces and moments acting on the system that
enforce the constraints associated with the pin joint were presented in Equation
(8.23). Because these quantities are prescribed by Lagrange’s prescription, and be
cause of our choice of coordinates, we can use Approach II and immediately write
the equations of motion for this system
∗
:
d
dt
∂
˜
T
∂ ˙ u
A
−
∂
˜
T
∂u
A
=
A
(A= 1, . . . , 7) , (11.5)
where
A
=
2
¸
K=1
F
K
·
∂ ¯ v
K
∂ ˙ u
A
+M
K
·
∂ω
K
∂ ˙ u
A
.
It is emphasized that F
c1
, F
c2
, M
c1
, and M
c2
do not contribute to
1
, . . . ,
7
. Indeed,
it is a good exercise to verify that
F
c1
·
∂ ¯ v
1
∂ ˙ u
A
+F
c2
·
∂ ¯ v
2
∂ ˙ u
A
+M
c1
·
∂ω
1
∂ ˙ u
A
+M
c2
·
∂ω
2
∂ ˙ u
A
= 0 (A= 1, . . . , 7) .
∗
Technically, we have shown that Approach II works only for a single rigid body, a single particle,
and a system of particles. As mentioned previously, the corresponding result for a system of rigid
bodies was established in [30], and we make use of it here.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
11.4 A SingleAxis Rate Gyroscope 351
You may have noticed that these identities are special cases of (11.4). Once the ap
plied forces and moments are prescribed and
˜
T calculated, the seven secondorder
differential equations of (11.5) sufﬁce to determine the motion of the system of two
rigid bodies. The resulting system of differential equations can be formidable, and
often additional constraints and geometric simpliﬁcations are employed.
An alternative formulation of the equations of motion would be to consider the
balance of linear and angular momenta for each of the rigid bodies:
m
1
¨ x
1
= F
1
,
m
2
¨ x
2
= F
2
,
˙
H
1
= M
1
,
˙
H
2
= M
2
. (11.6)
This would yield a set of coupled equations for the constraint reactions µ
1
, . . . , µ
5
[cf. (8.23)] and ¨ u
1
, . . . , ¨ u
7
. The advantages of Lagrange’s equations of motion in its
automatic selection of linear combinations of Equations (11.6) should be obvious.
11.4 A SingleAxis Rate Gyroscope
As shown in Figure 11.1, a singleaxis rate gyroscope is mounted on a platform P.
The gyroscope consists of a gimbal G that is free to rotate relative to P through an
angle α about the axis e
1
and a rotor R that is free to rotate relative to G through an
angle β about the axis e
2
. In addition to the pair of bearings connecting the gimbal
and platform, a spring of stiffness K and a dashpot with a damping coefﬁcient c are
suspended between G and P.
α
β
K
c
P
R
G
e
1
e
2
r
1
r
2
r
3
t
1
t
2
t
3
Figure 11.1. Schematic of a singleaxis rate gyroscope. This devices consists of a rotor R that
is suspended by a gimbal G over a platform P. When the platform rotates about t
3
the rotor
will want to dip, and this tendency is resisted by the spring and dashpot shown in the ﬁgure.
P1: FYX
mastercup cuus273/Oliver M. O’Reilly 978 0 521 87483 0 June 2, 2008 13:58
352 Introduction to Multibody Systems
The principle of operation of the gyroscope is as follows: Suppose that the plat
form is given an angular velocity about the vertical t
3
direction; then the rotor
will seek to rotate about a horizontal direction e
1
. This tendency to rotate will be
counterbalanced by the spring, but will result in an angle of deﬂection α = α
0
. As
we shall subsequently see,
∗
α
0
∝ .
Consequently, if the gyroscope is calibrated, then a measurement of α
0
can be used
to determine . In practice, however, α
0
is sensitive to many other effects that can
often serve to corrupt the measurement of . A wonderful discu