Advanced

Classical
Mechanics
Advanced
Classical
Mechanics

Bijan Kumar Bagchi

CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742

2017 by Taylor & Francis Group, LLC

CRC Press is an imprint of Taylor & Francis Group, an Informa business

No claim to original U.S. Government works

Printed on acid-free paper

Version Date: 20170330

International Standard Book Number-13: 978-1-4987-4811-7 (Hardback)

This book contains information obtained from authentic and highly regarded sources. Reasonable
efforts have been made to publish reliable data and information, but the author and publisher cannot
assume responsibility for the validity of all materials or the consequences of their use. The authors and
publishers have attempted to trace the copyright holders of all material reproduced in this publication
and apologize to copyright holders if permission to publish in this form has not been obtained. If any
copyright material has not been acknowledged please write and let us know so we may rectify in any
future reprint.

Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced,
transmitted, or utilized in any form by any electronic, mechanical, or other means, now known or
hereafter invented, including photocopying, microfilming, and recording, or in any information
storage or retrieval system, without written permission from the publishers.

For permission to photocopy or use material electronically from this work, please access
www.copyright.com (http://www.copyright.com/) or contact the Copyright Clearance Center, Inc.
(CCC), 222 Rosewood Drive, Danvers, MA 01923, 978-750-8400. CCC is a not-for-profit organization
that provides licenses and registration for a variety of users. For organizations that have been granted
a photocopy license by the CCC, a separate system of payment has been arranged.

Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and
are used only for identification and explanation without intent to infringe.
Visit the Taylor & Francis Web site at
http://www.taylorandfrancis.com
and the CRC Press Web site at
http://www.crcpress.com
In my fathers memory
Contents

Preface xi

1 Conceptual basis of classical mechanics 1

1.1 Newtons three laws . . . . . . . . . . . . . . . . . . . . . . . 2

1.2 Homogeneity and isotropy . . . . . . . . . . . . . . . . . . . 7
1.3 Solution process . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.4 Conservative forces . . . . . . . . . . . . . . . . . . . . . . . 11
1.5 Simple harmonic motion . . . . . . . . . . . . . . . . . . . . 12
1.6 Damped and forced oscillator . . . . . . . . . . . . . . . . . . 15
1.6.1 Damped oscillator . . . . . . . . . . . . . . . . . . . . 15
1.6.2 Forced oscillator . . . . . . . . . . . . . . . . . . . . . 17
1.7 The simple pendulum problem . . . . . . . . . . . . . . . . . 20
1.8 Conservation principles . . . . . . . . . . . . . . . . . . . . . 22
1.8.1 Conservation of linear momentum . . . . . . . . . . . 22
1.8.2 Conservation of angular momentum . . . . . . . . . . 23
1.8.3 Conservation of energy . . . . . . . . . . . . . . . . . . 24
1.9 Perturbative analysis and the quartic oscillator . . . . . . . . 28
1.10 Rewriting Newtons second law in terms of kinetic and potential
energy in a conservative system . . . . . . . . . . . . . . . . 30
1.11 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

2 Central force problems 35

2.1 Inertial and gravitational mass: Principle of equivalence . . . 36

2.2 Derivation of Keplers three laws . . . . . . . . . . . . . . . . 38
2.3 Properties and equations of orbits . . . . . . . . . . . . . . . 40
2.4 Integral representations . . . . . . . . . . . . . . . . . . . . . 42
2.5 A general class of power law potentials . . . . . . . . . . . . 42
2.6 Mapping the general class of potentials: Orbit equation for the
inverse square law problem . . . . . . . . . . . . . . . . . . . 45
2.7 Coulomb and isotropic oscillator potentials . . . . . . . . . . 45
2.8 LaplaceRungeLenz vector . . . . . . . . . . . . . . . . . . 49
2.9 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

vii
viii Contents

3 Lagrangian formulation in mechanics 53

3.1 Constraints and generalized coordinates . . . . . . . . . . . . 54

3.2 Formulation of DAlemberts principle . . . . . . . . . . . . . 58
3.3 Kinetic energy of a holonomic system . . . . . . . . . . . . . 60
3.4 Lagranges equations of motion . . . . . . . . . . . . . . . . . 62
3.5 Lagranges equations for some simple systems . . . . . . . . 66
3.5.1 Plane pendulum . . . . . . . . . . . . . . . . . . . . . 66
3.5.2 Spherical pendulum . . . . . . . . . . . . . . . . . . . 67
3.5.3 Binary star system . . . . . . . . . . . . . . . . . . . . 68
3.5.4 A system with four degrees of freedom . . . . . . . . . 69
3.5.5 The problem of a damped oscillator . . . . . . . . . . 70
3.5.6 A conservative scleronomic system . . . . . . . . . . . 71
3.6 Ignorable coordinates: Rouths procedure of solution . . . . . 72
3.7 Liouvilles class of Lagrangians . . . . . . . . . . . . . . . . . 76
3.8 Small oscillations . . . . . . . . . . . . . . . . . . . . . . . . 80
3.9 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

4 Hamiltonian and Poisson bracket 91

4.1 The Hamiltonian . . . . . . . . . . . . . . . . . . . . . . . . . 92

4.2 Hamiltonian canonical equations of motion . . . . . . . . . . 94
4.3 Poisson bracket . . . . . . . . . . . . . . . . . . . . . . . . . 101
4.4 Properties of Poisson bracket . . . . . . . . . . . . . . . . . . 102
4.5 Poisson theorem . . . . . . . . . . . . . . . . . . . . . . . . . 104
4.6 Angular momentum . . . . . . . . . . . . . . . . . . . . . . . 105
4.7 Liouvilles theorem . . . . . . . . . . . . . . . . . . . . . . . 107
4.8 The case of singular Lagrangians . . . . . . . . . . . . . . . . 109
4.9 Higher derivative classical systems . . . . . . . . . . . . . . . 111
4.10 The PaisUhlenbeck oscillator . . . . . . . . . . . . . . . . . 112
4.11 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

5 Dynamical systems: An overview 117

5.1 Basic notions and preliminaries . . . . . . . . . . . . . . . . . 118

5.2 Simple examples from classical mechanics . . . . . . . . . . . 122
5.3 Analysis of a linear system . . . . . . . . . . . . . . . . . . . 125
5.4 Nonlinear systems: Process of linearization . . . . . . . . . . 133
5.5 LotkaVolterra model . . . . . . . . . . . . . . . . . . . . . . 138
5.6 Stability of solutions: Lyapunov function . . . . . . . . . . . 141
5.7 Van der Pol oscillator and limit cycles . . . . . . . . . . . . . 145
5.8 Bifurcations . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
5.9 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
 Contents ix

6 Action principles 161

6.1 The principle of stationary action . . . . . . . . . . . . . . . 161

6.2 Corollaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
6.3 Continuous systems: Uniform string problem . . . . . . . . . 169
6.4 Normal modes of oscillation . . . . . . . . . . . . . . . . . . 172
6.5 Extended point transformation and variation . . . . . . . 174
6.6 and variations . . . . . . . . . . . . . . . . . . . . . . . . 177
6.7 Brachistochrone problem . . . . . . . . . . . . . . . . . . . . 180
6.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182

7 Motion in noninertial coordinate systems 185

7.1 Rotating frames . . . . . . . . . . . . . . . . . . . . . . . . . 186

7.1.1 Basic equations . . . . . . . . . . . . . . . . . . . . . . 186
7.1.2 Some remarks on the Coriolis force . . . . . . . . . . . 190
7.1.3 Effective gravitational constant . . . . . . . . . . . . . 191
7.1.4 Foucaults pendulum . . . . . . . . . . . . . . . . . . . 192
7.2 Nonpotential force . . . . . . . . . . . . . . . . . . . . . . . . 194
7.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
7.4 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196

8 Symmetries and conserved quantities 199

8.1 Condition of invariance and Noethers theorem . . . . . . . . 200

8.2 Operator approach . . . . . . . . . . . . . . . . . . . . . . . . 205
8.2.1 Symmetry operator . . . . . . . . . . . . . . . . . . . . 206
8.2.2 Parity transformation . . . . . . . . . . . . . . . . . . 207
8.2.3 Time-reversal symmetry . . . . . . . . . . . . . . . . . 208
8.3 Virial theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 209
8.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210

9 HamiltonJacobi equation and action-angle variables 213

9.1 Canonical transformation . . . . . . . . . . . . . . . . . . . . 215

9.2 Symplectic property . . . . . . . . . . . . . . . . . . . . . . . 218
9.3 Idea of a generating function . . . . . . . . . . . . . . . . . . 221
9.4 Types of time-dependent canonical transformations . . . . . 223
9.4.1 Type I canonical transformation . . . . . . . . . . . . 223
9.4.2 Type II canonical transformation . . . . . . . . . . . . 224
9.4.3 Type III canonical transformation . . . . . . . . . . . 225
9.4.4 Type IV canonical transformation . . . . . . . . . . . 225
9.5 Infinitesimal canonical transformations . . . . . . . . . . . . 228
9.6 HamiltonJacobi equation . . . . . . . . . . . . . . . . . . . 229
x Contents

9.6.1 Time independent HamiltonJacobi equation:

Hamiltons characteristic function . . . . . . . . . . . 232
9.6.2 Other variants of HamiltonJacobi equation . . . . . . 235
9.7 Action-angle variables . . . . . . . . . . . . . . . . . . . . . . 235
9.7.1 Motion of a particle in a 2-dimensional
rectangular well . . . . . . . . . . . . . . . . . . . . . . 242
9.8 Possible trajectories . . . . . . . . . . . . . . . . . . . . . . . 243
9.8.1 Periodic trajectories . . . . . . . . . . . . . . . . . . . 243
9.8.1.1 Some explicit examples for periodic
trajectories . . . . . . . . . . . . . . . . . . . 244
9.8.2 Open trajectories . . . . . . . . . . . . . . . . . . . . . 247
9.8.3 Special trajectories when the billiard ball
hits a corner . . . . . . . . . . . . . . . . . . . . . . . 247
9.9 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248

References 253

Index 257
Preface
The book has evolved from my teaching of the subject of classical mechanics
for many years. It is designed to serve as a textbook for the young postgraduate
students, researchers and teachers of theoretical physics/applied mathematics
and also students of engineering who need a good background and understand-
ing of classical mechanics. We have therefore placed emphasis on the logical
ordering of topics and appropriate formulation of the key mathematical con-
cepts with a view to imparting a clear knowledge of the basic tools in the
subject and improving the problem solving skills of the students. Complexi-
ties of the mathematical structure are kept to a minimum but greater stress
is laid on the application-side of the subject. We have tried to develop new
ideas as a smooth continuation of the preceding ones. The book is expected
to give a systematic and comprehensive coverage of the methods of classical
mechanics.
The book is organized into nine chapters and begins, in Chapter 1, with the
conceptual basis of classical mechanics to provide the necessary background
for the later chapters. This is expected to enable the students to have a grasp
on the perspectives behind the development of classical mechanics. In Chap-
ter 2 a treatment of central force problems is presented and certain typical
issues like Keplers laws, power law potentials and LaplaceRungeLenz vec-
tor are studied in much detail. Chapter 3 is concerned with the treatment
of Lagrangian dynamics. A large variety of problems is solved in this chap-
ter. Considerable emphasis is placed on the basic ideas keeping in mind the
difficulties that a student could face when being exposed to these principles
for the first time. Brief discussions of ignorable coordinates and Liouvilles
class of problems are given here. We have also included the topic of small
oscillations. In Chapter 4 a formulation of Hamiltonian dynamics is devel-
oped and Hamiltons equations of motion are derived. Also discussed in this
chapter are the Poisson brackets, their various properties and utilities. The
case of singular Lagrangians is dealt with in detail and a section is devoted
toward the treatment of higher derivative Lagrangians. Chapter 5 considers
an overview of dynamical systems and covers topics such as an analysis of lin-
ear systems, Lotka-Volterra models, Lyapunov systems, van der Pol oscillator,
limit cycles and the theory of bifurcations. Chapter 6 focuses on a detailed
treatment of action principles and discusses extended point transformations
along with different types of variations. The brachistochrone problem is stud-
ied here. Rotating frames and velocity dependent potentials are introduced in
Chapter 7 along with a treatment of the non-potential force. Its relevance in
physical phenomena is highlighted. The topic of Foucaults pendulum is also
touched upon. Chapter 8 takes up the role of symmetries and conservation
laws in mechanical systems. A discussion of the operator approach is given.

xi
xii Preface

Chapter 9 is concerned with the canonical transformations and their role in

physical problems. Topics like the HamiltonJacobi equation and the theory of
action-angle variables are treated extensively. A notable feature of this book is
that each chapter contains a wide range of worked out problems to clarify the
basic ideas involved. It is believed these will help further the understanding
of the subject.
First I thank Prof. Rupamanjari Ghosh, Vice Chancellor of Shiv Nadar
University for sustained encouragement during the entire process of writing
this book. I am also obliged to thank Prof. Sankar Dhar, head of the De-
partment of Physics, School of Natural Sciences, Shiv Nadar University for
excellent cooperation toward successful completion of the book.
It was the late Jnanendra Gopal Chakraborti who introduced me to the
subject of classical mechanics and greatly impressed me with his unique teach-
ing methods. I am indebted to him.
I profoundly thank Dr. Anindya Ghose Choudhury and Dr. Priya Johari
for reading certain portions of the manuscript and giving their insightful re-
marks and candid opinions. I express my sincere appreciation to Dr. Santosh
Singh for his help in finalizing the draft and kind interest. I am thankful
to my colleagues Prof. Tanuka Chattopadhyay, Dr. Soumen De, Prof. Sami-
ran Ghosh, Prof. Partha Guha, Dr. Swarup Poria, and Prof. Arabinda Roy for
their comments at various stages. Thanks are also due to many of my students
for their assistance during the writing of the book. I take this opportunity to
make a special mention of Dr. Abhijit Banerjee, Mr. Partha Mandal, Ms. Sa-
heli Mitra, and Mr. Tarun Tummuru. The references listed at the end have
deeply influenced me in making up the material of the book and I recommend
them for further reading.
I gratefully acknowledge the assistance from the librarians of the Inter-
University Center for Astronomy and Astrophysics, Pune and the Indian In-
stitute of Technology, Kanpur for access to their libraries and providing a
convenient ambiance for work.
I would also like to express my appreciation to Ms. Aastha Sharma, Com-
missioning Editor, CRC Press, Taylor & Francis Group for not only helping
me to initiate the project but also for her excellent cooperation in turning the
book into a reality.
I owe thanks to the late Jayanta Kumar Bagchi, my uncle, who constantly
urged me to complete the book but unfortunately did not live to see its pub-
lication.
Finally, I thank my wife Minakshi, and daughter Basabi, for their patience
and support.

Shiv Nadar University

Greater Noida Bijan Kumar Bagchi
Chapter 1
Conceptual basis of classical
mechanics

1.1 Newtons three laws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.2 Homogeneity and isotropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3 Solution process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.4 Conservative forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.5 Simple harmonic motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.6 Damped and forced oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.6.1 Damped oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.6.2 Forced oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.7 The simple pendulum problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.8 Conservation principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.8.1 Conservation of linear momentum . . . . . . . . . . . . . . . . . . . . . . 22
1.8.2 Conservation of angular momentum . . . . . . . . . . . . . . . . . . . . 23
1.8.3 Conservation of energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.9 Perturbative analysis and the quartic oscillator . . . . . . . . . . . . . . . . . 28
1.10 Rewriting Newtons second law in terms of kinetic and potential
energy in a conservative system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.11 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

Classical mechanics is the study of physical laws that control the motion of
material objects which are under the action of a force or system of forces. It
provides the basis for the growth of modern science. It has applications that
cover areas such as physics, chemistry, applied mathematics, biology and en-
gineering sciences. In particular, it seeks to address and explain the dynamics
of particles and rigid bodies, general classes of interactive systems, rotating
Earth problems, motion of charged objects, planetary motions around the Sun
and modeling of biological systems. Classical mechanics has an extraordinarily
rich history that began about the time of Galileo (15641642) although its ba-
sic foundations were laid later by Newton (16421727) in his famous treatise,
the Principia. He enumerated a set of three axioms which became the corner-
stone in explaining most of the qualitative features of classical mechanics. In
this chapter we discuss these laws and provide a multi-layered perspective on
them.

1
2 Advanced Classical Mechanics

1.1 Newtons three laws

Newtons three laws may be stated as follows:
1. Every body continues in its state of rest or of uniform motion in a
straight line unless it is compelled by external forces to change the state.
2. The rate of change of momentum is proportional to the impressed force
and takes place along the direction in which the force acts.
3. To every action there is an equal and opposite reaction.
The first law provides the concept of inertia. It is also referred to as the
Law of Inertia. The idea of absoluteness of space (Euclidean) and time is inher-
ent in the Newtonian scheme of things. Both are looked upon as independent
entities. As Newton noted in his Principia: Absolute space, in its own nature,
without regard to anything external, remains always similar and immovable.
About absoluteness of time, which is measured vis-a-vis the motion of differ-
ent objects and which evolves continuously, Newton opined that Absolute,
true, and mathematical time, of itself, and from its own nature, flows equably
without regard to anything external, and by another name is called duration.
Despite the independence of space and time, their correlation cannot be de-
nied in that for a material point characterized only by its location without
any reference to time is not of much significance. Conversely, the happening
of an event at a particular point of time or for a certain duration with a priori
no knowledge of its location looks conceptually absurd.
There are actually two aspects to the first law that may be necessary to
be kept in mind. It at once seeks to address both a stationary object and a
moving one. For the stationary object it says that an object at rest continues
to do so unless acted upon by an external force. For a moving object the law
holds that a particle in motion remains in motion (in a straight line) with the
same speed and direction unless of course it is interacted upon by an external
force.
Newtons first law also defines a class of reference frames which are called
inertial frames. These reference frames are not supposed to undergo any type
of acceleration. In other words, an inertial frame moves with a constant ve-
locity or is at rest with respect to any other inertial frame. In 1886 Ludwig
Lange, a German physicist, defined an inertial frame to be: A reference frame
in which a mass point thrown from the same point in three different non-
coplanar directions follows rectilinear paths each time it is thrown, is called
an inertial frame. An inertial frame may be looked upon as the one that is
fixed relative to the average position of a fixed star or that is moving with a
constant velocity and without any rotation relative to it. According to New-
ton, absolute space and time form a convenient background against which we
visualize occurrence of physical phenomena but the background itself cannot
be influenced by physical events themselves. To describe a physical event,
 Conceptual basis of classical mechanics 3

a fixed reference coordinate frame (inertial) must be chosen with respect to

which the location of a material object is defined.
A non-inertial frame is the one that is not inertial. A rotating frame is a
common example of a non-inertial frame. Although for short scales of time and
distance we can consider the planet Earth to be an inertial frame, truly speak-
ing, it is rotating about an axis passing through its geographical poles. Hence,
technically, it does not constitute an inertial frame. A rotating frame, as we
will find later, generates fictitious or pseudo forces. The Coriolis force is such
a type of force and we need to carefully deal with it along with the so-called
centrifugal force to make a proper use of Newtons laws in rotating frames.
A physical event with coordinates ~r = (t, ~x) is confined to the direct prod-
uct of the t-axis with the time parameter t R and space variables ~x R3 .
The latter is equipped with a fixed Euclidean structure called the Galilean
space. Such a set of space-time coordinates signifies an event.
Consider specifically a one-dimensional motion in which the x-coordinates
of two observers define two different moving frames S and S 0 . Let |~u| denote
the magnitude of uniform velocity of S 0 relative to S along their common x and
x0 axis with the same spatial origin coinciding at time t = t0 = 0. Due to the
inherent inertial character Newtons laws are unaffected by transformations
of the type

x0 = x + ut, y 0 = y, z 0 = z, t0 = t (1.1)
The one-dimensional transformation as given above relates the coordinates
of any event as noted in the frame S in terms of those as recorded in the frame
S 0 . In Newtonian mechanics there is a single universal time t and the last equa-
tion of (1.1) expresses precisely this, namely, the absolute character of timea
feature that is independent of the relative motion of the two velocities.
Differentiating by t leads to

U0 = U + u (1.2)
0
0
where U = dx dx
dt and U = dt . The above equation represents the classical
velocity-addition formula.
We can generalize (1.1) to three dimensions for two systems labeled by the
coordinates ~r (x, y, z) in the frame S and r~0 (x0 , y 0 , z 0 ) in the frame S 0
where the latter is moving uniformly with velocity V~ (u, v, w) in relation to
S. To this end, we write down

~ t x0 = x + ut,
r~0 = ~r + V y 0 = y + vt, z 0 = z + wt (1.3)
with t0 = t meaning time to be absolute in that it is independent of the relative
motion of the two observers in S and S 0 . In Equation (1.3), (u,v,w) are the
components of the constant velocity V~ . Such transformations are the standard
or pure Galilean transformations and the motion is said to be invariant under
them.
4 Advanced Classical Mechanics

It is possible further to extend (1.3) by the inclusion of a static rotation

and an arbitrary spatial shift
~ , i.e.,

~t+
r~0 = R~r + V ~, ~ R3
(1.4)
along with a constant temporal translation , i.e.,

t0 = t + , R (1.5)
where R stands for a 3 3 orthogonal rotation matrix and is given by

cos cos cos sin sin cos cos sin + sin cos cos sin
R = sin cos cos cos sin sin cos sin + cos cos sin sin
sin cos sin sin cos
(1.6)
where , and are the three Eulerian angles of rotations. The domains of
these angles are 0 < 2, 0 and 0 < 2. The rotation matrix
R satisfies |R| = 1 and RRT = RT R = I.
Evidently the generalized Galilean transformations (1.4) are a set of linear
equations. Given an inertial frame S, it can be carried over to another inertial
frame S 0 in 10 possible ways. These correspond to three for spatial transla-
tions, three for rotations, one for time translation and three for boosts denoted
by the constant velocity V~ such as in (1.3). All these constitute a 10-parameter
Galilean group. Representing an element of such a group as h(R, V ~ ,
~ , ), the
composition rule is defined as

h3 (R3 , w, 3 , 3 ) = h2 (R2 , v, 2 , 2 ) h1 (R1 , u, 1 , 1 ) (1.7)

where the various entries stand for

R3 = R2 R1 , w = v + R2 u, 3 = 2 + R2 1 + v1 , 3 = 1 + 2 (1.8)

In other words, successive operations of two rotations is equivalent to a single

rotation and reveals the essential feature that makes a group.
Newtons second law considers the momentum variable to be fundamental
and is concerned with the rate of change in it when an external force is applied.
Operationally the force acting upon an object of constant mass m is given by
a quantity which is the product of m times its acceleration ~a:

F~ (~r, ~r, t) = m~r = m~a (1.9)

where ~a = ~r is the acceleration and the force function is assumed to depend on
the position, velocity and time. Thus, an application of a force causes the body
to accelerate. Without it the body stays at rest and the motion is restricted by
the equation m~r = 0. m is called the inertial mass: it is the ability of the body
to resist being accelerated. With F~ known, a larger m points to a smaller ~a.
 Conceptual basis of classical mechanics 5

At the heart of the second law lies the postulate that the forces are lin-
early additive and behave as vectors. So, in an interacting system, if n forces
F~1 , F~2 , ...F~n to act on a body of mass m, it produces an acceleration

1 ~ F~
~a = (F1 + F~2 + ... + F~n ) = (1.10)
m m
(1.10) is an extended version of Equation (1.9) and embodies the central idea
of the principle of linear superposition. The main point of the second law is
that an application of a force brings about a change in the velocity of the
object which in turn causes it to accelerate. Conversely, a time-change in
velocity accounts for the force. Knowing the force function F~ along with the
initial conditions x(t) = x0 and x(t) = x0 at time t = t0 allows one to draw a
conclusion about the trajectory of the particle that the particle actually traces
out.
Classical mechanics is characterized by two underlying principles governing
it, namely:
(i) NewtonLaplace principle of determinacy (or the deterministic law)
which states that the state of the system (comprising the position and veloc-
ity of the particle) at time t completely determines its behavior for all future
(> t) and past (< t) times; in other words, the laws of physical phenomena
are entirely deterministic and

(ii) Galilean principle of relativity which states that the law of motion
(1.9) has the same form in every inertial frame.
Suppose that a physical state of a mechanical system is known at the initial
time t0 as given by the coordinate ~r(t0 ) = r~0 and velocity ~r(t0 ) = r~0 = v~0 .
The principle of determinacy tells us that the motion given by the function
~r = ~r(t, r~0 , r~0 ) is uniquely known for all t > t0 and t < t0 . Assuming such a
function to be sufficiently smooth a couple of differentiations with respect to
t yields the form r(t0 ) = f~(t0 , r~0 , r~0 ) in which t = t0 has been set. Since t0
can be chosen arbitrarily, the equation

r(t) = mf~(t, ~r, ~r) = F~ (t, ~r, ~r) (1.11)

is the one that describes the motion of a particle of mass m and is valid for
all t. Notice that in writing down the above differential equation we have not
specified any frame of reference. If the frame of reference is identified with an
inertial frame, then (1.11) is identical with Newtons equation of motion (1.9)
with F~ as the external force acting on the particle with mass m.
Conversely, given (1.11) and the initial conditions ~r = r~0 , ~r = r~0 = v~0 at
time t = t0 with the function f assumed to be sufficiently smooth we know
that there exists a unique solution in the form ~r = ~r(t, r~0 , r~0 ). This implies
the principle of determinacy. Note that the kind of law by which the system
is deterministic in only one direction (say, for instance, the future) but not in
the other (the past) or conversely is forbidden.
6 Advanced Classical Mechanics

Turning to Newtons third law, it states that if a body exerts a force on

another object then the latter returns the force with an amount equal and
opposite to what it receives. In the case of the rigid body, the forces between
the constituent particles that make up the rigid body are then equal and
opposite. These are the internal forces and do not contribute to producing
any effect on the motion of the body. Taken with the first law, we can say
that the deviation from the state of rest or uniform motion is caused only by
the external forces.
Consider the specific case of the component F~i in (1.10) when the ith
particle exerts a force on the body of mass m. By Newtons third law, it has
to be equal and opposite to what m exerts on the ith particle but it is tacitly
assumed that other particles (excluding the ith particle) in the system are
just spectators and their role in the contribution to this mutual force-play is
taken to be too insignificant to affect the linearity of the relation (1.10). The
principle of linear superposition disregards all kinds of self-interactions and
other many-body interactions.
We now move to an elementary consequence. Consider three particles i,
j and k with respective masses mi , mj and mk existing in an isolated envi-
ronment. Labeling F~ij and F~kj as, respectively, the force that the ith particle
exerts on the j th particle and the one that the k th particle exerts on the j th
particle, we can write for the j th particle

mj~aj = F~ ij + F~ kj (1.12)
Similarly for the k th particle, which would be influenced by the ith and j th
particles, we would have

mk~ak = F~ ik + F~ jk (1.13)
By Newtons third law the forces F~ kj and F~ jk are equal and opposite and
so when we add (1.12) and (1.13) the effects of these forces cancel leaving us
with the relation

mj~aj + mk~ak = F~ ij + F~ik (1.14)

If the accelerations ~aj and ~ak are now taken to be the same, say equal to
~a, then (1.14) points to an equation of motion of a single mass M according
to the form

M~a = F~ ij + F~ ik (1.15)
j k
where M = (m + m ) and the right-hand side expresses the total force that
the ith particle exerts on the j th and k th particles. The above result is easily
generalizable to the case of N interacting particles.
 Conceptual basis of classical mechanics 7

1.2 Homogeneity and isotropy

An important class of invariance in the Newtonian theory is the one that
is due to the time translation. This means that if (1.9) admits of a solution
of the type ~r = f~(t), then the class denoted by ~r = f~(t + ) also stands for
a solution for R. In other words, (1.9) depicts an autonomous character,
i.e., there can be no explicit time-dependence in F~ :

m~r = F~ (~r, ~r ) (1.16)

The implication is that every point of time is as good as any other moment.
Apart from the time translation, another class of invariance pertains to
space-translations in R3 . A consequence is that if (1.10) takes in a solution of
the type r~i = f~i (t), i = 1, 2, ...n, corresponding to the motion of a system
of n particles, then r~i = f~i (t) + ~i , i = 1, 2, ...n, where ~i is a constant
vector, also passes for an equally acceptable class of solutions. The message
is that every point in the Universe is equivalent to every other point. Qual-
itatively the concept of homogeneity implies that the properties of the body
do not change or vary from one point in the body to another. Indeed on
sufficient large scales, Newtonian space and time are both homogeneous and
isotropic.
Homogenity in space also implies that the force function F~i on the ith
particle in an inertial coordinate system can be only a function of the difference
of two coordinates like r~j r~i and the relative velocities r~j r~i . Thus

mi r~i = F~i (r~j r~i , r~j r~i ) (1.17)

where i, j = 1, 2, ...n.
Newtonian space is additionally isotropic implying that there is no prefer-
ential direction. To put it simply, if R represents an orthogonal transforma-
tion in the three-dimensional Euclidean space R3 , then isotropic invariance
expresses invariance under rotations as given by

F~ (R~r, R~r ) = RF~ (~r, ~r ) (1.18)

where R is an orthogonal matrix operating on each component of ~r. Of course
homogeneity suggests isotropy but the opposite is not necessarily true. Con-
sider a smooth flat region with a sea of uniformly green all around and one
is occupying a position in an elevated site somewhere in the middle of it.
Whichever way you look it will appear the same in all directions. We say that
the space is isotropic. However, because of ones privileged location one can-
not take for granted the space to be homogeneous as well. For a homogeneous
region one point in it can be substituted for another and so it is left to our
choice as to which one is preferred irrespective of which direction it is located.
8 Advanced Classical Mechanics

All
All
All
All
All All
All

All All
All

FIGURE 1.1: Two coordinate frames S and S 0 which are in translatory

motion with respect to each other. P is the representative point whose coor-
dinates with respect to the origins O and O0 are, respectively, ~r and r~0 .

Returning to the frames S and S 0 discussed previously, let us think of a

translatory motion of S 0 with respect to S. Such a motion does not change
the directions of the three spatial axes of S 0 relative to the corresponding ones
of S. In other words, they remain parallel to one another.
Let us consider now the motion of a particle at a point P whose coordi-
nates with respect to the origins O and O0 are, respectively, ~r and r~0 . See the
configuration in Figure 1.1 in which |d|~ is the distance between O and O0 . It
follows from the vector addition formula namely, ~r = d~ + r~0 , that in addition
to velocity addition rule (1.2) a similar one for the acceleration also holds if
we further differentiate it with respect to t. It follows that

~a = d~ + a~0 (1.19)

where d~ could be interpreted as the acceleration of S 0 with respect to S while
v and a~0 stand, respectively, for the velocity and acceleration of the particle
~ 0

at P with respect to S 0 .

For a non-accelerating frame S 0 , clearly d~ vanishes, implying from (1.19)
that the two accelerations coincide: ~a = a~0 . Consequently the law of motion is
similar in the two frames S and S 0 , i.e., F~ = m~a = ma~0 , reflecting the inertial
character of Newtonian mechanics. On the other hand, for an accelerating

moving frame the term d~ cannot vanish and we have a modified form of the
second law of motion F~ 0 = ma~0 where the modified force F~ 0 is defined by

F~ 0 = F~ md~ . In the latter form, the additional term -md~ arising from the
0
accelerating character of the coordinate frame S is generally responsible for
the occurrence of a fictitious force term.
 Conceptual basis of classical mechanics 9

1.3 Solution process

Determining the solutions for a given physical system involves basically
two fundamental steps. The first is to choose an appropriate reference frame
with respect to which the given problem is defined unambiguously while the
other is to set up the differential equation as guided by Newtons second law.
Concerning the first step, the Cartesian system of coordinates is often the
simplest reference frame to work with. If Fx , Fy , Fz are the components of an
external force F~ along the x, y, z directions, respectively, then for a particle of
constant mass m, we have according to Newtons second law, the component
equations, Fx = mx, Fy = my, Fz = mz.
For an N -point particle system, we can similarly write down relations of
the type

m1 x1 = Fx1 , m2 x2 = Fx2 , ... mN xN = FxN , (1.20)

m1 y1 = Fy1 , m2 y2 = Fy2 , ... mN yN = FyN , (1.21)

m1 z1 = Fz1 , m2 z2 = Fz2 , ... mN zN = FzN , (1.22)

In more compact terms we have

mi xi = Fxi , mi yi = Fyi , mi zi = Fzi i = 1, 2, ..., N (1.23)

It should be mentioned that when we switch over to an arbitrary reference
frame the transparency of the force-acceleration relationships, as provided by
the Cartesian form, may be lost. As a specific example, we may think of the
two-dimensional plane polar coordinates (r, ) system. Here the forms of the
velocity and acceleration are

~v = rr + r (1.24)

~a = (r r2 )r + (r + 2r) (1.25)
where r, are unit vectors along the directions of r and . We see that Fr 6= mr
and F 6= m, Fr and F being, respectively, the components of the external
force in the r and directions.
Next, let us probe into the interplay of (1.24) and (1.25). Consider the
simple case of a particle moving uniformly along a circle with a constant
angular velocity = . Then the acceleration turns out to be simply ~a =
- 2 ~r. The negative sign implies that the acceleration is directed toward the
origin with a constant magnitude. The occurrence of acceleration is due to the
fact that the direction of the velocity vector as given by (1.24) is continuously
10 Advanced Classical Mechanics

changing at a constant rate. Such a kind of acceleration is referred to as the

centripetal acceleration.
Concerning the use of other reference frames, there are indeed situations
depending on the type of problem at hand, for instance the central force prob-
lem, where a great deal of simplification results if the plane polar coordinates,
(r, ), are employed. Some well-known examples of the central force are the
inverse Newtons square law of gravitation and Coulombs electrostatic force
between two charges.
Apart from the Cartesian reference frame, use of a spherical polar system
having coordinates (r, , ) or a cylindrical system having coordinates (, , z)
can be convenient. The relations in terms of (x, y, z) and the corresponding
Laplacian 2 are given by

x = r sin cos , y = r sin sin , z = r cos (1.26)

1 2 1 1 2
2 = (r ) + (sin ) + 2 (1.27)
r2 r r r2 sin r2 sin 2
and

x = cos , y = sin , z=z (1.28)

1 1 2 2
2 = ( ) + 2 + (1.29)
2 z 2
There also exist other coordinate frames such as the parabolic system
(, , ) wherein the correspondences are defined by
p p 1
x= cos , y = sin , z = ( ). (1.30)
2

4 4 1 2
2 = ( ) + ( ) + (1.31)
+ + 2
Proceeding now to the second step, a differential equation that follows
from (1.9) has to be formulated. In this regard, we have to identify the guiding
forces and make sure that the conditions prescribed in the given problem are
the appropriate ones for which Newtons law is applicable. In general if the
constraints are present, which are so ordinarily, then the external forces are
not known completely. To tackle this issue we have to go for an analytical
formalism that will be explained in Chapter 3.
We have already referred to the consideration of the inertial frame. Others
are
(i) The magnitude of the masses and time-distance scales should neither be
too small (like the dimensions we deal with at the microscopic level of atoms
and nuclei where the principles of quantum mechanics are applicable) nor too
large (as we encounter in the solar system or a galaxy wherein the underlying
 Conceptual basis of classical mechanics 11

dynamics are governed by the more sophisticated Einsteins theory of general

relativity) and
(ii) The magnitude of the velocity of the particle must be very small com-
pared to the velocity of light c. It is well known that dynamics of objects
moving with high velocities, comparable to the order of c, come under the
purview of the formulae of the special theory of relativity. In the latter for-
malism, Galilean transformations are replaced by the Lorentz transformations
that do not hold the time variable t to be absolute but is treated at par with
the space coordinates.

1.4 Conservative forces

The conservative force represents a special class of force such that the work
done by it, as the system evolves from one configuration to another, depends
on its initial and final positions only. Its occurrence in nature is quite com-
monplace. Gravity and elastic spring force are the two familiar examples of
conservative force. Friction, however, is a non-conservative force. A conserva-
tive force can be distinguished by any one of the following features:
(i) The work done by the force is path independent. In other words, no
matter what route is chosen, the work done by the force acting upon the
particle will be the same as long as it is conservative.
(ii)Around a closed path or loop in a simply connected region the work
done by the force is zero: F~ .d~r = 0 for any closed loop C.
H

(iii) If (Fx , Fy , Fz ) are the Cartesian components of the force F~ , then the
sum Fx dx + Fy + Fz dz is an exact differential.
(iv) F~ is only a function of position and x F~ = 0.
(v) A potential energy function V exists that has a definite value at every
point.
(vi) T + V = constant where T is the kinetic energy.
From (iv) we can write F~ as the gradient of some scalar function V . It also
points to the freedom that an arbitrary constant can be added to V without
altering F~ . Since the gradient points to the direction of increasing potential
and forces cause the system to move to a lower potential, a negative sign is
chosen to express F~ = V : in other words, the force is the negative gradient
of some potential function V . In one dimension it is simply Fx = dV dx . When
integrated it gives immediately for the work done, W = [V (b) V (a)] =
4V . We are therefore led to the following principle that the work done on
the system (positive work) increases the potential energy while the work done
by the system (negative work) decreases the potential energy.
Sometimes a force can be a combination of both conservative and non-
conservative parts. Then, of course, the work done by the non-conservative
12 Advanced Classical Mechanics

part will only contribute to the change in energy as the motion is executed
between two different points. Let us now give some examples of conservative
and non-conservative forces.

Example 1.1

The force of gravity between two objects, of masses M and m and |~r| as
the separating distance between them, is given by

GM m
F~ i = ~r, r 6= 0
r3
Around a closed path

I I
GM m GM m GM m
~r.d~r = dr = = 0, r 6= 0
r3 r 2 r
Thus gravity is a conservative force.

Example 1.2

The frictional force, given an infinitesimal displacement d~r, always acts

oppositely: F~ .d~r < 0. As a result F~ .d~r < 0. Therefore, friction is a non-
H

conservative force.

Example 1.3

Consider a force F~ with components (y, x, z) and C to be a circle defined

by x2 + y 2 = c2 , z = h. With the parametrization x = c cos and y = c sin
for the circle, it is easy to work out

Z2
(c sin , c cos , h).(c sin , c cos )d = 2c2 6= 0
0

So the force F~ here is a non-conservative one.

1.5 Simple harmonic motion

The simple harmonic motion (SHM) is perhaps the simplest and most
elegant of all dynamical systems. It considers the problem of a mass-spring
system being attracted to a given fixed point by a force which by Hookes law is
 Conceptual basis of classical mechanics 13

assumed to be directly proportional to the distance from the point. Obviously

the SHM, which is also referred to as the simple harmonic oscillator, is a
conservative system. In a one-dimensional setting, along the x-axis, the force
is given by F (x) = kx, where k > 0, which is linear and of a restoring
type. F (x) being an odd function of x is negative when x > 0 and positive
when x < 0. For such a force we run into the differential equation as given by
Newtons second law
r
2 k
mx + kx = 0 x + 0 x = 0, 0 = (1.32)
m
where the overhead dots represent derivatives with respect to the time t and
1
0 is the natural frequency (angular) having a dimension (time) . A sys-
tem whose dynamics is governed by such an equation is called a harmonic
oscillator.
Equation (1.32) admits the following trigonometric solution:

x(t) = A cos(0 t + ) (1.33)

The coefficient A is the amplitude of the motion and is the initial phase at
t = 0. To get an estimate of the period of the motion T, the solution being
of a trigonometric type, we have to consider the repeated intervals of time
for which the motion is similar due to the cosine nature of the solution. The
value of T is obtained by noting that each time the phase angle changes by
2, both x and the velocity v = dxdt undergo a complete cycle of variable. In
this way corresponding to the pairs, (1 , t1 ) and (1 + 2, t + T ), we have

1 = 0 t1 + , 1 + 2 = 0 (t1 + T) + (1.34)

By subtraction we get 0 T = 2 which yields

r
2 k
T= = 2 (1.35)
0 m
The inverse of the period T gives the frequency measured in Hertz, i.e.,
cycles per second: = T1 . Because of (1.35) it is clear that the period or
frequency is independent of the amplitude. This is a very important property
as far as the simple harmonic motion is concerned. Such a system of bounded
periodic motion is referred to as an isochronous system. In the literature only
a limited number of isochronous systems are known.
Having a general solution (1.33) involving two arbitrary constants A and
, we need certain initial conditions to fix them. For instance, the initial
conditions may correspond to the initial position x0 and initial velocity v0 of
the particle at t = 0. Since
dx
v= = 0 A sin(0 t + ) (1.36)
dt
14 Advanced Classical Mechanics

All
Kind Kind

All

FIGURE 1.2: The parabolic form of the potential V (x) = 12 kx2 .

it turns out that

x0 = A cos , v0 = 0 A sin (1.37)
The quantities A and can now be solved to get
s
v2 v0
A = (x0 )2 + 02 , = arctan( ) (1.38)
0 0 x0

We now make a few remarks on the character of the potential function

V (x) associated with the SHM. We say that the particle is in an equilibrium
(or stationary) state if it is not subject to any force. This implies dV dx = 0
and we observe that an equilibrium state of the particle pertains to being
a minimum or maximum of the potential. To make the subtle distinction
between when the equilibrium is stable or unstable we have to check whether
forcing small deviations from the equilibrium point x = x0 causes the particle
to return to the point. Mathematically, this means that the requirement of
stable equilibrium corresponds to the inequality dF dx < 0 at x = x0 : in other
words, if the particle steps to the right then the force acts to the left while the
opposite is true when the particle is displaced to the left. On the other hand,
if dF
dx > 0 the contrary features hold and we run into an unstable equilibrium
state. Hence because F = dV dx , a stable (unstable) equilibrium point signals
d2 V
the inequality dx2 > (<)0 and so corresponds to the minimum (maximum)
d2 V
of the potential function V (x). The third possibility dVdx = 0 = dx2 points to
a neutral equilibrium point.
For the simple harmonic oscillator the force term implies that the potential
is given by V (x) = 12 kx2 and so the total mechanical energy of the system,
being the sum of of kinetic energy and potential energy, is
1 1 dx 1 2
E= mv 2 + V (x) = m( )2 + kx (1.39)
2 2 dt 2
The positivity of the kinetic energy restricts E > 0 implying E needs to be
greater than V . For E = 0, the particle is at rest at the lowermost point
 Conceptual basis of classical mechanics 15

O of the parabola as shown in Figure 1.2. It is always stable there: small

displacements from O inevitably cause the particle to return to O. For E > 0,
however, the inequality V (x) E has to hold, i.e., the particle remains within
the confines of the parabola: E 12 kx2 . The roots of the latter give the turning
q q
points, a = 2E k and b = 2E
k . The regions beyond these turning points
are classically inaccessible.
Consider a smooth and continuous function V (x). Taylor series expansion
about an arbitrary point x = x0 gives

1
V (x) = V (x0 ) + (x x0 )V 0 (x0 ) + (x x0 )2 V 00 (x0 ) +
2
1
+ (x x0 )n V (n) (x0 ) + Rn+1 (1.40)
n!
where Rn+1 corresponds to the remainder after (n + 1) terms.
If x0 is an equilibrium point, i.e., V 0 (x0 ) = 0 and if we retain terms up to
O((x x0 )2 ), then for small displacements from the equilibrium
1
V (x) V (x0 ) + (x x0 )2 V 00 (x0 ) (1.41)
2
At the point O where the potential has the minimum, V 00 (x0 ) > 0. Should we
redefine the zero to set V (x0 ) = 0, the potential V (x) simplifies to
1
V (x) (x x0 )2 V 00 (x0 ) (1.42)
2
which has the profile
q of a harmonic oscillator potential having an angular
00 (x )
frequency given by V m 0
. From such a general consideration, we therefore
conclude that any physical system near its stable equilibrium point can always
be approximated by a harmonic oscillator potential. In fact, for many physical
phenomena, whenever we consider the motion of a particle being subjected
to a potential that has one or more local minima, the harmonic oscillator
approximation gives the initial clue to the understanding of the behavior of
the system.

1.6 Damped and forced oscillator

1.6.1 Damped oscillator
In reality, oscillations tend to get overpowered by frictional forces that
have the character to reduce or dampen the amplitude causing the system to
come ultimately to a halt. A damped oscillator is a more complicated case
compared to the simple harmonic motion.
16 Advanced Classical Mechanics

In the presence of a frictional term which is proportional to the velocity of

motion, i.e., the case of viscous damping, (1.32) can be extended to the form
mx + kx + x = 0. The parameter is referred to as the damping constant.
In the following we reexpress this equation in terms of a quantity

x + 02 x + 0 x = 0 (1.43)
where we have put = m0 relating the damping factor with the natural
frequency parameter 0 .
The linearity of Equation (1.43) suggests that we can try a solution of the
type x(t) et . Substitution in (1.43) implies that should be restricted by
r
0 2
= , = 0 1 (1.44)
2 4
A look at the form for reveals that three different cases arise according to
whether (i) < 2, (ii) = 2 or (iii) > 2. Let us consider these cases sepa-
rately.

Case (i): < 2

Here is imaginary and the general solution of (1.43) can be written as

r
0 t 2
x = Ae 2 cos[(0 1 t + ] (1.45)
4
where A and are integration constants. An appropriate choice of the ini-
tial conditions enables one to fix these constants. An imaginary corresponds
to the case of underdamping. The solution (1.45) represents q an oscillation of
0 t 2
amplitude Ae 2 with a frequency as given by 0 1 4 . The am-
plitude can be seen to decay exponentially with time while the frequency is
lower than the natural frequence 0 of the standard harmonic oscillator. Ac-
tually Equation (1.43) can be converted to the form z + 0 2 z = 0 by making
0 t
use of the transformation x Be 2 z. This is an SHM but the oscillation
happens with a reduced or damped frequency . The process is carried on
until eventually the equilibrium state is reached.

Case (ii): = 2

This case corresponds to = 0. With the general solution given by

0 t
x = x0 e 2 (1.46)
we have here a critical damping case. Note that the system decays exponen-
tially to the equilibrium position.
 Conceptual basis of classical mechanics 17

Case (iii): > 2

Such a possibility for implies to be real and the system is overdamped.

Equation (1.43) admits the following class of general solution:
0 t
x=e 2 (Cet + Det ) (1.47)
where C and D are constants of integration to be determined from the initial
conditions. Clearly of the two terms in the right-hand side, one is smaller in
magnitude in comparison to the other and gets to a steady state at an earlier
time. The point to note is that it is achieved slower than the critically damped
case.

1.6.2 Forced oscillator

(a) Forced harmonic oscillator

Let us next consider the case of a forced harmonic oscillator subject to a

time-varying periodic external force F (t) given by F (t) = F0 cos(t) on the
undamped motion of the simple harmonic oscillator:
F0
x + 02 x =cos(t) (1.48)
m
where denotes the driving frequency. The general solution of (1.48) can be
expressed in the form

F0 /m
x(t) = A cos(0 t ) + A cos(t), A = (1.49)
(02 2 )
where the first term in the right-hand side corresponds to a solution of the
simple harmonic motion, 0 being its natural frequency. The solution (1.49)
is a composition of two cosine-like terms valid at different frequencies 0 and
. We need to distinguish between the two cases when 6= 0 and = 0 .
The case 6= 0 is a simple one. We notice that the second amplitude A
in (1.49) has a finite value when 0 while it approaches zero as .
However, the situation changes drastically when approaches 0 . Here we
have to replace the solution (1.49) by a different form
F0
x(t) = P cos(0 t) + t sin(0 t) (1.50)
2m0
To arrive at (1.50) we took the representation x(t) = P cos(0 t) + Qt sin(0 t)
as a plausible solution of (1.48) and fixed the arbitrary constants P and Q
through matching of the two sides. We notice that the second term in the
right-hand side of (1.50) blows up as t . Also, because of the presence
F0 F0
of the sine term it oscillates between - 2m 0
and 2m 0
and dominates the os-
cillation of the first term as t grows. Thus by driving the system to a certain
18 Advanced Classical Mechanics

frequency value we are met with a large oscillation behavior. Such a feature
is referred to as the resonance.

(b) Forced harmonic oscillator with linear damping

What happens if a frictional term is added in (1.48)? We will then have to

deal with an explicitly velocity-dependent differential equation with a time-
varying source term:
Fo
x + 02 x + 0 x = cos(t) (1.51)
m
To handle such an equation the usual trick is to complexify it and identify it
as its real part. Toward this end we introduce the complex variable z = x + iy
and notice that Equation (1.51) along with its partner equation containing a
sine-like external force term, namely,
Fo
y + 02 y + 0 y = sin(t) (1.52)
m
are the respective real and imaginary components of the following single equa-
tion:
Fo it
z + 02 z + 0 z = e (1.53)
m
where we have made use of the trigonometric identity

ei(t) = cos(t) + i sin(t) (1.54)

to arrive at the right-hand side of (1.53).

We now seek a particular solution of (1.53) as

z = z0 eit (1.55)

We immediately notice that such a solution is perfectly tenable provided the

following condition is fulfilled:
Fo 1
z0 = (1.56)
m [ 2 + i0 + 02 ]
The modulus value of z0 is recognized as
Fo 1
|z0 | = (1.57)
m [( 2 + 02 )2 + 2 2 02 ] 12

signalling the proportionality of the amplitude |z0 | to F0 .

To write down the general solution of (1.51) we first of all observe that it
is possible to express z0 as a product of the amplitude |z0 | and a phase factor
ei0 , i.e., z0 = |z0 |ei0 , where 0 can be put as
0
0 = tan1 (1.58)
( 2 02 )
 Conceptual basis of classical mechanics 19

Implicit in the above is the feature that 0 0 as approaches while

0 0 as approaches 0 from the positive or the negative side.
In this way, we are led to the following expression for z(t):

Fo
z(t) = 1 ei(t+0 ) (1.59)
m[( 2 + 02 )2 + 2 2 02 ] 2

The general solution of the linear differential equation (1.51) is the super-
position of the complementary function noted earlier in (1.45) for the damped
oscillator problem and the one obtained by projecting out the real part from
(1.59). Thus we arrive at
r
0 t 2 Fo
x(t) = Ae 2 cos[0 1 t+]+ 1 cos(t+0 )
4 m[( 2 + 02 )2 + 2 2 02 ] 2
(1.60)
A look at the above complete solution reveals that the first term, which con-
tains the initial conditions, but being appended with a damping exponential
term eventually dies out at large times and is therefore transient in character.
Therefore, the motion is ultimately governed by the steady solution which
being a particular solution of the differential equation is independent of the
initial conditions:
Fo
xsteady = 1 cos(t + 0 ) (1.61)
m[( 2 + 02 )2 + 2 2 02 ] 2

The amplitude of the steady state solution is given by |z0 | and has a max-
imum value when the quantity [( 2 + 02 )2 + 2 2 02 ] reaches a minimum.
The required condition for this to happen amounts to 2 = R 2 = 12 (22 )02
which is positive for < 2. Near = R results in a large response, in other
words, very large oscillations and the behavior of resonance is noticed.

(b) Forced harmonic oscillator with nonlinear damping

Finally, we consider the problem of a periodically forced oscillator sub-

jected to a nonlinear damping according to the equation1
Fo
x + 02 x + 0 |x|x = cos(t) (1.62)
m
We proceed in a somewhat different way than pursued in the previous case.
First of all let us assume a solution like

x(t) = A cos(0 t + ) (1.63)

1 A. Li, L. Ma, D. Keene, J. Klingel, M. Payne and X-J. Wang, Forced oscillations with

linear and nonlinear damping, Am. J. Phys. 84(2016)32.

20 Advanced Classical Mechanics

where A is the amplitude. Substituting it in the above equation yields the

expression

A(02 2 ) cos(t + ) 0 A2 2 | sin(t + )| sin(t + )

Fo
= [cos(t + ) cos + sin(t + ) sin ] (1.64)
m
We now exploit the above relation by multiplying both sides by sin(t + )
and integrate from 0 to 2 i.e., over a complete period. In this way we arrive
at the relation
80 2 2 Fo
A = sin (1.65)
3 m
where we have used the definition | sin | = sin along with the trigonomet-
ric identity 4 sin3 = 3 sin sin 3. In the above equation the + and signs
correspond, respectively, to the intervals 0 < < and < < 2.
On the other hand, multiplication of both sides of (1.64) by cos(t + )
gives on integration from 0 to 2
Fo
A(02 2 ) = cos (1.66)
m
The above two relations put us in a position to determine the amplitude A
and the phase angle in the forms

s r
3 2 642 02 Fo 2 80 A
A= + ( ) , tan = (1.67)
80 2 4 9 2 4 m 3

where is given by = [1 ( 0 )2 ]2 .

1.7 The simple pendulum problem

As shown in Figure 1.3, in the simple pendulum case, a bob P of mass m
is hung from a fixed point O by an inextensible rod of length l and is allowed
to swing freely in a vertical plane only. Air resistance is neglected because it
can be taken to be too small to influence the motion. Of the two forces at
work, the tension is directed toward O while the force of gravity always acts
in the downward direction. The former does no work because it is operative
in a direction perpendicular to the direction of motion of the bob.
The force driving the bob is given by mg sin where the negative sign
is appended because g acts downward. To write out the equation of motion,
we note that for the displacement y of the bob from the equilibrium state,
we need to equate, by Newtons second law of motion, the driving force with
 Conceptual basis of classical mechanics 21

All
All
All
All

Kind
All
All

FIGURE 1.3: A simple pendulum swinging in the vertical plane.

the mass times acceleration y of the displacement. This gives the relation
my = mg sin . Now since is the angular displacement of the bob with the
vertical axis, as it sweeps out a circular arc while it swings from one position
to another, we have the relationship y = l from the arc length formula. Hence
one deduces the equation of motion
g
+ sin = 0. (1.68)
l
A linearlized version of the above equation, when sin is approximated by , is
the form of the simple harmonic oscillator whose general solution is sinusoidal
in nature as noted in (1.33). Its period of oscillation is
s
l
T = 2 (1.69)
g
Since T here is independent of the amplitude, the isochronicity of the simple
pendulum is indicated in a first approximation.
Comparing (1.69) with the period of oscillation of the SHM as furnished
by (1.35) we notice the correspondence
r r
k g
(1.70)
m l
Numerically for the standard value of g = 9.8062m/s2 and length of 1 meter,
(1.69) yields a period of approximately 2 seconds. The general solution of
(1.68) is governed by the Jacobi elliptic function to which we shall return
later.
The potential energy of the simple pendulum reads
Zh
V = (mg)dx = mgh (1.71)
0
22 Advanced Classical Mechanics

where, as shown in Figure 1.3, h corresponds to the height or the distance

from the point of suspension to the resting or equilibrium position of the bob.
Since h = l l cos , V can be expressed as

V () = mgl(1 cos ) (1.72)

where is the angle between OP and the vertical line OC.
The total energy is given by
1 2 d 2
E=ml ( ) + mgl(1 cos ) (1.73)
2 dt
where the first term in the right-hand side represents the kinetic energy term.
This is in conformity with the energy relation (1.39) of the harmonic oscillator
taking into account the replacement (1.70), the connection of the angular
2
displacement with y namely, y = l and approximating cos 1 2 .

1.8 Conservation principles

1.8.1 Conservation of linear momentum
Let a system of N particles be acted upon by a set of external forces. We
also take in the possibility that the particles may interact internally with each
other because of the mutual forces operating between themselves. Ignoring the
self forces the latter actions may be represented by F~ij where F~ii = 0, i, j =
1, 2, ...N . In this way we can project out the equation of motion of the ith
particle in the manner
d~
pi E X
= mi r~i = F~i + F~ij , i, j = 1, 2, ...N (1.74)
dt
i6=j

E
where p~i is the linear momentum of the ith particle, F~i is the total external
force on the ith particle and the terms i6=j F~ij denote the remaining ones
P
due to the internal interactions: j 6= i implying that self-interactions are being
discarded.
Summing (1.74) over all the particles we get
X d~
pi X X E XX
= mi~ri = F~i + F~ij , i, j = 1, 2, ...N (1.75)
i
dt i i i i6=j

Using Newtons third law, i.e., mutually interacting particles exert equal and
opposite force forces upon themselves implying conditions like

F~ij = F~ji i, j = 1, 2, ...N (1.76)

 Conceptual basis of classical mechanics 23

it follows that the double-sum in the right-hand side of (1.75) vanishes. We

are therefore left with an equation like

dP~ X X E
= mi~ri = F~i = F~ , i = 1, 2, ..., N (1.77)
dt i i

where P~ is the total linear momentum of the particles, pointing to the equality
of the rate of change of the total linear momentum as equal to the total
external force F~ . If further we ignore the collective effect of the total external
forces, then the last equation integrates P to the result for the constancy of the
linear momentum for a closed system: i mi v~i = a constant vector, where
~vi = ddtr~i is the ith velocity.

1.8.2 Conservation of angular momentum

~ be the resultant angular momentum of a system of particles
Let

~ =
X d(mi r~i )
~ri (1.78)
i
dt
where the angular momentum of each particle has been summed over. We now
~ to write down
compute the total derivative of

~
d
~ri F~iE
X X X
= ~vi mi v~i + ~ri mi r~i = (1.79)
dt i i i

where the first term in the right-hand side of the first equality vanishes be-
cause of the cross-product between similar vectors. For the second term which
receives contributions from both the parts F~iE and i6=j F~ij , i, j = 1, 2, ...N ,
P

~r F~E because other terms

P
yields, when summed over, only the quantity i i i
being of the type ~r1 F~12 + ~r2 F~21 reduce to just (~r1 ~r2 ) F~12 . But, by
Newtons third law of motion, the sum of such vector products has to vanish
because the force F~12 acts along the line joining the particles 1 and 2.
The vector quantity in the right-hand side of (1.79) expressed in terms
of the external force is called the vector moment or torque. It stands for the
time rate of change of the angular momentum about a fixed point expressed
in terms of the torques of all the external forces acting on the system. For a
truly closed system the external forces have no role upon the system and one
~
is led to the result that ddt vanishes. This implies that the angular momentum
is conserved.
A couple of remarks are in order. First, use of a different vector ~r1 which
is in the same line of action of F~i will not change the torque because of the
null condition (r~1 r~i ) F~1i = 0. Second, if there are two equal and opposite
forces F~i and F~i acting along the same line, then also we have trivially
~r1 F~i + ~r1 (F~i ) = 0.
24 Advanced Classical Mechanics

1.8.3 Conservation of energy

The criterion (vi) for the classification of a conservative force easily follows
from Newtons second law of motion. Indeed we find for a particle of mass m
moving with velocity ~v

Z Z Z Z Z
d~v
m~v . dt = F~ .~v dt = F~ .d~r = (V ).d~r = dV (1.80)
dt

where the force F~ = V in terms of the potential energy function V. The

above equation then points to the constancy of energy:
1
m|~v |2 + V = constant = E (1.81)
2
where the constant is the total energy which is the sum of kinetic and potential
energies. Like we saw in the case of the harmonic oscillator in (1.39), the energy
equation (1.81) for a general class of potential V gives a useful insight on the
restriction of a particle motion.
In one dimension, for a general V (x), we can reexpress (1.81) in the form
p
p = mx = 2m(E V ) (1.82)
It gives the integral
r Z
m dx
t= p + constant (1.83)
2 E V (x)
which provides t as a function of x.
As an application of the above equation let us consider the simple case
of a constant force acting on a particle under a potential given by V = cx
where the constant c > 0. We find from (1.83)
Zx
1
r
m dx
t= = 2m( E + cx E + ca) (1.84)
2 E + cx c
a
where a is the position of the particle at t = 0. With the help of the energy
equation E = 12 mv02 ca, where v0 is the initial velocity, we straightforwardly
1
obtain the usual distance-time relation x(t) = a + v0 t + 2m ct2 .
Since the kinetic energy of a moving particle is always positive, have the
constraint V (x) < E for the motion be finite. Here the particle is confined (i.e.,
trapped) in the region for which such an inequality holds. However, should
V (x) equal E, the particle comes to rest and for subsequent dynamics the di-
rection of motion is reversed and the process is continued leading to periodic
oscillations. Naturally solving for the equation V (x) = E yields the turn-
ing points and the latter effectively prescribe the limits of the finite motion.
Beyond V (x) = E the motion leads to untrapped trajectories. Trapped and
untrapped trajectories are separated by curves called separatrices.
 Conceptual basis of classical mechanics 25

Kind

All All Kind All

All All All All

FIGURE 1.4: Restriction of the particle motion in the region AB under the
influence of the potential V (x). The dotted line corresponds to V (x) = E.

In Figure 1.4 we have graphically illustrated the allowed region along with
the turning points A and B. It is easy to realize that for the motion bounded in
the region AB, the particle moves back and forth between the turning points
a and b and the motion is oscillatory in character. The period T is given by
the time taken by the particle to travel from A to B along the potential curve
and back. By symmetry this is twice the time from a to b and hence from
(1.83) we obtain
Zb
dx
T = 2m p (1.85)
E V (x)
a

where the ranges of the integral correspond to the values of the roots of the
equation V (x) = E assuming E to be known.

Example 1.4

For a particle under the influence of a quartic potential V (x) = x4 , the

period of oscillation is
1
E4
Z
dx
T = 2 2m (1.86)
E x4
0
26 Advanced Classical Mechanics
1
Substituting by turn x = E 4 z and t = z 4 , it is straightforward to reduce the
above integral to a tractable form
r Z1
m 1 3 1
T= E4 t 4 (1 t) 2 dt (1.87)
2E
0

The integral in the right-hand side is in the form of the beta function integral
R1
B(x, y) = tx1 (1 t)y1 dt with x = 14 and y = 12 . But since the beta
0
(x)(y)
function is expressible in terms of the -function through B(x, y) = (x+y) ,
the time period for the quartic oscillator is given by

1 2m 1 ( 14 )
r
T= E4 3 (1.88)
2 E ( 4 )

Consider again the problem of the simple pendulum. Here the total me-
chanical energy is conserved because the only external force, the tension T,
does no work. Note that gravity is considered an internal force. More specif-
ically, the conservation can be seen by calculating the time derivative of E
from (1.73) which gives

dE g
= ml2 ( + sin ) (1.89)
dt l
The right-hand side vanishes by using the equation of motion of the pendulum
implying that E is constant with time.
Let us use the trigonometric identity 1 cos = 2sin2 ( 2 ) to write the
energy equation as
1 2 d 2
E() = ml ( ) + 2mgl sin2 ( ) (1.90)
2 dt 2
As the pendulum oscillates in the vertical plane let 0 be the highest point
reached in the motion. Then 0 = 0 and we get from above
0
E(0 ) = 2mgl sin2 ( ) (1.91)
2
Thus since the energy is conserved we can equate E() and E(0 ) to be
led to the constraint
d 2 4g 0
( ) = (sin2 sin2 ) (1.92)
dt l 2 2
For small values of 0 , (1.92) can be approximated to

d 2 4g 02 2
( ) ' ( ) (1.93)
dt l 4 4
 Conceptual basis of classical mechanics 27

which points to the equation of a circle

2 + ( p g )2 ' 0 2 (1.94)
l

in the plane of and g .

l
To obtain an estimate of the time period that follows from (1.93) let us
first integrate it over one-fourth of the time period as the pendulum swings
from the highest point 0 to the zero value of at the vertical position
T /4
Z s Z0
l d
dt = p (1.95)
g 2
0 2
0 0
q
The integral being of the form l
g sin1 ( 0 ) gives
2 when evaluated over the
end-points. We are therefore led to the same result of T as obtained in (1.69).
In the next approximation it can be shown that
1 0
T = T0 (1 + sin2 ) (1.96)
4 2
where T0 is the value given by (1.69) (see Exercise 10).
The exact solution of (1.92) can be given by Jacobi elliptic functions.
Toward this end let us introduce two variables and defined by
0
= sin( ), = sin2 ( ) (0) = (1.97)
2 2
Since
d 1 d d 1 d
= cos , ( )2 = (1 2 )( )2 (1.98)
dt 2 2 dt dt 4 dt
(1.92) is readily converted to the form in terms of the variable

d 2 g 2
( ) = (1 2 )(1 ) (1.99)
dt l
Defining = gl t and =

one ends up with the differential equation

( 0 )2 = (1 2 )(1 2 ), 0<<1 (1.100)

where a prime denotes differentiation with respect to . The general solution

of (1.100) is one of Jacobi elliptic functions, namely, sn(; ) with (0) = 1
and 0 (0) = 0 and is the modulus ranging between 0 and 1. For the end
values of , the Jacobi elliptic function reduces to sin for = 0 and tanh
for = 1.
28 Advanced Classical Mechanics

1.9 Perturbative analysis and the quartic oscillator

The method of approximations in connection with the traditional pertur-
bation approach has a long history dating back to the times of Poincare in the
early part of the twentieth century. However, in the approximations of higher
orders (i.e., truncating the series solution by a finite number of terms) one runs
into difficulties because of the appearance of the secular or divergent terms.
For instance, consider the simple example of the differential equation (1.43)
of the damped oscillator. The exact solution for the underdamping case gov-
erned by < 2 is furnished in (1.45) where the integration constants are de-
termined from the initial conditions. If we look at as a small parameter that
could be treated perturbatively, then x(t) can be expanded in a series form

x(t) = x0 (t) + x1 (t) + 2 x2 (t) + ... (1.101)

For the zeroth-order and first-order of we obtain the following respective

equations:

0 (zeroth-order) : x0 + 02 x0 = 0 x0 = A cos (0 t + ) (1.102)

1
(first-order) : x1 + 02 x1
= 0 x0 = A02
sin (0 t + )
At
x1 = cos (t + ) (1.103)
2
Thus to O() we have the following approximate expression for x(t)
 
t
x = x0 + x1 = A 1 cos (t + ) (1.104)
2
When compared with the exact solution (1.45), a disturbing feature of
(1.104) is the presence of the term At2 cos (t + ) which, due to the attach-
ment of the factor t in the coefficient, behaves as a secular or diverging term.
It increases linearly with time. It is evident that such a perturbative treatment
will inevitably be confronted by the appearance of such terms that blow up
asymptotically. One way to avoid the secular term is to employ a power series
to represent the nonlinear frequency as a function of and then renormalize
the coefficients for its removal. Such a procedure of renormalization is known
as the LindstedtPoincare method.
Let us apply such an approach for the quartic oscillator case. We note
that the SHM deals with the case of an idealized spring when the system
is guided by a linear force term. When the system suffers large amplitude
oscillations there is indeed a deviation from such a situation. In effect this
amounts to considering the higher effects in the expansion of the potential
energy function given by (1.40) of which the quadratic term corresponds to
the linear restoring force. Requiring the potential to be still symmetric under
space inversion x x, the quartic term in the potential is the next choice.
 Conceptual basis of classical mechanics 29

It corresponds to a cubic force that naturally makes the underlying equation

of motion nonlinear:
x + 02 x = x3 (1.105)
(1.105) is called the equation of the Duffing oscillator. In the following we
assume > 0.
To proceed further we adopt a change in variable:

= $t + (1.106)

where is a constant. It transforms according to

x(t) z( ) z($t + ) (1.107)

As a result (1.105) takes the form

d2 z
$2 z + 02 z = z 3 , z = (1.108)
d 2
Writing down expansions of both z( ) and the nonlinear frequency $ in
the form of a power series we obtain the first order in

z( ) = z0 ( ) + z1 ( ) + O( 2 ) (1.109)

$ = $0 + $1 + O( 2 ) (1.110)
Setting, as is customary in the power series approach, $0 = 0 so that $2 =
02 +20 $1 +O( 2 ), we get from (1.108) the following sequence of equations:

0 (zeroth-order) : 0 2 (z0 + z0 ) = 0 (1.111)

1 (first order) : 0 2 (z1 + z1 ) = z03 20 $1 z0 (1.112)

etc.
The zeroth-order resembles a SHM-like equation. The solution of (1.111)
is of the form

z0 = a cos = a cos($t + ) (1.113)

where a is the amplitude. When the above solution is substituted in (1.112),
it gives
a3
z1 + z1 = 2
cos cos3 (1.114)
40 40 2
where the quantity stands for 3a3 8a0 $1 a 6= 0. A particular solution
of the above equation reads

a3
z1 = 2
cos + 2
cos 3 sin (1.115)
160 320 80 2
30 Advanced Classical Mechanics

Taken with (1.113), the general perturbative solution z( ) up to O() is given

by

a3
 

z( ) = a cos + cos + cos 3 sin (1.116)
160 2 320 2 80 2

where the last term is the growth term.

In the LindstedtPoincare approach, is set equal to zero and the elim-
ination of the secular term fixes $1 = 83 0 a2 . It implies the dependence of
the frequency on the amplitude. For the second-order, it is easily checked that
one encounters three types of divergent terms sin , sin 3 and 2 cos . The
first two terms vanish for $1 = 83 0 a2 and the second one for $2 = 256 3
0
a4 .
Thus, the elimination of the secular terms is done in each step of the power
series in a recursive manner and renormalized expressions derived. However,
a shortcoming of the perturbative method, which was developed mainly for
astronomical calculations, is the difficulty one faces with the convergence of
the power series employed although it must be said that such a disadvantage
is not serious for a physical problem. Actually, what happens is that higher
$i s produce higher powers in a, i.e., $1 a2 , $2 a4 etc., which means that
for a > 1 the series diverges and the expansion does not make sense. In other
words, the domain a > 1 cannot be covered within the canonical approach.

1.10 Rewriting Newtons second law in terms of kinetic

and potential energy in a conservative system
Newtons second law of motion gives the relationship between the force,
mass and acceleration. However, to put the law in practice it is often helpful
to translate it in terms of the kinetic and potential energies of the system.
For a configuration of N particles with masses m1 , m2 ,...mN situated at the
points whose Cartesian coordinates are given by the respective sets (x1 , y1 ,
z1 ),(x2 , y2 , z2 ),...,(xN , yN , zN ), the total kinetic energy has the explicit form

1 1
T = (m1 x1 2 + m2 x2 2 + ... + mN xN 2 ) + (m1 y1 2 + m2 y2 2 + ... + mN yN 2 )
2 2
1 2 2 2
+ (m1 z1 + m2 z2 + ... + mN zN ) (1.117)
2
Expressed as a sum, T reads
N
1X
T = (mi xi 2 + mi yi 2 + mi zi 2 ) (1.118)
2 i=1
 Conceptual basis of classical mechanics 31

On the other hand, in a conservative system, the force acting on the system
being derivable from a potential function V (x) as its negative gradient, we
can write for each component
V V V
Fx1 = , Fx2 = , ..., F xN = (1.119)
x1 x2 xN
V V V
Fy1 = , Fy2 = , ..., F yN = (1.120)
y1 y2 yN
V V V
Fz1 = , Fz2 = , ..., FzN = (1.121)
z1 z2 zN
In a short-hand notation these are simply
V V V
Fxi = , Fyi = , Fzi = , i = 1, 2, ..., N (1.122)
xi yi zi
Employing the Cartesian components of the force-potential relationship as in
above, Newtons laws for the x, y and z components can be put out in the
manner
V V V
mi xi = , mi yi = , mi zi = i = 1, 2, ..., N (1.123)
xi yi zi
From (1.118), since
T T T
= mi xi , = mi yi , = mi zi (1.124)
xi yi zi
we can equivalently express (1.123) in terms of a set of the following three
equations:
d T V d T V d T V
( )= , ( )= , ( )= i = 1, 2, ..., N
dt xi xi dt yi yi dt zi zi
(1.125)
In this way we have been able to express the Cartesian forms of Newtonian
equations of motion in terms of the energies of the system rather than the
forces themselves. To ensure a more compact representation of the above
equations it is instructive to introduce a Lagrangian function defined as the
difference of the kinetic energy (which is a function of velocities) and po-
tential energy (which is a function of position in a conservative system) i.e.,
L = T V . L is thus a mixed function of coordinates and velocities.
Through the use of L, Equation (1.126) acquires the following forms:
d L L d L L d L L
( )= , ( )= , ( )= i = 1, 2, ..., N (1.126)
dt xi xi dt yi yi dt zi zi
Note that like the Newtonian equations, Lagrangian formulation also yields a
set of second-order differential equations. The concept of a Lagrangian func-
tion is fundamental and taken as a reference point in the development of an
32 Advanced Classical Mechanics

analytical treatment of classical mechanics. Lagrangian equations can be de-

rived in a more general setting by making use of DAlemberts principle and
getting rid of the so-called unknown forces of constraints (like the tension force
in the simple pendulum problem which we identify as a constraint force) by in-
voking the principle of virtual work. This is the main advantage of Lagrangian
mechanics. Note that Newtons laws do not distinguish the constraint forces
as such. The avoidance of constraint forces and use of generalized coordinates
(that could have a wide range of options given the nature of the system we
are dealing with) are the key features in the Lagrangian description of things.
For instance, a generalized coordinate could refer to an angle or an arc length
apart from various other possibilities (the Cartesian or polar or cylindrical is
one of many such possibilities) of definition. This of course leads to the min-
imization of the number of coordinates depending on the number of degrees
of freedom that a physical system avails of.
Before concluding this section, let us revisit the problem of a simple pen-
dulum as a typical application of Lagrangian dynamics. From (1.73) we can
identify the kinetic energy and potential energy as
1 2 d 2
T = ml ( ) , V = mgl(1 cos ) (1.127)
2 dt
Here acts as the generalized coordinate and the Lagrangian function accord-
ing to the definition L = T V is
1 2 d 2
L= ml ( ) mgl(1 cos ) (1.128)
2 dt
Calculating the partial derivatives, namely,
L L
= mgl sin , = ml2 (1.129)

and substituting them in a similar equation like (1.126) but interpreted in
terms of the generalized coordinate , the familiar equation of motion (1.68)
for the simple pendulum straightforwardly emerges.
A detailed discussion of the Lagrangian formulation of classical mechanics
will be taken up in Chapter 3 of the book.

1.11 Summary
By enunciating Newtons three laws of motion we addressed various as-
pects of classical mechanics that are inherent in these laws such as the con-
cept of inertial frame, Galilean invariance and the homogeneous and isotropic
character of the Newtonian Universe. We discussed the process of seeking the
 Conceptual basis of classical mechanics 33

solutions of a physical system highlighting different coordinate frames that

could be suitably employed to describe it. We looked, in particular, at the
particular class of force called the conservative force which can be represented
by a potential energy function. Use of the latter facilitates a great deal of
simplification and its role was illustrated by turning to several problems that
are of great importance in classical mechanics: the simple harmonic motion,
the damped oscillator along with its various manifestations and the pendu-
lum problem. Subsequently, the formulation of conservation principles was
considered for the specific cases of linear momentum, angular momentum and
energy. A general form of the time period formula was derived in terms of Ja-
cobi elliptic function. We also introduced a treatment of perturbative analysis
and discussed LindstedtPoincares technique of avoiding the secular terms in
approximating a periodic solution of a differential equation focusing on the
specific case of the quartic oscillator. Finally, by the introduction of kinetic
and potential energies, we showed how a Lagrangian function can be defined
for a conservative system in a Cartesian frame. A detailed treatment of the
Lagrangian dynamics will be taken up in Chapter 3.

Exercises
1. For the Earth whose radius at the equator is R = 6400 kms and angular
velocity rotating on its axis given by = 7.3 105 rad/s, show that the
centripetal acceleration is 0.034 m/s2 .
2. A bird takes a spiral path to its nest in a manner that the radial distance
reduces at a constant rate r = a bt, while the angular speed increases
at apconstant rate = t. Show that the speed as a function of time is
v = b2 + (a bt)2 2 t2 .
3. Generalize the result (1.15) to the case of N -interacting particles.
4. Show that the expressions of the kinetic energy in spherical polar and
cylindrical coordinates are respectively given by
1
T = m(r2 + r2 2 + r2 sin2 2 ) (1.130)
2
1
T = m( 2 + 2 2 + z 2 ) (1.131)
2
5. Show that a particular solution of Laplaces equation 2 = 0 which is
independent of has a general solution

X
= (An rn + Bn rn1 )Pn (cos ) (1.132)
n=1

where Pn (cos ) are the usual Legendre polynomials. Discuss the case when
6= 0.
6. Identify the components of the momentum and angular momentum
which are conserved in the field of an infinite homogeneous plane, infinite
34 Advanced Classical Mechanics

homogeneous cylinder and infinite homogeneous semiplane which is bordered

by one of the axis.
7. For a system of N particlesPhaving an angular velocity , determine the
n
kinetic energy in the form T = i=1 12 mi |~ r~i |2 .
8. Show that in the case of the undamped oscillator the approximate ex-
pression of the total energy shows that it decays at half the rate at which
amplitude does.
9. Find the period of oscillation for the potentials (i) V (x) =
V0 cosh2 (kx), (ii) V (x) = |x|n and (iii) V (x) = V0 tan2 (kx).
10. If denotes the amplitude of oscillation, show for the pendulum prob-
lem that the first correction to the time period of oscillation is given by
s
l 2
T = 2 (1 + + O(4 )) (1.133)
g 16
3
[Hint: Set = gl and use the expansion sin = 6 to write (1.68)
p
2 2
as ddt2 + 2 ( 6 ) ' 0. Try a solution = cos t + 3 cos 3t to deduce
q
2
= 1 18 2 and = 192 . Hence derive the amplitude to this order of
approximation.]
11. A particle of mass m moving in the x direction has its kinetic energy
given by T = m 2
2 x and potential energy V (x). Show that if x = 0 is a position
of equlibrium, then for small values of x, the particle oscillates q harmonically
00
about the equilibrium point with an angular frequency = V m(0) .
12. Show that the equation for the damped linear oscillator can be ex-
pressed in a coupled equation form

x = y, y = x 2 y (1.134)

Interpret the function (x, y) = 12 (my 2 + kx2 ) where = m k

.
13. A particle of mass m moves along the positive x-axis. It is acted upon
by a constant force directed toward the origin with magnitude C and and in-
verse square law repulsive force with magnitude x2 . Find the potential energy
function U (x) and the equilibrium position of the particle.
14. Consider the quartic oscillator potential V (x) = 12 m 2 x2 + 4 x4 . Show
3a2
for small the time period is given by T = 2 (1 6 2 ) where a is the
amplitude of oscillation.
15. Consider the Lagrangian in the Cartesian coordinates
1
L= m(x2 + y 2 + z 2 ) mgz (1.135)
2
Write down the equations of motions and all the conserved quantities that
you can identify.
Chapter 2
Central force problems

2.1 Inertial and gravitational mass: Principle of equivalence . . . . . . . . 36

2.2 Derivation of Keplers three laws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.3 Properties and equations of orbits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.4 Integral representations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.5 A general class of power law potentials . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.6 Mapping the general class of potentials: Orbit equation for the
inverse square law problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.7 Coulomb and isotropic oscillator potentials . . . . . . . . . . . . . . . . . . . . . 45
2.8 LaplaceRungeLenz vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.9 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

The historical basis for central force problems comes from Keplers three laws
of planetary motion. These also go by the names of law of equal areas, law
of ellipse and law of harmonics. Kepler announced the first two laws in 1609
while the third was stated in 1619 to get an insight into the behavior of the
then known planets orbiting around the Sun. The statements of the three
Keplers laws are
1. Equal areas are swept out in equal intervals of time by the line segment
drawn from a planet to the Sun.
2. The locus of a planetary orbit is an ellipse with the Sun at one of the
foci.
3. The square of the periodic times of the planets is proportional to the
cubes of the major axes of their orbits.
But we owe to Newton for a fuller understanding of these laws as explained
in his Principia in 1687. Taking them in order, the law of equal areas focuses
on the speed of the planet: it is fastest when it is closest to the Sun while it
is slowest when it is most distant from the Sun. The law of ellipse points to
the elliptical path that a planet traces out while moving around the Sun with
the Sun in one of its foci. The law of harmonics provides a comparison of the
orbital period (T) and radius of the orbit (R) as compared to other planets
T2
in the family. Quite interestingly, the ratio R 3 turns out to be of nearly the

same value of unity for every planet.

In this chapter we will be interested in the problem of the central force in
which the force function depends only on the distance r from the origin thus
embodying the character of spherical symmetry. In some of the well-known

35
36 Advanced Classical Mechanics

examples of the central force like Newtons inverse square law of gravitation
and Coulombs electrostatic force between two charges, the force function F (r)
is proportional to the inverse square of the distance and is negative. However,
in the case of the spherical harmonic oscillator, F (r) is linearly proportional to
the distance and negative as well. By Bertrands theorem1 it can be established
that these two force functions are the only possible forms of the central force
fields having orbits that are all stable and closed. In the central force problem
the underlying force acts in a direction that is toward or away from a fixed
point called the force center. As such, the torque ~rF~ on the particle about
the force center vanishes resulting in the constancy of the angular momentum
~ = ~r(m~v ) = ~r~
p.

2.1 Inertial and gravitational mass: Principle

of equivalence
We already discussed, in the previous chapter, the concept of the inertial
mass. Let us compare the second law motion with Newtons inverse square
law of gravitation between a body of mass m at the position ~r and n different
others specified by their masses mi occupying positions ri , i = 1, 2, ..., n. The
force of attraction is given by
X GM mi (~r r~i )
F~ = , i = 1, 2, ..., n (2.1)
i
|~r r~i |3

where G is the gravitational constant approximately equal to 6.673

1011 N m2 kg 2 and measured in Newtons (N ), kilograms and meters.
What happens if we follow the second law of motion of Newtons to identify
the m in (2.1) as an inertial mass and write F~ as simply F~ = m~a ? In such a
situation m cancels out and we are left with an expression for the acceleration
of a particle in a gravitational field:
X mi (~r r~i )
~a = G , i = 1, 2, ..., n (2.2)
i
|~r r~i |3

To dig into this issue a little further, let us consider specifically two parti-
cles m1 and m2 in the presence of the gravitational field of a certain particle
of mass m. We assume that both m1 and m2 are positioned at the same dis-
tance r from m. In such a case, these particles would experience a force due to
1 The essence of Bertrands theorem is that the gravitational or electrostatic potential and

the radial oscillator potential are the only types belonging to central force problems with the
feature that all bound orbits are also closed. For a fuller discussion see E.T. Whittaker, A
Treatise on the Analytical Dynamics of Particles and Rigid Bodies, Cambridge University
Press, London (1988).
 Central force problems 37

the gravitational field of m. We distinguish the role of mass m, which exerts

a force, as an active gravitational mass and label it as ma while the receiver
masses m1 and m2 , which experience the force, as passive gravitational masses
and label them as m1p and m2p . From (2.1) the two forces are given by
ma m1p ma m2p
F1 = G , F2 = G (2.3)
r2 r2
In the context of Newtons second law of motion defining an inertial (I)
mass, given the above two forces, we could define the corresponding accelera-
tions a1 and a2 as given by
F1 F2
a1 = , a2 = (2.4)
M1I M2I
in terms of two inertial masses M1I and M2I .
At this point let us invoke the idea of the equivalence principle (weak)
which states that if objects are dropped from some height from say, the top
of a building, then so long as their motion is not affected by air resistance or
other disturbances, will arrive at about the same time on the ground i.e., they
fall at the same rate with equal accelerations. This curious feature is known
from Galileos time. (Mathematically, corresponding to ~g , i.e., the acceleration
due to gravity, assumed the same for particles near the Earths surface, the
force experienced by a particle of m is F~ = m~g . It is with this force that
a particle coming under the influence of Earths gravitational field will be
attracted toward the Earth).
We therefore put a1 = a2 and divide out the two relations (2.3) and (2.4).
It is straightforward to see that the active gravitational mass cancels out and
we get the ratio
m1p m2p
= (2.5)
M1I M2I
pointing to the proportionality between the passive and inertial masses.
In point of fact, the active and passive gravitational masses are also pro-
portional among themselves. This can be seen2 using Newtons third law of
motion. Consider two particles at points P1 and P2 at respective distances
r~1 and r~2 from a fixed point O (see Figure 2.1). Then the force F1 that the
particle at P1 exerts on P2 and the force F2 that the particle at P2 exerts on
P1 are, respectively,

(r~2 r~1 )
F1 = Gm2p m1a , (2.6)
|r~1 r~2 |3
(r~1 r~2 )
F2 = Gm1a m2p , (2.7)
|r~1 r~2 |3
2 We follow M.V. Berry, Principles of Cosmology and Gravitation, Cambridge University

Press, London (1989).

38 Advanced Classical Mechanics

All

All All

All
All

FIGURE 2.1: Two particles at points P1 and P2 with respect to the fixed
point O.

The directions of F1 and F2 are opposite in accordance with Newtons third

law but they have to be equal in magnitude as well. This requires the following
ratio to hold if we divide the above two relations:
m1p m1a
= (2.8)
m2p m2a
In other words, passive and active gravitational masses are proportional. In-
deed there is no difference between the gravitational mass and inertial mass
as experimental results have time and again shown. In his theory of General
Relativity, Einstein also did not distinguish between gravitational and inertial
mass. He asserted ...we shall (therefore) assume the complete physical equiv-
alence of a gravitational field and a corresponding acceleration of the reference
system. He put out his idea of the principle of equivalence in the following
way: We consider two systems 1 and 2 in motion. Let 1 be accelerated in
the direction of its X-axis, and let be the (temporally constant) magnitude
of that acceleration. 2 shall be at rest, but it shall be located in a homoge-
neous gravitational field that imparts to all objects an acceleration in the
direction of the X-axis. As far as we know, the physical laws with respect to
1 do not differ from those with respect to 2 ; this is based on the fact that
all bodies are equally accelerated in the gravitational field.

2.2 Derivation of Keplers three laws

We focus on the two-body system of the Sun and Earth whose respective
masses are given by say, M and m. An implication of the identification of m
as the inertial mass is that from (2.2) we can express ~a, for such a system, as
GM
~a = 3 ~r (2.9)
r
Clearly (2.9) is the outcome of the inverse square force law of gravitation.
 Central force problems 39
d~
r
r
d(~ )
Since dt
dt
= ~r ~a and ~a being given by above as directly proportional
to ~r , we are led to the vanishing of the time derivative of the quantity ~r d~
r
dt
implying that the latter is a constant vector C: ~

~
~ = lE
C (2.10)
m

where ~lE is the angular momentum of Earth about the Sun. Thus ~lE is a
conserved quantity. As such

d~r ~
~r ~ = lE
=C (2.11)
dt m
The sectorial area A being given by 12 |~r d~r|, it follows from (2.10) and (2.11)
that
dA ~
2 =C~ = | lE | (2.12)
dt m
In other words, equal areas are swept out in equal intervals of time which
is Keplers first law. That the motion lies on a plane is because ~r remains
perpendicular to ~lE :
d~r ~lE
~r [~r ] = ~r = 0. (2.13)
dt m
Next from (2.9) and (2.11) we can write

~
d(C d~
r
dt ) ~ ~a = GM d~r
=C ~r (~r ) (2.14)
dt r2 dt
Using the vector triple product formula

~ (B
A ~ C
~ = B(
~ A~ C)
~ C(
~ A~ B)
~ (2.15)
on integrating (2.14) we obtain

~ d~r = ( GM ~r + D)
C ~ (2.16)
dt r

where D ~ is some constant vector. Taking now a dot product with ~r, (2.16)
becomes
dr ~ ~
(~r ) C = GM r + ~r D (2.17)
dt
Since D~ is arbitrary, we can choose its direction to coincide with the polar
~ = rD cos . Using now (2.11) we easily derive
axis. Thus we can represent ~r D
the polar representation of a conic as given by
%
r= (2.18)
1 + e cos
40 Advanced Classical Mechanics
2
lE D
where the parameters %, e are % = GM m2 and e = GM . For a planetary orbit,
(2.18) is a closed curve such as an ellipse with the focus at the origin and e its
eccentricity. In this way we arrive at a mathematical form of Keplers second
law.
The points = 0 and = are the ends where (2.18) crosses the polar
axis. Adding the corresponding values of we determine the length 2a of the
major axis of the ellipse to be

1 1 C2
2a = %( + )=2 (2.19)
1+e 1e GM (1 e2 )

where % has been expressed in terms of C by using (2.10). Now the area of an
ellipse being
A = ab, where b is related to the length a of the semi-major axis
by b = a 1 e2 and, from (2.12), the period for one complete revolution to
be given by T = 2A
C , we obtain, on substituting for C from (2.19), the formula
r
a3
T = 2 (2.20)
GM
2
The result (2.20) has the implication that T
a3 behaves as a constant given by
4 2
GM which corresponds to Keplers third law.

2.3 Properties and equations of orbits

For a general force law F~ which is central, with the particle moving in a
plane, it acts in a direction which is toward or away from a fixed point called
the force center. As such, the torque ~r P~ on the particle about the force
center vanishes resulting in the constancy of the angular momentum ~l:
~l = ~r (m~r) (2.21)

The conservation of the angular momentum has the implication that the
orbital plane, given by the equation ~l ~r = 0, is the one on which the motion
is confined and is normal to the angular momentum vector ~l. Of course to
determine the motion completely we also need to define the initial position
vector ~r(0) and the initial velocity vector ~v (0) on such a plane.
Let us assume |~l| 6= 0. If (r, ) are the coordinates of the particle in the
orbital plane with respect to a fixed origin, then Newtons equations of motion
in plane polar coordinates read

m(r r2 ) = F (r), m(r + 2r) = 0 (2.22)

 Central force problems 41

The second equation integrates to

1 d
(mr2 ) = 0 (2.23)
r dt
pointing to the constancy of the angular momentum l = mr2 with respect
to time where = . In terms of , the radial equation becomes r = r 2 +
1
m F (r) and looks like a typical representative of a one-dimensional motion but
restricted on the half line 0 < r < .
It is often instructive to employ a change of variable r = u1 to effect the
transformations
l du l2 u2 d2 u
r = , r = 2 (2.24)
m d m d2
which cast the radial equation in the form

d2 u m
2
+ u = F (u1 ) 2 2 (2.25)
d l u
(2.25) provides the differential equation of the orbit for a particle under the
control of a central force of a general form P ( u1 ).
Sometimes it is useful to display the angular momentum variable explicitly.
In the presence of a potential V (r) given by F (r) = dV
dr , the radial equation
acquires the form

d l2 dU (l, r)
mr = [ + V (r)] mr = (2.26)
dr 2mr2 dr
and we see that the motion is influenced by an effective potential

l2
U (l, r) = + V (r) (2.27)
2mr2
The first term in the right-hand side of (2.27) is the effect of a centrifugal
l2
barrier whose magnitude is given by mr 3.

Integrating (2.26) with respect to time t, the total energy E that is constant
in time is given by

1 2 l2 1
E= m|~r| + + V (r) = m|~r |2 + U (l, r) (2.28)
2 2mr2 2
2
l
where 2mr 2 is the centrifugal term Vcf . For the Newtonian inverse square law,

V (r) = r (which is negative for an attractive force, > 0), a graphical

description of U (l, r) and Vcf is illustrated in Figure 2.2. We see that while
Vcf goes as r12 , V (r) is always negative. As a result, U (l, r) can descend to a
minimum having a finite negative value.
42 Advanced Classical Mechanics

All

All

AllAll

All

FIGURE 2.2: A graphical description of the function U (l, r) against Vef and
V (r).

2.4 Integral representations

In terms of the angular momentum parameter l, the energy formula (2.28)
can be easily represented in a derivative form
r
dr mr2 2
= [E U (l, r)] (2.29)
d l m
The corresponding representations for u are
r
du 2m 1 l 2 u2
= EV( ) (2.30)
d l u 2m
The choice of signs is guided by the initial conditions that are relevant to the
system. Before we turn to the evaluation of the above integrals let us first
consider a general class of power law potentials.

2.5 A general class of power law potentials

The entire class of power-law potentials can be described by the general
form3

V (r) = = 2+2 , 6= 0, 6= 0 (2.31)
r r
3 J. Daboul and M. M. Nieto, Quantum bound states with zero binding energy, Phys.

Lett. A 190 (1994) 357.

 Central force problems 43

where = 12 ( 2). The above potential is attractive for > 0 but repulsive
for < 0. Corresponding to (2.31), U (l, r) is given by

l2
U (l, r) = (2.32)
2mr2 r
1/(2)
l2
  
1 1 2m
i.e. U () = , a= (2.33)
2ma2 2 l2

where = ar and a is determined from the condition Up (l, a) = 0.

In terms of a, V (r) assumes the form

l2 a2 1
V (r) = (2.34)
2m r
We speak of a bounded motion if r = rmin and r = rmax exist where r =
0. From (2.28), since E = 12 mr2 + U (l, r) for all r, we have the condition
U (l, r) E for a physical bounded motion. In Figure 2.2, the curve U (l, r) is
seen to descend to a minimum with a finite negative value implying a range
of bounded orbits.
Among the orbits the circular ones are, from the mathematical point of
view, the simplest to pick. For a particle in circular motion of radius r = r
2
with a constant acceleration vr toward its center, v being the magnitude of its
velocity given by v = r, it is clear that the following balancing equation holds:

F (r) + mr2 = 0
l2
i.e. F (r) + = 0 (2.35)
mr3
where F (r) is the force at r = r.
For the stability of the circular orbit we enquire for a minimum of the
effective potential U (l, r) and hence for the following two conditions to hold:

dU (l, r) d2 U (l, r)
U 0 (l, r) = 0, U 00 (l, r) >0 (2.36)
dr dr2
at r = r. Note that (2.35) is also consistent with seeking a derivative of U (l, r)
with respect to r and putting r = r remembering that the force F (r) is given
by the negative gradient of the potential.
Writing
d2
mr = m 2 (r r) = U 0 (l, r) (2.37)
dt
dU (l,r)
where U 0 (l, r) = dr , let us expand the right-hand side as

U 0 (l, r) U 0 (l, r) + (r r)U 00 (l, r)

= (r r)U 00 (l, r)
44 Advanced Classical Mechanics

This gives
d2
m (r r) (r r)U 00 (l, r) (2.38)
dt2
which we recognize as the familiar equation of the undamped oscillator, im-
plying stability of the circular orbit for U 00 (l, r) > 0. Since

l2 3l2
U 0 (l, r) =
3
+ V 0 (r), U 00 (l, r) = + V 00 (r) (2.39)
mr mr4
we then have a small amplitude oscillation with an angular frequency given
by
r r
U 00 (l, r) V 00 (r) + 3l2 /mr4
= = (2.40)
m m
If we eliminate l from the condition U 0 (l, r) = 0 then we have another repre-
sentation for
s 
3V 0 (r)

1
= V 00 (r) + (2.41)
m r
The period of the orbit is given by
2
T =r h i (2.42)
1 00 (r) 3V 0 (r)
m V + r

As an application let us examine the stability condition for the circular

orbits of the entire class of power law potentials as given by (2.31).
In such a context the force corresponding to V (r) must be attractive, i.e.,

dV
F = < 0.
dr
and implies r1 < 0. So we must have > 0.
The condition of stability of a circular orbit as just deduced is U 00 (l, r) > 0.
0
From (2.40) and (2.41) it gives V 00 (r) + 3V r(r) > 0. As a consequence we work
out the condition
3
( + 1)r2 + ()r1 > 0
r
or, [( + 1) 3] > 0
or, ( 2) > 0
Since > 0 we are led to the condition < 2.
 Central force problems 45

2.6 Mapping the general class of potentials: Orbit

equation for the inverse square law problem
We start with the general case of the power law potentials as given by
(2.31). We deduce4 from (2.29) the corresponding integral form for (r):
Z
dr
(r) = q (2.43)
r2 2m l 2 (E + r ) 1
r 2

Making a change of variable by setting u = r in (2.31) where is a real

and arbitrary constant gives dr = (u1+ )du. As a result the integral is
transformed to
Z
du
(u) = q (2.44)
2m
u2 l2 (Eu 2+2
+ u2+2 )
1
u2

in terms of the u-variable. Hence to put (2.44) into a correspondence with

(2.30), we have to assume that the constants and satisfy the constraint

=2
2 (2.45)

Note that it can also be re-expressed in an equivalent factorized form (2
)( + 2) = 4. As a result (2.44) assumes the representation
Z
du
(u) = q (2.46)
u2 l2 (Eu + ) u12
2m

The above form is concerned with the power law potential u . Up to certain
adjustments of the constants it compares interestingly with the representation
(2.43) pointing to a kind of duality between the potentials r and u .

2.7 Coulomb and isotropic oscillator potentials

The orbit equation for the Coulomb potential, V (r) = r , which cor-
responds to = 1 in (2.31), is straightforwardly determined by integrating
(2.26). We can equivalently integrate (2.25). It gives the following differential
equation for u:
d2 u m
2
+u= 2 (2.47)
d l
4 A. K. Grant and J. L. Rosner, Classical orbits in power-law potentials Am. J. Phys.

62(1994)310.
46 Advanced Classical Mechanics

Its general solution is given by

m
u = A cos( 0 ) + (2.48)
l2
in terms of two constants A and 0 . Without loss of generality we can set
0 = 0 by assuming it to be measured from the pericentron (or pericenter).
Pericentron is the term used for the shorter apsis while apocentron (or apoc-
enter) stands for the longer apsis while dealing with the motion of orbits.
dr
Apsis is the point where d = 0.
Inverting u in terms of r yields an equation of a conic in the standard polar
form
l2 1
r= (2.49)
m 1 + e cos
where the parameter e stands for the eccentricity of the orbit. For the peri-
centron ( = 0) and apocentron ( = ) we have, respectively, the relations

l2 1 l2 1
rper = , rapo = (2.50)
m 1 + e m 1 e
Different cases of orbits arise depending on the various possibilities of e.
Before we distinguish them it is worthwhile to have a look at the energy
equation (2.28). For the Coulomb potential the expression for E reads

1 2 l2
E= m|~r| + 2
(2.51)
2 2mr r
and implies an effective potential

l2
U (l, r) = (2.52)
2mr2 r
It gives for the rate of change of radial velocity
r
dr 2
= (E U (l, r)) (2.53)
dt m
We therefore conclude that the maximum radial velocity is attained when
U (l, r) is a minimum.
Writing dr dr d
dt = d dt , one can transform E to a form

l2 1 dr m
E= [ ( )2 + 1 2 2 r] (2.54)
2mr2 r2 d l
dr
Evaluating d from (2.49) which gives

dr l2 e sin
= (2.55)
d m (1 + e cos )2
 Central force problems 47

and using the explicit form for r, a little calculation yields a rather simple
representation for E:
m 2 2
E= (e 1) (2.56)
2l2
Notice no -dependence in (2.56) which is as it should be because energy
is conserved for the system under consideration. The interplay between the
eccentricity e and the energy is thus evident in determining the character of
the conic. We summarize the different possibilities for the orbits:
Ellipse (bounded or trapped orbits) : e < 1 = E < 0,
Circle (special case of an ellipse) : e = 0 = E < 0,
Hyperbola (unbounded orbits) : e > 1 = E > 0,
Parabola (open orbit) : e = 1 = E = 0.

Let us consider them in turn. To this end it is often useful5 to transform

(2.49) to Cartesian coordinates. With this aim in mind we put x = r cos , y =
sin which for the ellipse results in the standard canonical form encountered
in coordinate geometry
l2 e 2
(x + m 1e2 ) y2
+ =1 (2.57)
a2 b2
l2 l2
( )2 ( )2
where a2 = (1em 2 m
2 )2 and b = (1e2 ) . Since for the ellipse e < 1, it follows that

b2 < a2 . Note that the geometric center of the closed orbit of the ellipse is not
l2 e
at the origin but shifted to the point x = m 1e2 . We have here a periodic
motion in which the radial distance moves between rapo and rper . It can be
shown that one of the foci is at the origin (see Exercise 3).
From (2.56) we can read off for the range of the energy

m 2
< E < 0. (2.58)
2l2
Note that the circle is an extreme case of the ellipse to which the latter degen-
2
erates to when e = 0. Here E = m 2l2 . In the case of gravitation the circular
orbit is due to a balance between the centrifugal force and the gravitational
force. The radial velocity has a zero-value and there is only the transverse
circular velocity having the value a , where a is the semi-major axis of the
p

ellipse. See Exercise 4.

Unlike an ellipse, the hyperbola describes the case of an open orbit. Its
equation reads
l2 e 2
(x m e2 1 ) y2
+ =1 (2.59)
a2 b2
5 D. Tong, Lecture notes on dynamics, Cambridge lecture notes (unpublished).
48 Advanced Classical Mechanics
l2 l2
( )2 ( )2
where a2 = (e2m 2 m
1)2 and b = (e2 1) . The eccentricity value being greater than
unity implies E > 0.
Finally for the parabola the Cartesian form of the equation is given by
l2 l2 2
y 2 = 2 x+( ) (2.60)
m m
With the eccentricity value of unity it corresponds to zero-energy orbits. A
parabolic orbit is also open like the hyperbolic one.
We next address the three-dimensional oscillator having the same fre-
quency in every direction. Such an oscillator is called the isotropic oscillator
and is guided by the potential
1
V (r) = m 2 r2 (2.61)
2
It induces an effective potential
1 l2
U (l, r) = m 2 r2 + (2.62)
2 2mr2
The minimum of U (l, r) is estimated in the usual way. It turns out that the
lowest point of U (l, r) is reached at
|l|
rmin = (2.63)
m
2
where d Udr(l,r)
2 > 0, rmin is the point of minimum. The particle has a natural
tendency to be stable at such a point. The energy at r = rmin is Emin = 12 |l|.
All other values of energy are double-valued and greater than Emin . These
values could be solved from (2.28) by putting r2 = t yielding the equation
m2 2 t2 2mEt + l2 = 0 (2.64)
The two solutions of t for such a quadratic equation are
s r s r
E 2
Emin E 2
Emin
t+ = (1 1 ), t = (1 + 1 ) (2.65)
m 2 E2 m 2 E2
To derive an orbit equation for the isotropic oscillator we can employ either
(2.29) or (2.30). We find it preferable to use (2.30).
Note that for the isotropic oscillator potential we can express
l2 u2 E2 lu2
r
1 1 1 2 E m 2
EV( ) = 2 [(m 2 m ) ( ) ] (2.66)
u 2m u 2l 2 2m l 2
2
E 1 2
Setting for the quantity m 2l 2 2 m = 2 and making a change of variable
2
lu E
pm
w = 2m - l 2 , the integral in (2.30) can be transformed to
Z
1 1 1 w
= p dw = arccos( ) (2.67)
2 2
w 2 2
 Central force problems 49

up to an additive constant. With w = cos 2 we get for r2 the expression

l
2m 1
r2 = q = q (2.68)
E mE 2 l2 2
pm 1 2 mE
l 2 + 2l2 2 m cos 2 l2 (1 + 1 E2 cos 2)

which corresponds to the polar equation of an ellipse having its center at the
origin and semi-axes given by the pair (t+ , t ) shown in (2.65).

2.8 LaplaceRungeLenz vector

While the angular momentum and energy are natural candidates for con-
served quantities in a Keplerian system, the latter also incorporates an addi-
tional hidden symmetry that goes by the name of LaplaceRungeLenz (LRL)
vector. An interesting history on the LRL vector can be found in the papers
of Goldstein and Leach and Flessas.6 Here we simply note that the LRL vec-
tor came into notability from Paulis work7 on hydrogen atoms in which such
a vector was exploited to derive the latters energy levels employing certain
quantum mechanical calculations. Pauli used Lenzs name in such a context
although it must be said, as Goldstein8 pointed out, that the presence of
LRL-vector was implicit in the 1845 landmark paper of Hamilton entitled
Applications of Quaternions to Some Dynamical Quantum.
d~l
Since ~l is a constant of motion, i.e., dt = 0, we can express the product
m~r ~l as
d
m~r ~l = (~p ~l), p~ = m~r (2.69)
dt
Thus from the radial equation of motion m~r = F (r) ~rr and using ~l = ~r p~ we
find

d ~r
p ~l) = F (r) (~r p~)
(~ (2.70)
dt r
Now using the standard formula for a vector triple product the right-hand
side turns out to be

F (r)
=m [~r(~r ~r) r2~r ] (2.71)
r
6 P.G.L. Leach and G.P. Flessas, Generalizations of LaplaceRungeLenz vector, J. Non.

Math. Phys. 10(2003) 340.

7 W. Pauli, Uber das Wasserstoffspektrum vom Standpunkt der neuen Quantenmechanik,

Z. Phys. 36 (1926) 336.

8 H. Goldstein, More on the prehistory of the Laplace or RungeLenz Vector, Am. J. Phys.

44 (1976) 1123.
50 Advanced Classical Mechanics

Now the product ~r ~r being simply rr, we obtain

d
p ~l) = mF (r)[r~r r~r ]
(~ (2.72)
dt
where unfortunately the right-hand side cannot be written as a total derivative
for a general form of F (r). The best we can achieve is to express (2.72) as

d d ~r
p ~l) = mF (r)r2 ( )
(~ (2.73)
dt dt r
The non-integrability is evidently due to the presence of an arbitrary force
term F (r) in the right-hand side of (2.73). However, for the specific case of
a repulsive force which varies inversely as the square of the distance as given
by F (r) = r2 , the system is rendered solvable. In fact, the above equation
reduces to a total derivative form, namely,

d ~r
p ~l m ) = 0
(~ (2.74)
dt r
~
implying conservation of a vector A defined by

~ = p~ ~l m~r
A (2.75)
~ is called the LRL vector. In particular if ~l points to the z-direction desig-
A
~ in the xy-plane are
nated as lz , then the components of A
x y
Ax = py lz m , Ay = px lz m (2.76)
r r
where px and py are the components of the momentum in the xy-plane.
A couple of observations are in order:
(i) From the form of A ~ ~l = 0. Hence A
~ it follows that A ~ is orthogonal to
~l pointing that ~l lies in the orbital plane.
(ii) We can express the energy E explicitly in terms of the magnitudes of
the A~ and ~l vectors. To this end we square the expression of A ~ to write

A2 = |~p ~l m~r|2 = p2 l2 + m2 2 2m l2 (2.77)
r
where we have used the vector identities (~a ~b)(~c d) ~ = (~b d)(~
~ a ~c)(~b~c)(~a d)
~
~ ~
and ~a (b ~c) = (~a b) ~c. Noticing that for our inverse square law for the
p2
force implies a potential V (r) = r , we have the total energy as E = 2m r
2
which can be reexpressed in terms of A as

A2 m2 2
E= (2.78)
2ml2
This gives the relation between the energy and the square of the magnitude
of the LRL vector.
 Central force problems 51

2.9 Summary
In this chapter we looked at the general class of central force problems
noting the distinction between inertial and gravitational mass and the related
issue of the principle of equivalence. We briefly provided a derivation of Ke-
plers three laws and then gave a discussion of the properties and equations
of orbits. A study of the general class of power law potentials was our next
point of inquiry in which we examined, in particular, the stability condition
of the entire class of power law potentials and mapping of the general class
of potentials. The specific cases of the Coulomb and isotropic potentials were
illustrated and finally a treatment of LaplaceRungeLenz vector was given.

Exercises
1. Demonstrate the equivalence of the integral (2.46) with (2.44).
2. Show from (2.43) that the case = 2 leads to orbits for a Coulomb
potential. Also address the dual case of the oscillator potential.
3. Show for the ellipse equation as written down in (2.57) one of the foci
is at its origin.
4. Determine the transverse velocity of the circular orbit which results due
to a balance between the centrifugal and the gravitational forces.
p ~l) = (~r p~) ~l = l2 , show that the
5. Using the vector identity ~r (~
expression (2.75) represents a conic by taking a dot product with ~r.
Chapter 3
Lagrangian formulation in mechanics

3.1 Constraints and generalized coordinates . . . . . . . . . . . . . . . . . . . . . . . . . 54

3.2 Formulation of DAlemberts principle . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.3 Kinetic energy of a holonomic system . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.4 Lagranges equations of motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.5 Lagranges equations for some simple systems . . . . . . . . . . . . . . . . . . 66
3.5.1 Plane pendulum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.5.2 Spherical pendulum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.5.3 Binary star system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3.5.4 A system with four degrees of freedom . . . . . . . . . . . . . . . . . . 69
3.5.5 The problem of a damped oscillator . . . . . . . . . . . . . . . . . . . . 70
3.5.6 A conservative scleronomic system . . . . . . . . . . . . . . . . . . . . . . 71
3.6 Ignorable coordinates: Rouths procedure of solution . . . . . . . . . . . 72
3.7 Liouvilles class of Lagrangians . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
3.8 Small oscillations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
3.9 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

In this chapter we take up the mathematical formulation of the Lagrangian

approach to classical mechanics that provides a basis for its analytical foun-
dation. The main aim is to do away with the forces of constraint which could
be present in the Newtonian equations of motion. These are unknown quan-
tities there and require to be solved for. The presence of constraints limits
the freedom of the system and reduces its degrees of freedom to describe
the motion. The independent coordinates (not necessarily Cartesian) account-
ing for the number of the degrees of freedom are termed generalized coordi-
nates. By invoking the principle of virtual work and making a judicious use
of DAlemberts principle, a completely new form of the equations of motion
emerges which replaces the Newtonian ones by a set of equations called La-
granges equations. In place of force and acceleration, these equations are de-
pendent on the kinetic and potential energies of the system. In the Lagrangian
scheme of things we do not have to worry about the forces of constraints. Of
course, no new physics emerges but the Lagrangian approach gives a funda-
mental starting point toward building up of a solid analytical structure of
classical mechanics which proves very powerful in applications to a wide vari-
ety of problems. In the following we begin with the constraints and the idea
of generalized coordinates.

53
54 Advanced Classical Mechanics

3.1 Constraints and generalized coordinates

For a physical system consisting of a finite number of particles moving
under the action of a prescribed force, we usually find that the particles have
their coordinates restricted in one way or another. We can think of numerous
examples in this regard. The simplest one that comes to mind is the case of
a simple plane pendulum. As already discussed in Chapter 1, such a system
is composed of a bob of mass m attached to a massless, inextensible string of
length l, the other end of the string being suspended from a pivot, without
friction. Its position can be described by the angle that the string makes
with the vertical line through the support or the horizontal component of the
displacement of the bob. We say that the degree of freedom for the plane
pendulum is one.
There are systems that require two or more independent coordinates for the
description of their motion. These are called muliti-degree-of-freedom systems.
For instance, if we replace the plane pendulum by taking some flexible string
or if we attach a string to a rigid body at a point different from its center of
mass, then we encounter a system with two degrees of freedom. This means
that we need two independent coordinates to fully understand the motion of
the system.
Consider the case of a rigid body which is a material body composed of an
aggregate of particles, the relative distance between every pair of points in it
is invariable. A rigid body does not yield to any deformation of shape arising
out of expansion or contraction although it can move from one position to
another as a whole as observed by an external observer. It may be interpreted
as a continuous distribution and that we can arbitrarily subdivide it into
small parts thus facilitating the use of laws of motion and integral calculus
to understand their changes of position. The number of degrees of freedom of
a rigid body in space is estimated easily. First of all, its orientation is fully
determined when any three points of it are known. Now a point in space can be
represented by three coordinates implying a total of nine coordinates for these
three points of the rigid body. Next, the three lines joining the three points
have to have their lengths invariable since the mutual distances of the points
in a rigid body remain fixed. This puts three restrictions. Hence, a rigid body
can have six degrees of freedom. Looked at in a different way, the six degrees
of freedom account for the fact that a rigid body can have three translations
along with three rotary motions. The degree of freedom is therefore defined as
the number of independent coordinates (excluding the time variable) required
to completely specify the position and configuration of the system.
Let us now turn to a mechanical system composed of N particles. Newtons
equation of motion for the ith particle is given by

mi r~i = F~i , i = 1, 2, ..., N (3.1)

 Lagrangian formulation in mechanics 55

Since each particle can be specified by three coordinates, we have at hand 3N

coordinates which are subjected to, say, k( 3N ) equations of constraints.
The number of degrees of freedom is then given by n = 3N k.
A constraint is some kind of a restriction on the motion of a particle. The
forces responsible for the restriction are called the forces of constraint. The
forces of constraint are initially unknown and require to be solved for. For
instance, the tension of the string in the plane pendulum problem is the force
of constraint which is to be determined by solving the equations of motion. If
the forces other than the forces of constraint are designated as applied forces
then F~ appearing in (3.1) can be split up in the manner
(a) (c)
mi r~i = F~i + F~i , i = 1, 2, ..., N (3.2)
(a) (c)
where F~i are the applied forces and F~i are the constraint forces.
Generally, a constraint is of the form

~ (r~j , v~j , t) = 0, j = 1, 2, .., N (3.3)

where v~j ~rj are the velocities. In the absence of velocities, (3.3) is called a
finite or geometric constraint:

~ (r~j , t) = 0, j = 1, 2, .., N (3.4)

Otherwise, (3.3) is a differential or kinematical constraint.

A particular class of differential constraints is of the linear form
N
X
a~j .v~j + F = 0 (3.5)
j=1

where the coefficients a~j are not all vanishing and F is a scalar function of ~r
and t. In the stationary case, t~ = 0 in (3.4) while in (3.5) F = 0 and a~j are
functions of position only. A system is called scleronomic if it is subjected to
only stationary constraints; otherwise, it is called rheonomic.
Apart from the type (3.3) which are called bilateral constraints, there can
be unilateral constraints which appear as inequalities.

~ (r~j , v~j , t) 0
(3.6)

As an example of (3.6), one can think of a volume of gas confined in a box

of lengths a, b and c. Then the motion of the gas particles is restricted to the
regions 0 x a, 0 y b and 0 z c.
Constraints may be further distinguished into two classes. We can have
the ones that are expressible as an algebraic equation involving the coordi-
nates and time but independent of velocities. Such constraints are holonomic
constraints. The corresponding system is a holonomic system. On the other
hand, there can exist a larger class of constraints which are nonintegrable and
56 Advanced Classical Mechanics

certainly not reducible to the holonomic types. Such constraints are therefore
nonholonomic and the system subjected to them is a nonholonomic system.

Some examples given below will help us to clarify the above issues.

Example 3.1

For the simple plane pendulum problem the equations of constraints are
x2 + z 2 = l2 and y = 0 where l is the length of the string and the pendulum
bob is restricted to swing in the xzplane.
Here the constraint is holonomic and scleronomic. If, however, the length
of the pendulum changes with time due to seasonal effects, then we have a
time-dependent rheonomic constraint.

Example 3.2

Suppose a particle moving in a three-dimensional space has its motion

restricted to a circle whose radius changes with time. The constraint has the
form:
xdx + ydy + zdz = r(t)dt
In general the right-hand side is not integrable in a closed form and hence it
is a case of a nonholonomic constraint.

Example 3.3

A velocity-dependent constraint has a typical form Ax + B y + C z = 0.

This of course can be reexpressed as

Adx + Bdy + Cdz = 0

If the left-hand side is an exact differential, i.e., df , where f f (x, y, z), a sim-
ple integration converts it into a holonomic constraint. A sufficient condition
for the integrability is
f f f
A= , B= , C=
x y z
For the above criterion to hold the necessary and sufficient conditions are
given by the set of equalities
A B B C C A
= , = , =
y x z y x z
Otherwise, the system under consideration is a nonholonomic one.
 Lagrangian formulation in mechanics 57

All

All
All

All

FIGURE 3.1: The rolling disc.

Example 3.4

We remarked earlier that the volume of a gas confined in a box of lengths

a, b, c is subjected to the unilateral constraints 0 x a, 0 y b, 0
z c. Here the constraints are obviously nonholonomic.

Example 3.5

Let us consider a rolling disc problem. With the angles defined as shown
in Figure 3.1, we can write down the constraint equations

x sin y cos = 0
x cos y sin = r
The first one speaks of no lateral motion while the second one points to
pure rolling. In differential terms these are dx = r cos d, dy = r sin d
which are nonintegrable. We conclude that the problem is nonholonomic.
Once the constraint equations have been correctly identified it proves useful
to set up a set of n independent coordinates, qi (i = 1, 2, .., n), called gener-
alized coordinates, to describe the configuration of a physical system. These
generalized coordinates, whose total number equals the number of degrees
of freedom available for the system, are quite general in character and need
not always conform to any special type like the Cartesian or polar or say,
parabolic coordinates. For the plane pendulum problem, where the constraints
are x2 + z 2 = l2 and y = 0, any one of x or z or (the angle which the string
makes with the vertical) may serve as the generalized coordinate.
There is no general rule for adopting of a particular set of generalized
coordinates. As to which one needs to be employed depends a great deal upon
an educated guess and also upon the conditions of a problem.
58 Advanced Classical Mechanics

In the following section let us derive an expression for the kinetic energy
in terms of the generalized coordinates q1 , q2 , ..., qn .

3.2 Formulation of DAlemberts principle

Before embarking upon the concept of virtual work, let us consider some
typical situations when the forces of constraint do no work.
First consider the case of a simple plane pendulum whose length l is con-
stant. It is obvious that as it swings, the bob traces out an arc of a circle in
the xzplane due to the constraint x2 + z 2 = l2 . The displacement of the bob
is normal to the direction of the force of constraint, namely, the tension T ,
which acts along the length of the string. Hence, work done by the force of
constraint is zero.
Next consider the sliding of an object down a frictionless inclined plane.
Apart from the force of gravity, which is the only applied force acting upon
the object, there is the normal reaction of the plane which is the force of
constraint. The latter acts perpendicular to the direction toward which the
object slides. It is therefore clear that such a force of constraint does no work.
Now take the case of a rigid body. Work done to move the ith particle of
the rigid body is X
Wi = F~ij .d~
ri (i 6= j) (3.7)
j

where d~ri stands for the displacement and F~ij represents the constraint force
on the ith particle due to the j th particle and we ignore the self-forces.
To consider all the particles in the rigid body, we need to sum over i in
(3.7) and get for the total work done
X XX
W = Wi = F~ij .d~
ri (3.8)
i i j

Interchanging i and j in (3.8) and utilizing Newtons third law,

F~ij = F~ji

we arrive at the following form for W

1 XX ~
W = ri dr~j )
Fij .(d~ (3.9)
2 i j

For a rigid body since the inter-particle distances are fixed we have (~ ri
r~j )2 = constant. Taking differential it yields (~ri r~j ).(d~
ri dr~j ) = 0. From
Newtons third law, F~ij acts along the relative vector r~i r~j which stands for
 Lagrangian formulation in mechanics 59

the direction along the line joining the particles i and j. It is thus implied
from (3.9) that W = 0. Hence, the total work done by the forces of constraint
in a rigid body is zero.
In all the above examples we found that the total work done by the force(s)
of constraint is zero. However, the forces of constraint can do work in certain
situations like the following one. Consider the case of the simple pendulum
whose length l is changing with time: l = l(t). Here it is obvious that the bob
traces out a different route than the usual circular arc. The displacement of
the bob is therefore not normal to the direction of T , the force of constraint.
So the work done by T is non-zero.
How to treat to time-independent as well as time-varying constraints in a
consistent framework ? Fortunately we have a way out. We invoke the concept
of an infinitesimal virtual displacement which is consistent with the forces and
constraints operating on the system at a given instant of time. This leads to
the corresponding definition of virtual work. Here the word virtual has the
underlying meaning that no passage of real time is involved during the dis-
placements taken. In other words, we freeze the system at a certain point
of time and think of virtual displacements r~i (i = 1, 2, ..., ) that are consis-
tent with the conditions of the constraints. Some arbitrariness is, of course,
involved in the choice of the direction of r~i . This is exploited by the principle
of virtual work which we state below:
c
The total virtual work done by forces of constraint F~i is zero for any vir-
tual displacement.

Mathematically it means
N
X
W c = F~jc . r~j = 0 (3.10)
j=1

Let us take a dot product of both sides of the equation of motion (3.2) by the
virtual quantity r~i . It gives
N
X N
X N
X
mi r~i . r~i = F~ia . r~i + F~ic . r~i (3.11)
i=1 i=1 i=1

By virtue of the postulate (3.10) the second term in the right-hand side drops
out and we are left with the sum
N
X
(F~ia mi r~i ). r~i = 0 (3.12)
i=1

The expression (3.12) summarizes DAlemberts principle:1 in it the con-

straint forces do not appear. Perhaps the biggest success of DAlemberts
1 Jean DAlembert published in 1743 the principle that states that a combination of

reversed effective forces that act on each particle of the body and external forces keep the
system in equilibrium. The form as given by (3.12) was given by Joseph Lagrange later.
60 Advanced Classical Mechanics

principle is its ability to get rid of the constraint forces. In the absence of
any constraint, one can look upon r~i as independent which means that their
coefficients have to vanish in (3.12). The resultant equations are nothing but
Newtons equations of motion for unconstrained motion. However, if the mo-
tion is a constrained one, as is indeed so in the presence of the forces of
constraint, then r~i are not independent and we cannot have the coefficients
to be zero. We then have to deal with the single equation (3.12) whose nature
can be very complicated.
As an application of DAlemberts principle let us again focus attention
on the plane pendulum problem. The generalized coordinate is chosen as .
So the work done by the applied force mg is (mg sin )(l) where is the
virtual displacement that undergoes. The acceleration of the bob is l. We
thus have from (3.12)

(mg)(l sin ) ml(l) = 0 (3.13)

 g g
or = sin = (3.14)
l l
for small . This is the usual equation of the simple harmonic motion.
Next suppose that the length of the string is changing with time. Here
the only change from (3.12) is that the pendulum bob has a component of
acceleration (l + 2l) in the direction. Hence the work done by the inertial
force is m(l + 2l)l. So (3.12) is modified to the form

(mg)(l sin ) m(l + 2l)(l) = 0

yielding
d
(ml2 ) = mgl sin (3.15)
dt
The left-hand side of (3.15) is the rate of change of the angular momentum of
the bob about the point of support which, in the absence of gravity, remains
constant even when the length of the pendulum is changing with time.

3.3 Kinetic energy of a holonomic system

A constraint-free system of N -particles has 3N coordinates and hence 3N
degrees of freedom. Let us impose k holonomic constraints upon such a system.
The degrees of freedom get reduced to 3N k which now stand for the number
of independent coordinates. We thus represent the 3N coordinates as functions
of n = 3N k independent generalized coordinates as follows:

xi = xi (q1 , q2 , ..., qn ; t), yi = yi (q1 , q2 , ..., qn ; t), zi = zi (q1 , q2 , ..., qn ; t)

(3.16)
 Lagrangian formulation in mechanics 61

These equations are equivalent to the vector form

r~j = r~j (q1 , q2 , ..., qn ; t), j = 1, 2, ..., n (3.17)

By definition, the kinetic energy of a system of n particles is

n
1X
T = mj |r~j |2 (3.18)
2 j=1

where r~j is given by

X r~j r~j
r~j = qi + (3.19)
qi t
Substituting the expression (3.19) into (3.18) we are led to the following rep-
resentation for T
n n
1 X X
T = aik qi qk + ai qi + a0 (3.20)
2 i=1
i,k=1

where the coefficients aik , ai , a0 are

n
X r~j r~j
aik = mj .
j=1
qi qk
n
X r~j r~j
ai = mj .
j=1
qi t
n  2
1X r~j
a0 = mj (3.21)
2 j=1 t

and i, k = 1, 2, ..., n.
An interesting off-shot of (3.21) is that, in the scleronomic case, both the
coefficients ai and a0 drop out and we are left with a homogeneous function of
the second degree of the generalized velocities for the kinetic energy function:
n
1 X
T = aik qi qk T2 (say) (3.22)
2
i,k=1

Actually T2 can be shown to be always degenerate:

det(aik )ni,k=1 6= 0 (3.23)

For if the above determinant were to vanish, we would be faced with a set
of homogeneous linear equations of the type
n
X
aik xk = 0, i = 1, 2, ..., n (3.24)
k=1
62 Advanced Classical Mechanics

which has a non-zero real solution. Now if we multiply the left-hand side of
(3.24) by xi and sum over i we would obtain

N N n
!2
X X X r~j
0= aik xi xk = mj xi (3.25)
j=1 i=1
qi
i,k=1

where (3.21) has been used. Therefore we conclude that

n
X r~j
xi = 0, j = 1, 2, ..., n (3.26)
i=1
qi

Since r~j (xj , yj , zj ), (3.26) reflects that the columns of the following Jaco-
bian matrix J are linearly dependent:
x x1 x1

q
1
q ... qn
y11 y21 y
q1 q2 ... qn1

z1 z1 z1
q
1 q2 ... q
n
... ... ... ...
[J]
...
(3.27)
x ... ... ...
xn xn
q1 q2 ... qn
n

yn yn y
q1 q2 ... qnn

zn zn
q1 q2 ... z n
qn

In other words the rank of [J] is less than n. Thus we arrive at a contra-
diction since according to the definition of generalized coordinates we have n
degrees of freedom and now we find that < n. Hence (3.23) holds. Further
T2 0 since T2 stands for the kinetic energy in the stationary situation with
the equality sign occurring when qi = 0, i = 1, 2, ..., n.

3.4 Lagranges equations of motion

Joseph-Louis Lagranges work on analytical mechanics was first published
in 1788. Although it appears that Lagranges equations, which are also some-
times referred to EulerLagranges equations, were known to Leonhard Euler
since the 1740s, it was Lagrange who first formulated an analytical derivation
of Eulers results. In this section, we are going to set up Lagranges equations
of motion (or the generalized equations of motion) for a holonomic N particle
system possessing n degrees of freedom. A great advantage with Lagranges
approach is that the number of unknowns exactly equals the number of degrees
of freedom. This is achieved by invoking DAlemberts principle so that con-
straint forces are automatically done away with. Lagrangian approach opens
 Lagrangian formulation in mechanics 63

up a new procedure for handling particle dynamics: the main difference with
its Newtonian counterpart is that the energies of the system are addressed
rather than the forces themselves.
Looking at the expression (3.12) of DAlemberts principle we see that the
works done by the applied forces and inertial forces under a virtual displace-
ment r~j (j = 1, 2, ...N ) are respectively given by
N
X (a)
W (a) = F~j . r~j (3.28)
j=1

and
N
X
W (in) = (mj r~j ). r~j (3.29)
j=1

In a virtual variation only displacements of the coordinates are considered

and time is taken to be frozen, i.e., no variation of t is involved. We there-
fore have in terms of n independent generalized coordinates q1 , q2 , ..., qn the
relations
N
X r~i
r~i = qj , i = 1, 2, ..., n (3.30)
j=1
q j

Plugging (3.30) into (3.28) and (3.29), W (a) and W in acquire the forms
n
X
W (a) = Qj qj (3.31)
j=1

where
N
X (a) r
~i
Qj = F~i . (3.32)
i=1
q j

and
n X
N
X r~i
W in = (mi r~i ). qj (3.33)
j=1 i=1
qj

Since in the right-hand side of (3.31), Qj appears attached to qj , which

are virtual displacements for generalized coordinates, one defines Qj to be the
generalized force. However, unlike the vectorial character of the conventional
force, it is scalar.
a
If further the system under consideration is conservative, F~i can be ex-
a
pressed as a negative gradient of some potential function, i.e., F~i = ~ iV
and V = V (r~1 , r~2 , ..., r~N ). As a result we can write
V
Qj = (3.34)
qj
In other words Qj s are derivable also from the same potential function: V =
V (q1 , q2 , ..., qn ).
64 Advanced Classical Mechanics

To interpret the right-hand side of (3.33) we need to prove two lemmas:

Lemma 1: The kinetic energy T obeys the equation
N
T X r~i
= mi r~i . , j = 1, 2, ...n (3.35)
qj i=1
q j

Proof: It is straightforward to deduce from (3.19) that

r~i r~i
= (3.36)
qj qj
Moreover, from the form of T given by (3.18) we have
N
T X r~i
= mi r~i . (3.37)
qj i=1
qj

Combining (3.36) and (3.37), (3.35) follows.

d
Lemma 2: The operators dt and q j are interchangeable in the sense
   
d r~i d~
ri
= (3.38)
dt qj qj dt
Proof: The left-hand side of (3.38) can be expanded as
n
2 r~i
X 2 r~i
= qm
+
m=1
qm qj tqj
" n #
X r~i r~i
= qm +
qj m=1 qm t
 
d~
ri
= = right-hand side of (3.38)
qj dt
d
Next operating by dt upon (3.35) and using (3.36) we get
  N   
d T X
r~i d r~i
= mi r~i . + mi r~i .
dt qj i=1
qj dt qj
N  
X r~i
= mi r~i . + mi r~i . (r~i )
i=1
qj qj
N
" N
#
X
r~i X 1  2
= mi r~i . + mi r~i 
i=1
qj qj i=1 2
N
X r~i T
= mi r~i . + (3.39)
i=1
qj qj

where we have used (3.38).

 Lagrangian formulation in mechanics 65

Finally we make use of (3.39) in (3.33) to arrive at

n   
X T d T
W (in) = qj (3.40)
j=1
qj dt qj

The representations (3.31) and (3.40) enable us to restate DAlemberts prin-

ciple (3.12) in the following form:
n   
X T d T
Qj + qj = 0 (3.41)
j=1
qj dt qj

The quantities qj being arbitrary and independent, it transpires from

(3.41) that the coefficients of each qj must vanish separately. In consequence
it must be true that
 
d T T
= Qj , j = 1, 2, ...n (3.42)
dt qj qj

which actually describes a set of n second-order differential equations involving

n generalized quantities and their velocities.
For a conservative system when (3.34) holds, (3.42) can be expressed in
the form
 
d L L
=0 (3.43)
dt qj qj

where L = T V and V is independent of the velocities q1 , q2 , ..., qn . L is

called the Lagrangian and the Equations (3.42) or (3.43) are referred to as
Euler-Lagranges equations of motion.
It is to be stressed that the unknown forces of constraint are absent from
Lagranges equations. Further, unlike Newtons equations of motion, there is
no direct reference to quantities like the force or acceleration. In contrast, only
a knowledge of kinetic and potential energies is required to set up L. However,
the appearance of the generalized forces Qj in the right-hand side of Equation
~(a)
(3.42), which in turn are related to the applied forces Fi , signals that the
kinetic energy T needs to be evaluated in an inertial frame. The reason is that
~(a)
Fi have their origins in Newtons laws, which are valid in inertial frames
only.
The form (3.43) admits of the addition of a total derivative term to L with-
out affecting the equations of motion. For if we construct a new Lagrangian
L0 from L according to
d
L0 (qj , qj , t) = L(qj , qj , t) + , j = 1, 2, ...n
dt
66 Advanced Classical Mechanics

where is any differentiable function of positions and time, then it is trivial

to check that d
dt gives a vanishing contribution to the equations of motion:
    
d d d
= (3.44)
dt qj dt qj dt
Pn
To justify (3.44) we simply have to note that ddt =

j=1 qj qj + t and the
result immediately follows. We therefore conclude that both L and L0 lead to
the same equations of motion.
Sometimes, depending upon the nature of the problem, a more general
representation of Qj than the one given in (3.34) is called for. Suppose a
velocity dependent potential exists, namely U (qj , qj , t), such that Qj s are
derivable in the manner
   
d U U
Qj = , j = 1, 2, ...n (3.45)
dt qj qj
then a similar set of equations as (3.43) follows from (3.42) if L is defined
according to L = T U . In Chapter 7 we shall see that a velocity dependent
potential has relevance in setting up of a Lagrangian for rotating frames.

3.5 Lagranges equations for some simple systems

3.5.1 Plane pendulum
The plane pendulum problem has been discussed before. Treating to be
the generalized coordinate, the kinetic and potential energies are
1 2 2
T = ml , V = mgl cos
2
As a result the Lagrangian for the plane pendulum problem is
1 2 2
L=T V = ml + mgl cos
2
Such a Lagrangian yields the partial derivatives
L
= mgl sin

L
= ml2

resulting in the following equation of motion:
d2 g
+ sin = 0
dt2 l
 Lagrangian formulation in mechanics 67

All

All All
All

All
All

FIGURE 3.2: Spherical pendulum.

already referred to in Chapter 1. We therefore find that L

stands for the
angular momentum of the mass about the point of support and L represents
the torque.
We remark that is not the only choice of the generalized coordinate for
the plane pendulum problem. We can also employ the horizontal displacement
or the vertical displacement of the bob as a candidate for the generalized
coordinate. However, use of the angular displacement appears to be the
most convenient one.

3.5.2 Spherical pendulum

The bob of a spherical pendulum can swing in any direction in the three-
dimensional space. As a result, the mass m traces out a sphere of constant
length l. See Figure 3.2. Using the polar coordinates and , which work
as the generalized coordinates, we find for the present problem the following
forms of the kinetic and potential energies:
1 2 2
T = ml ( + 2 sin2 )
2
V = mgl cos

These imply for the Lagrangian

1 2 2
L = ml ( + 2 sin2 ) + mgl cos
2
68 Advanced Classical Mechanics

Working out the following partial derivatives corresponding to and

L
= ml2 sin cos 2 mgl sin

L
= ml2

L
= 0

L
= ml2 sin2

the resulting equations of motion for and take the forms
g
= sin cos 2 sin
l
d
(ml2 sin2 ) = 0
dt
The second equation states that the component of angular momentum in
the direction is a constant of motion. More specifically we have
A
=
ml2 sin2
where B is a constant.
On the other hand, in the -equation of motion, the second term in the
right-hand side represents an equivalence of the gravitational torque while the
first term is a fictitious force. On integrating we get the form
A2
2 = 2g +B
m2 l2 cot2 + l cos
where B is a constant.

3.5.3 Binary star system

Consider a binary star composed of two masses m and M . If O denotes
the fixed origin and N the center of mass, then according to Figure 3.3
~ + r~0 ,
r~1 = R ~ + r~0
r~2 = R
1 2

~ is the center of mass and given by

where R

~ = mr~1 + M r~2
R
m+M
For the present binary system, the kinetic and potential energies being
1  2 1  2
T = m r~1  + M r~2 
2 2
GmM
V =
|r~1 r~2 |
 Lagrangian formulation in mechanics 69

All
All

All All
All
All
All All

FIGURE 3.3: Binary star system.

the Lagrangian assumes the form

    2 
1  2 GmM
L = m r~1  + M r~2  +
 
2 |r~1 r~2 |
1  2
= (m + M )R ~ 2 + 1 mM ~r  + GmM
2 2m+M |~r|

where ~r = r~1 r~2 = r~10 r~20 .

Onecan see that that the coordinate R does not appear explicitly implying
d L
= 0. Hence the quantity (m + M )R ~ is a constant signifying that the
dt ~
R
kinetic energy of the system as a whole (see the first term in the Lagrangian)
remains constant. Of course, such a term assumes no significance when we
study the internal motion of the system.

3.5.4 A system with four degrees of freedom

For a system having four degrees of freedom suppose that the Lagrangian
is of the type
4
X
L = m[q4 2 q1 2 q2 2 q3 2 ]1/2 + e Ak qk
k=1

where A s are functions of coordinates alone and the parameters m, e are

constants.
Using the above form of L we can show that the equations of motion could
be given by the following sets
4  
d X Aj Ak
m (qj ) = e qk , j = 1, 2, 3
dt qk qj
k=1
4  
d X Ak A4
m (q4 ) = e qk
dt q4 qk
k=1

where = (q4 q1 q2 q3 2 )1 .
2 2 2 2
70 Advanced Classical Mechanics

To this end, we write down for the Lagrangian L the partial derivatives
with respect to qj and qj

L
= mqj (q4 2 q1 2 q2 2 q3 2 )1/2 + eAj ,
qj
= mqj + eAj ,
4
L X Ak
= e qk
qj qj
k=1

where j = 1, 2, 3. Inserting the above expressions in Lagranges equation of

motion given by (3.43) we get
4
d X Ak
(mqj + eAj ) = e qk , j = 1, 2, 3
dt qj
k=1

dA P4 Aj
Further writing dtj = k=1 qk qk , Aj being functions of position only, it
follows from the above that
4  
d X Aj Ak
m (qj ) = e qk , j = 1, 2, 3
dt qk qj
k=1

where 2 = (q4 2 q1 2 q2 2 q3 2 )1 .
On the other hand, for q4 , we have
L
= mq4 + eA4
q4
4
L X Ak
= qk
q4 q4
k=1
P4
Here writing dA
dt =
4
k=1
A4
qk qk and using Lagranges equation of motion for
q4 we find the expression
4  
d X Ak A4
m (q4 ) = e qk
dt q4 qk
k=1
.

3.5.5 The problem of a damped oscillator

Let us consider the Lagrangian
m 0 t 2
L= e (q 02 q 2 ), <2
2
 Lagrangian formulation in mechanics 71

where m, , 0 are all positive constants. Noting that L q = mqe

0 t
and
L 2 0 t
q = m 0 qe , Lagranges equation of motion is easily found to be

q + 02 q + 0 q = 0
As discussed in Chapter 1, the above equation represents the one for a damped
0
oscillator. If we set q = e 2 t z, it can be transformed to the harmonic
2
oscillator form z + 2 z = 0 where 2 = (1 4 )02 > 0 for < 2.
It needs to be pointed out that just by adding a damping term m0 q to
the Lagrangian of the harmonic oscillator, namely, LHO = m 2 2 2
2 (q 0 q ) will
not produce the equation of motion for the damped oscillator. The reason is
that the additional term m0 q is just a total time-derivative of the quantity
m 2
2 0 q and hence leads to the same equation of motion. The correct La-
grangian must contain an explicit time-dependence in an overall exponential
factor as given above.

3.5.6 A conservative scleronomic system

For a conservative and scleronomic system we can write

n
d X L
qj L
dt j=1 qj

n    X n  
X L d L L L
= qj + qj qj + qj
j=1
qj dt qj j=1
qj qj

= 0 (3.46)
 
d L L
where we have exploited Lagranges equations dt qj = qj and noted that
L
for a scleronomic system t = 0. Hence we conclude that
n
X L
qj L = constant of motion (3.47)
j=1
qj

Next we may express

n
X L
qj L
j=1
qj
n
X T V
= qj (T V ) ( =0 for a conservative system)
j=1
qj qj
= 2T (T V ) (by Eulers theorem of homogeneous functions)
= T +V (3.48)
72 Advanced Classical Mechanics

Combining above with (3.47) we

Pnarrive at the result that for a conservative,
scleronomic system the quantity j=1 qj L
qj L is a constant of motion and
represents the total energy T + V is a constant.

3.6 Ignorable coordinates: Rouths procedure

of solution
It often happens that for a system described by a set of n generalized
coordinates, a few of them say, q1 , q2 , ..., qk , are not explicitly present in the
Lagrangian L although the corresponding velocities q1 , q2 , ..., qk appear in it.
Such coordinates which are absent from L are called ignorable (or cyclic)
coordinates. We have already encountered ignorable coordinates in some of
the problems we have come across: for example, in the central force prob-
lem is the ignorable coordinate, in the spherical pendulum is absent from
the Lagrangian and as such it is ignorable while for the binary star problem
the coordinate R does not appear explicitly in the Lagrangian. It should be
clear that corresponding to any ignorable coordinate the corresponding La-
granges equation reveals an associated constant of motion. Our task here
would be to set up a modified Lagrangian addressing the remaining coordi-
nates qk+1 , qk+2 , ..., qn from which the equations of motion can be derived. The
modified Lagrangian for the explicit coordinates present in the Lagrangian is
called the Routhian.
We first of all observe that for the ignorable coordinates q1 , q2 , ..., qk La-
granges equations are
 
d L
= 0, r = 1, 2, ...k
dt qr
(3.49)

Then the k first integrals are given by

L
= r , r = 1, 2, ...k (3.50)
qr
where 1 , 2 , ..., k are arbitrary constants of integration.
Define now the quantity
k
X L
RL qr (3.51)
r=1
qr

called the Routhian. Noting that the equations (3.50) are linear in q1 , q2 , ...qk ,
we can always express q1 , q2 , ...qk in terms of the coordinates qk+1 , qk+2 , ..., qn ,
 Lagrangian formulation in mechanics 73

the remaining velocities qk+1

, qk+2
, ...qn and the quantities 1 , 2 , ...k . Thus,
we can write R as
R = R[qk+1 , qk+2 , ...qn ; qk+1
, qk+2
, ...qn ; 1 , 2 , ...k ] (3.52)
To proceed further we take the increment of both sides of (3.51), i.e.,
" k
#
X L
R = L qr (3.53)
r=1
qr

in which we notice that L being ignorable in the coordinates q1 , q2 , ..., qk gives

for its virtual variation
n k n
X L X X L
L = qr + ( + ) qr (3.54)
qr r=1
qr
r=k+1 r=k+1

where t is treated as frozen because of the virtual variation. Therefore the

variation of the second term in the right-hand side of (3.53) gives
" k # k k
X L X L X
qr = qr + qr r (3.55)
r=1
qr r=1
qr r=1

Combining (3.54) and (3.55) we have for R

n n k
X L X L X
R = qr + qr qr r (3.56)
qr qr r=1
r=k+1 r=k+1

where we have employed (3.50).

For the left-hand side of (3.53) considering R to be the function of the
arguments as specified in (3.52) gives
n n k
X R X R X R
R = qr + qr + r (3.57)
qr qr r=1
r
r=k+1 r=k+1

Comparing (3.56) and (3.57) and noting that the variations are arbitrary
independent we get the following set of consistency conditions:
L R
= , r = k + 1, k + 2, ...n
qr qr
L R
= , r = k + 1, k + 2, ...n (3.58)
qr qr
R
qr = , r = 1, 2, ...k
r
The first two equations of (3.58) can be put together to read
 
d R R
= , r = k + 1, k + 2, ...n (3.59)
dt qr qr
74 Advanced Classical Mechanics

while the last one implies

Z
R
qr = dt, r = 1, 2, ...k (3.60)
r

We easily see from (3.59) that from a knowledge of the Routhian R, the
coordinates qk+1 , qk+2 , ...qn can be determined in terms of t. Having got them,
the remaining ones can be obtained from (3.60).

Example 3.6

In a dynamical system the kinetic and potential energies are

1 q1 2 1
T = 2 + q2 2 , V = c + dq22
2 a + bq2 2

Determine q1 (t) and q2 (t) by Rouths process of ignoration of coordinates.

From the above forms of T and V , the Lagrangian reads

1 q1 2 1
L= 2 + q2 2 c dq22
2 a + bq2 2

in which q1 can be recognized to be the ignorable coordinate. As such

L q1
= =
q1 a + bq22

where is a constant.
The Routhian is given by
L
R = L q1
q1
L
Inserting the forms for L and q1 ,
R takes the form
 
1 1 1
R = q2 2 d + b 2 q22 c a 2
2 2 2
which involves the coordinate q2 and velocity q2 only.
For such an R, Equation (3.59) gives

q2 + (2d + b 2 )q2 = 0

which has the solution

h i
q2 = A sin (2d + b 2 )1/2 t + 

where A and  are the constants of integration.

 Lagrangian formulation in mechanics 75

The coordinate q1 can be obtained from (3.60)

Z
R
q1 = dt

Z
= (a + bq22 )dt

Substituting the above solution of q2 the integration can be done in a straight-

forward way and we get for q1

1 bA2
q1 (a + bA2 )t sin[(2d + b 2 )1/2 t + ] + B
2 4(2d + b 2 )1/2

where B is a constant of integration.

Example 3.7

Solve the planetary problem by Rouths process of ignoration of coordi-

nates.

We addressed the planetary problem in Chapter 2 in the context of the

orbit equation for the inverse square law problem. To employ Rouths method
we note that in this case the kinetic and potential energies are

m 2
T = (r + r2 2 )
2

V =
r
yielding for the Lagrangian
m 2
L=T V = (r + r2 2 ) +
2 r
Since L does not contain , it is an ignorable coordinate:
L
= constant = l(say)

or, ml2 = l

which implies the conservation of angular momentum.

We can now write down the Routhian
L
R = L
 2

m 2 2 l
= r r 2 4 +
2 m r r
76 Advanced Classical Mechanics

which contains only r and r. We then have from (3.59)

l2
mr = 2
mr3 r
whose first integral gives the conservation of total energy E :

1 2 1 l2
mr + + V = constant = E
2 2 mr2
as we have already know from the central force problem. The energy equation
can be expressed in the integral form
Z r
dr
t=
1/2
l2
2
m E V 2mr 2
r0

where r0 is the initial value of r.

On the other hand, Equation (3.60) gives
Z
R
= dt
l
Z t
dt
= l 2 (t)
+ 0
0 mr

where we can determine explicitly if r is known.

3.7 Liouvilles class of Lagrangians

The Liouvilles class of Lagrangians is the one for which the kinetic and
potential energies appear in the special forms
1
T = [u1 (q1 ) + u2 (q2 ) + ... + un (qn )][v1 (q1 )q1 2 + v2 (q2 )q2 2 + ... + vn (qn )qn 2 ],
2
V = [w1 (q1 ) + w2 (q2 ) + ... + wn (qn )]/[u1 (q1 ) + u2 (q2 ) + ... + un (qn )] (3.61)

where ur , vr , wr are functions of the coordinates qr only, r = 1, 2, ...n for a

system having n degrees of freedom and the forces are conservative in nature
which means that they are derivable from a potential function. A great ad-
vantage with the problems of Liouvilles type is that these admit separation
of variables and hence can be solved completely.
Let us effect a change of variables from the coordinates q1 , q2 , ..., qn to a
new set of variables Q1 , Q2 , ..., Qn defined by
2 2 2
Q1 = v1 (q1 )q1 2 , Q2 = v2 (q2 )q2 2 , ..., Qn = vn (qn )qn 2 (3.62)
 Lagrangian formulation in mechanics 77

which in turn imply

p
dQi = vi (qi )dqi , i = 1, 2, ..., n (3.63)

These allow us, in principle, to integrate the above equations to obtain Qi as

a function of qi alone for i = 1, 2, ..., n.
This enables us to transform ui (qi ) and wi (qi ) to their new forms in terms
of Qi , i = 1, 2, ..., n which we specify by Ur (Qr ) and Wr (Qr ), respectively.
Thus, T and V can be represented by
n
1 X 2
T = U Qj (3.64)
2 j=1
W
V = (3.65)
U
Pn Pn
where U = j=1 Uj (Qj ) and W = j=1 Wj (Qj )
In terms of the variables Qi , which we look upon as a new generalized
coordinate, Lagranges equations read
 
d T T V
=
dt Qi Qi Qi
n
d 1 U X 2 V
i.e. (U Qi ) ( Qj ) = , i = 1, 2, ..., n (3.66)
dt 2 Qi j=1 Qi

where we use the same notations for T and V .

To tackle Equation (3.66) we need to multiply both sides by U Qi and use
(3.64) to get
d 1 2 2 U V
( U Qi ) T Qi + U Qi =0 (3.67)
dt 2 Qi Qi
We have already seen that for a conservative, scleronomic system, if the
kinetic energy is a homogeneous quadratic function of velocities then the total
energy is constant: T + V = h, h is a constant. As such (3.67) can be put in
the form
   
d 1 2 2 U
U Qi Qi h (U V ) = 0
dt 2 Qi Qi
   
d 1 2 2 U W
i.e. U Qi Qi h = 0
dt 2 Qi Qi
   
d 1 2 2 Ui Wi
or U Qi Qi h = 0, i = 1, 2, ..., n (3.68)
dt 2 Qi Qi

Since, by using the chain rule of partial derivatives, we can express

dUi Ui dWi Wi
= Qi , = Qi (3.69)
dt Qi dt Qi
78 Advanced Classical Mechanics

we arrive at the constants of motion

1 2 2
U Qi hUi + Wi = Ci , i = 1, 2, ..., n (3.70)
2
where Ci s are the constants of integration.
Denoting
di = Ci + hUi Wi , i = 1, 2, ..., n (3.71)
and restoring now to the old quantities u, wi , qi the above constants of motion
take the forms
1 2
u i qi 2 = di (qi ), i = 1, 2, ..., n (3.72)
2
In consequence we have
s
i (qi ) 2
d(qi ) = dt, i = 1, 2, ...n (3.73)
di (qi ) u

or more explicitly
Z s Z s Z s
1 (q1 ) 2 (q2 ) n (qn )
d(q1 ) = d(q2 ) + 1 = ... = d(qn ) + n1
d1 (q1 ) d2 (q2 ) dn (qn )
(3.74)
In this way the variables become separated. Multiplying (3.74) by u1 , u2 , ..., un
for each respective value of i = 1, 2, ...n and adding we get
n Z Z P
X r
j uj
uj dqj = 2 dt + c = 2t + c (3.75)
j=1
dj u

where c is the constant of integration.

Equations (3.74) and (3.75) provide the complete solution to Liouvilles
problem. Note that the complete solution is subject to C1 + C2 + ... + Cn = 0.
This can be seen very easily from (3.72) which because of (3.71) can be written
as
1 2
u vi (qi )qi 2 = Ci + hui (qi ) wi (qi ), i = 1, 2, ..., n (3.76)
2
Summing up over i we get
n n n n
1 2X 2
X X X
u vj (qj )qj = Cj + h uj wj (3.77)
2 j=1 j=1 j=1 j=1
P
In other words, Cr = u(T + V h) = 0 where we have used the definitions
of T and V furnished by (3.61).
 Lagrangian formulation in mechanics 79

Example 3.8

Show that the dynamical system for which 2T = r1 r2 (r1 2 + r2 2 ) and

V = r11 + r12 can be expressed as one of Liouvilles types.
Let us put r1 = q1 + q2 and r2 = q1 q2 . T and V become
T = (q12 q22 )(q1 2 + q2 2 )
2q1
V =
q1 q22
2

and the problem can be recognized to be of Liouvilles type.

Example 3.9

A particle moves in a plane under the action of two Newtonian centers of

0
attraction at the points (c, 0) and (c, 0), the attractions being r2 and r02 , re-
spectively; r, r0 being the distances from (c, 0) and (c, 0), respectively. Show
that the problem is of Liouvilles type.

Here
1 1
T = mv 2 = (x2 + y 2 )
2 2
0 0
V = 0 = p p
r r (x c)2 + y 2 (x + c)2 + y 2
Let us set r0 = q1 + q2 and r = q1 q2 . Then T and V can be seen in the
forms
q1 2 q2 2
 
1 2
T = (q1 q22 ) 2 +
2 q1 c2 c2 q22
 
q1 q2
V = ( + 0 ) 2 + ( 0
)
q1 q22 q12 q22
So the problem is Liouvilles type.

Example 3.10

If T = 12 (q12 + q22 )(q1 2 + q2 2 ) and V = 1

q12 +q22
, solve the problem completely
using Liouvilles approach.

For this problem one can easily identify

u1 (q1 ) = q12 , u2 (q2 ) = q22 , v1 (q1 ) = 1, v2 (q2 ) = 1, w1 (q1 ) = 1, w2 (q2 ) = 0
Then from (3.74) we write
Z s Z s
v1 (q1 ) v2 (q2 )
dq1 = dq2 +
d1 (q1 ) d2 (q2 )
80 Advanced Classical Mechanics

where d1 (q1 ) and d2 (q2 ) are given by

d1 (q1 ) = C1 + hu1 (q1 ) w1 (q1 ) = C1 + hq12 1

d2 (q2 ) = C2 + hu2 (q2 ) w2 (q2 ) = C1 + hq22 ( C1 + C2 = 0)
Z s Z s
1 1
dq1 = dq2 +
C1 + hq12 1 C1 + hq22
Integration gives
q1 q2
cos1 q cos1 q = constant = C0 (say)
1C1 C1
h h

We thus arrive at the form

a2 q12 + b2 q22 + 2abq1 q2 cos C0 = sin2 C0

q q
h h
where the quantities a and b stand for a = 1C 1
and b = C1 .

3.8 Small oscillations

In the theory of small oscillations one considers the perturbed behavior
near the equilibrium configuration of a mechanical system and inquires if
the system has a tendency to return to its original position given a slight
disturbance from the position of equilibrium. The system is said to be in
stable equilibrium if such a feature holds. The case of a suspended pendulum
in which a point mass, tied to an inextensible string of negligible mass and
hanging vertically down, furnishes one such example. On the other hand, if a
slight disturbance causes a significant deviation of the system from its original
position the system is said to be in a state of unstable equilibrium. A rod
standing on its one end is an example of this type.
Consider a holonomic dynamical system with n degrees of freedom. Let
q1 , q2 ,...,qn be the generalized coordinates and the forces derivable from a
potential function V . The equilibrium position of the system is obtained by
solving the following set of equations:
V V V
= = ... = =0 (3.78)
q1 q2 qn
Let us subject the system to a small perturbation so that all the coordi-
nates and velocities in the subsequent motion remain small. Taking, without
 Lagrangian formulation in mechanics 81

loss of generality, the coordinates to be located at the zero position, we carry

out a Taylor expansion of V about the point q1 = q2 = ... = qn = 0 to write
V 1 2V
V = V0 + |0 qi + |0 qi qj + ... i, j = 1, 2, ..., n (3.79)
qi 2 qi qj
where the suffix (0) denotes that the corresponding quantity is to be deter-
mined at the equilibrium position and it is implied that repeated indices are
summed over. The first term in the right-hand side is a mere constant and
does not affect the physics of the problem while the second term vanishes at
the equilibrium position. If we neglect third and higher order terms, V reads
simply
V = dij qi qj , i, j = 1, 2, ..., n (3.80)
where dij are symmetric quantities and stand for the second-order derivatives
1 2V
2 qi qj |0 evaluated at the equilibrium point and so are constants.
With the kinetic energy being quadratic in velocities and given by the form
T = cij qi qj , where cij s (i, j = 1, 2, ..., n) are symmetric and functions of the
coordinates q1 , q2 ,...,qn only, we can expand T to write
cij
T = cij |0 qi qj +|0 qi qi qj + ... i, j = 1, 2, ..., n (3.81)
qi
The second term corresponds to higher order terms and can be neglected.
Henceforth omitting the suffix (0) we have for T the term

T = cij qi qj , i, j = 1, 2, ..., n (3.82)

to lowest order.
The Lagrangian takes the form

L = cij qi qj dij qi qj , i, j = 1, 2, ..., n (3.83)

which results in the equations of motion

cij qj + dij qj = 0, i = 1, 2, ..., n (3.84)

where the index j has been summed over and the equations are linear and
homogeneous with constant coefficients cs and ds.
Looking for solutions

qi = Ai et , i = 1, 2, ..., n (3.85)
where Ai s are constants which could be complex, yield the following full set
of n equations:
(c11 2 + d11 )A1 + (c12 2 + d12 )A2 + ... + (c1n 2 + d1n )An = 0
(c21 2 + d21 )A1 + (c22 2 + d22 )A2 + ... + (c2n 2 + d2n )An = 0
....
(cn1 2 + dn1 )A1 + (cn2 2 + dn2 )A2 + ... + (c2n 2 + dnn )An = 0 (3.86)
82 Advanced Classical Mechanics

Note that the determinant has to vanish to ensure a nontrivial solution

c11 2 + d11 c12 2 + d12 ... c1n 2 + d1n

c21 2 + d21 c22 2 + d22 ... c2n 2 + d2n
det =0 (3.87)
... ... ... ...
cn1 2 + dn1 cn2 2 + dn2 ... c2n 2 + dnn

Such a determinant is called a secular determinant and gives a polynomial in

2 of degree n.
While the kinetic energy is a positive definite function, so is V since it is
zero at the position of the equilibrium and we are interested in carrying out a
perturbation around the stable equilibrium point. Thus seeking an oscillatory
type of solution we can write
= 2 (3.88)
where 2 is real giving the simple harmonic form

qi (t) = Ai eit , i = 1, 2, ..., n (3.89)

with varying amplitudes but having the same period. The reality of 2 can be
checked by substituting the above form of qi in the equations of motion (3.84)
directly, splitting Ai s into real and imaginary parts and finding straightfor-
wardly that the imaginary part vanishes. In fact, for a positive definite po-
tential energy, is real too but it becomes purely imaginary for a negative
definite potential energy.
Using (3.89) the equations of motion emerge as
n
X
(dij 2 cij )Aj = 0, i = 1, 2, ..., n (3.90)
j=1

where Aj play the role of an eigenvector while = 0 stands for

det(dij 2 cij ) = 0, i, j = 1, 2, ..., n (3.91)

which has n positive semi-definite solutions for 2 . We label these as i2 ,

i = 1, 2, ..., n. Knowing these we can determine the different eigenvectors from
the previous equation and normalize them appropriately.
Take two different eigenvalues labeled by A and B. Then the corresponding
equations of motion assume the form
n n
(A) (B)
X X
2 2
(dij A cij )Aj = 0, (dij B cij )Ai =0 (3.92)
j=1 i=1

Noting that both the matrices dij and cij are symmetric, if we multiply from
(B) (A)
the left the first equation by Ai and the second equation by Aj and sum
 Lagrangian formulation in mechanics 83

respectively over i and j we observe that the terms involving dij drop out on
subtracting leaving us with the relation
n X
n
(A) (B)
X
2 2
(A B ) (Ai cij Aj ) = 0 (3.93)
i=1 j=1

Since the eigenvalues are different the double sum has to vanish. When the
eigenvalues are equal the double sum can be normalized to unity implying the
simple result
n X n
(A) (B)
X
(Ai cij Aj ) = ij (3.94)
i=1 j=1

This is a relation for cij .

The one corresponding to dij is given by
n X
n
(A) (B)
X
2
(Ai dij Aj ) = B ij (3.95)
i=1 j=1

(B)
where the equation of motion for qj is used in the above relation to eliminate
cij .
Thus we have arrived at the conclusion that both cij and dij admit of si-
multaneous diagonalization. At this point we can revert to normal coordinates
by defining a new set of variables Qs
n
(A)
X
qi = Ai Q(A) (3.96)
i=1

This converts the Lagrangian to the normal form

n
X
L= (Q2A A
2 2
QA ) (3.97)
i=1

To put the above scheme in practice we discuss the following examples.

Example 3.11: The double pendulum

The double pendulum is a combination of two simple pendulums positioned

in the same vertical plane and connected in such a way that the lower one
hangs from the bob of the upper one. The bob of the upper one of mass m1 is
suspended from a ceiling by an inextensible string of negligible mass having
a length l1 while the lower bob of mass m2 is tied to the bob of the upper
one by an inextensible string of negligible mass having length l2 . The setup is
illustrated in Figure 3.4.
84 Advanced Classical Mechanics

All

All All
All

All All
All

FIGURE 3.4: The double pendulum.

By choosing a suitable frame of reference we can introduce for the hori-

zontal and vertical coordinates the respective relations

X = l1 sin 1 , x = l1 sin 1 + l2 sin 2

Y = l1 cos 1 , y = l1 cos 1 + l2 cos 2

This results in the following forms of the kinetic and potential energies
m1 2 m2 2 m1 + m2 2 2
T = (X + Y 2 ) + (x + y 2 ) = l1 1
2 2 2
m2 2 2
+ l 2 + m2 l1 l2 cos(1 2 )1 2
2 2
V = m1 gY m2 gy
= (m1 + m2 )gl1 cos 1 m2 gl2 cos 2

yielding for the Lagrangian

m1 + m2 2 2 m2 2 2
L=T V = l1 1 + l 2 + m2 l1 l2 cos(1 2 )1 2
2 2 2
+ (m1 + m2 )gl1 cos 1 + m2 gl2 cos 2

Using Lagranges form of the equations of motion we easily deduce the

pair
m2 l2 m2 l2 2
l1 1 + cos(1 2 )2 + sin(1 2 )2 + g sin 1 = 0
m1 + m2 m1 + m2
2
l1 cos(1 2 )1 + l2 2 l1 sin(1 2 )1 + g sin 1 = 0

The above equations constitute a set of nonlinear equations. For small os-
cillations we will be interested only in their linearized versions and as such
 Lagrangian formulation in mechanics 85

retain terms only of small 1 and 2 and their time-derivatives but neglect
their squares and higher powers. We are thus led to the forms
g m 2 l2
1 + 1 = 2
l1 (m1 + m2 )l1
g l1
2 + 2 = 1
l2 l2
which can be viewed as a system of two coupled differential equations.
To determine the normal modes we look into the simple case when l1 =
l2 = l. The above equations then assume a greatly reduced form

ml 1 + m1 g1 = m2 g(2 1 )
m2 l(1 + 2 ) + m2 g2 = 0 (3.98)

and can be reexpressed in the manner

d2
   
1 1
=M
dt2 2 2

where the two-dimensional matrix M is given by

m1 +m2
m
 
g m1 m1
2

M =
l m1m+m
1
2 m1 +m2
m1

We look for normal mode solutions for 1 2

1
   
1 it
=e
2 2
Since the determinant of the coefficient matrix has to vanish to ensure a non-
trivial solution, this means

2 + gl m1m+m gl m
 2 2

m1
det 1
=0
gl m1m+m
1
2
2 + gl m1m+m
1
2

which translates into a quadratic equation in 2 :

s
g m1 + m 2 g m2 (m1 + m2 )
2 = =0
l m1 l m1
The normal frequencies then are
s
g 1
=
l 1
m2
where stands for the ratio m1 +m2 .
86 Advanced Classical Mechanics

Actually if we employ the normal coordinates defined by

= 1 + 2

Equation (3.98) takes the form

g
(1 + )+ + (1 ) + (+ + ) = 0
l
g
(1 + )+ + ( 1) + (+ ) = 0
l
and gives a pair of decoupled equations

+ + 2+ + = 0
+ 2 = 0
g 1
where the quantities 2 = l 1 are called the normal frequencies of oscil-
lation.

Example 3.12: Coupled harmonic oscillators

We consider a system of two coupled oscillators as described by two iden-

tical harmonic oscillators that are coupled through an elastic spring. An ap-
propriate Lagrangian is of the form
1 1
L= m(x1 2 + x2 2 ) 02 (x21 + x22 ) + x1 x2
2 2
Such a system has two degrees of freedom with x1 , x2 representing two gen-
eralized coordinates for its description and is a coupling constant. The
coordinates x1 , x2 account for the displacements of the particles from their
equilibrium positions.
For simplicity setting m = 1, the equations of motion resulting from the
above Lagrangian are easily found to be

x1 02 x1 + x2 = 0
x2 02 x2 + x1 = 0

These equations can be put in the matrix form

d2
   
x1 x1
= M
dt2 x2 x2
where M is a 2 2 matrix given by

02
 

M =
02
 Lagrangian formulation in mechanics 87

As before we are concerned with the normal mode solutions like

   
x1 it x1
=e
x2 x2
and a nontrivial solution is ensured provided
 2
0 2


det =0
02 2
The vanishing of the determinant yields a quadratic equation whose eigen-
value solutions lead to the following accompanying eigenvectors:
   
x1 1 1
q
2
1 = 0 , =
x2 2 1
   
x1 1 1
q
2 = 02 + , =
x2 2 1
These show a symmetric mode and an antisymmetric mode of the normal
mode of oscillations. The former case reveals an oscillation in the same di-
rection while the latter points to an out of phase oscillation. Note that if the
eigenvector solutions are 1 , 2 then these may be expressed in terms of x1 , x2
through 1 = x1+x
2
2
, 2 = x1x
2
2
.
For the weak coupling case when << 02 , we have the approximations

1 = 0 (1 ), 2 = 0 (1 + )
202 202
Corresponding to these approximations the two normal mode solutions can
be superposed to give for x1 and x2 the following results:

0 (1 2 )t 0 (1+ 2 )t
(x1 , x2 ) = e 20
e 20
(3.99)

where we suppressed an overall constant factor. Apart from a phase term e0 t ,

the two solutions consist of factors cos 0 t and sin 0 t, where 0 stand for the
ratio 2 0 , reflecting an out of phase behavior.

3.9 Summary
In this chapter we reviewed the principles of the Lagrangian approach to
classical mechanics whose analytical structure is described by a set on n ordi-
nary second-order differential equations in terms of a Lagrangian function. Af-
ter discussing the constraints, the principle of virtual work and DAlemberts
principle we derived the Lagrangian equations of motion. We then considered
88 Advanced Classical Mechanics

their applications to several simple systems. We gave a special emphasis on the

role of ignorable coordinates in physical systems. Apart from offering simplifi-
cation of the corresponding equation of motion ignorable coordinates furnish
the important result that the generalized momenta corresponding to them are
constants of motion. We described Rouths procedure of solving equations of
of motion when some of the coordinates are ignorable. Subsequently we inves-
tigated Liouvilles class of Lagrangians and provided a method toward their
solvability. We also carried out a general treatment of the problem of small
oscillations about a stable equilibrium.

Exercises
1. The Lagrangian for a mechanical system of two degrees of freedom is
given by

ma2 mga 1 2 5 2
L= (16q1 2 + 20q1 q2 + 25q2 2 ) ( q + q )
72 2 3 1 6 2
where a is a constant. Solve the system subject to the initial conditions
q1 = q2 = 0 and q1 = q2 = q.

2. The kinetic and potential energies of a mechanical system with two de-
grees of freedom are T = 12 (q1 2 + q2 2 ), V = f (q1 q2 ) where f is an arbitrary
function of the given argument. By making a change of variables solve the
problem in terms of an integral involving f .

3. The P
kinetic and potentialPn energies of a mechanical system are in the
n
forms T = 1 fi (qi )qi 2 , V = 1 Vi qi for a system with n degrees of freedom.
Show that the problem can be solved by the method of quadrature.

4. A particle of mass m slides down under the action of gravity on the

inside of a smooth surface given by the equation z = x2 + y 2 . Use cylindrical
coordinates and employ Rouths procedure to convert the problem to one de-
gree of freedom.

5. The Lagrangian of a mechanical system with two degrees of freedom

has the form
q1 2 1
L= + q2 2 + 2q23 + cq2
aq2 + b 2
where a, b, c are constants. Find an integral for q2 .

6. The kinetic and potential energies of a mechanical system with two de-
q1 2
grees of freedom are T = 12 a+bq 2
+ 12 q2 2 q22 , V = c + dq2 where a, b, c, d are con-
stants. Show that the coordinate q2 satisfies the equation (q2 )(q2 + 2)2 =
(t t0 )2 , where , are constants and t = t0 is the initial value of time.
 Lagrangian formulation in mechanics 89

7. The kinetic and potential energies of a particle are T = 12 (x2 + y 2 ), V =

0
r r0 ,
where , 0 are constants. Further, r and r0 denote the respective dis-
tances of the particle from the points whose coordinates are (a, 0) and (a, 0).
Reduce the problem to Liouvilles form.

8. A particle of mass m is located in a one-dimensional potential field

V (x) = x2 x , where , are positive constants. Show that the period of
small oscillations
q that the particle performs about the equilibrium position
2m3
will be 4 4 .

9. Find the eigenfrequencies and eigenvectors of the normal modes for a

coupled system of oscillators given by the following Lagrangian:
1 1
L= (m1 x1 2 + m2 x2 2 ) (12 x21 + 22 x22 ) + x1 x2
2 2
where m1 , m2 , 1 , 2 are the respective masses and angular frequencies.

10. A bead of mass m slides down under the action of gravity on the inside
of a frictionless cyclodial wire given by the usual equation in the parametric
form x = a( sin ), y = a(1 + cos ), where 0 2. Set up the
Lagrangian and find the equation of motion. Writing z = cos( 2 ), show that
2 g
the equation of motion can be reexpressed in the form ddt2z + 4a z = 0 implying
q
that the bead oscillates with the period 2 4a g .
Chapter 4
Hamiltonian and Poisson bracket

4.1 The Hamiltonian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

4.2 Hamiltonian canonical equations of motion . . . . . . . . . . . . . . . . . . . . . 94
4.3 Poisson bracket . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
4.4 Properties of Poisson bracket . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
4.5 Poisson theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
4.6 Angular momentum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
4.7 Liouvilles theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
4.8 The case of singular Lagrangians . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
4.9 Higher derivative classical systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
4.10 The PaisUhlenbeck oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
4.11 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

Like the Lagrangian, we can define another fundamental quantity, the Hamil-
tonian, associated with the evolution of a mechanical system. While in the
Lagrangian formalism we are required to solve a set of n second-order differ-
ential equations for a system of particles with n degrees of freedom, Hamiltons
equations, in essence, provide a factorized version of them in terms of 2n first-
order partial differential equations in terms of the generalized coordinates qi
and generalized or conjugate momentum pi . The almost symmetrical appear-
ance of these dynamical variables facilitates development of formal theories
based on canonical transformations, action-angle variables and so on to which
we shall come later. We refer to (qi , pi ) as the set of conjugate or canonical
variables. The coordinates (q1 , q2 , ..., qn ) and the momenta (p1 , p2 , ..., pn ) con-
stitute a 2n-dimensional space called the phase space. Hamiltonian equations
define completely the evolution of the system in phase space. The Lagrangian
L(q1 , q2 , ...qn ; q1 , q2 , ...qn ; t), on the other hand, is interpreted in the configura-
tion space with qi and qi denoting, respectively, the coordinates and velocities
at a specific point at time t.

91
92 Advanced Classical Mechanics

4.1 The Hamiltonian

Consider the total time rate of change in the Lagrangian L(qi , qi , t), i =
1, 2, ...n
n n
dL X L X L dqi L
= qi + + (4.1)
dt i=1
qi i=1
qi dt t
Since the second term in the right-hand side can be written as
n n
! n
!
X L dqi d X L d X L
= qi qi (4.2)
i=1
qi dt dt i=1 qi dt i=1 qi

we have
" n # n   
d X L X L d L L
qi L = qi (4.3)
dt i=1 qi i=1
qi dt qi t

If the coordinates qi s obey Lagrangian equations of motion, then the above

equation reduces to " n #
d X L L
qi L = (4.4)
dt i=1 qi t
We now define the generalized or conjugate momentum pi associated with
the generalized coordinate qi to be
L
pi = , i = 1, 2, ...n (4.5)
qi
then (4.4) reads " n #
d X L
pi qi L = (4.6)
dt i=1 t
Note that since the kinetic energy in terms of the Cartesian velocities is T =
1 2 2 2
2 m(x + y + z ), the components of the linear momentum in terms of T
are simply T T T T
x , y and z . It is therefore appropriate to view qi as a kind
of generalized momentum which when the potential is a function of position
only represents L qi as in (4.5).
A word about the dimensions of the generalized momentum pi . Since the
generalized coordinate qi can have dimensions given in units of length or
angle or whatever is found suitable, the units of the generalized or conjugate
momentum are to be interpreted accordingly. Indeed we can express the units
of pi and the force Fi in terms of the dimensions of a generalized coordinate
as in the following:

M L2 1 M L2 1
[pi ] = [Fi ] = (4.7)
T [qi ] T 2 [qi ]
 Hamiltonian and Poisson bracket 93

One sees that when the generalized coordinate qi has the dimension of length,
pi has the dimension of momentum and Fi has the dimension of force. When
the generalized coordinate represents a rotation, it is dimensionless implying
2
that pi has the dimension of the angular momentum, i.e., ( MTL ) and Fi has
2
the dimension of torque, i.e., ( MTL2 ).
The construction of the Hamiltonian from a given Lagrangian relies upon
a Legendre transformation mapping the coordinates qi to pi . We introduce a
function H defined in the manner
n
X
H= pi qi L (4.8)
i=1

so that from (4.6)

dH L
= (4.9)
dt t
The function H is known as Hamiltons function or more precisely the Hamil-
tonian. It is at once implied from (4.9) that if t does not appear explicitly in
L then H is reduced to a constant in time.
We already
Pn noted earlier that for a conservative, scleronomic system the
quantity i=1 qi L
qi L is a constant of motion and that if the kinetic energy is
a homogeneous quadratic function of the velocities, the total energy is constant
as well. In mathematical terms this means
n n
X X L
pi qi = qi
i=1 i=1
qi
n
X (T V )
= qi
i=1
qi
n
X T
= qi
i=1
qi
= 2T

where L = T V and we have employed Eulers theorem in the last step. As

such the Hamiltonian

H = 2T (T V )
= T +V

corresponds to the total energy of the system. Sometimes H is also interpreted

as the generalized energy.
94 Advanced Classical Mechanics

4.2 Hamiltonian canonical equations of motion

We can generate Hamiltonian canonical equations of motion from the def-
inition of the Hamiltonian. First of all, taking the differential of (4.8) we get
n    
X L L L
dH = qi dpi + pi dqi dqi dt
i=1
qi qi t
n  
X L L
= qi dpi dqi dt (4.10)
i=1
qi t

where we used the definition (4.5) of pi .

Next, looking upon H as a function of qi , pi and t, i.e., H = H(qi , pi , t), i =
1, 2, ..., n, we can express dH as
n  
X H H H
dH = dqi + dpi + dt (4.11)
i=1
qi pi t

By comparing (4.10) and (4.11) one gets the set of consistency equations

H
qi =
pi
L H
pi = =
qi qi
H L
= (4.12)
t t
These are Hamiltonian canonical equations of motion and were set up by
William Rowan Hamilton in 1835. Hamiltons canonical equations govern the
trajectories of particle motion in phase space. Hamiltons approach is in some
sense a reformulation of Lagranges second-order differential equations but
in terms of a coupled pair of first order partial differential equations. Our
reference to these equations as canonical equations arises from the fact that
the investigation of motions subjected to the influence of a potential is just
being reduced to the examination of simple differential equations of the form
(4.12).
Interestingly the set of equations (4.12) is invariant under the replacements
qi pi and pi qi . To make a judicious use of such equations it is necessary
to know the given form of the Hamiltonian H. If a direct knowledge of H is not
known, then it may be ascertained from the Legendre transformation (4.8) by
setting up the Lagrangian of the system. Note that in a scleronomic system,
the equations for the 2n variables (qi , pi ) are subject to the pair
 Hamiltonian and Poisson bracket 95

dqi H
qi =
dt pi
dpi H
pi = (4.13)
dt qi
Because of the above equations the time rate of the Hamiltonian simplifies to
n  
dH X H H H
= qi + pi +
dt i=1
qi pi t
H
= (4.14)
t
showing that if H does not depend explicitly upon t then it is a constant of
motion.

Example 4.1

If all the coordinates of a system are cyclic prove that the coordinates may
be found out by integration. Prove further that if the system is scleronomous
then the coordinates are linear functions of time.

Since the coordinates are cyclic, the Hamiltonian is only a function of pi ,

i = 1, 2, 3... and t
H = H(pi , t), i = 1, 2, ...n
and therefore
H
pi = = 0, i = 1, 2, ...n
qi
So we have
pi = constant = i
where i are constants. Hence in the present case we deal with a free-particle
system.
Further
H
qi = = i (p1 , p2 , ...pn ; t) (say)
pi
= i (1 , 2 , ...n ; t)
Z
qi = i (1 , 2 , ...n ; t)dt + i

where i s are constants of integration. Thus the coordinates can be found

out by integration.
96 Advanced Classical Mechanics

For a scleronomous system the time t does not appear in i yielding

qi = i (1 , 2 , ...n )t + i , i = 1, 2, ...n

The coordinates are therefore linear in time. This examples provides an ex-
ample of an integrable system.

Remark:

For the kinetic energy function homogeneous in the second degree of the
generalized velocities, the Lagrangian reads
n
1 X
L= aik qi qk V (q) (4.15)
2
j,k=1

where q = (q1 , q2 , ...qn ) and the coefficient matrix A is symmetric: aik = aki .
The canonical momenta are given by the equations pj = L qj =
Pn
i=1 aij (q)qj . In a matrix form they are p = Aq. Inversion leads to

qi = a1
ij pj (4.16)
where a1 1
ij (in general aij 6= (aij )
1
) are the elements of the matrix A1 :
1 T
L=q Aq V (q) (4.17)
2
where T is the transpose and we have suppressed the summation sign.
The Legendre transformation gives for the Hamiltonian

H = pT q L (4.18)
which can be expressed as
1
H = pT q [ q T Aq V ]
2
T 1 1
= p A p [ q T Aq V ]
2
1 T 1
= p A p p A AA1 p + V
T 1
2
1 T 1
= p A p+V
2
= T +V

Thus H stands for the total energy.

 Hamiltonian and Poisson bracket 97

Example 4.2

Solve the plane pendulum problem using the Hamiltonian approach.

As discussed in Chapter 3, the Lagrangian for such a system reads

1 2 2
L= ml + mgl cos
2
where is the only generalized coordinate. Therefore the generalized momen-
tum corresponding to is
L
p = = ml2

From the Legendre transformation we can construct the Hamiltonian as
1 2 2
H = p L = ml mgl cos
2
However it needs to be defined in terms of proper variables and p . Conse-
quently we eliminate in favor of p to recast it as

p2
H= mgl cos
2ml2
where we have used p = ml2 . The accompanying Hamiltonian equations
which follow are
d H p
= =
dt p ml2
dp H
= = mgl sin
dt
We notice that t does not appear explicitly in H. So H is a constant of motion
and leads to the energy equation

p2
E= mgl cos
2ml2
where the energy E is the constant value of the Hamiltonian.

Example 4.3

Examine the motion of a particle sliding on a parabolic wire.

Let the parabolic wire be defined in the xy plane along which the particle
is sliding. We ignore any effect of friction and assume that the particle is acted
upon by gravity only.
Taking x to be the generalized coordinate we take the parabola to be given
by an equation y = 12 x2 .
98 Advanced Classical Mechanics

The Lagrangian is
1
L = m(x2 + y 2 ) mgy
2
1 mg 2
= m(1 + x2 )x2 x
2 2
L
p = = m(1 + x2 )x
x
Hence the Hamiltonian is given by

p2 mg 2
H = T +V = + x
2m(1 + x2 ) 2

which implies by the use of Hamiltonian equations

p2
 
p
x = , p = x + mg
m(1 + x2 ) m(1 + x2 )

Note that p 6= mx.

Example 4.4

The Lagrangian for a free particle in terms of paraboloidal coordinates (, , )

is
1 1
L = m( 2 + 2 )(2 + 2 ) + m 2 2 2
2 2
Set up the Hamiltonian.

The paraboloidal coordinates were defined in Chapter 1. The canonical

momenta conjugate to the generalized coordinates (, , ) are

L
p = = m( 2 + 2 )

L
p = = m( 2 + 2 )

L
p = = m 2 2

Hence the Hamiltonian has the form

H = p + p + p L
" #
1 p2 + p2 p2
= + 2 2
2m 2 + 2
 Hamiltonian and Poisson bracket 99

Example 4.5

Solve the planetary problem.

Here the kinetic and potential energies are

1
T =m(r2 + r2 2 )
2

V =
r
1
L = T V = m(r2 + r2 2 ) +
2 r
As such we can write for the canonical momenta
L
pr = = mr
r
L
p = = mr2

These give for the Hamiltonian
1
H = T + V = m(r2 + r2 2 )
2 r
2
 
1 p
= p2r + 2
2m r r
We therefore deduce
H 1 H p2
r = = pr , pr = = 3 = 2
pr m r mr r
H p
= = , p = 0
p mr2
Now p = 0 implies
d
(mr2 ) = 0
dt
or, mr2 = constant = l(say)

Eliminating pr , p and from the above equation we arrive at the form

l2
mr = 2
mr3 r
which has already been solved in Chapter 3 in connection with Rouths pro-
cedure of ignoration of coordinates.
100 Advanced Classical Mechanics

Example 4.6

Write down Hamiltonian equations in spherical polar coordinates for the

Lagrangian
1 k
L = m~r 2 +
2 r
where k is a constant.
In spherical polar coordinates the Lagrangian is of the form
1 k
L= m(r2 + r2 2 + r2 sin2 2 ) +
2 r
So the matrices A and A1 , defined earlier, are given by

m 0 0
A = 0 mr2 0
2 2
0 0 mr sin
1
m 0 0
A1 = 0 mr 1
2 0
1
0 0 mr 2 sin2

Consequently the Hamiltonian reads

 
1 1 1 k
H= p2r + 2 p2 + 2 2 p2
2m r r sin r

These give for the Hamiltonian equations

H H H
pr = , p = , p =
r
which lead to
2 k d cos
mr = 2, (mr2 ) = p2 , p = 0
mr3 r dt mr2 sin3
p2
where 2 = p2 + sin2
and pr = mr, p = mr2 , p = mr2 sin2 from

pr r
p = A
p
 Hamiltonian and Poisson bracket 101

4.3 Poisson bracket

Consider some function f (q, p, t) of the canonical variables qi and pi . Then
for i = 1, 2, ...n
n  
df X f dqi f dpi f
= + +
dt i=1
qi dt pi dt t
n 
X f H 
f H f
= + (4.19)
i=1
q i p i p i qi t

We introduce a quantity {u, v} involving the functions u = u(qi , pi , t) and

v = v(qi , pi , t) and defined by
n  
X u v u v
{u, v} = (4.20)
i=1
qi pi pi qi

A trivial consequence of (4.20) is the result

{qi , pj } = ij (4.21)

Further (4.19) can be expressed by the form

df f
= {f, H} + (4.22)
dt t
The above equation states that if f is a constant of motion, then f
t +{f, H} =
0. Furthermore, if f does not involve t explicitly, then {f, H} = 0.
A first integral of a Hamiltonian system is defined to be a quantity that
is a constant of the motion for the flow described by Hamiltons equations.
It follows that if the quantity X is a first integral of a Hamiltonian system
for the Hamiltonian H, then the Poisson bracket {H, X} = 0. Evidently the
Hamiltonian itself is a first integral. For a mechanical system with n degrees of
freedom described by the Hamiltonian H we call it to be completely integrable
if it has n first integrals including the Hamiltonian itself.
Along with (4.22) we have the relation
n  
X qi H qi H
{qi , H} =
j=1
qj pj pj qj
n
X H H
= ij = = qi (4.23)
j=1
pj pi

and
{pi , H} = pi (4.24)
102 Advanced Classical Mechanics

As a result Hamiltonian equations assume the forms

dqi
= {qi , H}
dt
dpi
= {pi , H} (4.25)
dt
where i = 1, 2, ...n. The quantity {u, v} defined by (4.20) is called the Poisson
bracket of two dynamical variables u(qi , pi , t) and v(qi , pi , t) and plays an
important role in Hamiltonian mechanics.

4.4 Properties of Poisson bracket

The Poisson bracket satisfies several interesting properties. Below we give
some of these with proofs.
1. Linearity: {u1 + u2 , v} = {u1 , v} + {u2 , v}
{cu, v} = C{u, v}, c a constant
2. Antisymmetry: {u, v} = {v, u}
3. Product rule: {u, vw} = {u, v}w + v{u, w}
4. Jacobi identity: {u, {v, w}} + {v, {w, u}} + {w, {u, v}} = 0
Proofs:

n  
X (u1 + u2 ) v (u1 + u2 ) v
(1) {u1 + u2 , v} =
i=1
qi pi pi qi
n     
X u1 u2 v u1 u2 v
= + +
i=1
qi qi pi pi pi qi
n   n
X u1 v u1 v X
= + [1 2]
i=1
qi pi pi qi i=1
= {u1 , v} + {u2 , v}
n  
X (cu) v (cu) v
{cu, v} =
i=1
qi pi pi qi
n  
X (u) v (u) v
= c c
i=1
qi pi pi qi
= c{u, v}
 Hamiltonian and Poisson bracket 103

n  
X u v u v
(2) {u, v} =
i=1
qi pi pi qi
n  
X v u v u
=
i=1
qi pi pi qi
= {v, u}

n  
X u (vw) u (vw)
(3) {u, vw} =
i=1
qi pi pi qi
n     
X u v w u v w
= w+v w+v
i=1
qi pi pi pi qi qi
n   n  
X u v u v X u w u w
= w+v
i=1
qi pi pi qi i=1
qi pi pi qi
= {u, v}w + v{u, w}

(4) {u, {v, w}} + {v, {w, u}}

= {u, {v, w}} {v, {u, w}}
n    
X v w v w u w u w
= u, v,
i=1
qi pi pi qi qi pi pi qi
n        
X v w w v u w w u
= u, u, v, + v,
i=1
qi pi qi pi qi pi qi pi
n        
X v w v w w v w v
= u, + u, u, u,
i=1
qi pi qi pi qi pi qi pi
n         
X u w u w w u w u
+ v, v, + v, + v,
i=1
qi pi qi pi qi pi qi pi
= X + Y (say)

where the product rule is employed and X, Y defined by

n        
X v w w v u w w u
X = u, u, v, + v,
i=1
qi pi qi pi qi pi qi pi
n
X v         
w w v u w w u
Y = u, u, v, + v,
i=1
qi pi qi pi qi pi qi pi
104 Advanced Classical Mechanics

Now
n        
X v u w w v u
X = u, v, u, v,
i=1
qi q i p i qi pi pi
n        
X v u w w v u
= u, + ,v u, + ,v
i=1
qi qi pi qi pi pi
n  
X w w w
= {u, v} . {u, v}
i=1
qi pi qi pi
= {{u, v}, w}

where we have used the results

   
v u
{u, v} = u, + ,v
qi qi qi
   
v u
{u, v} = u, + ,v
pi pi pi

which can be proved easily.

We next show that Y = 0. Indeed it is straightforward to see on expanding
the Poisson brackets that
X X 2 w  u u u u

2w

u u u v

Y = [ +
i j
pi pj qi qj qi qj qi qj pj pi pi pj
2w 2w
   
v u u u u u v u
+ + + + ]
qj pi qi pj qi pj qi pj qj pi qj pi

On interchanging the indices i and j the first and second terms vanish due to
antisymmetry while the third term becomes equal to the fourth term except
for a sign. We therefore conclude that Y = 0.
Hence
{u, {v, w}} + {v, {w, u}} = X = {{u, v}, w}
which stands for the Jacobi identity.

4.5 Poisson theorem

If corresponding to a given holonomic system u and v are two constants
of motion, then their Poisson bracket is also a constant of motion.
 Hamiltonian and Poisson bracket 105

Proof:

Using (4.20) we can write

d
{u, v} = {{u, v}, H} + {u, v}
dt t
Using the product rule and Jacobi identity we obtain on rearrangement
   
d u u
{u, v} = u, + , v {{v, H}, u} {{H, u}, v}
dt t t
   
u v
= + {u, H}, v + u, + {v, H}
t t
   
du dv
= , v + v,
dt dt
= 0 (4.26)
since du
dt = 0,
dv
dt = 0, u and v being constants of motion. Hence the theorem.
It is also known as the JacobiPoisson theorem. The Poisson theorem is useful
in uncovering new constants of motion.

4.6 Angular momentum

If lx , ly , lz are the components of the angular momentum vector ~l, then the
Poisson bracket of lx and ly may be obtained as
{lx , ly } = {ypz zpy , zpx xpz }
= {ypz , zpx } {ypz , xpz } {zpy , zpx } + {zpy , xpz }
= y{pz , z}px + x{z, pz }py
while the other terms vanish. Hence
{lx , ly } = lz (4.27)
Similarly

{ly , lz } = lx , {lz , lx } = ly (4.28)

If lx and ly are constants of motion then from Poissons theorem it follows
that their Poisson bracket namely lz is also a constant of motion.
We summarize some additional other results on the Poisson brackets in-
volving lx , ly , lz :
{x, lx } = 0 {x, ly } = z {x, lz } = y
{y, lx } = z {y, ly } = 0 {y, lz } = x
{z, lx } = y {z, ly } = x {z, lz } = 0 (4.29)
106 Advanced Classical Mechanics
~ = (Vx , Vy , Vz ) denotes a constant three-dimensional vector then
If V

~
{x, V ~l} ~ ~r)x
= Vy z Vz y = (V
~
{y, V ~l} ~ ~r)y
= Vz x Vx z = (V
~ ~l}
{z, V ~ ~r)z
= Vx y Vy x = (V (4.30)

The above results imply that

n o
~ ~l = V
~r, V ~ ~r (4.31)

Similarly we can derive n o

~ ~l = V
p~, V ~ p~ (4.32)

The transition from (4.31) to (4.33) involves two compensatory sign

changes: changing ~r and p~ produces a change of sign not only in the defini-
tion of the Poisson bracket but also in the definition of the angular momentum.

Example 4.7

A particle of mass m is acted upon by a constant force F . Using the

techniques of the Poisson bracket show that x and p are in the forms
p0 1F
x = x0 + t+ t
m 2m
p = p0 + F t

where x0 and p0 are the initial values of x and p.

Let g be a function of the canonical variables (q, p). A Taylor series expan-
sion gives the representation
   2  2
dg d g t
g = g0 + t+ + ...
dt 0 dt2 0 2!
where g0 is the initial value of g.
Now since

dg
= {g, H}
dt
2
 
d g dg
= , H = {{g, H}, H}
dt2 dt

and so on, we can represent g as

t2
g = g0 + {g0 , H}t + {{g0 , H}, H} + ...
2!
 Hamiltonian and Poisson bracket 107

For the given problem since the particle is acted upon by a constant force
F , we can identify the Hamiltonian as

p2
H= Fx
2m
implying for g = x the following results:

p2
 
1 p p
{x, H} = x, Fx = {x, p2 } = {x, p} =
2m 2m m m
p p2
 
F F
{{x, H}, H} = , F x = {p, x} = = constant
m 2m m m

So the series for g = x terminates beyond {{x, H}, H} whence it follows that

p0 1F 2
x = x0 + t+ t
m 2m
Similarly choosing g = p we find p = p0 + F t.

Example 4.8

Find the position x and momentum p at time t for the motion of a particle
p2
given by the Hamiltonian H = 2m + 12 m 2 x2 .

We proceed in a similar manner as in the problem above and obtain

   
1 2 2 p0 1 3 3 p0
x = x0 1 t + ... + t t + ... = x0 cos t + sin t
2! m 3! m
p = mx0 sin t + p0 cos t

4.7 Liouvilles theorem

Hamiltons canonical equations can be used to deduce Liouvilles theorem
which points to the preservation of volume in the phase space for a set of
points in it as the phase space evolves in time.
We focus on an n-particle system described in terms of the generalized
coordinates q1 , q2 , , qn having as their conjugate momenta the components
p1 , p2 , ..., pn . These coordinates and momenta constitute a 2n-dimensional vec-
tor in the phase space. Consider the following element of volume dA in such
a phase space:
dA = dq1 dq2 ...dn dp1 dp2 ...dpn (4.33)
108 Advanced Classical Mechanics

As the phase space evolves in time the representative point (qi , pi ), i =

1, 2, ..., n moves to a new location (qi , pi ) in time dt as given by

qi = qi + qi dt, pi = pi + pi dt, i = 1, 2, ..., n (4.34)

The differentials dqi and dpi change according to

qi pi
dqi = dqi + dqi dt, dpi = dpi + dpi dt, i = 1, 2, ..., n (4.35)
qi pi
The difference in the volume element can be expressed as
n
X qi pi
dA0 = dA + [ ( + )]dAdt (4.36)
i=1
q i pi

Using for qi and pi the Hamiltons equations of motion the terms in the squared
brackets in the right-hand side cancel out and we are left with the invariance
result dA = dA0 . These terms actually speak of the vanishing of the divergence
of the velocity vector V ~ having components (q1 , q2 , , qn ; p1 , p2 , ..., pn ), that is,
.V~ = 0.
To find the implication of the above result let be the density of the phase
fluid over the phase space. The following quantity

(q1 , q2 , , qn ; p1 , p2 , ..., pn )dA

can be looked upon as the probability that the system is residing in the in-
finitesimal volume dq1 dq2 ...dq3N dp1 dp2 , ...dp3N .
The continuity equation for
N
X (qi ) (pi )
+ ( + )=0 (4.37)
t i=1 qi pi

can be expanded to read

n N
X X qi pi
+ ( qi + pi ) + ( + )=0 (4.38)
t i=1 qi pi i=1
q i pi

Adjusting the second term in the left-hand side in terms of Poisson brackets
by using Hamiltons equations of motion which also enable us to do away with
the third term we arrive at Liouvilles equation
d
= + {, H} = 0 (4.39)
dt t
It shows that is stationary, i.e., the density of the phase fluid is unchanged.
 Hamiltonian and Poisson bracket 109

4.8 The case of singular Lagrangians

Sometimes we run into a possibility when not all the canonical
momenta are independent. Consider a set of n canonical momenta
(pi , p2 , ..., pi , ...pk , ..., pn ), where pk is given by

L(qi , qi )
pk = , i, k = 1, 2, ...n (4.40)
qk
qi and qi being the usual generalized coordinates and velocities.
A Jacobian matrix can be introduced according to

pk 2L
[ki ] = = (4.41)
qi qi qk
keeping in mind that the mapping from the Lagrangian to the Hamiltonian
form requires replacement of the velocities qs by the canonical momenta ps.
For the independence of the n momenta it is of course necessary that det() 6=
0. Indeed if we do not have an invertible matrix, then there is no unique
solution of the equations of motion. In point of fact, dependence of a few
momenta on others will cause the determinant to vanish.
From the EulerLagrange equation
 
d L(qi , qi ) L(qi , qi )
=0 (4.42)
dt qk qk

explicit evaluation of the time derivative gives

2L 2L L
qi + qi =0 (4.43)
qi qk qi qk qk

which is of the form

L 2L
ik qi = qi (4.44)
qk qi qk
This reads as a first-order differential equation in the velocities vi = qi

L 2L
ik vi = vi hi (qi , vi ) (4.45)
qk qi qk
where hi is a function of coordinates and velocities. Such an equation can be
looked upon as the transformed version of the EulerLagrangian equation.
Now if the determinant of the Jacobian matrix vanishes there would be
existence relationships among the momenta. These could exist as functions

j (p, q, t) = 0, j = 1, 2, ...m (4.46)

110 Advanced Classical Mechanics

Such connections speak of primary constraints of a Hamiltonian theory. For

instance if two momenta p1 and p2 turn out to be equal, resulting from some
choice of the Hamiltonian, an obvious choice for would be p1 p2 .
Before we turn to a concrete example dealing with such a situation let
us have a look at Diracs procedure of handling the Hamiltonian of singular
systems subject to constraints like (4.46). Dirac proposed an extended Hamil-
tonian of the form X
H = H + vk k (4.47)
i

where H corresponds to the Legendre transformed form

X
H = pi qi L (4.48)
i

and is free from the velocities which are present in the singular part of (4.29).
If we go for a virtual variation, H for the right-hand side of Legendres
form above would be
X X X L X L X
H = qi pi + qi pi qi qi = (qi pi pi qi ) (4.49)
i i i
qi i
qi i

with the qi terms canceling out. But being a function of the qs and ps the
variation of H in itself gives
X H H
H = ( qi + pi ) (4.50)
i
qi pi

Hence on comparison of the two expressions of H leads to

n
X H H
[(qi )pi (pi + )qi ] = 0 (4.51)
i=1
p i qi

On the other hand, variation of (4.46) shows

n
X j j
[ pi + qi ] = 0, j = 1, 2, ..., m (4.52)
i=1
pi qi

The last two equations are consistent if q and p are given by

n
H X l H
qi = + vl = = {qi , H}, i = 1, 2, ..., n (4.53)
pi i=1
pi pi

n
H X l H
pi = vl = = {pi , H}, i = 1, 2, ..., n (4.54)
qi i=1
qi qi
 Hamiltonian and Poisson bracket 111

which one has to remember that H, the total Hamiltonian, is given by the
extended form (4.47). Note that to build up a viable theory, the primary
constraints j are to be supplemented by the secondary restriction j = 0
because like the primary constraints the time derivative of these constraints
must vanish at all times.

4.9 Higher derivative classical systems

Development of higher-derivative classical systems has a long history dat-
ing back to an early work of Mikhail Vasilyevich Ostrogradski who extended
the HamiltonLagrange theory to a generalized solvable form. We give a brief
exposition to the procedure of deriving the Lagrangian form of any second-
order ordinary differential equation. Basically in the simplest case of a me-
chanical system with one degree of freedom, the point of enquiry is under what
condition a given family of differential equations of second-order, namely,

q = F (t, q, q) (4.55)

acquires a Lagrangian interpretation in the sense that (4.55) can be repre-

sented in the manner  
L d L
=0 (4.56)
q dt q
where L, the Lagrangian, is a function of q, q and t. When expanded (4.56)
reads
L 2L 2L 2L
q q 2 = 0. (4.57)
q t q q q q
When confronted with (4.45), the above equation takes the form

L 2L 2L 2L
q F (t, q, q) 2 = 0, (4.58)
q t q q q q
where q has been replaced by F for consistency. It then follows easily that if
we take a partial derivative with respect to q and slightly readjust the terms
then we arrive at the form
d F M
M+ M= + (M q) + (M F ) = 0, (4.59)
dt q t q q
112 Advanced Classical Mechanics

where the quantity M is defined by

2L
 
M (t, q, q) = , (4.60)
q 2
M is referred to as the Jacobi last multiplier (JLM ).
In terms of M , the Lagrangian of a second-order differential equation ad-
mits the following structure:
Z q Z q0
L(t, q, q) = M (t, q, q 00 )dq 00 dq 0 + A(t, q)q + B(t, q). (4.61)

The simplest choice M = 1 shows that the corresponding Lagrangian can be

at most a quadratic function of q:
1 2
L= q + A(t, q, q)q + B(t, q, q) (4.62)
2

4.10 The PaisUhlenbeck oscillator

The PaisUhlenbeck oscillator is a suitable special example obeying fourth-
order dynamical equation of motion
....
q + q + q = 0 (4.63)

where = (12 + 22 ) and = 12 22 are real and positive parameters. Let us

replace it by an equivalent system of two coupled differential equations that
follow from the usual first-order EulerLagrange equation.
Defining a new set of coordinates

Q1 := q, Q1 = q; and Q2 := q (4.64)

one can apply the usual EulerLagrange equation

d L L
( ) = 0, i = 1, 2 (4.65)
dt Qi Qi
along with the Lagrangian
1 1 1
L = Q22 Q21 Q1 Q2 + Q21 (4.66)
2 2 2
to get to a coupled second-order system

Q1 + Q2 + Q1 = 0, Q1 = Q2 (4.67)

Eliminating the variable Q2 gives the desired equation for PUO.

 Hamiltonian and Poisson bracket 113

The Lagrangian allows us to define the conjugate momenta in the usual

way
L
Pi = , i = 1, 2 (4.68)
Qi
With the Hamiltonian

H = Pi Qi L (4.69)
and canonical momenta given by

P1 = Q1 Q2 , P2 = Q1 (4.70)
a Hamiltonian for the PUO system can be extracted using L given by (4.66)
1 1 1
H = P1 P2 + P22 + Q22 Q21 (4.71)
2 2 2
From the standard definition of the Poisson bracket

Qi = {Qi , H}, Pi = {Pi , H}, i = 1, 2 (4.72)

the first-order Hamilton equations are given by

Q1 = P2 , Q2 = P1 + P2 ; P1 = Q1 , P2 = Q2 (4.73)

4.11 Summary
In this chapter we gave a brief description of the Hamiltonian formalism
in classical mechanics that forms the basis for working in a phase space. We
derived Hamiltonian canonical equations of motion in terms of the generalized
coordinates and generalized or canonical momenta. Unlike Lagranges equa-
tions of motion which hold in the configuration space, Hamiltons equations
are a set of two coupled partial differential equations representing the mo-
tion of the system in the phase space. We introduced Poisson brackets and
reviewed their main properties including recasting Hamiltonian equations in
terms of the Poisson bracket. We also discussed Liouvilles theorem. Further
we touched upon the subject of singular Lagrangians and gave an exposition
to higher derivative classical systems.
114 Advanced Classical Mechanics

Exercises
(1) If the Hamiltonian of a dynamical system is given by H = p1 q1 p2 q2
aq12 + bq22 where a, b are constants, show that p2 bq
q1
2
=constant.

(2) If H = qp2 pq + bp, find q and p as functions of t. Here b is a constant.

(3) The Lagrangian for the motion of a particle in a rotating frame is

1 2 1
L= m~r + m 2 (x21 + x22 ) + m(x1 x2 x2 x1 )
2 2
Find the Hamiltonian.

(4) For a linear transformation from the canonical variables (q, p) to the
pair (Q, P ) given by Q = a11 q + a12 p, P = a21 q + a22 p, find the conditions on
the numbers aij , i, j = 1, 2 so that {q.p} = 1 and pq = P Q + dFdt where F is a
function of q and p.

(5) Consider a particle of mass m attached to one end of a string placed

in a vertical position. The tension of the spring is subject to Hookes law.
The other end of the string is attached to a fixed point. Suppose that the
spring constant is k and its natural length is l. Write down the Lagrangian
and Hamiltonian.

(6) The point of suspension of a simple pendulum of length l and mass m

is constrained to move on a parabola given by the equation = ax2 , a > 0
in the vertical plane. Derive the Lagrangian and Hamiltonian governing the
motion of the pendulum and its point of suspension.

(7) A Hamiltonian of a mechanical system with one degree of freedom has

the form
p2 a 2 at k
H= qpeat + q e (a + eat ) + q 2
2a 2 2
where a, , k are constants, q is the generalized coordinate and p is the gener-
alized momentum. Find the Lagrangian and an equivalent Lagrangian which
is not explicitly time-dependent.

(8) Set up Hamiltonian equations for the Lagrangian

m 2 2
L(q, q, t) = (q sin t + q q sin 2t + q 2 2 )
2

(9) If X~ is defined as X
~ = c1~r + c2 p~ + c3 (~r p~) where c1 , c2 , c3 are arbi-
trary constants, then show for an arbitrary vector V ~ that the Poisson bracket
~ V
{X, ~ .~l} = V ~ where ~l is the angular momentum vector.
~ X,
 Hamiltonian and Poisson bracket 115

(10) Consider a particle of mass m and charge e moving in a magnetic field

~ and described by the Lagrangian
A
1 2 e ~
L= mr + r A
2 c
where c is a constant. Derive the following Poisson brackets:
e e e
{vx .vy } = Az , {vy .vz } = Ax , {vz .vx } = Ay ,
cm2 cm2 cm2
~
~ ec A
p
where ~v is given by ~v = m .
Chapter 5
Dynamical systems: An overview

5.1 Basic notions and preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118

5.2 Simple examples from classical mechanics . . . . . . . . . . . . . . . . . . . . . . . 122
5.3 Analysis of a linear system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
5.4 Nonlinear systems: Process of linearization . . . . . . . . . . . . . . . . . . . . . . 133
5.5 LotkaVolterra model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
5.6 Stability of solutions: Lyapunov function . . . . . . . . . . . . . . . . . . . . . . . . 141
5.7 Van der Pol oscillator and limit cycles . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
5.8 Bifurcations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
5.9 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158

A dynamical system is generally concerned with an evolutionary problem re-

lated to some physical, chemical or biological system in which the values of the
underlying variables are specified in an ordered way. A dynamical system can
describe a continuous evolution process addressing input and output variables
through a differential equation (linear or nonlinear) or a discrete one that
is controlled by difference equations or even deal with hybrid schemes which
integrate models of continuous and discrete dynamics. In this chapter we will
focus only on the continuous class of dynamical systems. One can consider such
systems analytically or subject them to numerical simulations when tracking
their behavior goes out of hand chiefly because of the uncontrollable nonlin-
ear influence. A dynamical system is deterministic or random depending on
when being subject to a specific evolution rule the system is taken uniquely
from a given definite state to a subsequent one or fails to be deterministic
in which case a stochastic differential equation can guide it. Mathematically,
nonlinear equations do not normally have solutions which would superpose
making the systems they represent rather complicated and difficult to ana-
lyze. In fact, most nonlinear systems are not analytically tractable and it is
impossible to predict for an arbitrarily long time the dynamical properties of
even weakly perturbative series. In a nonlinear theory multiple equilibrium
points can occur. One also frequently encounters such terms as the limit cy-
cles, high-period orbits and the phenomenon of chaos. For instance, a limit
cycle typically refers to an isolated closed trajectory in the sense that when
one looks at its neighboring trajectories they are not closed: either they tend
to diverge away from the limit cycle or spiral toward it. On the other hand,
a chaotic system is hypersensitive to initial conditions. In fact, a very slight

117
118 Advanced Classical Mechanics

difference in the initial conditions can cause the trajectories to follow different
lines of evolution.

5.1 Basic notions and preliminaries

In general, a dynamical system refers to an initial value problem of the
kind

dX~
= f~(X,
~ t, u) (5.1)
dt
~
where X(t) stands for an m-dimensional vector and f~ is a vector field which
can also depend on X~ and t apart from a set of auxiliary variables collectively
~ and f~ are represented by
denoted as u. More specifically X

x1 (t) f1
x2 (t) f2
~ , f~ ...

X(t) ... (5.2)
... ...
xm (t) fm
Prey-predator systems, bacterial growth and radioactive decay of atoms are
some simple examples of a dynamical system.
In principle f can depend upon several other parameters like the coupling
constant, amplitude, etc. but these have been omitted from giving an explicit
display. We assume that we have a deterministic system at hand and the
set of initial conditions (x1 (0), x2 (0), ...xm (0)) serves as judicious inputs to
determine uniquely the solution we are after. Normally these initial conditions
are always subject to uncontrollable fluctuations in the measurements because
no matter how good our intention is we can never make an infinitely precise
measurement. Thus, measurements are always subject to uncertainties so that
errors are bound to creep in. The question then becomes relevant as to how
sensitive are the solutions to small errors accompanying the initial conditions.
We take the evolution rule to be guided by a map from the underlying phase
space to itself. We parameterize such a map by the time parameter t and refer
to it by the function t (x).
Any differential equation is characterized by its order. Thus, (5.1) repre-
sents a system with order m. The simplest one corresponds to m = 1 which
is a straightforwardly integrable case:

Zt
x = x(0) + f (s, x(s), u)ds (5.3)
0
 Dynamical systems: An overview 119

for some initial condition x(0). The general solution for all t corresponding to
the set of initial conditions (x1 (0), x2 (0), ...xm (0)) is called a complete flow or
simply a flow and the set of points along a flow is called the trajectory or an
orbit. A flow at time t is the mapping t (x). We note that 0 (x0 ) = x0 and
that the composition rule for the flow, namely, t s = t+s holds. Here the
composition means t s = t (s (x)).
If t is not explicitly present in (5.1) the system is classified as an au-
tonomous system. Autonomous systems are thus stationary and in the absence
of the set of auxiliary variables u, X ~ is given by

dX~
= f~(x1 , x2 , ...xm ) (5.4)
dt
We say that the vector field f~ is subject to the Lipschitz condition on a set S
if for any pair of coordinates (X~ (i) , X
~ (j) ) S Rm there exists a constant
K such that

|f~(X
~ (i) ) f~(X
~ (j) )| < K|X
~ (i) X
~ (j) | (5.5)
for all K > 0. If the vector field f~ is subject to the Lipschitz condition in S,
then there exists a unique solution for the differential equation for a given set
of initial conditions.
The set of Hamiltons equations of motion defined in terms of the Hamil-
tonian H = H(x1 , x2 , ..., xm ; p1 , p2 , ..., pm ) but not consisting of t explicitly,
namely,

dxi H
xi =
dt pi
dpi H
pi = (5.6)
dt xi
(i = 1, 2, ..., m) is an example of a conservative autonomous dynamical system
with f~ standing for the column matrix
H
p1
H
p2

...


...


...

H
f~ pm (5.7)

H
x1

H


x2

...


...

...
H
xm
120 Advanced Classical Mechanics

More compactly, we can reexpress (5.5) in the manner

dX~ 
0 +Im

= Af~, A = (5.8)
dt Im 0

where X ~ is a column matrix like the one given in (5.2) whose first m entries are
the coordinates (x1 , x2 , ...xm ) and the remaining m entries are the momenta
(p1 , p2 , ...pm ). A is written in a block form with the object Im representing for
the (m m) identity matrix.
State variables are those that offer a complete description of the state
of a dynamical system while the state space corresponds to the set of all the
possible values of the state variables. The dimension of the dynamical system is
given by the number of state variables. Phase space normally refers to the state
space which is continuous and finite-dimensional. The phase plane consists of
trajectories obtained from the solutions of (5.1). For instance, taking m = 1,
a phase plane is guided by the pair (x, dx dt ) while for m = 2, the phase plane
corresponds to (x1 , x2 ) or simply (x, y) where x and y are a pair of Cartesian
coordinates. An ensemble of trajectories for a given system in the phase plane
or a phase space is called a phase portrait (see Figure 5.1). An isocline refers
to a curve in a phase plane or a phase space on which the trajectories have a
fixed gradient. In general, the set of initial conditions (x1 (0), x2 (0), ...xm (0))
dictates how a dynamical system evolves ultimately toward the final state
(which may be a point or an area or a curve depending on the manifold
the dynamical system operates in) in the phase space or state space. Such a
destination is called an attractor.
An equilibrium point or a stationary point is a solution resulting from the
vanishing of the right-hand side of (5.1). Thus, if ~x is an equilibrium or a fixed
point we have
, u) = 0
f~(t, ~x (5.9)
An equilibrium point or a stationary point is also referred to as a fixed
point or a critical point. The equilibrium may or may not be stable. If say,
the origin (0, 0) represents a state of stable equilibrium toward which the set
of all the trajectories is driven at, then the state of stable equilibrium of the
origin is termed the point attractor or a sink. If the opposite is the case, we
have a repeller or a source at the origin.
We now turn to the following examples.

Example 5.1

Consider a simple dynamical system

x = f (x), f (x) = sin x (5.10)

 Dynamical systems: An overview 121

FIGURE 5.1: Phase portrait.

The fixed points are at x = n, n = 0, 1, 2, .... The integration can be

performed easily starting with an initial value of x0 = 4 to arrive at the
following form for t:
1+ 2
t = ln | | (5.11)
csc x + cot x
While it is not immediately clear about the large-t behavior of x(t), it is
instructive to do a geometrical analysis to have an idea of the stable and un-
stable nature of the fixed point.

Example 5.2

Consider the damped pendulum problem where we modify the pendulum

equation (1.68) by including the effect of a frictional term which is proportional
to the velocity of the motion
g
+ + sin = 0 (5.12)
l
where is a damping constant. We observe that (5.12) can be put in the
framework of a dynamical system by casting it as a pair of coupled first-order
differential equations by setting x = and y = :
g
x = y, y = sin x y (5.13)
l
In this way we can convert almost any higher-order differential equation into
a system of first order ones by introducing suitable intermediate variables. In
122 Advanced Classical Mechanics

this particular problem, it is the time-derivative of x, namely, the variable y,

which serves the purpose. Of course such a procedure increases the dimensions
of the system. Note that in interpreting (5.12) in terms of (5.13) we have
enhanced the dimension by two.
The fixed points of (5.13) are identified by setting the right-hand side of
each equation to zero and are readily obtained as the set of points (n, 0),
n = 0, 1, 2, .... Note that while (0, 0) is a stationary solution which conforms
to the usual stable equilibrium position of the pendulum hanging straight
down, the solution (, 0) is stationary but unstable since they point to the
straight up position of the pendulum.
We therefore see that the fixed points could be stable or unstable. Our
next point of inquiry will be to look into certain stability aspects of some
simple systems of classical mechanics.

5.2 Simple examples from classical mechanics

The one-dimensional Newtons second law of motion in the presence of a
potential V (x) is described by a pair of dynamical equations
p dV
x = , p = (5.14)
m dx
where x is the position and p is the momentum of the particle. An extremum
of the potential occurs when p = 0.
If (x0 , p0 ) is an equilibrium position of the system then, as we saw in
(1.41), Taylor series expansion about x = x0 gives for small values of (x x0 )
an approximation of V (x)
1
V (x) V (x0 ) + (x x0 )V 0 (x0 ) + (x x0 )2 V 00 (x0 ) (5.15)
2
Such a representation of V (x) when plugged in the expression for the momen-
tum p as furnished in (1.91) gives
1
(p p0 )2 = 2m[E E0 (x x0 )2 V 00 (x0 )] (5.16)
2
where we assume that E has the value E0 at the point (x0 , p0 ) and use the
criterion that at an extremum point V 0 (x0 ) = 0.
Two possibilities immediately present themselves from the above expres-
sion. When the potential has a minimum at (x = x0 ), then V 00 (x0 ) > 0, the
system being subjected to an attractive force, and the locus of the represen-
tative point describes elliptic trajectories with center at (x0 , y0 )

(y y0 )2 (x x0 )2
= 1 (5.17)
b2 a2
 Dynamical systems: An overview 123

where a2 = 2(EE 0) 2
V 00 (x0 ) and b = 2m(E E0 ) are, respectively, the squares of the
semi-major and semi-minor axis of the ellipse. We call the point (x0 , p0 ) in
the phase plane an elliptic singularity for E > E0 which is in a state of stable
equilibrium.
On the other hand, when V 00 (x0 ) < 0, which corresponds to the possibility
of a repulsive force, the representative point traces out hyperbolic trajectories
as given by
(y y0 )2 (x x0 )2
2
= 1 + (5.18)
b a2
with respect to the center (x0 , y0 ). Note that ab are the slopes of the asymp-
totes that pass through the center and a is the fixed distance of the vertices
from the center. The point (x0 , p0 ) in the phase plane is called a hyperbolic
singularity or a saddle. The motion in the neighborhood of (x0 , y0 ) is in a
state of unstable equilibrium.
Take the case of the harmonic oscillator when the particle executes a simple
harmonic motion being attracted to a given fixed point by a constant force
F = kx, k > 0, which is proportional to the distance from the point.
The harmonic oscillator is a conservative system and as was already noted in
Chapter 1, it admits
q of a general trigonometric solution x(t) = A cos(0 t +
k
), where 0 = m, implying that p(t) = Am0 sin(0 t + ). With the
1 2 2
potential for such a conservative system being V (x) = 2 m0 x , the energy
2
p2
1 2 2 mA 02
E turns out to be E = 2m +2 m0 x = 2 . Squaring and adding the
expression for x and p we readily find that the representative point traces out
paths of ellipse
q inthe phase plane with semi-major and semi-minor axis of
lengths ( 0 2E
1
m , 2Em). We identify the point (0, 0) as the point of stable
equilibrium.
When the force is repulsive, i.e., F = kx, k < 0, a different picture
emerges. Here the trigonometric solutions are to be replaced by their hyper-
bolic ones and we run into the possibility of two different combinations of
solutions, namely, x(t) = A cosh(0 t) and p(t) = Am0 sinh(0 t) or the
pair x(t) = A sinh(0 t) and p(t) = Am0 cosh(0 t). In the first case, the
p2 mA2 2
energy is E = 2m 12 m02 x2 = 2 0 which is negative while in the second
mA2 2
case, the energy is E = + 2 0 which is positive. For the negative energy
solution we have at t = 0, x = A and p = 0 which means that the parti-
cle during the course of its motion from x = passes through the points
x = A and subsequently moves away in the future. For the positive energy
solution, however, the particle takes up the position x = 0 at t = 0 with a
finite value of p = Am0 on the center of repulsion in the phase plane. It is
clear that the asymptotes, which intersect at the point of unstable equilibrium
(0, 0), branch out the regions of positive and negative energy solutions.
124 Advanced Classical Mechanics

What happens in the case of the simple pendulum problem? The equation
of motion we wrote down in (1.68), namely,
g
+ sin = 0 (5.19)
l
has the accompanying Hamiltonian
1 2 g
H= cos (5.20)
2 l
From the above form of H we can identify different types of regions for the
motion. In the interval ( gl , gl ), the motion is finite and oscillatory in character
and corresponds to curves for the oscillatory pendulum motion swinging back
and forth about its vertical position. However, while H < gl does not allow
for any physical motion, being negative in this case, the other extreme
corresponding to H > gl points to an unbounded motion. H = gl corresponds
to the separatrix which separates the phase space, formed by the pair of
variables (, ), into two distinct regions. It passes through the point (, 0) in
the phase space.
To determine the equation of the separatrix we note that on it the Hamil-
tonian assumes the form
g 1 g
= 2 cos (5.21)
l 2 l
which gives for
r
g
= 2 cos (5.22)
l 2
The above equation can be easily integrated to yield
s
l 1
t= ln tan ( + ) (5.23)
g 4
and represents the equation of the
gseparatrix.
Writing (5.23) in the form e l t = tan( 4 + 4 ) with a similar one for its
inverse, and then adding the two it is straightforward to obtain the relationship
r
g
cos = sech t (5.24)
2 l
Thus from (5.22) we arrive at the form of the velocity profile
r r
g g
= 2 sech t (5.25)
l l
It represents a bell-shaped curve resembling a soliton-like solitary wave. The
latter decays asymptotically with time while its width of the curve goes in-
versely as the factor gl .
p
 Dynamical systems: An overview 125

5.3 Analysis of a linear system

Consider a system of n linear equations as given by

x1 = a11 x1 + a12 x2 + ... + a1n xn

x2 = a21 x1 + a22 x2 + ... + a2n xn
....
xn = an1 x1 + an2 x2 + ... + ann xn (5.26)

with constant coefficients aij , i, j = 1, 2, ...n. We assume det(aij ) 6= 0 of the

coefficient matrix Aij . Looking for solutions in the exponential forms

x1 = 1 et , x2 = 2 et , ...xn = n et (5.27)

with constant k s, k = 1.2....n and , the set of equations (5.13) is trans-

formed to

(a11 )1 + a12 2 + ... + a1n n = 0

a21 1 + (a22 )2 + ... + a2n n = 0
....
an1 1 + an2 2 + ... + (ann )n = 0 (5.28)

From the theory of linear algebraic equations, we know that (5.15) will ad-
mit of nontrivial solutions in k s provided the determinant of the coefficient
matrix vanishes:

a11 a12 ... a1n
a21 a22 ... a2n
det =0 (5.29)
... ... ... ...
an1 an2 ... ann

The above equation is referred to as the characteristic equation for . Its roots
are the eigenvalues of the matrix A.
Let us discuss the simplest case corresponding to n = 2 for the matrix A.
In the Cartesian (x, y) plane we address the two equations

x = a11 x + a12 y
y = a21 x + a22 y (5.30)

for < t < +, having the underlying determinant a11 a22 a12 a21 .
The characteristic equation for such a reduced case turns out to be a
quadratic in , namely,

2 (a11 + a22 ) + = 0 (5.31)

126 Advanced Classical Mechanics

If 1 and 2 are the two roots of this equation, then these are given by
1 p
1 , 2 = [T rA (T rA)2 4] (5.32)
2
In other words,

2 (a11 + a22 ) + ( 1 )( 2 ) (5.33)

with

T rA = (a11 + a22 ) = (1 + 2 ), = a11 a22 a12 a21 = 1 2 (5.34)

We call D to be the discriminant

D = (T rA)2 4 (5.35)

and it is easily found that D = (1 2 )2 .

Several possibilities present themselves depending upon the sign of the
discriminant D and the sign of the determinant . If > 0, then the following
obvious three cases turn up:
(a) = 1 2 >0, D>0
Here 1 and 2 are real and distinct eigenvalues keeping the same sign. If
T rA <0 and the exponentials in (5.14) act as damping factors (sink), then the
system is stable for asymptotic t. The solution is referred to as the stable node
(see Figure 5.2). If T rA >0 and the exponential factors in (5.14) grow with
time (source), the opposite situation is encountered: the system is unstable

FIGURE 5.2: Stable node.

 Dynamical systems: An overview 127

FIGURE 5.3: Unstable node.

for asymptotic t. The solution is referred to as the unstable node (see Figure
5.3). If T rA =0, then 1 =- 2 implying = 1 2 < 0 contrary to the chosen
sign of and hence is not a possible mode.
(b) = 1 2 >0, D<0
Here complex eigenvalues are signalled. Let these be

1 = + i, 2 = i (5.36)
where the imaginary part 6= 0. If T rA <0, then it implies < 0 and we have
an asymptotically stable case. The solution is referred to as the stable focus
or a stable spiral point (see Figure 5.4). The contrary behavior is noticed
for T rA >0 when > 0. Such a case corresponds to an unstable focus or
an unstable spiral point (see Figure 5.5). If T rA =0, for which = 0, the
solutions are guided only by a phase term eit suggesting only closed orbits
around the fixed point. The solution is stable in character and referred to as
the neutrally stable center rather than the asymptotically stable fixed point
(see Figure 5.6).
(c) = 1 2 >0, D=0
This is the case of equal roots 1 = 2 . We carry out the analysis as
follows: If there are two linearly independent eigenvectors, then (T rA) < 0
corresponds to 1 = 2 < 0 and we have asymptotic stability which is that
of a nodal point and referred to as the stable star node. In this connection it
needs to be noted that if any of the coefficients of our basic equation (5.17)
are fluctuated a little, then this may cause a slight displacement of the equal
roots to a pair of conjugate complex roots or a pair of real roots. Thus, the
position of a stable star node is intermediate between a nodal point or a focal
128 Advanced Classical Mechanics

FIGURE 5.4: Stable focus or a stable spiral point.

FIGURE 5.5: Unstable focus or unstable spiral point.

 Dynamical systems: An overview 129

FIGURE 5.6: Neutrally stable center.

point. If however (T rA) > 0, then the opposite behavior holds and we have an
unstable star node. The case (T rA) = 0 points to 1 =- 2 which cannot hold
unless both the eigenvalues vanish pointing to =0 which is contrary to the
assumption > 0. If there is only one independent eigenvector, then the role
of stable star node and unstable star node are replaced by stable degenerate
node and unstable degenerate node, respectively.
Next we turn to = 1 2 < 0. Here the eigenvalues are of opposite signs
showing that while one of the exponentials in (5.14) decays, the other grows.
We thus have a combination of trajectories one of which asymptotically (with
time) converges to the fixed point as represented by the negative eigenvalue
and the other diverges (hyperbolic in nature) as represented by the positive
eigenvalue. With the presence of one diverging term due to the effect of the
positive eigenvalue, the overall stability cannot be assured and so the fixed
point is an unstable saddle point (see Figure 5.7).
Finally we discuss the case when = 1 2 = 0. It follows that at least
one eigenvalue is zero and we get a line of fixed points. To clarify such an
issue we consider first the case of T rA <0 with 1 = 0 and 2 < 0. It gives
a line of neutrally stable (non-isolated) fixed points. On the other hand, for
the case T rA >0 the opposite feature holds and we have a line of unstable
130 Advanced Classical Mechanics

FIGURE 5.7: Unstable saddle point.

(non-isolated) fixed points. For T rA = 0, the two eigenvalues simultaneously

acquire a zero value. If there are two linearly independent eigenvectors, then
we have a plane of neutrally stable (non-isolated) fixed points while for only
one independent eigenvector, the plane of neutrally stable (non-isolated) fixed
points is replaced by a line of unstable (non-isolated) fixed points.
Some comments about the nature of the general solutions. For two distinct
real eigenvalues, the general solution can be expressed as

~x = c1 ~1 e1 t + c2 ~2 e2 t (5.37)
where c1 and c2 are constants and ~1 and ~2 are two independent eigenvectors.

When there are repeated real eigenvalues with two linearly independent
eigenvectors the general solution can be expressed as

~x = (c1 ~1 + c2 ~2 )et (5.38)

where c1 and c2 are constants, 1 = 2 = , and ~1 and ~2 are two indepen-
dent eigenvectors. For the case when there is only one independent eigenvector

~ , the general solution reads

x = (c1 ~ t)et
~ + c2 (5.39)
where c1 and c2 are constants.
 Dynamical systems: An overview 131

For the conjugate complex eigenvalues the general solution is expressed as

p + et (c1 0 cos t + c2 0 sin t)~q

x(t) = et (c1 cos t + c2 sin t)~ (5.40)

where c1 and c2 are constants, p~ and ~q are two independent eigenvectors and
c1 0 and c2 0 are another set of constants related to c1 and c2 .

We illustrate the above notions by means of the following and subsequent

examples.

Example 5.3

Let us consider the damped oscillator problem as considered earlier in

(1.43)

x + 02 x + 0 x = 0 (5.41)
where is proportional to the damping constant incorporating the effects of
the friction.
We can interpret the above second-order equation as a combination of two
first order equations
x = y, y = 02 x 0 y (5.42)
By equating to zero the above expressions for x and y, the fixed point is clearly
seen to be at the origin (0, 0).
The accompanying characteristic equation is evaluated from the determi-
nant  
1
det = 2 + 0 + 02 0 (5.43)
0 2 0
It gives for the eigenvalues
0 0 p 2
1 , 2 = 4 (5.44)
2 2
First of all, if = 0, then the system reduces to the simple harmonic
oscillator depicting a periodic motion. The eigenvalues are given by i0 . We
therefore have a center at the origin. Next with > 2 (overdamping), the fixed
point (0, 0) is a stable node. It includes also the case of critical damping for
= 2. However, in the range 0 < < 2 (underdamping), the fixed point (0, 0)
is a stable focus. Finally, corresponding to < 0, we have an unstable focus
and unstable node for the fixed point (0, 0) corresponding to the respective
cases of 2 < 4 and 2 4.
132 Advanced Classical Mechanics

Example 5.4

Consider a linear system as given by

x = x y, y = 7x + 5y 3 (5.45)

The fixed point is obtained by setting the derivatives x and y equal to zero.
We therefore get a unique solution ( 14 , 14 ). For a linear system the matrix A is a
constant. The eigenvalues are obtained by solving the characteristic equation
2
which for this system is easily found to be 6 + 12 = 0. The roots are
in conjugate complex pairs namely 3 i 3. It has a positive real part and
hence can be identified as an unstable focus or a spiral source. To find the
eigenvectors we solve for the matrix equation
    
1 1 0
= (5.46)
7 5 0

for 1,2 = 3 i 3. For the positive sign we obtain the following link between
and :
= (2 + i 3) (5.47)
implying that the accompanying two eigenvectors are
 
1
(1) = (5.48)
2 i 3

and replacing i by -i, the second one

 
(2) 1
(5.49)
2 + i 3

Hence the solution provided by the first eigenvector (1) is

 
1
= e(3+i 3)t (5.50)
2 i 3

which can also be put as

 
= e3t cos 3t + i sin 3t
(5.51)
(2 cos 3t 3 sin 3t) + i(2 sin 3t + 3 cos 3t)

where we have used Eulers formula ei = cos + i sin . Splitting the real and
imaginary components this is of the form 1 + 2 where
   
3t cos 3t 3t sin 3t
1 = e , 2 = e
(2 cos 3t 3 sin 3t 2 sin 3t + 3 cos 3t
(5.52)
 Dynamical systems: An overview 133

For the second eigenvector (2) , we would be similarly led to the form 1 2 .
Of course for a linear system the sum and difference of two solutions are also
solutions and so we conclude that both 1 and 2 are solutions too. These
are real solutions and linearly independent and hence the general solution is
obtained as
   
3t cos 3t 3t sin 3t
x(t) = c1 e + c2 e
(2 cos 3t 3 sin 3t 2 sin 3t + 3 cos 3t
(5.53)
where c1 and c2 are constants which can be determined from some given initial
conditions.

5.4 Nonlinear systems: Process of linearization

Let a pair of coupled differential equations be given by
x = ax + by + f (x, y), y = cx + dy + g(x, y) (5.54)
that has as its linear portion the equations
x = ax + by, v = cx + dy (5.55)
In (5.54) and (5.55), a, b, c and d are some real constants. The lineariza-
tion is carried out about a certain relevant point. Choosing it to be a
fixed point (x0 , y0 ), we assume that f and g are continuously, differentiable
functions about it. We also assume that the ratios f (x,y)
r and g(x,y)
r where
p
r = (x x0 )2 + (y y0 )2 are well behaved at the origin in the sense that
both of them approach 0 as r 0.
Let us consider a two-dimensional nonlinear system

dX~
= Af~ (5.56)
dt
~ and f~ are given by the column matrices
where X
   
~ x(t) ~ f1 (x, y)
X(t) , f (5.57)
y(t) f2 (x, y)
Suppose (x0 , y0 ) is a fixed point about which we carry out a Taylor expan-
sion for f1 (x, y) and f2 (x, y) as follows:
f1 f1
f1 (x, y) = f1 (x0 , y0 ) + (x x0 )( ) + (y y0 )( ) + R1 (x, y) (5.58)
x 0 y 0
f2 f2
f2 (x, y) = f2 (x0 , y0 ) + (x x0 )( ) + (y y0 )( ) + R2 (x, y) (5.59)
x 0 y 0
134 Advanced Classical Mechanics

where R1 (x, y) and R2 (x, y) are the remainder terms controlled by the condi-
tions
R1 (x, y) R2 (x, y)
lim [ ] = 0, lim [ ]=0 (5.60)
r0 r r0 r
p
where r = (x x0 )2 + (y y0 )2 .
Since (x0 , y0 ) is a fixed point, it should satisfy the stationary condition
and hence f1 (x0 , y0 ) and f2 (x0 , y0 ) (being just constants) have to vanish.
Transforming to new variables X and Y by performing a shift about the fixed
point namely
X = x x0 , Y = y y0 (5.61)
it readily follows that X and Y are given by
f1 f1
X = X( ) +Y( ) + R1 (X + x0 , Y + y0 ) (5.62)
x 0 y 0

f2 f2
Y = X( ) +Y( ) + R2 (X + x0 , Y + y0 ) (5.63)
x 0 y 0
On identifying
f1 f1 f2 f2
a=( ) ,b = ( ) ,c = ( ) ,d = ( ) (5.64)
x 0 y 0 x 0 y 0

and neglecting the remainder terms R1 and R2 we achieve a linear approx-

imation of the nonlinear system (5.62) and (5.63) but in terms of X and Y
variables. We indeed obtain
   
d X X
=M (5.65)
dt Y Y

where the two-dimensional matrix M given by

!
( f
x )0
1
( f
y )0
1

M = (5.66)
( f
x )0
2
( f
y )
2
0

is called the Jacobian matrix. The main task is to evaluate the Jacobian ma-
trix at the fixed point and go for the linearization. Let us take some simple
situations in the following examples.

Example 5.5

Suppose a dynamical system is described by the equations

x = 3x x2 xy, y = 2y xy 2y 2 (5.67)

By putting x = 0 and y = 0 we determine the fixed points to be (0, 0) and

(4, 1). Note that here, according to our notations, f1 = 3x x2 xy and
 Dynamical systems: An overview 135

f2 = 2y xy 2y 2 . About the point (0, 0), we get ( f f1

x )0 = 3, ( y ) =
1
0
0, ( f f2
x )0 = 0 and ( y )0 = 2. Hence the matrix M is diagonal and given by
2

 
3 0
M = (5.68)
0 2
It has positive eigenvalues 3 and 2. Hence the fixed point is an unstable node.
What about the other fixed point (4, 1)? Since at this point the values of
the partial derivatives are ( f f1 f2 f2
x )0 = 4, ( y )0 = 4, ( x )0 = 1 and ( y )0 =
1

2, the corresponding matrix M is given by

 
4 4
M = (5.69)
1 2
By solving the characteristic equation (5.17), weget a complex pair of roots
for with a negative real part as given by 3 i 3. The fixed point is there-
fore a stable focus.

Example 5.6

Let us consider the system

x = 2x xy 2x2 , y = 2x + xy (5.70)
Here f1 = 2x 2x2 xy and f2 = 2y + xy. The fixed points are easily
determined to be (0, 0) and (2, 6) by putting x = 0 and y = 0. For the fixed
point (0, 0), the eigenvalue 2 is repeated and the diagonal matrix turns out to
be
 
2 0
M = (5.71)
0 2
The general solution is of the type
   
1 0
c1 e2t + c2 e2t (5.72)
0 1
where c1 and c2 are constants. Here there are two linearly independent eigen-
vectors with a positive trace. Hence we have an unstable star mode.
For the other fixed point (2, 6), the partial derivatives are ( f
x )(2,6) =
1

4, ( f
y )
1
= 2, ( f f2
x )(2,6) = 6 and ( y )
2
= 0. The corresponding
(2,6) 2,6)
matrix M is given by
 
4 2
M = (5.73)
6 0

and the characteristic equation reads 2 4 12 = 0. Its roots are 6 and

2. These are of opposite signs and we conclude that the fixed point is an
unstable saddle point.
136 Advanced Classical Mechanics

Example 5.7

For the system

x = x 18xy + 10x2 + 7, y = 11y + 2xy + 9 (5.74)

we easily see by inspection that by putting x = 0 and y = 0 a fixed point

is located at the point (1, 1). To analyze its character we evaluate the par-
tial derivatives and find ( f f1
x )(1,1) = 3, ( y )
1
= 18, ( f
x )(1,1) = 2 and
2
(1,1)
( f
y )
2
= 9. Hence the corresponding matrix M is given by
1,1)
 
3 18
M = (5.75)
2 9

The charateristic equation turns out to be ( + 3)2 = 0 implying that the

eigenvalue 3 is repeated. The eigenvectors are obtained by solving the matrix
equation     
3 18 0
= (5.76)
2 9 0
at the point = 3. We obtain as a result the relationship = 3 implying
only one eigenvector is involved. Since the trace is negative we have a stable
degenerate node for this fixed point.
Suppose now we slightly perturb the elements of the matrix M in the
manner
 
3 18 + 
M = (5.77)
2 9

As a result of the presence of a small nonzero quantity  the characteristic

equation changes to 2 + 6 + 9 2 = 0 showing that the equal eigenvalues
3 branch out to different values 3 2. For  > 0, we get distinct nega-
tive eigenvalues implying a stable node while for  < 0, complex eigenvalues
appear with a negative real part signalling a stable spiral.

Example 5.8

Our heart is a rather complicated nonlinear system. Its periodic oscilla-

tory behavior in normal situations must be stable as our everyday experience
shows. A somewhat realistic heartbeat model was proposed in 1973 by E.C.
Zeeman which we now discuss briefly. As we all know the systolic (or the
period of chamber contraction) and diastolic (or the period of chamber relax-
ation) are the two phases of the heart. The pulsating of the heart is due to
the electrochemical gradient which taps each individual fiber. During systolic
contraction the heart squeezes signalling consumption of energy while during
relaxation energy is needed to overcome the bond (or the threshold) taking
place during contraction that moves out the calcium ions out of the systole.
 Dynamical systems: An overview 137

In this way the regulation of the pace of the heart goes on in a cycle with the
heart contracting from a diastole to a systole and then rapidly back again to
the original diastolic state which is a state of stable equilibrium (fixed point).
To make a qualitative study of heart functioning Zeeman considered the
following scheme in terms of the equations

x = x x3 y, y = x x0
where x stands for the length of a cardiac muscle fiber but has a small positive
coefficient  prefixed due to the fast eigenvalue of the system, y is a signature
of electrochemical activity and x0 is a constant corresponding to a typical
length in the muscle fiber in the diastolic state with x0 > 13 .
By putting x = 0 and y = 0 we determine the equilibrium point at (x0 , y0 )
where y0 = x0 x30 . To examine its stability behavior, we need to carry out
linearization and study the Jacobian matrix of this system.
Using the notations already employed earlier, we find here f1 = x x3
y and f2 = x x0 . The partial derivatives turn out to be ( f x )(x0 ,y0 ) =
1

1 3x20 , ( f
y )
1
= 1, ( f f2
x )(x0 ,y0 ) = 1 and ( y )
2
= 0. Hence the
(x0 ,y0 ) (x0 ,y0 )
corresponding Jacobi matrix M is given by

1 3x20
 
1
M = (5.78)
1 0

It yields the characteristic equation

2 (1 3x20 ) + 1 = 0
from which the two eigenvalues can be solved in the forms
1
q
1,2 = [(1 3x20 ) (1 + 3x20 )(x20 1)]
2
Interestingly both the eigenvalues are negative for the possibilities x0 > 13
or x0 < 13 . In each case the equilibrium point (x0 , y0 ) is stable. This will
be clear from the remarks made below.
A graphical illustration of the model in the yx-plane is shown in Figure 5.8.
The heartbeat cycle has two equilibrium states, namely, the diastole and sys-
tole. The main point to note is that as the chemical control variable y increases,
there is a gradual contraction in the muscle fiber until a threshold position is
reached when there is a rapid contraction and a systolic equilibrium state is
reached. After this there is a rapid relaxation of the muscle fibers until the
heart is returned to the original diastolic state and the cycle is completed.
138 Advanced Classical Mechanics

Available
Available

Loyal

FIGURE 5.8: The heartbeat model.

5.5 LotkaVolterra model

Study of interacting populations in a given environment is of great in-
terest in problems of mathematical ecology and constitutes an active area
of research. The interactions affect the growth rates of the organism in the
ecosystem and so for a balance between various types of species an appropri-
ate predator-prey relationship needs to be maintained. Among other things,
the following aspects are crucial to our understanding of the qualitative na-
ture of the system. Specifically one needs to know (i) how the interactions
among various species manifest themselves, (ii) how the species coexist in a
given environment, (iii) how nature strikes a balance between the two and
(iv) how a prey escapes being killed by a predator. It is natural that when the
preys are too many then the population of the predators increases but in the
reverse case when the population of the predators are too numerous then the
population of the preys shows signs of decreasing. On the other hand, while it
is evident that without preys the predators will reveal a negative growth rate,
a limited number of predators implies that the preys will tend to survive over
a longer period of time and ultimately reach a stage when an exponential rise
in their population will be noticed. In a practical scenario repeated increase
and decrease in the population sizes of preys and predators signals a cyclic
behavior of the survival of the two species.
The LotkaVolterra model is one of the simplest schemes of population
growth in a closed ecosystem (i.e., derecognizing any migration into or out
of the system) which aims at studying the inter-behavior of only two species,
the predator and the prey. It may not be out of context to mention here that
 Dynamical systems: An overview 139

the spirit of the LotkaVolterra model was contained in the 1920 paper by
Alfred Lotka, an American biologist and actuary, who proposed a predator-
prey model to understand the population variation of the interacting fish.
Such a study was of utmost relevance then since one already had some knowl-
edge on the percentage of the total catch of selachians (such as sharks, rays,
dogfish and skates) which were the predators of the fish (such as garfish, eel
and sardines) giving some qualitative idea on the predator-prey interactions.
The data collected by the Italian biologist Umberto dAncona accounted for
the period between 1914 and 1923 and showed a rather substantial increase
of the predators during the time of the First World War. Although people
initially thought that this could be due to less fishing due to the turmoil of
the war, it was also realized that the food fish also revealed marked changes
simultaneously. Umberto dAnconas father-in-law was Vito Volterra who also
worked on the same model as Lotkas around the same time. In 1925 he came
up with some explanations of the data collected by Umberto by giving an
interpretation of the cause-effect interactions between the selachians (preda-
tors) and the food fish (preys) and sought to explore, in general, the cyclic
relationship between the predator-prey species. Since then it has been cus-
tomary to refer to the two species predator-prey model as the LotkaVolterra
model.
In the LotkaVolterra model, a pair of coupled evolution equations is pro-
posed involving two unknown variablesone for the prey (x) and another for
the predator (y). The degree of interaction is assumed to be proportional to
the product xy. In this way the following two equations are set up:

x = ax gxy, y = by + gxy (5.79)

where the parameters a, b and g are all > 0. Here the parameter a signifies the
natural growth rate of the prey in the absence of any predator, the parameter
b is responsible for the natural decay rate of predators if there is no prey
to hunt for, and the parameter g stands for the interaction coefficient. It is
clear that if the predator is absent, i.e., y = 0, the prey will survive and show
an exponential growth. However, if there is no prey, i.e., x = 0, a negative
growth rate is signalled for the predators with the possibility that they may
even go extinct. The interaction term in (5.21) addresses the question of both
the preys being eaten up by their predators and the effect on the growing size
of the latter. In principle g could be different for the two equations but for
simplicity we have taken them to be equal. Note that by hunting the rate of
increase of prey is reduced (and so a is decreased) while the rate of decrease
of predator is increased (and so b is increased) but the interaction coefficient
g is not supposed to be affected.
LotkaVolterra equations can also be transformed to a one-parameter form
by effecting the following transfer of variables:
gx gy
s= , r= , = at (5.80)
d a
140 Advanced Classical Mechanics

We thus obtain
ds dr
= s(1 r), = r(s 1) (5.81)
d d
showing to be the only control parameter of the system.
Further, by defining the variables

p = logs, q = logr (5.82)

we can easily identify the Hamiltonian for (5.26) to be

H = (ep p) + (eq q) (5.83)

dq H
Indeed, it can be readily checked from the Hamiltonian equations d = p
dp
and d = H
q that relations (5.26) hold.
Turning to (5.21) a conservation law can be worked out by noticing that
the two equations yield on division
dy by + gxy
= (5.84)
dx ax gxy
which yields as the first integral the quantity

F (x, y) = log[|x|b |y|a eg(x+y) ] = C (5.85)

where C is a constant. F (x, y) = C represents the conservation law.
It is interesting to look for the nature of the fixed points of the Lotka
Volterra system. From Equation (5.21) the fixed points are obtained from
their stationary versions, namely,

ax gxy = 0, by + gxy = 0 (5.86)

The fixed points are identified to be located at
a b
(0, 0) and ( , ) (5.87)
g g
dy
Note that dx a
dt = 0 gives the horizontal isoclines (x = 0 or y = g ) while dt =0
b
gives the vertical isoclines (y = 0 or x = g ).
Now the Jacobian matrix for such a system reads

 
a gy gx
M = (5.88)
gy b + gx

At (0, 0), M has the form

 
a 0
M = (5.89)
0 b
 Dynamical systems: An overview 141

showing it to be a diagonal one with obvious eigenvalues a and b. Their

differing signs indicate that the fixed point (0, 0) is an unstable saddle point.
On the other hand, for the second fixed point ( ag , gb ), M is given by

 
0 b
M = (5.90)
a 0

It is an off-diagonal matrix with imaginary eigenvalues given by i ab. Hence
the fixed point is a stable center.

5.6 Stability of solutions: Lyapunov function

Consider a class of solutions x(t) of the dynamical system as given by
Equation (5.3). A fixed equilibrium point x(t0 ) at time t = 0 is said to be
Lyapounov stable if for all  > 0, there exists a (, t0 ) > 0 such that, if
around the equilibrium point |x(0) x(t0 )| < holds true at time t = 0, then
|x(t) x(t0 )| <  also holds true for all times t 0. More explicitly, if we
think of a neighborhood K around the equilibrium point x(t0 ) then such an
equlibrium is termed Lyapunov stable if we can find a K0 K such that for
any x K0 , the flow t (x) K for all times t 0. All periodic orbits display
stability of the Lyapunov type. Violation of the Lyapunov condition even for
at least one solution renders the equilibrium unstable. Note that the stability
is uniform if we can choose a independent of t0 , i.e., (). Apart from being
not only Lyapounov stable, if around the equilibrium point, |x(0) x(t0 )| <
holds true for > 0, then the equilibrium is said be asymptotically stable if
limt |x(t) x(t0 )| = 0 is fulfilled. Asymptotic stability has relevance for
those dynamical systems if when perturbed slightly around an equilibrium
point the system will have the tendency to return to it.
Let us extend the concept of the Hamiltonian function H(x, y) for an
autonomous system which is controlled by the pair of Hamilton equations
dx H dp H
= , = (5.91)
dt p dt x
to an arbitrary continuously differentiable function U (x, y) mimicking the role
of the Hamiltonian. In other words, U is subject to similar styled equations
as those given above
dx U dy U
= , = (5.92)
dt y dt x
where the partial derivatives are assumed to exist and are continuous. Then
it is evident that U also satisfies being a constant of motion:
142 Advanced Classical Mechanics

dU U dx U dy
= + =0 (5.93)
dt x dt y dt
dy
by substituting the above relations for dx
dt and dt . At an equilibrium fixed
~ ( U , U ), it
point both the partial derivatives of U vanish. Denoting y x
dy
also turns out that div = 0. Further it also holds that dx is given by
U
dy x
= U (5.94)
dx y
U
When the denominator y vanishes the solution curves terminate at a point.

Example 5.9

Consider
1
U (x2 + y 2 ) (5.95)
2
U U dx
then evaluating the partial derivatives x and y , we find dt = y and
dy dy
dt = x. Thus the fixed point is (0, 0). Now since dx = xy , when integrated
2 2 2
it gives the solution curve as circles x + y = c where c is a constant, for
(x, y) 6= (0, 0).
We now inquire into the possibility of replacing the partial derivatives U
x
dy
and Uy by some arbitrary functions say, u(x, y) and v(x, y). Then dx
dt and dt
would read
dx dy
= u(x, y), = v(x, y) (5.96)
dt dt
Hence any function of x and y, say, W (x, y), which depends upon the variable
t implicitly due to x and y being functions of t, will have its time derivative
go as
dW W dx W dy W W
= + = u(x, y) + v(x, y) (5.97)
dt x dt y dt x y
where the partial derivatives of W are assumed to exist and are continuous.
The above equation can be expressed in the form

dW (~x)
= W (~x) ~ (5.98)
dt
where ~x = (x, y) and ~ = (u, v). Note that W (~x) is called a Lyapunov function,
and it will be our endeavor to seek for such a function for the understanding
of the stability of a system. It must however be pointed out that for many
nonlinear systems constructing an appropriate Lyapunov function proves very
 Dynamical systems: An overview 143

difficult. Nonetheless, there are situations when a Lyapunov function exists in

a simple way.
Toward this end we impose upon W (~x) the criterion of being a positive def-
inite function while restricting its time derivative to be negative semi-definite:
W (~x) ~ 0. We then refer to W (~x) as a weak Lyapunov function. In the
case when the semi-definiteness can be replaced by a strict W (~x) ~ < 0,
W (~x) is referred to as a strong Lyapunov function. Let W0 be an equilibrium
point of a flow t . The Lyapunov stability theorem asserts that if W (~x) is
a weak Lyapunov function in some region K around the equilibrium point
W0 , then W0 is (Lyapunov) stable. On the other hand, asymptotic stability is
guaranteed if W (~x) turns out to be a strong Lyapunov function.
For a justification of the above statements we proceed as follows. For con-
creteness, we choose, without any loss of generality, the equilibrium point W0
to correspond to x = 0. We surround the point by a small region, say, K of
radius  such that K K. The region K is compact and a minimum of
W exists in it (i.e., in the region defined by |x| = ) which is positive because
W is positive. Since W decreases as x goes to zero, it is clear that we can find
a <  such that the value of W will drop below for x K . Now from
(5.98), since dWdt 0 for a weak Lyapunov function, W is nonincreasing along
orbits t and we can infer that W (t (x)) < corresponding to x K . With
< , W < on |x| =  as well pointing to t (x) K . We thus arrive at the
conclusion that the equilibrium point x = 0 is stable in the sense of Lyapunov.
To prove the asymptotic stability we make use of the BolzanoWeierstrass
theorem which states that every bounded infinite set of real numbers has
at least one limit point. Consider a sequence of time points t1 , t2 , ..., tn , ....
Then the compactness of K implies that the corresponding sequence of the
orbits ti (x) possesses limit points. We shall prove asymptotic stability by
contradiction and so we assume the absence of the point x = 0 from the limit
points of ti (x). In other words, we assume that as tn , tn (x) l 6=
0. Next we draw up a sequence of inequalities

W (tn (x)) > W (tn+1 (x)) > ... > W (l) (5.99)

from the nondecreasing condition of W and note that continuity implies

W (tn (x)) W (l). Since l is not a point of equilibrium, we can say that the
orbit of the limit point m (l) must be such that it is W (m (l)) < W (l) for any
positive m. By continuity it must hold that W (tn +m (x)) W (tn (l)) <
W (l). We can choose m in such a way that for n0 obeying tn0 > tn+m

W (tn0 (x)) < W (tn+m (x)) < W (l) (5.100)

which runs counter to the previous inequality. So all the limit points of the
flow are at the origin and x = 0 is asymtotically stable.
With l the limit point, for large n, x(tn ) can be made arbitrarily close to l.
We now turn to some examples to determine the Lyapunov function and
then examine the stability according to the criterion prescribed above.
144 Advanced Classical Mechanics

Example 5.10

The governing equation of the linear harmonic oscillator problem can be

expressed as
x = y, y = 02 x (5.101)
It turns out that the energy function E(x, y) = 12 02 x2 + 12 y 2 doubles up
as the Lyapunov function. Indeed identifying it with W (x, y), we notice that
W (0, 0) = 0 at the point of the minimum (0, 0) which is the equilibrium point.
Further, in the region in the deleted neighborhood of (0, 0), W (x, y) > 0.
Working out the time derivative of W (x, y) we easily find
W W
W = x + y = y02 x 02 xy = 0 (5.102)
x y
So we conclude that the system is Lyapunov stable.

Example 5.11

Let us take up the damped oscillator problem already considered earlier

in terms of two first order equations
x = y, y = 02 x 0 y, >0 (5.103)
1 2 2 1 2
We choose the Lyapunov function to be W (x, y) = 2 0 x + 2 y .
Since W (x, y)
is a sum of two squares, a unique minimum occurs at (0, 0) which is the
equilibrium point. For the quantity W (~x) , ~ we have W = 2 x and
x 0
W
= y and ~ has components y and 2 x 0 y. Substituting them we

y 0
find W (~x) ~ = 0 y 2 which is a strictly negative quantity for all y since
> 0. So the system is not only stable but asymptotically stable too.

Example 5.12

Let us address the pendulum problem as described by the equation of

motion + gl sin = 0. It can also be represented in the following form:
g
= y, y = sin (5.104)
l
We already know its fixed points to be (n.0), n = 0, 1, 2, .... Consider the
following candidate for the Lyapunov function:
l2 y 2
W (, y) = (1 cos )gl + (5.105)
2
Its features are (i) W (0, 0) = 0, (ii)W (, y) > 0 for 2 < < 2 for
(, y) 6= (0, 0) and (iii) W = gl sin + l2 y y = 0 when the above expressions
for and y are substituted. Hence we notice that the system is Lyapunov
stable but not asymptotically so because W is not strictly negative for all
(, y) 6= (0, 0).
 Dynamical systems: An overview 145

5.7 Van der Pol oscillator and limit cycles

Van der Pol oscillator is a mechanical model of a nonlinear system through
the inclusion of a friction-like term with varying coefficient. We already en-
countered in Chapter 1 the case of a damped oscillator in which the friction
term was accounted for both linearly and nonlinearly. In the Van der Pol case
the governing equation reads
x + af (x, x) + 02 x = 0 (5.106)
where the function f (x.x) = (x2 1)x incorporates the effects of friction. If
the coupling a = 0 one is left with the harmonic oscillator equation. It is
a second-order autonomous equation. The presence of a renders the system
non-conservative. It is to be noted that in the range 0 < x < 1 the oscillator
receives an additional impetus for the acceleration while for 1 > x > 0 the
volatility of the oscillator is reduced. Physically there is a contrasting role of
the energy: it is generated in low amplitudes while large ones dissipate it. An
interesting situation arises at some intermediate point when some kind of a
compromise is reached between energy generation and dissipation resulting in
oscillations forming around that state. The convergence toward such a state
is characterized as a limit cycle. Limit cycles appear in phase spaces of two
or more dimensions. The concept of an attractor was introduced earlier in the
context of trajectories in a dynamical system approaching the final state in the
phase space while evolving in time. A limit cycle is a second type of attractor
exhibiting steady closed oscillations: it is an isolated closed curve which is
stable or unstable depending upon whether the neighboring trajectories are
attracted upon it or get diverged away. There is also another class of limit
cycles called semi-stable limit cycles which attract nearby trajectories from
one side but repel those on the other. Stable and unstable limit cycles are
illustrated in Figure 5.9 and Figure 5.10.
To factorize (5.106) into a pair of first order equations, the usual trick is
to set
x = r cos(t + ), x = r sin(t + ) (5.107)
where it is to be noted that we have introduced two time-dependent quantities
r and which are, respectively, the amplitude and the phase. The second
equation does not represent the time derivative of the first but in order to be
so we evaluate the time derivative of the first and look for consistency with
the second. This yields the condition
r cos(t + ) r sin(t + ) = 0 (5.108)
On the other hand if we take the time derivative of the second equation in
(5.107) and compare with (5.106) we arrive at the requirement
r sin(t + ) + r cos(t + ) + af = 0 (5.109)
146 Advanced Classical Mechanics

All

All

FIGURE 5.9: Stable limit cycle.

All

All

FIGURE 5.10: Unstable limit cycle.

 Dynamical systems: An overview 147

where we have set 0 = 1 for simplicity. It is now easy to solve for r and
from (5.108) and (5.109) by first multiplying the two equations by cos(t + )
and sin(t + ) and subtracting and then multiplying the two equations by
sin(t + ) and cos(t + ) and adding to obtain, respectively,
a
r = a sin(t + )f, = cos(t + )f (5.110)
r
Equations (5.110) are an exact mapped version of (5.106) in terms of the
variables r and . So it has indeed been possible to factorize (5.20). In the
von der Pol case a specific form of f is used which is
f (r, t) = (1 r2 cos2 t)r sin t (5.111)
We now explore the situation when the quantities r and are slowly
varying functions of time as is indeed the case when the system approaches
the limit cycle. The averaged amplitude and phase are then given by
Z 2
a 2
Z
dr a d
= f (r, ) sin d, = f (r, ) cos d (5.112)
dt 2 0 dt r 0
where the integration is taken over the period of oscillation.
Using the form (5.111) for f the integrations are straightforward to perform
using the well-known results
Z 2 Z 2 Z 2
1 2 1
2
cos d = , 2
sin cos d = , and sin2 sin cos d = 0
0 2 0 8 0
(5.113)
We obtain
dr r r2 d
= a (1 ), =0 (5.114)
dt 2 4 dt
1
The r integration is easily performed by setting r = u 2 yielding the form
at
2e 2
r(t) = p (5.115)
4 + r02 eat r02
where r0 is given an initial condition.
It is worthwhile to inquire into the nature of the fixed points of the first
equation in (5.114). By inspection the fixed points are at r = 0 and r = 2.
Performing a slight perturbation around r = 0, i.e., setting r = 0 + , where
is a small quantity gives
a
= + O( 3 ) (5.116)
2
at
Therefore goes as e 2 which blows up asymptotically. If we do the same
around r = 2, i.e., set r = 2 + the situation turns out to be entirely different.
We find
a
= + O( 2 ) (5.117)
2
at
Therefore goes as e 2 signalling r = 2 to be a stable equilibrium point.
148 Advanced Classical Mechanics

Now concerning the solution (5.115) it is evident for large t it approaches

the circle of constant radius 2 from both the inner side and the outer side of
the radius for all values of the initial condition r0 . We therefore identify it as
a limit cycle. The critical value of r = 2 which is a point of stable equilibrium
has thus an interesting feature of attracting other points in the phase space
as time evolves. Limit cycles are not too infrequent in nature. Consider the
following example.

Example 5.13

Show that the following nonlinear system given by the pair of equations

x y
x = y + p [1 (x2 + y 2 )], y = x + p [1 (x2 + y 2 )]
x2 + y2 x2 + y2
has a limit cycle.

Transforming to polar coordinates by setting x = r cos and y = r sin

gives the following forms in the (r, ) plane:

r = 1 r2 , = 1
On integrating we find

ce2t 1
r(t) = , (t) = 0 t
ce2t + 1
1+r0
where c = 1r 0
and (r0 , 0 ) are a set of initial conditions. We readily notice
that as t increases r approaches a circle of radius 1 with constant -value. In
particular, when r > 1 it is seen that r < 0 while for r < 1 it is found that
r > 0. In the former case the trajectories spiral clockwise inward while in the
latter case the trajectories spiral clockwise outward. Thus, r(t) is a clear case
of a limit cycle.
Limit cycles are an intrinsic property of nonlinear systems. However, limit
cycles can also arise by piecing together linear equations. We follow the treat-
ment presented in [2]. We consider the damped oscillator problem on two
sides of the zero-boundary of the velocity variable given by the following pair
of linear equations:

x + 2g x + 02 x = 0 x < 0 (5.118)

x + 2g x + 02 (x ) = 0 x > 0 (5.119)
where is a real constant. We also restrict g to the interval 0 < g < 0 .
While the general solution of (5.118) is given by

x(t) = Aegt cos(t + ) (5.120)

 Dynamical systems: An overview 149

where 02 = 2 + g 2 , the one for (5.119) has the form

x(t) = + Begt cos(t + ) (5.121)

where A and B are the amplitudes.
Projecting out (5.119) in an equivalent form

x = y, y = 0 2 (x ) 2gy (5.122)
it is easily seen that the point (, 0) is a stable focus.
Let us now follow the motion in the phase plane defined by the coordinates
(x, y). Consider the lower half plane. Here a typical point will be guided by
the equation (5.118). If the motion starts from say, the point (x1 , 0) on the
x-axis at time say, t1 , then it obeys according to

x1 (t1 ) = Aegt1 cos(t1 + ) x1 (5.123)

When the trajectory again cuts the x-axis moving in the clockwise direction
at the point after a lapse of time say, (t1 + ), (5.120) this time gives

x1 (t1 + ) = Aeg(t1 + ) cos(t1 + +) = Aeg(t1 + ) cos(t1 +) (5.124)

Comparison of (5.123) and (5.124) shows that the factor Aegt1 cos(t1 + )
can be canceled out leaving us with the relation

x1 (t1 + ) = x1 (5.125)

g
where the coefficient factor ( e ) is smaller than unity because g
=
g
and 0 < g < 0 .
(0 g)(0 +g)
Next, as the representative point x1 enters the upper half plane, it comes
under the influence of Equation (5.119). The relevant equations as the point
travels from (t1 + ) until it meets the x-axis again at (t1 + 2
) are

x1 (t1 + ) = Beg(t1 + ) cos(t1 + + ) = Beg(t1 + ) cos(t1 + )

(5.126)

2
x1 (t1 + ) = Beg(t1 + ) cos(t1 + + ) = Beg(t1 + ) cos(t1 + )

(5.127)
Utilizing (5.125) a combination of the above two equations gives
2
x1 (t1 + ) = (1 + ) + 2 x1 (5.128)

150 Advanced Classical Mechanics

Relation (5.128) can now be iterated to write down the coordinates of the
chain of the intersections of the particle trajectory with the positive x-axis in
terms of the starting coordinate x1 only as the particle alternately moves in
the lower and upper half phase plane:

x1 , (1 + 2 ) + 2 x1 , (1 + 2 + 3 ) + 4 x1 , ..., (1 + 2 + ... + 2N 1 ) + 2N x1 , ...

(5.129)
where N denotes the number of complete revolutions. Since < 1, 2N x1 0
as the number of revolutions grow. So the coefficient of is left with the form
of a geometric series. The series of intersecting points thus converges to the

limit 1 .
To understand the significance of this result we observe that in Equation
(5.128) the criterion of the closed orbit would imply x1 (t1 + 2 ) = x1 (t1 )

resulting in the unique solution x1 = 1 . It is exactly the convergence point
of the geometric series as just now noticed. Thus our finding is that all the
trajectories asymptotically approach a unique attracting limit cycle.
PoincareBendixson theorem which is often found to be useful in the con-
text of a limit cycle is of much use in dynamical systems. It is essentially a
two-dimensional result. It states (and we state it without a proof) that if a
flow is confined within a closed and bounded domain D which does not con-
tain any fixed point, then D holds at least one periodic orbit which is a limit
cycle. Note that it speaks only of sufficient conditions. For practical utility of
this theorem, one needs to find a trapping region D containing no fixed point
and in which trajectories enter but do not move out. The following example
will make this issue and also the applicability of PoincareBendixson theorem
clear.

Example 5.14

Show that a physical system given by the equation x(13x2 2x2 )+x = 0
has a limit cycle.

We can express the above system as

x = y, y = y(1 3x2 2x2 ) x

It transpires by putting equal to zero the right-hand sides that the origin
(0, 0) is the only fixed point. Transforming to polar coordinates by setting
x = r cos and y = r sin gives the following forms of the equations in the
(r, ) plane

r = r(12r2 r2 cos2 ) sin2 , = 1+(13r2 cos2 2r2 sin2 ) sin cos

This implies that r > 0 for 1 > 2r2 + r2 cos2 > 0 or for
1 1
r2 < <
2 + cos2 3
 Dynamical systems: An overview 151

while r < 0 for 1 < 2r2 + r2 cos2 > 0 or for

1 1
r2 > 2
>
2 + cos 2
Comparing the two bounds we arrive at the definition of a domain D given
by
1 1
D = {(r, ) : < r < }
3 2
in which there is no fixed point, the latter existing only at the origin as ob-
served earlier. D thus corresponds to an annulus or the trapping zone which
excludes the fixed point (0, 0) and is bounded by the circles of radii 13 and
1 . Furthermore, this domain corresponds to a region for which r > 0 on the
2
one side but r < 0 on the other. This means that a trajectory once inside D
stays within it and cannot escape. By PoincareBendixson theorem there is a
limit cycle in D.

5.8 Bifurcations
The change in the nature of the fixed point leading to a drastic change
in the behavior of the trajectories in the neighborhood of the fixed point
gives rise to bifurcation. Let us have a look at the damped oscillator problem
considered earlier in this chapter. We found a change in the nature of the fixed
point from a stable focus to a stable node as moved out from the interval
0 < < 2 to the region > 2 . Since stability persists we cannot call = 2 to
be a bifurcation point. However, the other fixed point namely = 0, is indeed
a bifurcation point since there is a sudden change in the nature of stability:
while at = 0, the real part of the eigenvalue vanishes and we are left with a
center at the origin, for < 0, there is an unstable focus and unstable node
for the fixed point (0, 0) corresponding to the respective cases of 2 < 4 and
2 4.
Bifurcation is thus concerned with the qualitative sudden changes with the
model, namely, the phase portraits changing in a non-continuous way, as the
underlying parameters controlling the system undergo variation. Bifurcations
play a leading role in the theory of nonlinear dynamics. A complete treatment
of such a subject is beyond the scope of this book. We give a few illustrative
examples to show how bifurcations occur in physical systems. To this end we
consider four types of bifurcations:
(i) Saddle-node bifurcation
(ii) Transcritical bifurcation
(iii) Pitchfork bifurcation
(iv) Hopf bifurcation
152 Advanced Classical Mechanics

The first three types deal with real eigenvalues while in the fourth one
complex eigenvalues are relevant.
We consider an autonomous differential equation
dx
= f (x, ), x R, R (5.130)
dt
involving some parameter . To detect bifurcation points we write down the
linearized flow and search for those parameter values for which the fixed point
has a zero or purely imaginary eigenvalue:

either Re = 0 or detM = 0 (5.131)

where M stands for the Jacobi matrix and denotes an eigenvalue. Note that
it is only for non-hyperbolic fixed points that bifurcations can occur. In other
words, we set

f (x, )
f (x, ) = 0, |(x,) = = 0 (5.132)
x
where x is a bifurcation point and is the critical value of the parameter
where the bifurcation arises.
We now turn to a discussion of the four types of bifurcations as mentioned
above.

(i) Saddle-node bifurcation

A saddle-node bifurcation is characterized by the following features at the

point (, x):

f f 2f
= = 0, 6= 0, 6= 0, (5.133)
x x2
A typical model equation admitting a saddle-point bifurcation is
dx
= f (x) = x2 , x R, R (5.134)
dt
For > 0, the equilibrium or fixed points are at

x+ = , x = (5.135)
while for < 0, there is no fixed point because of the reality of x. Of the two

fixed points, it can be very easily checked that the one at is a node and the

one at is a repeller. What happens at the point = 0? We notice that
there is an abrupt jump in the number of fixed points as passes through the

zero value and a bifurcation occurs at = 0: the solution x+ = is a stable

one, in contrast, the other one corresponding to x+ = is unstable. The
bifurcation diagram is shown in Figure 5.11.
 Dynamical systems: An overview 153

All All

All

All All

FIGURE 5.11: Saddle-node bifurcation. The solid line represents the stable
portion and the dashed line represents the unstable portion.

(ii) Transcritical bifurcation

A transcritical bifurcation has the following features at the point (, x):

f 2f 2f
= = 0, 2
6= 0, 6= 0, (5.136)
x x x
Here a typical model equation admitting a transcritical bifurcation is
dx
= f (x) = x x2 , x R, R (5.137)
dt
The equilibrium or fixed points are at the points 0 and . To examine their
characters we find the following features:

< 0 : x = (unstable), x = 0(stable) (5.138)

> 0 : x = (stable), x = 0(unstable) (5.139)

The two signs of convey two opposite behaviors: both the fixed points 0 and
switch from being unstable(stable) to stable(unstable). Figure 5.12 gives
the bifurcation diagram which is of transcritical type.
154 Advanced Classical Mechanics

All

All

FIGURE 5.12: Transcritical bifurcation. The stable and unstable portions

are distinguished by solid and dashed lines.

(iii) Pitchfork bifurcation

The following features govern a pitchfork bifurcation. We have at the point

(, x):

f 3f 2f
= = 0, 6
= 0, 6= 0, (5.140)
x x3 x
For a pitchfork bifurcation a typical model equation is
dx
= f (x) = x x3 , x R, R (5.141)
dt
The equilibrium or fixed points are at

x+ = , x = , x0 = 0 (5.142)
We immediately notice that for < 0, the system has only one fixed point
at x0 = 0. It is stable. On the other hand, for > 0 the number of fixed

points are two, namely, and , both of which are stable but in this
case x0 = 0 is unstable. Figure 5.13 gives the supercritical bifurcation dia-
gram which is of pitchfork type. It is to be observed that the curve for = 0
simply passes through the origin and does not cut the abscissa anywhere else.
However for a positive the cubic meets the abscissa at three points. The sub-
critical pitchfork bifurcation diagram, for which the model equation stands as
dx 3
dt = f (x) = x + x , is illustrated in Figure 5.14.
 Dynamical systems: An overview 155

S' S'

X j

1 I ------------3> J1

I 1
I
FIGURE 5.13: Supercritical pitchfork bifurcation.

........ _ i
c----- ____ r
'. -------------- ? J-L

-~-
-- - - - - -/-----::1
, 1
.- ----1

FIGURE 5.14: Subcritical pitchfork bifurcation.

156 Advanced Classical Mechanics

(iv) Hopf bifurcation

Hopf bifurcation constitutes an important class of bifurcations in nonlin-

ear dynamics. It arises in situations when a complex pair of eigenvalues at a
fixed point sheds its real part and becomes purely imaginary. Thus its occur-
rence has relevance for systems defined in phase space with dimension m 2.
In assessing its occurrence we are particulary interested in a theorem called
Hopf bifurcation theorem whose basic formulation is concerned with sufficient
conditions related to observation of bifurcations to cycles and the orbital sta-
bility of the latter. Without getting into the details of the underpinnings of
the theorem, let us only note that if a differential equation of the autonomous
type
dX~
= f~(X,
~ ) (5.143)
dt
is endowed with the conditions that the associated Jacobi matrix has an iso-
lated fixed point at which a pair of conjugate-complex eigenvalues 1,2 appear
(whose appearence of course depends on some control parameter assuming
a specific value) such that (i) at a critical value of say, , the real part van-
ishes but the imaginary part remains nonzero and (ii) at the same point the
derivative of the real part is nonzero as well, then the system of differential
equations (5.143) admits periodic solutions which are both stable and unsta-
ble. Before we turn to the applicability of the theorem when the criterium of
its conditions are fulfilled, we note that the nature of Hopf bifurcation can
be supercritical or subcritical. In the supercritical case a stable limit cycle is
formed around an unstable equilibrium point whereas in the subcritical case
it is an unstable limit cycle that is formed around a stable equilibrium point.
Let us focus on the following examples. First take the Van der Pol oscillator
equation
x + (x2 )x + x = 0 (5.144)
where is a control parameter. Expressing it in the pair

x = y, y = (x2 )y x (5.145)

the equilibrium point can be seen to be located at (0, 0), the origin. The
underlying Jacobi matrix for the linearized system is given by
 
0 1
M = (5.146)
1

Its eigenvalues are easily determined which have a common real part 2 and
p
imaginary parts 2i 4 2 . The real part is seen to vanish at = 0 at which
point the imaginary part is nonzero: Im = 1. Further the derivative of the
real part with respect to at = 0 is 12 which is nonzero. Hence the two
conditions of Hopf bifurcation theorem are satisfied. Now, as just noted, the
real part of the eigenvalues is positive for > 0 and negative for < 0. So
 Dynamical systems: An overview 157

the equilibrium is an unstable focus for > 0 but stable for < 0. However,
there is a stable periodic orbit (limit cycle) if > 0 for the system (5.144) (see
Exercise 10) and hence our conclusion is that at = 0 there is a supercritical
bifurcation.
Another case of supercritical bifurcation occurs for the system

r = r( r2 ), = 1, r0 (5.147)

The Cartesian version of the above equation reads

x = x y x(x2 + y 2 ), y = x + y y(x2 + y 2 ) (5.148)

which straightforwardly yields (5.49) by transferring to polar coordinates x =

r cos and y = r sin .
The only equilibrium point for the above system is at the origin (0, 0), i.e.,
at r = 0. Its behaviors are (i) for > 0, an unstable focus (complex eigenvalues
having a positive real part), (ii) for < 0, a stable focus (complex eigenvalues
having a negative real part) and (iii) for = 0, a center (eigenvalues are
purely imaginary). For < 0, since the governing equation shows r < 0, it
indicates that the trajectories spiral asymptotically toward the origin in an
anticlockwise manner due to the positive sign of . However, for > 0, r

changes sign as one steps out of the interval from (0, ) to ( , ), there
being a stable limit cycle at r = . Since the origin is an unstable focus we
encounter a supercritical Hopf bifurcation.
We now provide an example of a subcritical bifurcation. Consider a system
which is given by

r = r + 2r3 r5 , = 1, r0 (5.149)

In Cartesian coordinates the equivalent form is

x = xyx(x2 +y 2 )(2x2 y 2 ), y = x+y+y(x2 +y 2 )(2x2 y 2 ) (5.150)

For small values of r it is clear that the equilibrium point exists at the
origin (0, 0), i.e., at r = 0: it is a stable focus for < 0 but an unstable one
for > 0. For = 0, however, the origin is unstable.
Let us focus on the interval 1 < < 0. The radial equation admits of a
factorization in the following form:

r = r(r 1 + z)(r + 1 + z)(r 1 z)(r + 1 z) (5.151)

where z = 1 + . Note that in the interval 1 < < 0, both 1 + z and
1 zare positive quantities.
We therefore identify two limit cycles of radii
r1 = 1 + z and r2 = 1 z. Further, r changes sign as follows: it is negative
in 0 < r < r1 , positive in r1 < r < r2 and again negative for r > r2 . We
therefore observe that the limit cycle r1 = 1 z is unstable but r2 = 1 + z
is stable.
158 Advanced Classical Mechanics

Turning to the case > 0 now, the factorization is to be carried out

somewhat differently, namely,

r = r(r 1 + z)(r + 1 + z)(r2 + z 1) (5.152)
The reason is that in this case while 1 + z is positive, the quantity 1 z is
negative. This means that we only have one limit cycle of radius r2 = 1 + z.
Concerning the signs of r, we see that it is positive in the interval 0 < r < r2
but negative for r > r1 . Hence it is stable. We therefore have a subcritical
Hopf bifurcation because in the interval 1 < < 0 the originwhich is a
limit cycle of radius r1 = 1 z and
stable focus has around it an unstable
a stable limit cycle of radius r2 = 1 + z whereas for > 0 it is the origin
whichis an unstable focus but has around it a stable limit cycle of radius
r2 = 1 + z.

5.9 Summary
In this chapter we conducted a broad survey of some of the major issues
of a dynamical system. As is widely recognized, a modern treatment of clas-
sical mechanics is incomplete without an insight from the standard ideas and
techniques of dynamical systems. Among them are essentially the theory of
stability, limit cycles and bifurcations. (We refrained from giving any treat-
ment on chaos which is beyond the scope of this book). To approach these
ideas we first provided the definitions of various concepts that frequently ap-
pear in dynamical systems. We placed a greater emphasis on the linear systems
and discussed how the process of linearization can be carried out for different
classes of nonlinear systems under certain suitable conditions. In this regard
many guided examples were worked out. We discussed the LotkaVolterra
model which is a popular model for prey-predator species. We considered sta-
bility of solutions and the role of the Lyapunov function. The concept of limit
cycles was explained by means of models including the one of Van der Pol os-
cillator. Several aspects of bifurcations were treated through model examples.

Exercises
1. Show that the origin of the system
x = x y 3 x2 y 2 , y = x y + x2 + y 2
is a stable focus.
2. Obtain the Lyapunov function for the system
x = xy y 3 x3 y 2 , y = xy x2
What is the character of the origin?
 Dynamical systems: An overview 159

3. Noting that a linear harmonic oscillator equation can be written as a

pair x = y, y = 02 x, examine the stability and asymptotic stability of the
modified system x = y, y = 02 x y 3 (1 + x2 ) when > or < 0.
4. Determine the fixed points and analyze the stability properties of the
Duffing oscillator

x + x x + x3 = 0
5. For the Brusselator model

x = 1 (1 + b)x + x2 y, y = bx x2 y
show by carrying out an analysis based on the Jacobian determinant method
for b < 2 there is a stable spiral mode where in the range 2 < b < 4 we have
a spiral repellor.
6. Find an appropriate Lyapunov function for the system

x = x + 2y 3 2y 4 , y = x y + xy
to show that the origin is an asymptotically stable point.
7. Show for the system

x = y + x(1 x2 y 2 ), y = x + y(1 x2 y 2 )
there exists an annulus bounded by the circles of radii 12 and 2 that contains
no fixed point. By the help of PoincareBendixson theorem deduce that there
is at least one limit cycle in this annulus.
8. Determine the equation of the flow for the above equation in polar
coordinates in the form

r2 e2t 1
t (r, ) = [( ) 2 , + t]]
1 r2 + r2 e2t
and also show that the unit circle describes a periodic orbit with period 2.
9. Use the PoincareBenedixson theorem to show that the system
1 1
x = y + x(1 2x2 2y 2 ), y = x + y(1 x2 y 2 )
4 2
admits a limit cycle inside the trapping region 12 r 2. Show that there is
a smaller region given by 12 r 1 containing the orbit.
10. Show that a stable periodic orbit exists for the system

x + (x2 )x + x = 0

11. Show that the Rayleigh equation

x + (x2 )x + x = 0
160 Advanced Classical Mechanics

can be converted to the Van der Pol form by differentaiting it with respect to
time and putting y = x. Show that both the equations have the same first-
order representation.
12. Analyze the following equations with regard to their bifurcation prop-
erties:

(i)x = f (x) = 1 + x + x2

(ii)x = f (x) = cosh x

(iii)x = f (x) = x x(1 x)

(iv)x = f (x) = x 4x3

x
(v)x = f (x) = x +
1 + x2
13. Study the pitchfork bifurcation for the system

x = x + y + sin x, y = x y
where is a parameter.
14. Analyze the system

3 3
x = x y (x + y)(x2 + y 2 ), y = x + y + ( x y)(x2 + y 2 )
2 2
for Hopf bifurcation.
15. Consider the Brusselator model involving an additional parameter

x = (1 + b)x + x2 y, y = bx x2 y
Discuss the stability of the fixed points and the existence of a limit cycle.
Comment on the Hopf bifurcation if = 1.
Chapter 6
Action principles

6.1 The principle of stationary action . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161

6.2 Corollaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
6.3 Continuous systems: Uniform string problem . . . . . . . . . . . . . . . . . . . 169
6.4 Normal modes of oscillation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
6.5 Extended point transformation and variation . . . . . . . . . . . . . . . . 174
6.6 and variations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
6.7 Brachistochrone problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
6.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182

An action principle is basically a variational principle with the central idea

derived by defining a functional, for two states of a physical system, corre-
sponding to their initial and final configurations as the system evolves in time,
which yields a stationary value when taken over the actual path which links
these configurations, as compared to the neighboring varied paths having the
same configurations, provided the total energy remains the same in the var-
ied motion as in the actual motion. In simple terms, the action is defined in
terms of the time integral of the Lagrangian between two fixed values of time
that are identified with the initial and final position of the particle during
the course of its motion. Action principles are of paramount importance in
classical mechanics in that almost all formulations of physics admit an ac-
tion principle interpretation while conversely any formalism resulting from an
action principle is termed as well defined.We first introduce the principle of
stationary action.

6.1 The principle of stationary action

Consider a physical system with n degrees of freedom. Let P denote a
point particle described by a set of n generalized coordinates q1 , q2 , ..., qn .
We call it the representative point of the system. As the system evolves, the
coordinates q1 , q2 , ..., qn change with time and P moves in the configuration
space as dictated by the forces acting upon it. As shown in Figure 6.1, C
denotes the actual path between points P0 and P1 along which the motion

161
162 Advanced Classical Mechanics

S'

S'

S'
S'

FIGURE 6.1: Two adjecent trajectories between representative points P0

and P1 .

is executed. P0 and P1 correspond to the representative points of the system

at times t = t0 and t1 , respectively. As the system evolves the representative
point traces out the curve C joining the points P0 and P1 . The curve is called
the trajectory of the system. The actual path is the one all along which the
equations of motion are satisfied. Thus C is a dynamically allowed path and
is distinguished from the neighboring ones (which may even lie infinitesimally
close to C) in that these are only geometrically possible but dynamically
impossible. In other words, the values of the coordinates and velocities that
could exist on the curve C 0 will not satisfy the equations of motion defined C.

We already encountered the notion of virtual variations qi which are con-

sistent with the conditions of the constraints. Note that is a shorthand
notation for the variation. Let C 0 be a geometrically allowed path which is
virtually displaced from the actual path C. Recall that in a virtual displace-
ment no passage of real time is involved. While the actual path is traced out
by the coordinate qi , the neighboring virtual path can be specified by the
displaced coordinate qi + qi . The corresponding velocities for the actual and
virtual paths would be, respectively, qi and qi + qi . The comparison of the
two paths, the actual and virtual, is, of course, done at the same point of time.
Since
d d
(qi + qi ) = qi + (qi )
dt dt
we readily identify
d
(qi ) = qi (6.1)
dt
from the remark just made. It is then implied that the time derivative operator
and the variation operation are interchangeable bringing out the commutative
character of these two quantities: d = d.
 Action principles 163

We now look into the consequences of (6.1) for the change in the La-
grangian. We find
n  
X L L
L = qi + qi
i=1
qi qi
n    n
X L d L d X L
= qi + qi
i=1
qi dt qi dt i=1 qi
n
d X L
= qi (6.2)
dt i=1 qi

where we have used Lagranges equations of motion which hold on the actual
trajectory. The appearance of total time derivative in the right-hand side of
(6.2) implies that if we integrate from an initial time t0 to the final time t1 ,
then Z t1 n
X L
Ldt = qi |tt10 (6.3)
t0 i=1
q i

Now writing Z t1
S[qi ] = L(qi , qi , t)dt, i = 1, 2, ..., n (6.4)
t0

we immediately notice that S corresponds to the left-hand side of (6.3). This

is because due to (6.1) we can write
Z Z Z Z Z
S = (L)dt + L(dt) = (L)dt + Ld(t) = Ldt

since for a virtual variation t = 0.

The action1 of the path is given by the functional S[qi ]: it is the time
integral of the Lagrangian between the terminal values of t0 and t1 along a
particular path qi (t). We can represent (6.3) as
n
X
S = pi qi |tt10 , i = 1, 2, ..., n (6.5)
i=1

using the usual definition of the canonical momenta pi = L qi . Now if both

the actual and virtual paths coincide at t0 and t1 , which are initial and final
times, respectively, the virtual displacements qi vanish at t0 and t1 implying
from (6.5)
S = 0 (6.6)
We are thus in a position to state the principle of stationary action which
says that the actual path chosen by a physical system, between end points
P0 (q0 , t0 ) and P1 (q1 , t1 ), is such that along it the action (6.4) is stationary as
1 Note that the action has the dimensions of energy time or force length time.
164 Advanced Classical Mechanics

compared with neighboring virtual paths (i.e., which are geometrically possi-
ble) having the same terminal points (namely, P0 and P1 ) as the actual trajec-
tory. In other words, the time integral of the Lagrangian is an extremum: (6.6)
is also called Hamiltons action principle. It was first proposed by Hamilton in
183435 and later extended to nonstationary constraints by Ostrogradsky in
1848. The principle of stationary action owes its formulation to Pierre Louis
Maupertuis who proposed it in 1744 in analogy to Fermats Principle of Least
Time and Leonhard Euler who established Maupertuiss idea in the same
year.
We have thus far exploited Lagranges equations of motion to arrive at
Hamiltons principle of stationary action (6.6). The converse also works. Since
Z t1 Z t1 n   
X L d L
S = Ldt = qi dt
t0 t0 i=1
qi dt qi
where qi are arbitrary independent variations, S = 0 provides Lagranges
equations of motion
 
d L L
= 0, i = 1, 2, ..., n
dt qi qi

by equating to zero the coefficient of each qi (i=1,2,...,n) which are arbitrary

and independent.
The principle of stationary action has an axiomatic status in classical me-
chanics. Lagranges equations and the principle of stationary action are equiv-
alent from an information content point of view. The action principle has its
roots in the calculus of variations where the differential equations of the form
as above are referred to as Eulers equations and emerge as a necessary con-
dition for the existence of an extremum for the variational problem. The set
of Lagranges equations is also called EulerLagranges equations.

Proposition:
Rt Rt  dF

The functionals t01 L(q, q, t)dt and t01 L(q, q, t) + dt dt lead to the same
equations of motion.
Proof:

Z t1  
0 dF
S = L(q, q, t) + dt
t0 dt
Z t1
= L(q, q, t)dt + [F (t1 ) F (t0 )]
t0
S 0 = S + [F (t1 ) F (t0 )]
 Action principles 165

It is obvious that the second term in the right-hand side vanishes. Hence by
the principle of stationary action S 0 = S = 0 and we are led to a similar set
of equations of motion from both S and S 0 .
We conclude this section by making a few remarks on the passive and
active points of view. In the passive point of view, if C represents the path
of an actual motion, then the observer in the unprimed reference frame will
write his action as Z t1
S[C] = L(q, q, t)dt
t0

On the other hand, the observer in the primed reference frame will write his
action as Z t01
S[C] = L0 (q 0 , q0 , t0 )dt0
t00
0
Since dt0 is not expected to be equal to dt, i.e., dtdt 6= 1, the functional form
of L and L0 would, in general, be different. However, the form of Lagranges
equations, obtained from the stationary character of the action, one for the
unprimed system and one for the primed system, would be similar. This is
known as covariance:
 
d L L
= ,
dt q q
d L0 L0 dq 0
 
0
= 0
, q 0 0
dt q q dt

In the active point of view, for the actual path C, S[C] is

Z t1
S[C] = L(q, q, t)dt
t0

It is to be stationary in comparison to other neighboring paths. If we think of

transformations which carry actual paths to actual paths, namely from C to
C 0 , then such transformations are called invariance transformations. Thus for
invariance, the action for C 0
Z t01
0
S[C ] = L(q 0 , q 0 , t0 )dt0
t00

should be stationary as compared with neighboring paths to C 0 . Here L re-

mains the same. We shall deal with invariance transformation in some detail
in Chapter 8 in connection with symmetries and conserved quantities of a
physical system.
166 Advanced Classical Mechanics

6.2 Corollaries
(a) Hamiltons principle from DAlemberts principle
Rt
The integral t01 (T V )dt is stationary for an actual trajectory in compar-
ison with neighboring paths having coordinates of the end points fixed along
with the terminal time instants.
Proof:

We know from DAlemberts principle that for a set of applied forces F~ia
the following holds:
XN
(F~a mi r~i ). r~i = 0i
i=1

Since d
dt ( r
~i ) d
= dt ri ) = r~i , it can also be expressed in the form
(~

N   X N
X d
mi (r~i . r~i ) r~i . r~i = F~ia . r~i
i=1
dt i=1

Identifying thePsecond term in the left-hand side as the variation of the kinetic
N
energy T = 12 j=1 mj |r~j |2 and the right-hand side as the virtual work W
for the applied forces we have the relation
N
X d
mi [ (r~i . r~i )] = T + W (6.7)
i=1
dt
Integrating between t0 and t1 and noting that the coordinates of the end
points are fixed there, the left-hand side clearly vanishes yielding
Z t1
(T + W )dt = 0 (6.8)
t0

The above result holds in general encompassing non-conservative systems as

well.
It is to be stressed that while the virtualP
work W for the applied force is
N
a well-defined quantity as given by the sum i=1 F~ia . r~i , the same cannot be
said of a finite W which, in general, is a path-dependent entity. However, in
the case of conservative systems where the existence of a potential function V
is assured we can express
N N
X X V
W = F~ia . r~i = r~i = V (6.9)
i=1 i=1
r~i
 Action principles 167
Rt Rt Rt
As a consequence, we recast t01 (W )dt as - t01 (V )dt, i.e., - t01 V dt. We are
therefore led to Hamiltons principle
Z t1
(T V )dt = 0 (6.10)
t0

which in terms of the Lagrangian L = T V reads

Z t1
Ldt = 0 (6.11)
t0

signifying the extremum character of the time integral of the Lagrangian.

In the case of Ra free particle for which V = 0, we can express the action
integral
R as simply vdt = 0 for a suitable time interval. This can be translated
to ds = 0 which stands for the principle of shortest path.
(b) Hamiltons canonical equations from the stationary action
principle

We can write
Z t1 Z t1 Z N
t1 X
Ldt = Ldt = (pi qi H)dt
t0 t0 t0 i=1

using the definition of the Hamiltonian. Taking the variation we can express
the right-hand side as
Z N 
t1 X   
H H
pi qi + qi pi qi dt
t0 i=1
pi qi

d
But qi = dt (qi ) implying that we can write
Z t1 Z t1 Z t1
d
(pi qi )dt = pi (qi )dt = pi .qi dt
t0 t0 dt t0

because qi vanishes at the terminal time points t0 and t1 . Hence

Z t1 Z N 
t1 X     
H H
Ldt = + pi qi + qi pi dt
t0 t0 i=1
qi pi

The quantities qi and

R t pi being arbitrary and independent, the principle of
stationary action t01 Ldt = 0 gives Hamiltonian canonical equations:

H H
qi = , pi =
pi qi
168 Advanced Classical Mechanics

Example 6.1

Use Hamiltons principle to write down the equations of the motion in

spherical polar coordinates.

In spherical polar coordinates the kinetic energy T has the form

1 2 
T = r + r2 2 + r2 sin2 2 , (m = 1 has been set )
2
Therefore, Hamiltons principle reads
Z t1
1
[ (r2 + r2 2 + r2 sin2 2 ) V ]dt = 0
t0 2

where the potential V is V = V (r, , ). Writing

V V V
V = ( r + + )
r
and noting that integrating by parts gives
Z t1 Z t1
r rdt = rrdt
t0 t0
Z t1 Z t1
d 2
r2 dt = (r )dt
t0 t0 dt
Z t1 Z t1
d d 2 2
r2 sin2 ()dt = (r sin ) dt
t0 dt t0 dt
where we have discarded terms evaluated at the boundaries t0 and t1 because
of the fixed end points, we arrive at the expression
Z t1
V d V
[(r r2 r sin2 2 + )r + { (r2 ) r2 sin cos 2 + }
t0 r dt
d V
+ { (r2 sin2 ) + }]dt = 0
dt
Since r, and are arbitrary and independent variations we find on equat-
ing to zero the respective coefficients
V
r r2 r sin2 2 = ,
r
1 d 2 1 V
(r ) r sin cos 2 = ,
r dt r
1 d 2 2 1 V
(r sin ) =
r sin dt r sin
which are the respective equations of motion for r, and variables.
 Action principles 169

6.3 Continuous systems: Uniform string problem

Thus far in this book we have handled systems with a finite number of
generalized coordinates indexed by the label i = 1, 2, ..., n. In other words,
we deal with systems with a finite number of degrees of freedom. However, in
problems like that of a uniform string which constitute a continuous system
we need to construct equations that are local and concern infinite degrees
of freedom. Thus the generalized coordinates become continuous functions of
space and time.
Suppose we have a tightly stretched uniform string under a tension T
which is fixed at both ends. Let the string have a length l in its equilibrium
position and mass per unit length. We take the x-axis along the length of
the string. In equilibrium the string occupies the portion from x = 0 to x = l.
If the string is set in vibration we can write for the kinetic energy at time t
Z l
1 u 2
( ) dx (6.12)
0 2 t
where u(x, t) denotes the vertical displacement of the string from its equilib-
rium position to the point x at time t.
To find the potential energy, we note that for a slight displacement x
(that causes the length of the string, initially held tight, to increase by a
little amount say, l), work done against the force of tension T is given by an
element dW

dW = T (ds dx) (6.13)

where ds is an element of the arc length given by

ds2 ' dx2 + du2 (6.14)

For small vibrations we can approximate

ds 2 u(x, t) 2
( ) =1+( ) (6.15)
dx x
enabling us to write

r
u(x, t) 2 1 u(x, t) 2
dW = T [ 1 + ( ) 1]dx = T ( ) dx (6.16)
x 2 x
Thus the total potential energy which is the work done while the string is
stretched to its displaced state from the equilibrium postion is
Z l
1 u(x, t) 2
V = T( ) dx (6.17)
0 2 x
170 Advanced Classical Mechanics

The Lagrangian for the string can be written as

Z l
1 u 1 u(x, t) 2
L= [ ( )2 T ( ) ]dx (6.18)
0 2 t 2 x
which can be reexpressed in terms of the Lagrangian density L
Z l
L= Ldx (6.19)
0
where

1 u 2 1 u(x, t) 2
L=
( ) T ( ) ] (6.20)
2 t 2 x
To derive the equations of motion for such a Lagrangian we can use Hamil-
tons principle by considering the variation u(x, t) on u for fixed end points
in time

u(x, t0 ) = u(x, t1 ) = 0 (6.21)

The variations, of course, also vanish at the fixed end points of the string.
Turning to the action which is given by
Z t1 Z l
S= Ldx (6.22)
t0 0

its variation is given by

Z t1 Z l
L L 0 L
S = [ u + u + u]dx (6.23)
t0 0 u u0 u
where a dot and a prime denote, respectively, partial derivative with respect
to t and x. Interchanging as usual the operations of a variation and the partial
derivative and integrating by parts with respect to t in the first term and with
respect to x in the second term of the integrand leaves us with the result
Z t1 Z l
d L d L L
S = [ ( ) ( )+ ]dxdtu(x, t) (6.24)
t0 0 dt u dx u0 u
The variations u(x, t) being arbitrary and independent, for an action to be an
extremum the integrand has to vanish giving the equations of motion
d L d L L
( ) ( )+ =0 (6.25)
dt u dx u0 u
For the specific case of the Lagrangian of the string given above we find the
form

2u 2
2 u T
= c , c2 = (6.26)
t2 x2
 Action principles 171

which gives the transverse equation of motion of a uniform string. Note that c
has the dimension of velocity and is the propagation velocity of the transverse
waves. Equation (6.26) represents a one-dimensional linear wave equation.
To look for a general solution of (6.26) we invoke a change of variables
specified by = x ct, = x + ct which transform it to the form u = 0.
It transpires immediatey that its solution has a separable form consisting of
arbitrary function of and . Taking cue from it we can express the general
solution of (6.26) in the form

u(x, t) = F (x ct) + G(x + ct) (6.27)

where F and G are arbitrary functions of their arguments.

Let us consider an initial value problem given by the initial conditions:

u(x, 0) = f (x), gt (x, 0) = g(x) (6.28)

where f and g are given functions corresponding to initial displacement and

initial velocity, respectively. Note that we are dealing with an evolution process
and that (6.27), (6.28) together constitute the so-called Cauchy problem for
the one-dimensional wave equation (6.26). The lines x ct = constant and
x+ct = constant are called the characteristics in the terminology of differential
equations.
Given the above initial conditions we infer from (6.27) the following con-
ditions that are needed to be satisfied:

F (x) + G(x) = f (x), cF 0 (x) cG0 (x) = g(x) (6.29)

where the prime represents a derivative. Solving for F and G we find

Z x+ct
1 1 1
u(x, t) = f (x + ct) + g(x ct) + g(s)ds (6.30)
2 2 2c xct

(6.30) is called DAlemberts solution for the wave equation. It shows that the
solution depends only upon the initial value functions specified at x ct and
x + ct and on the integral of g evaluated between these two points. In the case
of zero initial velocity we have a simple superposition
1 1
u(x, t) = f (x + ct) + g(x ct) (6.31)
2 2
showing two waves travelling to the right and left with the same speed c.
To show that DAlemberts solution is well-posed we take a set of two
solutions u1 and u2 corresponding, respectively, to two initial specifications of
functions f1 , f2 and g1 , g2 . Suppose that

|f1 (x) f2 (x)| < , |g1 (x) g2 (x)| < (6.32)

172 Advanced Classical Mechanics

where is an arbitrary small preassigned number. Since from (6.30)

1
u1 u2 = [f1 (x ct) f2 (x + ct) + g1 (x ct) g2 (x + ct)]
2
Z x+ct
1
+ [g1 (s) g2 (s)]ds (6.33)
2c xct
then for all x IR and 0 t , it follows by (6.32)
1 1
|u1 u2 | < ( + ) + (x + ct x + ct) < (1 + ) <  (6.34)
2 2c

where has been taken < 1+ for a given  > 0. We therefore conclude that
small changes in the initial conditions produce a corresponding small change
in the solutions implying that the Cauchy problem is well-posed.

6.4 Normal modes of oscillation

To find normal modes of oscillations, we note that the string equation is
subject to the boundary conditions
u(0, t) = 0, u(l, t) = 0 (6.35)
If we seek a solution of the form
u(x, t) = (x) cos(t + ) (6.36)
which is in a separable form involving an arbitrary function (x) and a trigono-
metric factor of time chosen due to the linearity of the differential equation,
then substitution in (6.30) shows that satisfies the differential equation
d2 2
+ 2=0 (6.37)
dt2 c
The accompanying boundary conditions as guided by (6.34) are
(0) = 0, (l) = 0 (6.38)
In the general solution of (6.37) which is of the form
x x
(x) = A cos + B sin (6.39)
c c
where A and B are arbitrary constants, A turns out to be zero while is
restricted by the constraint l = nc, where n is an integer. As a result the
nth normal mode of the string is given by
nx
un (x, t) = Bn sin cos (n + ) (6.40)
l
 Action principles 173

Hence u(x, t) is expressible in the form

X nx nct nct
u(x, t) = sin [Cn cos + Dn sin ] (6.41)
n=1
l l l

where Cn and Dn are arbitrary constants and can be fixed from the initial
conditions.
To this end we make use of the first condition in (6.28) to write down

X nx
u(x, 0) = f (x) = Cn sin (6.42)
n=1
l
which on inversion gives
Z l
2 nx
Cn = f (x) sin dx (6.43)
l 0 l
On the other hand, if we differentiate with respect to t and use the second of
(6.28) then

X nc nx
ut (x, 0) = g(x) = Dn sin (6.44)
n=1
l l
On inverting this gives for Dn
Z l
2 nx
Dn = g(x) sin dx (6.45)
nc 0 l
We wish to remark that in the case of zero initial velocity the solution
(6.41) because of (6.45) reduces to

X nx nct
u(x, t) = Cn sin cos (6.46)
n=1
l l

where Cn is given by (6.43). Making use of the trigonometric identity

sin A cos B = 12 [sin(AB)+sin(A+B)], we deduce from (6.46) the expression
for u(x, t)

X n(x ct) n(x + ct)
u(x, t) = Cn [sin + sin ] (6.47)
n=1
l l

which can be recognized to be in DAlemberts form (6.31).

174 Advanced Classical Mechanics

6.5 Extended point transformation and variation

There are certain physical situations in which moving boundaries are rel-
evant. This is in contrast to the case of Hamiltons principle where the varied
path shares with the actual path the same end points which are fixed.
Curves with variable boundary points can appear as a result of extended
point mappings transforming (q, t) to the new forms (q 0 , t0 ) as given by q 0 =
q 0 (q, t), t0 = t0 (q, t). We restrict ourselves to the infinitesimal case
q 0 = q + q(q, t), t0 = t + t(q, t) (6.48)
where q and t are infinitesimal small changes. Note that is to be distin-
guished from in that we use for a virtual (time-frozen) change.
In the active point of view there is a single observer who observes the
evolution of the system paths C to C 0 . To him the difference in the action for
two neighboring paths appears as
Z t1  
dq 0
  
dq
S = S[C 0 ] S[C] = L q 0 , 0 , t0 dt0 L q, , t dt (6.49)
t0 dt dt
where (q 0 , t0 ) are infinitesimally different from (q, t).
We can reexpress (6.48) as
   
0 dq 0 dt
dq = dq + dt, dt = dt + dt (6.50)
dt dt
d
where dt = t + q q as usual. Since q and t are both infinitesimal we have
for the ratio
dq 0
 
dt dq dq
= 1 + (6.51)
dt0 dt dt dt
yielding the non-trivial result
dq 0 dq dq
0
6=
dt dt dt
i.e., change in the generalized velocities is not the same as the time derivative
of the change in the generalized coordinate. 
0
Using (6.50) and (6.51), L q 0 , dqdt0 , t0
can be expanded as

dq 0
      
dt dq dq dt
L q 0 , 0 , t0 dt0 = L q + q, 1 + , t + t 1 +
dt dt dt dt dt
    
dq dq dt dq dt
= L q + q, +{ }, t + t 1 +
dt dt dt dt dt
      
dq L dq dt dq L L dt
= L q, , t + q +{ } + t 1+
dt q dt dt dt q t dt
   
dq L L dq dt L dt
= L q, , t + q + q + t + L
dt q q dt dt t dt
 Action principles 175

where we have kept only the first order quantities in q, t and used Taylor
expansion

(x + x, y + y, z + z, ...) = (x, y, z, ...)

 

+ x + y + z + ... (x, y, z, ...) + ...
x y z

Thus (6.49) becomes

Z t1    
L L dq dt L dt
S = q + q + t + L dt
t0 q q dt dt t dt

The above expression can be simplified further because by integration by parts

we can write
Z Z  
L dq L d L
dt = q qdt
q dt q dt q
Z Z  
L dt L d L
q dt = qt q tdt
q dt q dt q
Z Z
dt dL
L dt = Lt tdt
dt dt
which yields the form
Z t1      t1
L d L L L
S = dt (q qt){ ( )} + q + L q t
t0 q dt q q q t0

where we have expressed the difference dL L L

dt t = q q . For the actual path
Lagranges equations are satisfied and hence we are left only with
   t1
L L
S = q + L q t (6.52)
q q t0

An interesting special case of (6.52) is that when the the system is conser-
vative and scleronomic having its kinetic energy as a homogeneous function of
velocities. Corresponding to the Hamiltonian H and the canonical momentum
p we then have
t
S = [pq Ht]t10 (6.53)
For N number of particles, (6.53) can be extended to
Z t1 N
X t
S Ldt = [pi qi Ht]t10 (6.54)
t0 i=1
176 Advanced Classical Mechanics

If we restrict to the following prescriptions:

(i) H does not depend explicitly on t and so H is conserved.

(ii) H is conserved not only on the actual path but also on the varied path
and
(iii) Varied paths are constrained such that qi (i = 1, 2, ..., n) vanish at the
end points but not t.
then (6.54) further simplifies to
Z t1
S Ldt = H(t1 t0 ) (6.55)
t0

Now, the action integral reads

Z t1 Z t1 "Xn
# Z t1 n
X
Ldt = pi qi H dt = pi qi dt H(t1 t0 ) (6.56)
t0 t0 i=1 t0 i=1

Taking variation of both sides we have

Z t1 n
Z t1 X
Ldt = pi qi dt H(t1 t0 ) (6.57)
t0 t0 i=1

If we compare with (6.55) we find

Z t1 n
X
pi qi dt = 0 (6.58)
t0 i=1

The integral in (6.58) is generally referred in old books as the action or

the action integral and (6.58) as the principle of least action. Historically it
was Fermat who made an application of it in the case of refraction of light.
In recent times it is more customary to refer to the integral in Hamiltons
principle as the action. Recall Hamiltons principle is the variational principle
which states that the physical trajectory is the one for which the action is
stationary.
If L = T V and H = T + V then
n
X
pi qi = H +L
i=1
= T + V + (T V )
= 2T

Hence another form of (6.58) is

Z t1
2T dt = 0 (6.59)
t0
 Action principles 177

6.6 and variations

For the terminal time points t0 and t1 we already know that (t0 ) = (t1 ) =
0. Consider a function f (t). We can expand it as
f1 (t + t) f (t) = f1 (t) + tf1 f (t) = [f1 (t) f (t)] + tf1 = f + tf1
Writing f1 = f + f we obtain
f = f + tf (6.60)
where obviously f = f if t = 0. Differentiating (6.60) with respect to t
we have
d d d d
(f ) = (f ) + (t)f + t (f)
dt dt
 dt  dt
d d
= f + t (f ) + (t)f

dt dt
d
= f + (t)f (6.61)
dt
where we have used f = f + tf analogous to (6.60).
We thus arrive at the general formula
d d
(f ) = f + (t)f (6.62)
dt dt
d d d
showing that dt and are not interchangeable: dt () 6= dt .

Parametric representation of (6.59)

We now proceed to express (6.59) as a parametric equation. Toward this

Rt
end writing A = t01 2T dt, we have
Z t1
A = 2T dt
t0
Z t1
= 2T 2T dt
t0
Z t1
r
p X ds
= 2(H V ) m( )2 dt
t0 dt
Z t1 p qX
= 2(H V ) mds2
t0
Z P1 p r
X ds
= 2(H V ) m( )2 d
P0 d
where is an arbitrary parameter.
178 Advanced Classical Mechanics

Thus A can be written in an equivalent form

Z P1
A= Id (6.63)
P0

p qP
ds 2
where I is given by I = 2(H V ) m( d ) . Thus I may be looked upon
as a function of the generalized coordinates (q1 , q2 , ..., qN ) and the derivatives
( dq 1 dq2 dqN 0 0 0 0 dqi
d , d , ..., d ) denoted by (q1 , q2 , ..., qN ) where qi d , i = 1, 2, ..., N .

Proposition:

The quantity I obeys the differential equations

 
d I I
= (6.64)
d qi qi

where i = 1, 2, ..., N .

Proof:

From (6.59), i.e., A = 0, we have on using (6.60) and (6.63)

Z P1
0 = Id (as I is independent of t)
P0
Z n 
P1 X 
I I
= qr + 0 qr0 d (6.65)
P0 r=1
qr qr

Now since we can express

Z P1 Z P1
I 0 I d
0
q r d = 0
(qr )d
P0 qr P0 qr d
 P1 Z P1
I d I
= 0
qr ( 0 )qr d
qr P0 P0 d qr
Z P1
d I
= ( 0 )qr d (as qr = 0 at P0 and P1 )
P0 d qr

substituting it in (6.65) we get

Z P1 n  
X I d I
( 0 ) qr d = 0
P0 r=1
qr d qr

Since the variations qr are arbitrary and independent, the proposition follows.
 Action principles 179

Example 6.2:

A particle of unit mass is projected so that its total energy is E in a field

of force whose potential is (r) at distance r from the origin. Deduce the
differential equation of the path to be
 
dr
C 2 r2 + ( )2 = r4 [h (r)]
d

where C is a constant.

Here H = E, V = (r) and in the (r, ) plane ds2 = dr2 +r2 d2 . Therefore

Z t1
A = 2T 2T dt
t0
Z t1
r
p X ds 2
= 2(H V ) m( ) dt
t0 dt
Z t1 p qX
= 2(E ) mds2
t0
Z P1 p r
X ds
= 2(E ) m( )2 d
P0 d
Z P1 p r
dr d
= 2(E ) ( )2 + r2 ( )2 d
P0 d d

where is a parameter other than t. Therefore I can be identified to be

p p
I = 2(E ) r02 + r2 02

where r0 d
dr
, 0 d
d d I
. Since is an ignorable coordinate we deduce d ( 0 ) =
I

0 from (6.64). As a result 0 = constant. Let the constant be 2C implying

p r2 0
2(E ) = 2C
r02 +r 2 02

which yields on some rearrangements the desired expression

 
dr 2
C r + ( ) = r4 [E (r)]
2 2
d
180 Advanced Classical Mechanics

S'

S'

S'
S'

S'

FIGURE 6.2: Brachistochrone curve.

6.7 Brachistochrone problem

The brachistochrone problem is due to Johann Bernoulli who in 1696 at-
tempted to find a plane curve that joins two points such that the time required
for a particle travelling under the influence of gravity takes the minimum time
to move from one point to the other. The problem was subsequently solved
among others by Leibniz, LHospital and Newton.
In Figure 6.2 we take the x-axis to be horizontal and z-axis as vertically
upward. We consider a small element of arc P Q corresponding to the initial
and final positions P and Q of the particle, respectively. If the velocity and
height of the particle at P be (v, z) and the same at Q be (v, z) then it is clear
that
1
total energy at P = mv 2 + mgz
2
1
total energy at Q = mv 2 + mgz
2
Conservation of energy requires these to be equal so that
v 2 = v 2 + 2g(z z) (6.66)
ds
Now we can express v = where ds stands for the element PQ and
dt
dt specifies the short interval of time. Hence the above equation can be ex-
pressed as
Z Q
 1
ds v 2 + 2g(z z) 2

t =
P
1
z0
1 + ( dx
 2 2
dz )
Z
= 1 dz (6.67)
z [v 2 + 2g(z z)] 2

where we have put ds = dx2 + dz 2 .
 Action principles 181

The integral above has the same form as that of an action with x playing
the role of a generalized coordinate and z that of time. The integrand
defines the following Lagrangian:
1
1 + ( dx
 2 2
dz )
L = 1
[v 2 + 2g(z z)] 2

1
1 + x02 2 dx
= 1 , x0 =
[v 2 + 2g(z z)] 2 dz
= L(x, x0 , z) (6.68)
Note x is an ignorable coordinate here.

The path of minimum time is obtainable by taking an extremum of the

integral Z z0
L(x, x0 , z)dz = 0 (6.69)
z
which yields
d L L
( 0) = =0 (6.70)
dz x x
From this we deduce
L
= constant
x0
x0
i.e., 1 1 = C (say)
(1 + x02 ) 2 [v 2 + 2g(z z)] 2
1
x02 (1 + x02 )1 v 2 + 2g(z z) = C2

or (6.71)
which gives
 
1 1 C 2 v 2 + 2g(z z)
=
x02 C 2 [v 2 + 2g(z z)]
dx bz
or, ( )2 = (6.72)
dz a+z
where the constants a and b are

1 C 2 v 2 2gC 2 z (v 2 + 2gz)
a= , b= (6.73)
2gC 2 2g
Setting
ab a+b
z = ( )( ) cos (6.74)
2 2
facilitates integration since we get the form
Z r
a+b 1 + cos
dx = ( ) sin d (6.75)
2 1 cos
182 Advanced Classical Mechanics

which implies
a+b
x= ( sin ) + constant (6.76)
2
The variables z and x provide the curve for minimum time, the brachis-
tochrone, in parametric form with both x and z appearing as functions of
. The curve can be easily recognized to be a cycloid.

6.8 Summary
We gave a general treatment of an action principle by taking recourse to
variations which puts it in an elegant integral representation with the La-
grangian serving as the integrand between two terminal points of time. What
the action principle says is that of all the possible paths, as the system traces
out a trajectory from a given initial time t0 to a final time t1 , the actual path,
which is a solution of the equation of motion, is the one for which the action is
an extremum. Interesting corollaries that follow from the action principle are
the Lagrangian equations of motion and the Hamiltons canonical equations
of motion. We considered the extended point transformations which are rele-
vant for moving boundary problems and derived a particular variant of action
principle. We also addressed briefly the brachistochrone problem.

Exercises
1. Consider the integral
Z t1 p
f (x) 1 + x2 dt (6.77)
t0

involving a function f (x). Show that the extremum of the integral leads to
the differential equation for f (x)

1 + x2 = cf (x)2 (6.78)

where c is a constant.
2. A particle oscillates in a straight line about a center of force which varies
as the distance. By integrating from t0 to t deduce the action in the form

(x20 + x2 ) cos (t t0 ) 2xx0
(6.79)
2 sin (t t0 )

where is the constant of proportionality.

 Action principles 183

3. Consider a particle of unit mass moving in the xy plane under a central

acceleration proportional to the radial distance. By integrating from 0 to t
determine the action to be
n
[(x2 + x2 + y02 + y 2 ) cos nt 2(xx0 + yy0 )] (6.80)
2 sin nt 0
where n2 is the constant of proportionality.
4. Show that if time is coordinated as an additional coordinate of a me-
chanical system, then the momentum associated with time is the Hamiltonian
of the system with its sign changed. Deduce that the total energy of a sclero-
nomic system is a constant of motion.
5. Consider the integral Z t1
I= f dt (6.81)
t0

for the motion along an actual path and the integral

Z t1 +t1
I1 = f1 dt (6.82)
t0 +t0

for a varied path. Show that the difference I = I1 I is given by the integral
representation Z t1
d
I = [f + (f t)]dt (6.83)
t0 dt
Chapter 7
Motion in noninertial coordinate
systems

7.1 Rotating frames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186

7.1.1 Basic equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
7.1.2 Some remarks on the Coriolis force . . . . . . . . . . . . . . . . . . . . . 190
7.1.3 Effective gravitational constant . . . . . . . . . . . . . . . . . . . . . . . . . 191
7.1.4 Foucaults pendulum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
7.2 Nonpotential force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
7.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
7.4 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196

In Chapter 1 we introduced the idea of an inertial frame and noted that

any coordinate frame moving uniformly with respect to it is inertial. We also
encountered Galilean transformations that leave invariant the force-free equa-
tions of motion as derived from Newtons second law. In fact, use of the second
law is limited only to inertial frames. However at certain times it is necessary
for certain physical problems, to deal with non-inertial reference frames as in
the case of motion occurring near the surface of the rotating Earth. Here in-
troduction of a non-inertial frame simplifies the situation. Non-inertial frames
arise, in general, for rotating frames and accelerated coordinate systems. The
aim in this chapter is to study particle dynamics in a rotating frame of coordi-
nates and provide a detailed derivation of forces that need be considered along
with the external forces that exert on the system. The time independent har-
monic restoring force, namely, the centrifugal force and the time-dependent
Coriolis force, are two typical forces which appear in rotating reference frames.
In particular, the role of the latter is to rotate the velocity vector making it
precess around the rotation vector direction.

185
186 Advanced Classical Mechanics

z'

0'
S'
y'
x'

FIGURE 7.1: Rotating coordinate system S fixed in the rigid body R and
inertial coordinate system S 0 fixed in space.

7.1 Rotating frames

7.1.1 Basic equations
A rotating frame of reference is a special case of a non-inertial system. Be-
cause of rotation it generates pseudo-forces as we shall presently see. Consider
two orthogonal coordinate frames of reference namely S:OXYZ attached to a
rigid body R with standard unit basis vectors i, j, k and S:OXYZ a fixed
one in space representing an inertial coordinate system (see Figure 7.1).
If the rigid body is rotating with an angular velocity ~ about a fixed axis
through O, then it is obvious that for a vector ~ fixed in R, an observer
positioned at O will see no change in the components (x , y , z ) of ~ relative
to OX, OY, OZ. However to an inertial observer at O0 , the time-rate of change
of ~ will appear as

d~ d~i d~j d~k

 
dx~ dy ~ dz ~
= i+ j+ k + x + y + z (7.1)
dt dt dt dt dt dt dt
~ ~ ~
where dtdi
=~ ~i, ddtj = ~ ~k are to be interpreted as induced
~ ~j, ddtk =
velocities caused by the angular velocity of the frame S relative to S 0 .
It therefore follows that
   
d ~ d
= ~ +
~ ~ (7.2)
dt fixed dt rot
d


where dt rot stands for the time-rate as measured from the rotating frame
S. Indeed Equation (7.2) furnishes the typical motion of a rigid body being
described by a combination of translation and rotation. The suffixes fixed
 Motion in noninertial coordinate systems 187

z'
z p

r
-r'

y'
0'
ii
0

X S'
y x'
s
FIGURE 7.2: Position of a particle at P viewed from two coordinate systems
S and S 0 .

and rot specify the roles of a fixed observer in S 0 and someone rotating with
the rigid body.
To derive the governing equations of a motion of a particle moving relative
to a rotating frame, let us imagine that the coordinate frame S 0 has been set
up in a fixed star and the rigid body is our Earth itself. We neglect the orbital
motion of Earth around the Sun and assume the Earth to be a perfect sphere.
Let P be the position of a particle at time t with OP ~ = ~r and O~0 P = r~0 . If
~
0 ~ 0
O O = ~a then it is clear from Figure 7.2 that r = ~r + ~a. As such the velocity
of the particle at P when measured from O0 would read
   
d ~0
d d~a
~u r = ~r + (7.3)
dt fixed dt fixed dt
d


To estimate dt fixed ~r we substitute the relationship (7.2) between the
fixed and a rotating frame replacing ~ by ~r. In consequence we get from (7.3)
 
d~r d~a
~u + ~ ~r + (7.4)
dt rot dt
The first term in the right-hand side of (7.4) gives the velocity of the particle
relative to the rotating frame S, the second term appears as a result of the
rotation of the OXY Z frame and the third term represents the so-called drag
velocity. The drag velocity can be ignored if the distance vector between the
points O and O0 does not change with time.
To derive an expression for the acceleration of the particle at P as measured
from O0 , we operate upon (7.4) by the derivative dt d


fixed :

 2
d2~a
       
d d ~ 0
d d~r d
~u r = + (~
~
r )+
dt fixed dt2 fixed dt fixed dt rot dt fixed dt2
(7.5)
188 Advanced Classical Mechanics

We now make use of ~

d~
r

the following results obtained by replacing successively
by the vectors dt rot and ~ ~r in (7.2):

     2   
d d~r d ~r d~r
= 2
+
~
dt fixed dt rot dt rot dt rot
   
d d
~r) =
(~ ~r)
(~ ~ (~
+ ~r) (7.6)
dt fixed dt rot
Substituting (7.6) in (7.5) we obtain

!
d2 r~0 d2~a
   
d~u d~

= + ~ (~
~r) + ~r
dt fixed dt2 dt2 dt rot
fixed    2 
d~r d ~r

+2~ + (7.7)
dt rot dt2 rot

Note that the first three terms in the right-hand side of (7.7) survive even
when P is stationary relative to the rotating frame
   2 
d~r d ~r
S: = =0 (7.8)
dt rot dt2 rot
2
Ignoring the ddt~2a term by assuming that the distance OO ~ 0 is not changing
with time and observing that in an inertial frame Newtons second law implies
that
 
d~u
m = F~ (7.9)
dt fixed
with F~ representing the vector sum of the forces acting on the particle, we
have from (7.7)
 
d~v
m = F~ 2m(~ ~v ) m~ (~
~r) m~ ~r (7.10)
dt
where we have set
   2   
d~r d~v d ~r d~

~v = , = and ~ =
(7.11)
dt rot dt dt2 rot dt rot
and dropped the suffixes fixed and rot without causing any confusion. It
should be clear that ~v is the velocity relative to the rotating frame.
Equation (7.10) is similar in form to Newtons equation. However, apart
from F~ , its right-hand side, is also influenced by fictitious forces, namely,
the Coriolis force 2m(~v ~ ), the centrifugal force m~ (~r
~ ) and the
transverse or azimuthal force due to the non-uniform rotation m ~ ~r. The
 Motion in noninertial coordinate systems 189

z'
z p

r
-r'

y'
0'
ii
0

X S'
y x'
s
FIGURE 7.3: Typical three-dimensional motion with rotation about the
zaxis.

latter can be neglected for a uniform rotation. Note that a fictitious force
is due to an accelerated frame of reference. The form of the Coriolis term
implies that it is always perpendicular to the direction of velocity and so it
can never change the speed of a particle (except, of course, for its direction).
As a result, the Coriolis force does not contribute to the energy equation. The
Coriolis force is also referred to as a deflecting force.
Writing F~ = V r , where V (~
~ r) is the potential, we can easily verify that
the Lagrangian for (7.10) is
1
L= m[v 2 + 2~v (~ ~r)2 ] U
~r) + (~ (7.12)
2
in which the velocity-dependent potential U is given by
1
U = V (~r) ~v (~ ~r)2
~r) m(~ (7.13)
2
Consider a typical three-dimensional motion with rotation about the
zaxis (see Figure 7.3) and assuming
~ constant (= ):

~ = ~k
(7.14)
= (cos er sin e ) (7.15)

where er , e are the unit basis vectors in the ~r and ~ directions, respectively.
Noting that

~v = rer + (r)e
d~v 1 d 2
= (r r2 )er + (r )e (7.16)
dt r dt
~ ~r
= (cos er sin e ) rer = r sin e
190 Advanced Classical Mechanics

where e is the basis vector in the direction, we can write

~ ~v = (r sin + r cos )e
(7.17)

~r) = r sin (sin er + cos e ) 2

~ (~ (7.18)
~ ~r, we have componentwise
Further since ~u = ~v +

ur = r, u = r, u = r sin (7.19)
Substituting (7.16), (7.17), (7.18), (7.19) in (7.10) we arrive at the following
expressions of the various components of acceleration in a non-inertial rotating
frame:

radial component of acceleration fr = r r2 r sin2 2

1 d 2
cross-radial component of acceleration f = (r ) r 2 sin cos
r dt
d 2 2
azimuthal component of acceleration f = (r sin ) (7.20)
r sin dt

7.1.2 Some remarks on the Coriolis force

From (7.10) the Coriolis force term is 2m~ ~v where ~v d~ r
dt . The
Coriolis force is due to the rotation of the Earth. Consider the case of a flat
rotating disc. For a particle moving across a disc under no forces, an inertial
observer (i.e., the one who is in a fixed frame) will see it moving across a
straight line (Figure 7.4) according to Newtons first law. However, in view
of the fact that the disc is rotating, an observer stationed on the disc will
view the particle taking a curved track (see Figure 7.5) due to the Coriolis

FIGURE 7.4: A particle moving across a disc under no forces.

 Motion in noninertial coordinate systems 191

FIGURE 7.5: An observer stationed on the disc.

force operating in a direction perpendicular to the motion of the particle.

Note that the Coriolis force causes the velocity vector to rotate clockwise in
the Northern hemisphere and counterclockwise in the southern hemisphere.
The Coriolis force causes hurricanes, tropical cyclones and such severe storms
to rotate in different directions in the Northern and Southern hemispheres.
The effect of the Coriolis force is minimum near the equator (in fact, it is
zero there and hence its effect is more profound for the longitudinal motion,
i.e., North-South wind, than the latitudinal counterpart, i.e., East-West wind)
and maximum toward the poles. Thus a hurricane will never happen near the
equator but gain in intensity as it approaches the poles.

7.1.3 Effective gravitational constant

2
Assuming ~ to be constant and neglecting the ddt~2a term we can rewrite
Equation (7.7) as
!
d2 r~0
 2   
d ~r d~r
=
+ 2~ + ~ (~ ~r) (7.21)
dt2 dt2 rot dt rot
fixed
For a particle moving under gravitation (i.e., the attractive force due to New-
tons law of gravitation) and also subjected to a non-gravitational force F~ ,
the equation of motion is
!
d2 r~0
m = m~g + F~ (7.22)
dt2
fixed
Thus from (7.21) and (7.22) we find

d2~r
   
~ 2m~ d~r
m = m~
g + F m~
(~
~r) (7.23)
dt2 rot dt rot
192 Advanced Classical Mechanics

In the laboratory when we measure the acceleration due to gravity what we

determine is actually g~ which is the effective gravitational constant defined
by the difference

g~ = ~g ~ ~r)
~ ( (7.24)
The right-hand side of (7.24) is a combination of gravitational and centrifugal
forces. Thus a ball released near the Earths surface will fall with acceleration
g~ . Because of the negative sign of the vector ~ ( ~ ~r) which points
radially outward, g~ in the Northern hemisphere will point to the south of the

Earths center.

The horizontal and vertical components of g~ are ghor = 2 r sin cos and

gver = g 2 r sin2 . At the pole g = g while on the equator g = g 2 r.
Substituting (7.24) in (7.23) and writing ~g in place of g~ we have the
formula

d2~r d~r
= m~g + F~ 2m~
m (7.25)
dt2 dt
where the suffix rot is dropped. The components of
~ being (0, sin , cos ),
where is the angle between the direction of Earths axis and g~ , the Coriolis
force is given by

d~r
2m~
= 2m(y cos z sin , x cos , x sin ) (7.26)
dt
d~
r
where dt (x, y, z).

7.1.4 Foucaults pendulum

A useful device for observing the effects of the Coriolis force is the so-called
Foucaults pendulum. The latter is a perfectly symmetric setup with the bob
which is hanging from a string of length l can swing freely in any vertical
plane. Because of its symmetric nature, the periods of oscillation of Foucaults
pendulum are equal in all directions. Neglecting the vertical component of the
Coriolis force, which is very small compared to g, the equations of motion in
the x and y directions are
g
x = x + 2 y cos (7.27)
l
g
y = y 2 x cos
l
where we have used (7.25) and (7.26). One can see that the right-hand side
of Equation (7.27) carries the contributions from the Coriolis acceleration in
addition to the usual ones of the simple pendulum motion. In writing down
(7.27) we also assumed that for small amplitude, the motion is nearly hori-
zontal, i.e., z = 0.
 Motion in noninertial coordinate systems 193

The underlying Lagrangian for the planar motion (7.27) is given by the
form
1 mg 2
L= m(x2 + y 2 ) + m cos (xy y x) (x + y 2 ) (7.28)
2 2l
A straightforward way to solve (7.27) is to set r = x + iy that results in
the following complex linear equation:

r + 2ik r + 02 r = 0 (7.29)

where k = cos and 02 = gl .

Seeking a periodic solution r = Aeit where A and are constants leads
to the constraint 2 2k 02 = 0. The two roots are given by the sum and
difference
= k 0 (7.30)
where 02 = 02 +
2
. As a result we have for r(t) the general expression
0 0
r(t) = c1 ei(k+ )t + c2 ei(k )t (7.31)

where c1 and c2 are arbitrary complex constants which can be determined

from the initial conditions of position and velocity.
An interesting particular case of (7.31) corresponds to the choice of the
integration constants c1 = c2 = 12 a, in which the small transverse motion is
omitted, and gives

r(t) = aeikt cos 0 t

i.e. x(t) = a cos kt cos 0 t, y(t) = a sin kt cos 0 t (7.32)

The square of the modulus of r(t) turns out to be |r(t)|2 = a2 cos2 0 t. This
speakspof a simple harmonic motion with amplitude a and frequency 0 '
0 = gl . The representations of (7.32) show that we have a superposition of
slowly varying amplitudes given by the quantities a cos t and a sin t.
During initial times (i.e., near t = 0) (or if being say close to the zero
value near the equator where = 2 ), the oscillation of the pendulum tends to
be simple harmonic entirely in the xdirection, i.e., in the north-south plane.
However, with passage of time (additionally can acquire nonzero values),
the amplitude of the ycoordinate grows reflecting an oscillation that has
picked up an east-west component.
Finally from (7.32) we can determine the angle of inclination of the mo-
y(t)
tion by evaluating the ratio tan x(t) which turns out to be tan t
implying = t. It means that the pendulum is rotating with an angular
velocity which is opposite to the direction of the rotation of the Earth.
More specifically, in the Northern hemisphere where > 0, we have a clock-
wise rotation of the plane of oscillation of the Foucault pendulum while in
the Southern hemisphere it is just the opposite: the rotation of the plane of
oscillation takes place in the counterclockwise direction.
194 Advanced Classical Mechanics

7.2 Nonpotential force

We now consider a situation in which a physical system is acted upon by
0
nonpotential forces denoted by Qj :
0 0
Qj = Qj (qk , qk , t), j, k = 1, 2, ...n (7.33)

in addition to the potential forces. It is clear that the nonpotential forces

depend on generalized velocities for otherwise they will come in the category
of potential forces. Indeed in the presence of (7.33) the generalized forces are
to be expressed as
V 0
Qj = + Qj (7.34)
qj
where V is the potential. As such Lagranges equations of motion take the
form  
d T T V 0
= + Qj , j = 1, 2, ...n (7.35)
dt qj qj qj
We look at the total energy E = T + V where the kinetic energy T is of
the form T = T (qj , qj , t). The time rate of change of T is
n  
dT X T T T
= qj + qj +
dt j=1
qj qj t
n n   
d X T X T d T T
= qj + qj +
dt j=1 qj j=1
qj dt qj t
n n  
d X T X V 0 T
= qj + Qj qj + (7.36)
dt j=1 qj j=1
q j t

where we have used (7.35)

As we learned in Chapter 3 the kinetic energy T can be split up as

T = T0 + T1 + T2 (7.37)

where T2 , T1 are quadratic and linear in generalized velocities, respectively,

while T0 is independent of it. Therefore, by Eulers theorem of homogeneous
functions
n
X T2
qj = 2T2
j=1
qj
n
X T1
qj = T1 (7.38)
j=1
qj
 Motion in noninertial coordinate systems 195

As a result (7.36) takes the form

n  
dT d X V 0 T
= (2T2 + T1 ) + Qj qj +
dt dt j=1
qj t
n
dT d T dV V X 0
= 2 (T1 2T0 ) + + Qj qj (7.39)
dt dt t dt t j=1

Pn V T0
where we have used dV dt =
V
j=1 qj qj + t and also qj = 0. The above
equation implies that the time rate of change of the total energy is
n
dE X 0 d
= Qj qj + (T1 + 2T0 ) (T V ) (7.40)
dt j=1
dt t

For a scleronomic system T1 = T0 = 0 and T t = 0. If the potential energy

n 0
too is not explicitly time-dependent then dE
P
dt = j=1 Qj qj . The latter is called
the power of the nonpotential forces. Nonpotential forces are called gyroscopic
Pn 0
if the power is zero: j=1 Qj qj = 0 and dissipative if the power is negative
Pn 0
j=1 Qj qj < 0. In the former we have systems for which the nonpotential
forces do not consume power while in the latter case the systems consume
power. In dissipative systems, dissipative forces like friction are included, even
through they sometimes do no work. The energy is generally lost through heat,
sound, etc.
On the other hand, for a scleronomic system, the Coriolis force is a gyro-
scopic force. From the form (7.26) it is clear that since the Coriolis force for
the j th particle is
 
~ cor dr~j
Fj = 2mj ~ = 2m(~ v~j ) (7.41)
dt

we therefore have
n
X
F~jcor .v~j = 0 (7.42)
j=1

We conclude that the Coriolis force is a gyroscopic force.

7.3 Summary
This chapter was devoted to the investigation of motion in a noninertial co-
ordinate system. After dealing with the general form of the equation of motion
for rotating frames, the roles of the Coriolis force and effective gravitational
constant were reviewed. We also discussed Foucaults pendulum which is a
196 Advanced Classical Mechanics

device for observing the effects of the Coriolis force and noted the typical im-
plications that follow. Finally we dealt with the situation when a system is
acted upon by nonpotential forces.

Exercises

7.4 Examples
1. Find the deflection of a freely falling body from the vertical caused by
Earths rotation.

Solution: For a particle dropped from rest from a height h above the ground
the motion is described by the equations
1
x = 0, y = 0, z = h gt2
2
where we have only considered gravity and neglected any effect of the
q Coriolis
force. The time that the particle will take to hit the ground is t = 2h
g . The
corresponding velocities are

x = 0, y = 0, z = g

To find the effect of Coriolis force on the equation of motion (7.25) we make
use of these velocities in (7.26) to arrive at the following component equations:

mx = 2mgt sin , my = 0, mz = gt

Using the above value of t we find on integration of mx the result

 3/2
1 2h
x = g sin
3 g
 3 1/2
1 8h
i.e. x = sin
3 g

where we used x = 0, x = 0 at t = 0.
We thus conclude that there will be an easterly deviation from the vertical
 3 1/2
by an amount 13 8hg sin .

2. A bead of mass m slides freely along a smooth circular wire which is

rotating with an angular velocity
~ about its fixed vertical diameter. Derive
 Motion in noninertial coordinate systems 197

A
mg

FIGURE 7.6: Motion of bead sliding along a smooth circular wire rotating
with angular velocity .

and discuss the energy conservation equation. Take a to be the radius of the
wire and set = ng
p
a where n is a parameter.
,

Solution: As shown in Figure 7.6 we have

~ = ez = ( cos er + sin e )
 
The centrifugal force has the magnitude = m 2 N~P  = m 2 a sin and acts
 
 
horizontally outward along N~P  . Obviously this force has a relevance in
 
rotating frames only.
From the second relation of (7.20), the cross-radial equation of motion
reads
1 d 2
m. (a ) = mg sin + (m 2 a sin ) cos
a dt
or, a = g sin + ng sin cos
where we have put a 2 = ng. Integration gives
1 mga
ma2 2 + (1 n cos )2 = constant
2 2n
This is essentially the energy conservation equation in a rotating frame with
1 2 2
2 ma representing the kinetic energy and the potential energy is given by

mga
V (n, ) = (1 n cos )2
2n
198 Advanced Classical Mechanics

Differentiations yield
dV
= mga sin (1 n cos )
d
d2 V
= mga[cos (1 n cos ) + n sin2 ]
d2
Therefore dV 1
d = 0 at = 0 (namely, the lowest point A) and also at cos = n
for n > 1. Now a minimum of V corresponds to the position of relative stable
equilibrium and a maximum of V corresponds to the position of relative un-
stable equilibrium.

Case 1: =0

Here
d2 V
= mga(1 n) > 0 if n < 1
d2
= 0 for n < 1 is a minimum for V

Hence = 0 is a position of relative stable equilibrium if n < 1. If however

n > 1 then = 0 is a maximum for V and so = 0 is a position of relative
unstable equilibrium.
Case 2: cos = n1 (n > 1)

We have
d2 V
 
1
=n 1 2 >0
d2 n
Therefore cos = n1 is a position of relative stable equilibrium. We conclude
that new stable solutions are created at cos = n1 as n exceeds beyond the
critical value of n = 1. Such a phenomenon is an example of bifurcation.
Chapter 8
Symmetries and conserved quantities

8.1 Condition of invariance and Noethers theorem . . . . . . . . . . . . . . . . . 200

8.2 Operator approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
8.2.1 Symmetry operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
8.2.2 Parity transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
8.2.3 Time-reversal symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
8.3 Virial theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
8.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210

Symmetries can be characterized1 in many ways depending on the problem

context. They can be continuous (translations, rotations or a boost are some of
the examples) which can be obtained by successive applications of infinitesimal
transformations or discrete (like space reflection or time reversal for both of
which Newtonian mechanics is invariant). They can be global (depending on
parameters which are constant) or local (having functional dependence upon
space and time). Symmetries can also be distinguished as those of geometrical
or dynamical types. Examples of a geometrical symmetry are homogeneity
and isotropy which arise due to the inner structure of spacetime. On the other
hand, we can think of symmetries as being produced by the form invariance
of the underlying equations. Dynamical symmetries are associated with such
invariances.
Let us look at the following example of the gravitational equation of mo-
tion:
X Gmi (r~j r~i )
r~j = , i = 1, 2, ..., n (8.1)
i
|r~j r~i |3
where mi and ~ri are the mass and position of the i-th particle while G is
Newtons constant. Consider the Galilean transformation
r~k r~k0 = r~k V
~t (8.2)
where V~ is a constant velocity. We easily see that under such a transformation
the form of the above equation remains unchanged:

~ X Gmi (r~j0 r~i0 )

rj0 = , i = 1, 2, ..., n (8.3)
i|r~0 r~0 |3
j i

1 K. Brading and E. Castellani, Symmetry in classical physics, Handbook of the Philoso-

phy of Physics, North-Holland, pp. 1331-1367, J. Butterfield and J. Earman, Eds.

199
200 Advanced Classical Mechanics

This is a typical case of a dynamical symmetry. In the following we will first

investigate the situation when a system exhibits continuous symmetry and
proceed to show that it can then have an associated conserved quantity or
charge. Discrete symmetries will be considered later.

8.1 Condition of invariance and Noethers theorem

In classical mechanics, Noethers theorem occupies a prominent position
because according to this theorem if a symmetry is found to exist in a dy-
namical problem then there is a corresponding constant of motion. It pro-
vides a connection between global continuous symmetry and the resulting
conservation law. The theorem for such an assertation was put forward by
Emma Noether in the paper Invariante Variationsprobleme which came out in
1918.
To have an idea of how the proposition goes, we will be concerned with
the invariance of a Lagrangian L(qi , qi , t), i = 1, 2, ..., n (where qi and qi are,
respectively, the generalized coordinates and velocities of a point particle in a
system with n degrees of freedom and t is the time variable) under a reversible
variable spacetime transformation of the type

qi0 = qi0 (q1 , q2 , ..., qn ; t), t0 = t0 (t), i = 1, 2, ..., n (8.4)

Let us suppose that the primed coordinates and their velocities generate a
dq 0
Lagrangian L0 (qi0 , qi0 , t0 ), i = 1, 2, ..., n, where qi0 dt0i with respect to the new
time variable t0 :

L(qi , qi , t) L(qi0 , qi0 , t0 ), i = 1, 2, ..., n (8.5)

We deal with those classes of suitable symmetry transformations that leave

the form of the equations of motions invariant in the new primed system. Such
transformations are referred to as the invariance transformations.
We already encountered in Chapter 3 the form invariance of equations of
motion when, for instance, L0 differs from L by a total time derivative term
d
dt , where is an arbitrary function of coordinates and time but indepen-
dent of velocities. Expressed in terms of primed coordinates we noticed that
corresponding to a relation

d(qi0 , t0 )
L0 (qi0 , qi0 , t0 ) = L(qi0 , qi0 , t0 ) + , i = 1, 2, ..., n (8.6)
dt0
 Symmetries and conserved quantities 201
d
the term dt0 gives a vanishing contribution to Lagranges equations of motion

because of the following set of relations:

d(q 0 , t) d(q 0 , t)
    
d
=
dt q 0 dt q 0 dt
  
d 0 d
or q + = 0
dt q 0 q 0 t q dt
 
d d
or = 0
dt q 0 q dt
Let us look at the action integrals S and S 0 induced, respectively, by the
Lagrangians L and L0 :
Z Z
0
S = L(qi , qi , )d, S = L0 (qi0 , qi0 , 0 )d 0 , i = 1, 2, ..., n (8.7)

Invariance of the actions implies S = S 0 and so we write

Z Z
S S 0 S = L0 (qi0 , qi0 , 0 )d 0 L(qi , qi , )d = 0 i = 1, 2, ..., n (8.8)

Using now the form of L0 as given by (8.6) we have

d(qi0 , 0 ) 0
Z Z Z
L(qi0 , qi0 , 0 )d 0 L(qi , qi , )d + d = 0 i = 1, 2, ..., n (8.9)
d 0
Seeking invariance of the action under small infinitesimal perturbations
where the coordinates qi , the velocities qi and transform according to

qi0 = qi + qi , qi0 = qi + qi , 0 = + i = 1, 2, ..., n (8.10)

(8.9) gives on substituting for the primed quantities

Z   Z
d
L(qi + qi , qi + qi , + ) 1 + d L(qi , qi , )d
d
Z
d(qi + qi , + )
+ d = 0 (8.11)
d
where i = 1, 2, ..., n and in the last term of the left-hand side only the
lowest-order infinitesimal is kept. We have also replaced by to indicate
that the Lagrangian changes by a total derivative term of infinitesimal order.
Further, at the same level, Taylor expansion leads to
Z  
L L L d d(qi , )
qi + qi + + L+ d = 0 (8.12)
qi qi d d
The validity of (8.12) is ensured if the following holds:
 
L L L d d(qi , )
qi + qi + + L d = d (8.13)
qi qi d d
202 Advanced Classical Mechanics

Observe that the left-hand side is required to vanish if we insist upon

the form-invariance of the Lagrangian. For instance, consider a pure space
N
L
P
translation by a constant amount ki . Then this implies ki q i
= 0 implying
i=1
N
P
pi = 0 which stands for the vanishing of the total momentum acting on
i=1
an N -particle system. On the other hand, for a constant time translation one
finds L
= 0 pointing to the time-independence of the Lagrangian. Thus for
the specific cases of a pure space translation and a constant time displacement
we have constancy of linear momentum and Hamiltonian, respectively.
The constancy of the total angular momentum following from the rota-
tional invariance of the Lagrangian can be readily derived for the general
setting of a physical system of N particles described by the Lagrangian
N N
X 1 1X
L= mi |~ri |2 Vij (|~ri ~rj |) (8.14)
i=1
2 2 i=1

Here L represents a closed system of N interacting particles under the influ-

ence of a potential Vij which operates between the i-th and j-th particle.
Consider an infinitesimal rotation by a quantity about the z-axis (note
that in the following we do not distinguish between and variations as t is
not explicitly involved in L). Then x and y coordinates change according to

x0 = x cos y sin x y
0
y = y cos + x sin y + x
(8.15)

Hence for an infinitesimal rotation about the z-axis we have

N   X N
X L L
(yi ) + (xi ) = [mi (yi xi + xi yi )] = lz (8.16)
i=1
xi yi i=1

where lz stands for the z-th component of the angular momentum. Infinites-
imal rotations about the x- and y-axis can be similarly written down. When
we consider rotations in all the three directions we have r~0 ~r = ~r = ~ ~r
implying
N N
X L X
.~ri = mi~ri . ~ ~ri = ~l. ~ (8.17)
~ri
i=1 i=1
PN
where ~l = i=1 ~ri mi~ri is the total angular momentum of the system and
represents a conserved Noether quantity.
d
More generally, we make use of the non-commutativity of d and oper-
ators through a relation already written down in (6.62), namely,
d d
(q) = q + q (t) (8.18)
dt dt
 Symmetries and conserved quantities 203

and adopt a similar strategy for derivation as we did to get (6.52) to arrive at
the constancy condition
 
L L
qi + L qi t + = constant (8.19)
qi qi
The above is the central result of Noethers theorem which states that asso-
ciated with an infinitesimal invariance transformation there is a constant of
motion and hence an associated conservation law exists.
With L
qi = pi , the generalized momentum, (8.19) can also be interpreted
in terms of the Hamiltonian H of the system for a conservative, scleronomic
system having its kinetic energy a homogeneous function of the velocities. As
we saw earlier, in such a case the quantity qi L
qi L is a constant and equals
to H. We thus have the result

pi qi Ht + = constant (8.20)

A few points are in order regarding the applicability of Noethers theorem.

First, it is relevant to systems enjoying a continuous class of symmetry (such
as rotations as considered above) rather than a discrete one like parity (which
seeks invariance under space reflection that reverses the sign of the position
and velocity of the particle). This means that even if dissipative systems ex-
hibit continuous symmetries there may not be any corresponding conservation
law. Second, it produces conservation laws as we saw it ought to because of the
underlying relation (8.19). Third, it insists upon the existence of a Lagrangian
for (8.19) to be a valid constraint.

Example 8.1:

Show that angular momentum is conserved for a Lagrangian given by

L = (~r )2 + ~r.~r + (~r)2

Under a rotation we have

~r ~ ~r, ~r ~ ~r

We see that the variation of the Lagrangian vanishes

L L
L = .~r + .~r
~r
~
r
       
= 2~r + ~r . ~ ~r + ~r + 2~r . ~ ~r
= 0

because of the vector identity ~a.(~b ~c) = ~b.(~c ~a).

204 Advanced Classical Mechanics

Example 8.2:

Consider the free particle problem described by the Lagrangian

1
mx2 L=
2
How does L transform under x x + (t)?
Since here
x0 = x + (t), x0 = x +
the Lagrangian in primed coordinate reads
1 1
L0 (x0 , x0 , t) =
m(x0 )2 = L(x0 , x0 , t) mx0 + m2
2 2
For an invariant transformation there exists some function (x0 , t) such
that
d
L0 (x0 , x0 , t) = L(x0 , x0 , t) +
dt
Writing
d 0
= x +
dt x0 t
and comparing the above two forms of L (x , x0 , t) we find
0 0

1
= m, = m2
x0 t 2
Consistency of the relations yields = 0. It implies (t) to be linear in time,
i.e.,

(t) = c1 + c2 t
1
(x, t) = mc2 x0 + mc22 t
2
where c1 and c2 are arbitrary constants. Thus x0 is of the form x0 = x +
c1 + c2 t which is a combination of spatial displacement (c2 = 0) and Galilean
transformation (c2 6= 0). It also follows that a transformation to a uniformly
accelerating frame does not constitute an invariant transformation.
 Symmetries and conserved quantities 205

8.2 Operator approach

In this section we address the formulation of dynamical symmetries in
terms of operators2 that act on the state space for the system. A classical
dynamical system is specified by a set of dynamical variables

S = f (r1 , r2 , ..., rn ) (8.21)

Note that the time t is not considered explicitly.
The set S evolves in time according to the equations of motion

rk = k (r1 , r2 , ..., rn ; t), k = 0, 1, 2, ..., n (8.22)

where k s are, as yet, some unspecified functions of coordinates and time. As
an example, if n=2, we can identify r1 = z as the particle position and r2 = v
as its velocity:

z = v, mv = F (z, v, t) (8.23)
where F (z, v, t) is some force function which depends on the coordinate, ve-
locity and time in a very general setting. For an autonomous system we have
the reduced form

rk = k (r1 , r2 , ..., rn ), k = 0, 1, 2, ..., n (8.24)

We now inquire into the time development of S from instants t = ti to
t = tf . Introducing the corresponding evolution operator U (tf , ti ), such an
evolution can be expressed as

U (tf , ti )[r1 (ti ), ..., rn (ti )] = [r1 (tf ), ..., rn (tf )] (8.25)
Should the equations of motion not depend upon time t explicitly, it is
evident that U (tf , ti ) would depend only on the difference tf ti and we
denote in such a case

U = U ( ) = U (, 0), = tf ti (8.26)
We are then led to an invariance under time translation:

U (tf + , ti + ) = U (tf , ti ) (8.27)

for a constant time interval .
Some simple expressions of the evolution operator resulting from an ap-
propriate choice of the force function are given by the following examples:

2 A.D. Boozer, Dynamical symmetries in classical mechanics, Eur. J. Phys. 33(2012) 73.
206 Advanced Classical Mechanics

(1). The case of the free particle for which the force function F (z; v) = 0.
Here U ( ) is given by

U ( )(z, v) = (z + v, v)

(2). The case of simple harmonic motion described by F (z; v) = m 2 z

for which U ( ) is given by
v
U ( )(z, v) = (z cos + sin , v cos z sin )

(3). The case of damped particle under the influence of F (z; v) = mv

where is a damping constant. Here U ( ) is given by
v
U ( )(z, v) = z + [(1 e ), ve ]

(4). The case of a uniformly accelerated Newtonian particle with an ac-

celeration a subjected to the force function F (z; v) = ma. Here U ( ) is given
by

a 2
U ( )(z, v) = (z + v + , v + a )
2

8.2.1 Symmetry operator

We can define for every symmetry transformation a corresponding time-
dependent coordinate transformation operator (t) and consider the evolution
of the set of dynamical variables (r1 , r2 , ..., rn ) according to some given equa-
tions of motion. The form invariance of the latter requires that

rk0 = k (r10 , r20 , ..., rn0 ; t), k = 0, 1, 2, ..., n (8.28)

Transforming to new dynamical variables (r10 , ..., rn0 ) under (t) we can
write

(t)(r1 , r2 ..., rn ) = (r10 , r20 , ..., rn0 ) (8.29)

Assuming an inverse of (t) to exist, it follows that

1 (t)(r10 , ..., rn0 ) = (r1 , ..., rn ) (8.30)

 Symmetries and conserved quantities 207

For the new set of dynamical variables (r10 , r20 , ..., rn0 ) let us define a corre-
sponding evolution operator U 0 (tf , ti ) such that

U 0 (tf , ti )[r10 (ti ), ..., rn0 (ti )] = [r10 (tf ), ..., rn0 (tf )] (8.31)
We therefore deduce

U 0 (tf , ti )(ti )[r1 (ti ), ..., rn (ti )] = (tf )[r1 (tf ), ..., rn (tf )]
= (tf )U (tf , ti )[r1 (ti ), ..., rn (ti )] (8.32)

where for the second equality we have used (8.25). Consistency requires

U 0 (tf , ti )(ti ) = (tf )U (tf , ti ) (8.33)

yielding the following connection between U 0 (tf , ti ) and U (tf , ti ):

U 0 (tf , ti ) = (tf )U (tf , ti ) 1 (ti ) (8.34)

Form invariance of the equations of motion requires that the evolution

operator for the old variables and the transformed ones be alike: U (tf , ti ) =
U 0 (tf , ti ) resulting in the invariance result

U (tf , ti ) = (tf )U (tf , ti ) 1 (ti ) (8.35)

For the time independent case when (t) = , we get the simple commutative
condition
U (tf , ti ) = U (tf , ti ) (8.36)
The operator method is a powerful approach for providing connections to
discrete symmetries. Let us try for the operations of parity and time-reversal.

8.2.2 Parity transformation

The parity operator P is associated with reflection in space. It reverses the
sign of the position and velocity of the particle:

P : z z, v v (8.37)

It is a discrete symmetry and for the invariance of a system under such an

operation the commutative condition holds as given below:

U (tf , ti )P = P U (tf , ti ) (8.38)

A free particle is invariant under parity transformation as can be directly

verified:
P U ( )[z, v] = P [z + v, v] = [z v, v] (8.39)

U ( )P [z, v] = P [z, v] = [z v, v] (8.40)

208 Advanced Classical Mechanics

However in the case of a Newtonian particle which is uniformly accelerated,

the symmetry under parity transformation is not preserved:

a 2 a 2
P U ( )[z, v] = P [(z + v + , v + a )] = [z v , v a ] (8.41)
2 2
but

a 2
U ( )P [z, v] = U [z, v] = [(z v + , v + a )] (8.42)
2

8.2.3 Time-reversal symmetry

The time-reversal operator reverses the sign of the time variable t according
to

T : t t (8.43)
Like the parity, the time-reversal transformation is discontinuous too and so
not amenable to the treatment of Section 8.1. Note that the discontinuous
character means that the underlying transformation cannot be developed from
a series of infinitesimal transformations as is true for the continuous case.
If the system undergoes a time-reversal change, the transformed equations
of motion will be given by a similar set as in (8.24):

rk0 = k (r10 , r20 , ..., rn0 ), k = 0, 1, 2, ..., n (8.44)

where because of no explicit time dependence k s remain unchanged.
At this point if we integrate (8.24) between an initial time ti and a final
time tf , then an operator can be defined depending on , which is the difference
tf ti , evolving the dynamical variables r1 , r2 , ..., rn . On the other hand, if we
integrate (8.44), we can think of another operator which evolves the variables
r10 , r20 , ..., rn0 . We therefore have the connection U 0 ( ) = U ( ).
If the link between the two sets (r10 , r20 , ..., rn0 ) and (r1 , r2 , ..., rn ) is provided
by the application of an operator T on the latter, i.e.,

T (r1 , r2 , ..., rn ) = (r10 , r20 , ..., rn0 ) (8.45)

then

U 0 ( ) = T U ( )T 1 (8.46)
We therefore arrive at the result

T U ( ) = U ( )T (8.47)
Time reversal implies that the role of time is reversed, in other words, there is
a reversal of motion. In Newtonian mechanics in which the second law is the
 Symmetries and conserved quantities 209

same with regard to the explicit replacement t t, time reversal means that
the particle retraces its path and ultimately arrives at the point it started from
with a reversed momentum. Note that time-reversal invariance is not obeyed
in our previously considered example of a damped particle.

8.3 Virial theorem

Virial theorem gives a relation for the average over time of the total kinetic
energy. The result can be derived as follows.
Consider an N -particle system as rest for which we define a quantity G
N
X
G= p~i .~ri (8.48)
i=1

where ~ri denotes the ith position of a particle with respect to a fixed origin.
Clearly G is restricted to be finite since a material particle has neither an
infinite momentum nor can it be infinitely away from the origin.
Translation of the origin by a finite amount ~r0 leaves G invariant:

N
X
0
G = p~i . (~ri ~r0 )
i=1
N
X
= G ~r0 p~i
i=1
= G

from the conservation of linear momentum.

The time rate of change of G is given by

N 
X 
G = p~i .~ri + p~i~ri
i=1
N
X
= 2T + F~i .~ri
i=1
= 2T + W

where T Pis the kinetic energy and W is called the virial of the system. Like
N
G, W = i=1 F~i .~ri too is independent of the choice of origin for the system
at rest.
210 Advanced Classical Mechanics

Let us go for the time average of 2T + W . We define

1
Z
=< 2T + W > = lim Gdt
0

G( ) G(0)
= lim

= 0

since G( ) G(0) is always finite. Hence

1
T = W (8.49)
2
This is the virial theorem.

Consider a two-particle system for which

W = f~12 .~r2 + f~21 .~r1 = f~12 . (~r2 ~r1 ) (8.50)

where f~ij , i, j = 1, 2 are the inter-particle forces and we have used Newtons

third law. If V12 = |~r2 ~ r1 | is the inter-particle potential ( a constant) then
f~12 is
(~r2 ~r1 )
f~12 = (8.51)
|~r2 ~r1 |3
It shows

f~12 .(~r2 ~r1 ) = = V12 (8.52)
|~r2 ~r1 |
As such W represents the total potential energy V (= V12 ) of the system
and implies
1 1
T = W = V (8.53)
2 2
From the conservation of total energy we can write the above result as
1
E = T = V (8.54)
2
Virial theorem is of great importance in many branches of physics especially
in statistical mechanics. It does not depend on the notion of the temperature
and holds even for systems that are not in thermal equilibrium.

8.4 Summary
The purpose of this chapter was to investigate the role of symmetry, essen-
tially the dynamical ones, in classical mechanics. We gave a simple derivation
of the condition of invariance from which we extracted the essence of Noethers
 Symmetries and conserved quantities 211

theorem which holds for a continuous class of symmetries. After working out
a number of examples we discussed the problem of dynamical symmetry fol-
lowing an operator approach in which the notion of an evolution operator was
employed. The operator method allows one to address discrete symmetries
such as the parity and time-reversal operations. We also considered briefly
the virial theorem.

Exercises
1. Consider a ray of light traveling from the point P0 to the point P1
through the point of reflection at the point (x, 0) on a mirror M. Calculate
the time taken and hence derive the law of reflection.
2. Consider two mediums S0 and S1 characterized, respectively, by the
index of refraction n0 and index of refraction n1 . Calculate the time taken for
the light traveling from the point P0 to the point P1 through the point (x, 0).
Hence, derive Snells law of refraction.
3. Show that the shortest distance between two points in a plane is a
straight line.
4. Show that the shortest distance between two points on the surface of a
sphere is a great circle.
5. Find the equation of the curve which makes the surface area of revolution
generated by rotating the curve y = y(x) around the x-axis.
Chapter 9
HamiltonJacobi equation and
action-angle variables

9.1 Canonical transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215

9.2 Symplectic property . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
9.3 Idea of a generating function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
9.4 Types of time-dependent canonical transformations . . . . . . . . . . . . 223
9.4.1 Type I canonical transformation . . . . . . . . . . . . . . . . . . . . . . . . 223
9.4.2 Type II canonical transformation . . . . . . . . . . . . . . . . . . . . . . . 224
9.4.3 Type III canonical transformation . . . . . . . . . . . . . . . . . . . . . . 225
9.4.4 Type IV canonical transformation . . . . . . . . . . . . . . . . . . . . . . 225
9.5 Infinitesimal canonical transformations . . . . . . . . . . . . . . . . . . . . . . . . . . 228
9.6 HamiltonJacobi equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
9.6.1 Time independent HamiltonJacobi equation:
Hamiltons characteristic function . . . . . . . . . . . . . . . . . . . . . . . 232
9.6.2 Other variants of HamiltonJacobi equation . . . . . . . . . . . . 235
9.7 Action-angle variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
9.7.1 Motion of a particle in a 2-dimensional rectangular well 242
9.8 Possible trajectories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
9.8.1 Periodic trajectories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
9.8.1.1 Some explicit examples for periodic
trajectories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
9.8.2 Open trajectories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
9.8.3 Special trajectories when the billiard ball hits a corner 247
9.9 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248

The analytical structure of classical dynamics describing the motion of a par-

ticle or a system of particles rests on two important but almost equivalent
principles, namely, those of Lagrangian and Hamiltonian mechanics. Let us
recall that for, say, a time-independent mechanical system of particles with
n degrees of freedom, the Lagrangian is a function of the generalized coor-
dinates qi , i = 1, 2, ..., n and generalized velocities qi , i = 1, 2, ..., n and ap-
pears as the difference of kinetic and potential energies of the particles. To
get a knowledge of the dynamics of the motion requires solving 2n second-
order differential equations subject to a set of appropriate initial conditions.
Although the formulation of Lagrangian mechanics does not yield any addi-
tional information as compared to what is generally known from Newtonian

213
214 Advanced Classical Mechanics

mechanics, the main advantage with it is that one does not have to worry with
the presence of constraint forces since the latter are eliminated by invoking
DAlemberts principle. If the system is not conservative, dissipative or driven
forces are taken into account by the non-potential terms. One important as-
pect which makes a Lagrangian system very useful is the presence of conserved
quantities whenever one or more generalized coordinates are absent from the
Lagrangian. On the other hand, the Hamiltonian, which can be derived from
the Lagrangian by effecting a Legendre transformation, is regarded as a func-
tion of the generalized coordinates qi , i = 1, 2, ..., n and generalized momenta
pi , i = 1, 2, ..., n, and appears, in a conservative system, as the sum of kinetic
and potential energies. The underlying equations of motion are given by a set
of n first order partial differential equations each for the time derivative of the
generalized coordinates and momenta. The Hamiltonian procedure is carried
out in a phase space which is a 2n-dimensional space made up of positions and
the accompanying canonical momenta and the Hamiltonian equations give the
clue as to how a system evolves in phase space. Both Lagrangian and Hamil-
tonian formalisms address the laws of mechanics without any preference to
the selection of any particular coordinate system.
The Hamiltonian procedure, however, does not always offer any simplifi-
cation to the concerned dynamical problem. But the almost symmetrical ap-
pearence of the coordinates and momenta in the guiding Hamiltons equations
facilitates development of formal theories such as the canonical transforma-
tions, HamiltonJacobi equation and action-angle variables. Our first purpose
in this chapter will be to understand how canonical transformations are con-
structed that offer considerable simplifications in writing down the equations
of motion. The task is to transform the Hamiltonian to a new form through the
introduction of a new set of coordinates and momenta such that with respect
to the new set of variables the transformed Hamiltonian also leads to the form
of canonical equations. The utility of canonical transformation lies in the fact
that it is often possible to adopt new sets of conjugate variables through which
the basic equations get much simplified thus generating solutions that may
otherwise be very complicated to determine. The special case when all the
coordinates are cyclic yields a single partial differential equation of first order
called the HamiltonJacobi equation. In this chapter we also deal briefly with
the elements of a certain class of canonical variables called the action-angle
variables defined over a bounded phase space that emerge from a complete
separability of the HamiltonJacobi equation. Some interesting properties of
such variables are explored.
 HamiltonJacobi equation and action-angle variables 215

9.1 Canonical transformation

For simplicity we will be frequently restricting to a system of one degree
of freedom described by a single pair of canonical variables (q, p). The phase
space is then of dimension two. The Hamiltonian equations are given by
H H
q = , p = (9.1)
p q
We ask the question whether a transformation from (q, p) to a new set of
variables (Q, P ) is feasible in a way that the above form of the Hamiltonian
equations is preserved
K K
Q = , P = (9.2)
P Q
We emphasize that our intention in looking for new variables (Q, P ) is to
inquire if these could also serve in a new but simplified means of describing
the system but at the same time not disturb the essential physics content. In
(9.2) K is a transformed version of H

H(q(Q, P ), p(Q, P )) K(Q, P ) K (9.3)

defined in terms of the variable Q and P . It is obvious that the equations
(9.2) may not always hold except for some special situations. Such restricted
transformations for which (9.2) are valid are called canonical transformations
implying that the new variables (Q, P ) too form a canonical set.
Noting that the Poisson bracket of any two functions F and G, defined in
terms of the old canonical variables q and p, is given by
F G F G
{F, G}(q,p) =
q p p q
Hamiltonian canonical equations for Q = Q(q, p), P = P (q, p) read

Q H Q H
Q = {Q, H}(q,p) =
q p p q
P H P H
P = {P, H}(q,p) = (9.4)
q p p q

Interpreting H(q, p) as K(Q, P ) in terms of the new variables (Q, P ) and

employing the chain rule of partial derivatives gives
H K Q K P
= +
q Q q P q
H K Q K P
= + (9.5)
p Q p P p
216 Advanced Classical Mechanics

This results in the following expressions for Q and P :

   
Q K Q K P Q K Q K P
Q = + +
q Q p P p p Q q P q
 
K Q P Q P
=
P q p p q
K
= {Q, P }(q,p)
P
K
P = {Q, P }(q,p) (9.6)
Q

These equations coincide with (9.2) provided we set {Q, P }(q,p) equal to unity:

{Q, P }(q,p) = 1 (9.7)

The Hamiltonian form of canonical equations are then valid for the new pair
of variables (Q, P ).
For a system of n degrees of freedom the condition for the canonical trans-
formation is given by the Poisson bracket conditions

{Qi , Qj }(q,p) = 0, {Pi , Pj }(q,p) = 0, {Qi , Pj }(q,p) = ij (9.8)

Invariance of Poisson bracket relations is a fundamental feature of canonical
transformations. In fact the above properties of the Poisson brackets serve as
the necessary and sufficient conditions for a transformation to be canonical.
Note that out of 2n variables, (qi , pi ) and (Qi , Pi ), i = 1, 2, ..., n, only 2n of
these are independent.
It needs to be pointed out that the Poisson bracket {Q, P }(q,p) is the same
as the Jacobian determinant:
Q P Q P
{Q, P }(q,p) =
q p p q
(Q, P )
= (9.9)
(q, p)

Conversely we also have

 1
(q, p) (Q, P )  1
{q, p}(Q,P ) = = = {Q, P }(q,p) (9.10)
(Q, P ) (q, p)

Consider a region D in the phase space plane (q, p) that is bounded by a

closed curve A. Using (9.9) we can express

Z Z Z Z
(Q, P )
dQdP = dqdp = [Q, P ]q,p dqdp = dqdp (9.11)
D D (q, p) D D
 HamiltonJacobi equation and action-angle variables 217

where we have exploited (9.7). In terms of closed line integrals we then have
by Stokess theorem
I I
pdq = P dQ (9.12)
A A
In general, for a system with n generalized coordinates qi together with
their associated momenta pi , the volume element transforms as
n
Y n
Y n
Y
dQi dPi = dqi dpi = dqi dpi (9.13)
i=1 i=1 i=1

where stands for the determinant of the Jacobian matrix

(Q1 , Q2 , ..., Qn ; P1 , P2 , ..., Pn )

= det[ ] (9.14)
(q1 , q2 , ..., qn ; p1 , p2 , ..., pn )
and we employed Liouvilles theorem (see Chapter 4, Section 4.7). Thus phase
space volume is preserved by canonical transformations.

Example 9.1:

Consider the transformation (q, p) (Q, P ) given by

r
2P
q= sin Q, p = 2mP cos Q
m
Inverting

1 p2
   
q
1 2 2
Q = tan m , P = + m q
p 2 m
Q mp Q mq
= 2 , = 2
q p + m2 2 q 2 p p + m2 2 q 2

As a result

Q P Q P
{Q, P }(q,p) =
q p p q
 
1 P P
= mp + mq
p2 + m2 2 q 2 p q
p
Since P
p = m ,
P
q = mq it follows that {Q, P }(q,p) = 1. Hence the trans-
formation is canonical.
218 Advanced Classical Mechanics

Example 9.2:

Let us focus on the specific case of the harmonic oscillator described by

1 2
the Hamiltonian H = 2m p + 12 m 2 q 2 . It is clear from the form of P in the
previous example that if employed the Hamiltonian takes a very simple form
H K = P , where K is the transformed Hamiltonian. The accompanying
Hamiltons equations for the new variables Q and P are easily read off
K
Q = =
P
K
P = =0
Q
Solving for Q and P we find Q = t + t0 , P = b where t0 and b are
constants of integration. Switching to the original variables we get for q and p
r
2b
q= sin (t + t0 )
m

p = 2mb cos (t + t0 )

which conform to their standard forms.

This example serves to illustrate the advantage of employing canonical
variables in a suitable manner as a result of which the basic equations get
simplified and obtaining the solutions becomes an easy task.

9.2 Symplectic property

Let us point out that the canonical transformations obey the symplectic
property. Indeed we observe that corresponding to the matrix of derivatives
of the canonical variables (Q, P ) namely
 
Q/q Q/p
J = (9.15)
P /q P /p

and in terms of an off-diagonal antisymmetric matrix A defined by

 
0 1
A = (9.16)
1 0

the relation A = JAJ T , J T being the transpose of J, is always valid. Indeed

a simple multiplication gives
 
{Q, Q}(q,p) {Q, P }{(q,p)
JAJ T = (9.17)
{Q, P }(q,p) {P, P }(q,p)
 HamiltonJacobi equation and action-angle variables 219

where the entries stand for the Poisson brackets. Using their properties as
furnished in (9.8) our assertion is seen to hold. We refer to J as a symplectic
matrix and the canonical transformations induced by (Q, P ) are said to be
symplectic.
In a general sense, Hamiltons equations for a 2n-dimension phase space,
defined by the generalized coordinates and generalized momenta which we
represent by a column vector X (q1 , q2 , ..., qn ; p1 , p2 , ..., pn )T (q, p)T , can
be written as
   
H d q q H(q, p, t)
X = A = AX H(X) V =A (9.18)
Xi dt p p H(q, p, t)

where A is a 2n 2n symplectic matrix given by

 
0 +In
A = (9.19)
In 0

with In representing the (n n) block identity matrix. The properties of A

are given by (A)1 = A = AT and A2 = 1.
Consider a transformation of coordinates on the phase space

X Yi (X) (9.20)
Yi
The Jacobian of such a transformation is Xj . Let us call it ij . The time
rate of X is then given by Y = X. Noting that
H H Yj
= = T (Y H) (9.21)
Xi Yj Xi
we get for X the equation
H
Y = X = A = AT (Y H) (9.22)
Xi
where we used (9.18).
Formally, a 2n 2n matrix A is said to be symplectic if it satisfies the
relation

AT = A (9.23)
We then see from (9.22) that the Hamiltonian equations are preserved. When
the Jacobian satisfies (9.23) in relation to the antisymmetric block matrix A
we say that the transformations are symplectic or canonical. The transformed
coordinates, occupying the top and bottom halves of the column matrix Y , in
view of the mapping (9.20), can be, respectively, identified as the new gener-
alized coordinates Q1 , Q2 , ..., Qn and generalized momenta P1 , P2 , ..., Pn .
220 Advanced Classical Mechanics

Example 9.3:

Determine the constants , , and  such that

Q1 = q1 + q2 , Q2 = q1 + q2 , P1 = 2p1 + p2 , P2 = p2
is a canonical transformation.
We can carry out the symplectic test to determine the constants , ,
and . To this end we first write down the relevant (4 4) Jacobian matrix
which reads

Q1 Q1 Q1 Q1

q1 q2 p1 p2
Q2 Q2 Q2 Q2
q1 q2 p1 p2
=

P1 P1 P1 P1
q1 q2 p1 p2
P2 P2 P2 P2
q1 q2 p1 p2

For the given relations it takes the form

0 0
 0 0
=
0

0 2 1
0 0 0 1

Now A being

0 0 1 0
0 0 0 1
A =
1

0 0 0
0 1 0 0

we find on multiplying the matrix AA to be given by

0 0 2 +
0 0 2 + d 
AA =
2 2 
(9.24)
0 0
 0 0

For the values = = 12 , = 0 and  = 1 we see that AT = A

implying that the transformation is canonical. The result is confirmed by
the Poisson bracket test which shows that {Q1 , Q1 } = {Q2 , Q2 } = 0,
{P1 , P1 } = {P2 , P2 } = 0 and {Q1 , P1 } = {Q2 , P2 } = 1. The transformed
coordinates Q1 and Q2 are Q1 = 12 q1 , Q2 = 12 q1 + q2 .
 HamiltonJacobi equation and action-angle variables 221

Example 9.4:

Show that canonical transformations form a group.

Apart from the identity transformation which is evident, canonical trans-

formations satisfy the product or composition law in that if X Y is a
canonical transformation due to the Jacobian of the transformation satis-
fying AT = A and Y Z is another canonical transformation due to the
Jacobian of the transformation satisfying AT = A, the product transfor-
mation taking X Z with the Jacobian given by is also canonical
because satisfies AT = AT T = AT = A. Moreover, if X Y
is a canonical transformation then the inverse transformation Y X is
also canonical. For this the determinant of the Jacobian is required to be
nonvanishing so that we can carry out the process of inversion. The inverse
transformation exists because of det = 1. Finally, the associative property is
valid which follows by taking three successive transformations characterized
by the Jacobians , and mapping successively X Y , Y Z and
Z T . The three transformations may be combined to be interpreted as a
single transformation taking X T by identifying it as a two-step process
either by X Z through followed by Z T through or by X Y
through followed by Y T through . The reason is due to the associa-
tive character of a matrix multiplication: () = (). Thus, all the four
properties of a group hold and hence canonical transformations form a group.

9.3 Idea of a generating function

That the quantity (pdq P dQ) is an exact differential can be established
by noting
 
Q Q
pdq P dQ = pdq P dq + dp
q p
 
Q Q
= pP dq P dp (9.25)
q p
from which the condition for an exact differential, namely,
   
Q Q
pP = P (9.26)
p q q p
is seen to hold because of {Q, P }(q,p) = 1.
Setting pdq P dQ = dG1 , we call G1 to be the generating function of the
transformation (q, p) (Q, P ). Such a class of generating functions is referred
222 Advanced Classical Mechanics

to as the Type I generating function. G1 being a function of q and Q [i.e.,

G1 = G1 (q, Q)] implies
G1 G1
dG1 = dq + dQ (9.27)
q Q
from which we find on matching with the left-hand side of (9.25)
G1 G1
p= , P = (9.28)
q Q

Consider Example 9.1. We first of all check whether pdq P dQ is a perfect

differential for this problem. For the generating function G1 , the independent
variables are q and Q. Expressing p and P in terms of these, namely, p =
mq cot Q, P = 12 mq 2 cosec2 Q we can express
 
1
pdq P dQ = d mq 2 cot (9.29)
2
which is indeed an exact differential. Hence G1 for this problem is G1 (q, Q) =
1 2
2 mq cot .

There can be other types of generating functions depending on what com-

binations of old and new canonical variables we choose. For instance a Type
II generating function G2 is a function of q and P and the counterparts of
(9.28) are
G2 G2
p= , Q= (9.30)
q P
For the Type III generating function G3 which is a function of p and Q
the underlying relations are
G3 G3
q= , P = (9.31)
p Q
Finally, a Type IV generating function G4 depends on the pair (p, P ) with
the complementary variables q and Q satisfying
G4 G4
q= , Q= (9.32)
p P
The transformations given in (9.28), (9.30), (9.31) and (9.32) hold for the
time-independent canonical transformations. These are restricted canonical
transformations. In the time-dependent case, which we discuss in the next
section, the generating functions possess additionally a function of the variable
t.
 HamiltonJacobi equation and action-angle variables 223

9.4 Types of time-dependent canonical transformations

Consider a transformation (q, p, t) (Q, P, t) for a general time-dependent
situation. Since adding a total differential does not change the essential dy-
namics of the system, we define a canonical transformation (q, p, t) (Q, P, t)
as the one guided by the following criterion:
n
X n
X
pdq Hdt = P dQ Kdt + dG1 (9.33)
i=1 i=1

with G1 = G1 (q, Q, t) to be the time-dependent Type I generating function of

the transformation. The time-independent generating function was considered
earlier from the exact differentiability of the quantity (pdq P dQ).
We now turn to four kinds of canonical transformations as induced by the
corresponding time-dependent generating functions.

9.4.1 Type I canonical transformation

A Type I canonical transformation treats the old and new coordinates qi
and Qi as independent variables. For G1 (qi , Qi , t) we have
n  
X G1 G1 G1
dG1 = dqi + dQi + dt (9.34)
i=1
qi Qi t

Putting this form in (9.33) we find on comparing the differentials

G1 (qi , Qi , t) G1 (qi , Qi , t)
pi = , Pi = (9.35)
qi Qi
along with
G1
K=H+ (9.36)
t
The first of (9.35) gives Qi in terms of qi and pi which when substituted in
the second equation determines Pi . Of course, we assume that the Jacobian
2
of the transformations det | q i Q
G1
j
|=
6 0). K is the new Hamiltonian.

It is worthwhile to note that the Type I canonical transformation induces

the exchange transformation. For instance if we choose G1 (q, Q, t) = qi Qi then
pi and Pi turn out to be Qi and -qi , respectively, while K = H reflecting that
the coordinates and momenta are exchanged.
224 Advanced Classical Mechanics

9.4.2 Type II canonical transformation

For a Type II canonical transformation the independent variables are qi
and Pi which stand, respectively, for the old coordinates and the new mo-
menta. In this case we express
n n
! n
X X X
Pi dQi = d Pi Qi Qi dPi (9.37)
i=1 i=1 i=1

which results in
n n n
!
X X X
pi dqi + Qi dPi Hdt = d Pi Qi Kdt + dG1 (9.38)
i=1 i=1 i=1

The accompanying generating function G2 can therefore be defined by

n
X
G2 = G1 + Pi Qi (9.39)
i=1

Viewing G2 as a function of qi , Pi and t we write

n  
X G2 G2 G2
dG2 (qi , Pi , t) = dqi + dPi + dt (9.40)
i=1
qi Pi t

which implies from (9.38) and (9.39)

G2 (qi , Pi , t) G2 (qi , Pi , t)
pi = , Qi = (9.41)
qi Pi
along with
G2
K=H+ (9.42)
t
The first of (9.41) gives Pi in terms of qi and pi which when substituted in
the second equation provides for Qi . The Jacobian of the transformation is
2 G2
assumed to be nonvanishing: det | q i Pj
|=
6 0 . K as given by (9.42) is the new
Hamiltonian.

The Type II canonical transformation has in its embedding both the iden-
tity as well as the point transformations. In fact, corresponding to G2 (q, P, t) =
qi Pi we see that pi = Pi , Qi = qi and K = H showing for the identity trans-
formation while for G2 (q, P, t) = (qi , t)Pi we have Qi = (qi , t) implying that
the new coordinates are functions of old coordinates.
 HamiltonJacobi equation and action-angle variables 225

9.4.3 Type III canonical transformation

In Type III canonical transformation the underlying generating function
G3 is a function of the independent variables pi , Qi which are, respectively,
the old momenta and new coordinates. G3 is defined by
n
X
G3 (pi , Qi , t) = G1 qi pi (9.43)
i=1

which leads to
G3 G3 G3
qi = , Pi = , K=H+ (9.44)
pi Qi t
Here the Jacobian of the transformation is assumed to be nonvanishing:
2
det | p i Q
G3
j
|=
6 0.

9.4.4 Type IV canonical transformation

In the Type IV canonical transformation the generating function is a func-
tion of old and new momenta and given by
n
X n
X
G4 (pi , Pi , t) = G1 + Qi Pi qi pi (9.45)
i=1 i=1

in which pi and Pi are treated as independent variables. We then have

G4 G4 G4
qi = , Qi = , K=H+ (9.46)
pi Pi t
Here the Jacobian of the transformation is assumed to be nonvanishing:
2 G4
det | p i Pj
|=
6 0.

Example 9.5

Consider the transformation

Q = p, P = q + p2

where is a constant. By the Poisson bracket test, namely, [Q, P ](q,p) = 1 we

conclude that the transformation is a canonical transformation.
The Type 1 generating function is determined by showing pdq P dQ to
be a perfect differential:

pdq P dQ = (Q)dq (q + Q2 )dQ

1
= d(qQ Q3 )
3
226 Advanced Classical Mechanics

Thus
1
G1 (q, Q) = qQ Q3
3
On the other hand, the Type 2 generating function is obtained by treating
q and P as independent variables. From (9.39) we have

G2 (q, p) = G1 + P Q
1
= qQ Q3 + (q + Q2 )Q
3
2 3
= Q
3
 3
2 P q 2
=
3

Check that
 
G2 1 1
= (P q) 2 = p
q P
 
G2 1 1
= (P q) 2 = Q
P q

which are as required.

Example 9.6

The transformations
p
Q = q cos sin
m
P = mq sin + p cos

are easily seen to be canonical due to {Q, P }(q,p) = 1. We also find

 
Q
p = m q cot
sin
 q 
P = m Q cot
sin
Hence the quantity pdq P dQ can be expressed as
 
1
pdq P dQ = d m(q 2 + Q2 ) cot mqQcosec
2

The Type 1 generating function G1 (q, Q) is identified as

1
G1 (q, Q) = m(q 2 + Q2 ) cot mqQcosec
2
 HamiltonJacobi equation and action-angle variables 227

On other other hand, the Type 2 generating function can be obtained from
G2 = G1 + P Q. We get
1
G2 (q, Q) = m(q 2 Q2 ) cot
2
To assign the right variable dependence on G2 , namely, q and P we note that
q P
Q= tan
cos m
by eliminating p from the given transformations. Substituting for Q, G2 turns
out to be
P2
 
qP 1
G2 (q, P ) = m q 2 + 2 2 tan
cos 2 m
We can verify that
 
F2 P
= mq tan = p
q P cos
 
F2 q P
= tan = Q
P q cos m

Example 9.7

We consider the harmonic oscillator Hamiltonian H which is invariant

under the set of canonical transformations considered in the previous example:
1 2 1
H(q, p) = p + m 2 q 2
2m 2
1 2 1 2 2
P + m Q = H(Q, P )
2m 2
The generating function G2 (q, P ) which was found to be

P2
 
qP 1
G2 (q, P ) = m q 2 + 2 2 tan
cos 2 m

gives for a partial derivative with respect to time

p2
    
G2 1 2
= qP sin m q + 2 2 sec2
t q,P 2 m
 2 
P 1 2
= + mQ
2m 2

= H(Q, P )

228 Advanced Classical Mechanics

Therefore, the transformed Hamiltonian reads

!

K(Q, P, t) = 1 H(Q, P )

K
which vanishes for = , i.e., = t. As a consequence Q = P = 0 and
P = K
Q = 0. Hence Q and P are constants which we write as

Q = Q0 , P = P0

where Q0 and P0 stand for the initial values of q and p, respectively.

Reverting to the old coordinates we get
P0
q(t) = Q0 cos t +sin t
m
p(t) = mQ0 sin t + P0 cos t

which give the time evolution for q and p.

9.5 Infinitesimal canonical transformations

We remarked earlier that the Type II canonical transformation is well
suited for the identity transformation. An infinitesimal canonical transforma-
tion (q, p) (Q, P ) about the identity is generated through an expression of
the form

G2 (q, P, t) = qi Pi + G(q, P ) (9.47)

where  is an infinitesimal quantity and results in

G2 (qi , Pi , t) G G2 (qi , Pi , t) G
pi = = Pi +  , Qi = = qi +  (9.48)
qi qi Pi Pi
with K = H. The departures from the exact identity are due to the presence
G G
of small quantities  qi
and  Pi
in pi and Qi , respectively. We can summarize
qi Qi qi and pi Pi pi as

G(q, P ) G(q, p) G(q, P ) G(q, p)

qi =  = +O(2 ), pi =  =  +O(2 )
Pi pi qi qi
(9.49)
As an immediate role of the above results we can think of the Hamiltonian
 HamiltonJacobi equation and action-angle variables 229

flow following the chain given by (q(t), p(t)) (Q(t), P (t)) (q(t + t), p(t +
t)):

H H
Qi (t) = qi (t) + qi t = qi (t) + t, Pi (t) = pi (t) + pi t = pi (t) t
pi qi
(9.50)
where we Taylor expanded qi (t + t) and pi (t + t) up to O(t).
qi and pi implied by (9.49) when compared with (9.50) leads to the
identification of the Hamiltonian H with the generating function G of the
infinitesimal canonical transformation, i.e., H = G while the increment t with
the small quantity , i.e., t = . Thus, the Hamiltonian can be looked upon
as the generator of the infinitesimal transformation and the time evolution as
a one-parameter canonical transformation.

9.6 HamiltonJacobi equation

The idea of deriving the HamiltonJacobi equation rests on effecting a
canonical transformation that maps one basis of known canonical variables to
a new set of coordinates and momenta such that the latter are cyclic for the
transformed Hamiltonian. What does it mean?
Consider the specific case of a Type II generating function G2 (qi , Pi , t), i =
1, 2, ..., n which facilitates the transformations according to (9.41) where the
time derivative of Qi and Pi stand as
K K
Qi = , Pi = , i = 1, 2, ..., n
Pi Qi
as defined by (9.2). Now if Qi and Pi are cyclic coordinates in the transformed
Hamiltonian K, then it follows that these have to be constants in time. Such
a validity is assured if we assume, without loss of generality, K to be zero. It
then gives from (9.42) the equation
G2
H(q1 , q2 , ..., qn ; p1 , p2 , ..., pn ; t) + =0
t
A few remarks are in order:
From (9.40) a total derivative of the generating function G2 (qi , pi , t) with
respect to t gives, in view of the above equation,
n n
dG2 X G2 X
= qi = pi qi H(qi , pi , t) = L (9.51)
dt i=1
qi i=1

where we employed (9.41) and noted that Pi s are cyclic coordinates in K

and hence taken to be constants. Thus, integrating between times, say, t1 and
230 Advanced Classical Mechanics
Rt
t2 the quantity S G2 defined by the action integral S = t12 Ldt satisfies
(9.41).
Replacing pi by the corresponding relation given in the first of (9.41) we
arrive at the form

S S S S S
pi = : H(q1 , q2 , ..., qn ; , , ..., ; t) + =0 (9.52)
qi q1 q2 qn t
which is called the time-dependent HamiltonJacobi equation. More com-
pactly it is also expressed as
S S
H(qi , , t) + =0 (9.53)
qi t
A function S which is a solution of the above is called the Hamilton principal
function.
Equation (9.53) is a first-order differential equation involving the n coor-
S
dinates qi s and t. It is not linear in q i
because the partial derivatives of S
appear in higher degree than the first. Associated with the (n + 1) variables (n
qi s and t) we expect (n + 1) constants of motion namely 1 , 2 , ..., n , n+1 .
However, we notice one curious thing in (9.53) which is that the dependent
variable S itself does not appear in it: only its partial derivatives do. So one of
the constants has no bearing on the solution, i.e., the solution has an additive
constant. Disregarding such an irrelevant additive constant we write for the
complete integral of (9.54) the form

S = S(q1 , q2 , ..., qn ; 1 , 2 , ..., n ; t) (9.54)

where it is ensured that none of the constants s is of additive nature to the
solution. Actually we identify these constants with the constant momenta Pi ,
i = 1.2..., n, which, according to our choice, were assumed to be cyclic.
The cyclic coordinates Qi s provide another set of constants i0 s, i =
1, 2, ..., n and read from the second equation of (9.41)
S
Qi = = i , i = 1, 2, ..., n (9.55)
i
We clarify the above issues by considering two examples: one for the free par-
ticle and the other that of the harmonic oscillator.

Example 9.8

The HamiltonJacobi equation for the free particle problem is obviously

1 S 2 S
( ) + =0
2m q t
 HamiltonJacobi equation and action-angle variables 231

Its complete integral can be easily ascertained by inspection which reads

S(q, , t) = 2mq t

where is a non-additive constant.

From (9.55)

S
= = k (say)
r

r
m 2E S
i.e. k= qt q= (t + k) and p = = 2m.
2 m q

which conform to their expected forms.

It is to be noted that HamiltonJacobi equation being a partial differential
equation can admit multiple solutions. For instance, in the above case, there
is also a legitimate solution given by

m(q )2
S(q, , t) =
2t
where is a non-additive constant. This implies
S m
= = (q )
t

giving the time-dependence: q(t) = m t where can be identified as the
initial value of q while is the negative of the momentum: = p.

Example 9.9

For the harmonic oscillator problem the HamiltonJacobi equation has the
form
1
p2 + m2 2 q 2


H=
2m
S
The relation p = q gives
" 2 #
1 S 2 2 2 S
+m q + =0
2m q t

To solve for the above equation we try separation of variables for S:

S(q : ; t) = W (q : ) t

where is a non-additive constant.The function W (q:) is the time-

independent part called Hamiltons characteristic function which we shall dis-
cuss in the next section.
232 Advanced Classical Mechanics

We get " #
2
1 W 2 2 2
+m q =
2m q
which gives the integral
r
Z
m 2 q 2
W = 2m dq 1
2
Such an integral implies for
r Z
S m dq
= = q t
2 2 2
1 m2q

which integrates to r
1 1 m 2
t + = sin q
2
Hence the solutions for q and p are
r
2
q = sin (t + )
m 2
S W p
p = = = 2m m 2 q 2 = 2m cos (t + )
q q

which are in agreement with their well-known forms.

9.6.1 Time independent HamiltonJacobi equation:

Hamiltons characteristic function
In the time independent case the Hamilton does not depend on time and
the system is conservative. The HamiltonJacobi equation reduces to the form
 
S S
H qi , + =0 (9.56)
qi t
S
where pi = q i
. Note that the time dependent character of S stays. A solution
of (9.56) is given in the separation of variables form

S(qi , i , t) = W (qi , i ) Et (9.57)

implying
   
W S W
pi = : H qi , = H qi , =E (9.58)
qi qi qi
 HamiltonJacobi equation and action-angle variables 233

where the Hamiltonian has the constant value E. W is called Hamiltons

characteristic function. Of the two equations in (9.58) the first one is the time-
independent HamiltonJacobi equation while the second one is the equation
for Hamiltons characteristic function.
The separability of the equation for Hamiltons characteristic function
means we should be able to express W as

W = W1 (q1 ; 1 , 2 , ..., n ) + W2 (q2 ; 1 , 2 , ..., n ) + ... + Wn (qn ; 1 , 2 , ..., n )

(9.59)
This has the effect of projecting the characteristic function equation as

Wi (qi ; 1 , 2 , ..., n )
Hi (qi , ) = E, i = 1, 2, ..., n (9.60)
qi
In general each Wi will contain the corresponding qi along with the set of s
but there may be situations when one Wi will contain one qi and one i for
i = 1, 2, ..., n as the following example demonstrates.
Consider a system with one degree of freedom having the Hamiltonian

p2
H=
+ V (q) (9.61)
2m
It results in the HamiltonJacobi equation
1 S 2 S
( ) + V (q) + =0 (9.62)
2m q t
Employing separation of variables we express

S(q, , t) = W (q, ) Et (9.63)

where the separation constant E is identified as the obvious non-additive
constant. It leads to an ordinary differential equation for W
1 dW 2
( ) + V (q) = E (9.64)
2m dq
which can be inverted to read in the integral form
Z p
W = 2m(E V (q))dq (9.65)

We do not need to determine W explicitly. What suffices is the knowledge

for the quantities q and which are obtainable from

r Z
W p W m dx
p= = 2m( V (q)), +t= = p
q E 2 (2m(E V (q)))
(9.66)
234 Advanced Classical Mechanics

We give below some illustrations how such a scheme works in practice.

Example 9.10

1 2 2
We again focus on the harmonic oscillator potential V (q) = 2 m q .
q
2E
Putting q = m 2 sin , an integration gives from (9.65)

E
W = ( + sin cos )

Further from (9.66) we obtain

W
p= = 2mE cos
q
W
+t= =
E

Example 9.11

Consider the vertical motion of a particle in a unform gravitational field as

given by the potential V (z) = mqz where we have identified the generalized
coordinate q with z. The Hamiltonian is

p2
H= + mgz = E =
2m
W
which for p = z gives the differential equation
1 dW 2
( ) + mgz =
2m dz
Exploiting the second equation of (9.66) gives
r
2p
g( + t) = E mgz
m
which on squaring leads to
1 E
z = g( + t)2 +
2 mg
On the other hand, for p we have from the first equation of (9.66)
W
p= = mg( + t)
z
If we use initial conditions that at t = 0, z = z0 , p = 0 then we have E = mgz0
and = 0 furnishing the standard expressions z = 12 gt2 + z0 and p = mgt.
 HamiltonJacobi equation and action-angle variables 235

9.6.2 Other variants of HamiltonJacobi equation

We derived the HamiltonJacobi equation by exploiting a Type II gen-
erating function. But we could as well have used the other types. Here is a
summary of HamiltonJacobi equations resulting from Type I, Type III and
Type IV generating functions:

G1 (q, Q, t) S S(qi , Qi , t)
Type I : H(qi , pi , t) + = 0 V H(qi , , t) + =0
t qi t
(9.67)

G3 (p, Q, t) S S(pi , Qi , t)
Type III : H(qi , pi , t)+ = 0 V H( , pi , , t)+ =0
t pi t
(9.68)

G4 (pi , Pi , t) S S(pi , Pi , t)
Type IV : H(qi , pi , t)+ = 0 V H( , pi , , t)+ =0
t pi t
(9.69)
Note that in both (9.68) and (9.69) the coordinates in the HamiltonJacobi
equations appear as partial derivatives which are rather complicated to han-
dle. So these do not have much utility. The form (9.67) is relatively simpler
compared to (9.68) and (9.69) and is sometimes used but one has to remem-
ber that the Type I and Type II generating functions are related as given in
(9.39).

9.7 Action-angle variables

We are familiar with ideas of finite motions that are periodic. For a system
with one degree of freedom implying that the phase space is of dimension two,
if the motion is described by a closed curve, then both q and p variables return
to the same point periodically in the manner

q(t + ) = q(t), p(t + ) = p(t) (9.70)

where is a constant quantity. We call the motion liberation or oscillatory in

character.
It can sometimes happen that while the p-variable is periodic, the period-
icity of the q variable is subject to a displacement by an amount q0 after each
period :
p(t + ) = p(t), q(t + ) = q(t) + q0 (9.71)
The rotation has such a feature.
236 Advanced Classical Mechanics

Let us consider a canonical transformation as guided by S in terms of the

characteristic function W (qi , i ). Let us define the quantities Ji by
I
JI = pi dqi , i = 1, 2, ..., n (9.72)

where for the integral we take the coordinate qi around a closed path while
other coordinates remain spectators. Replacing pi by the partial derivative
Wi (qi ;)
qi , we can express Ji as
I
Wi (qi ; )
Ji = dqi , i = 1, 2, ..., n (9.73)
qi
Since Wi depends on the variables qi s and s and qi s are integrated over,
Ji s finally become only a function of the constants 1 , 2 , ..., n :

Ji = Ji (1 , 2 , ..., n ), i = 1, 2, ..., n (9.74)

Ji is called the action variable for the ith degree of freedom.

Inverting (9.74) it follows that

i = i (J1 , J2 , ..., Jn ), i = 1, 2, ..., n (9.75)

We can thus write from (9.59)

W (qi , Ji ) = W1 (q1 ; J1 , J2 , ..., Jn ) + W2 (q2 ; J1 , J2 , ..., Jn ) + ...

+W1 (qn ; J1 , J2 , ..., Jn ) (9.76)

The coordinate i which is conjugate to Ji is called the angle variable. This

along with pi are given by
Wi (qi , Ji ) W
i = , pi = , i = 1, 2 =, ..., n (9.77)
Ji qi
The dimension of Ji is that of the action and W (qi , Ji ) is, in effect, the
generating function of the canonical transformation (qi , pi ) (i , Ji ).
The time independence of W makes the old and transformed Hamiltonians
to be the same. The Hamiltonian which is only a function of the constants
1 , 2 , ..., n is now a function of action variables J1 , J2 , ..., Jn because of
(9.75). The time derivatives of Ji and i read
H H
Ji = = 0, i = = i (Ji ), i = 1, 2 =, ..., n (9.78)
i Ji
An integration reveals that Ji s are constants while i s are linear functions
of time given by i = i (Ji )t + constant, involving the constants i s which in
turn depend upon Ji s.
Consider the pair (J, ). If now is chosen in such a way that it increases
by 2 as the motion undergoes one period, the time period will be given by
 HamiltonJacobi equation and action-angle variables 237
2
T = (J) where we call (J) to be the angular frequency. The phase curve will
be characterized by

q = q( + 2, J) = q(, J), p = p( + 2, J) = p(, J) (9.79)

reflecting the 2-periodic dependence on of p and q variables.

Example 9.12

Let us take the harmonic oscillator problem for which the Hamiltonian is
1 2
H = 2m p + 12 m 2 q 2 with H = E = . The Hamiltons characteristic function
obeys
1 dW 2 1
( ) + m 2 q 2 =
2m dq 2
From (9.65) W can be put in the form
Z r
2
W = m q 2 dq
m 2
Using it and the definition for the action variable
I
W (q; )
J= dq (9.80)
q
it
q is a straightforward task to perform the integration by substituting q =
2
m2 sin and taking (0, 2) as the range for . It gives for J

J = 2

As such we can express
J
H==
2
Taking partial derivative of H with respect to J we then find for the angle
variable

=
2
Note that being the angular frequency, the angle variable coincides with the
ordinary frequency of the oscillator.

Example 9.13

The potential of a two-dimensional oscillator with different angular fre-

quencies x and y is given by
1 1
V (x, y) = m2x x2 + m2y y 2
2 2
238 Advanced Classical Mechanics

The equation for the characteristic function reads

" 2  2 #
1 W W 1 1
+ + m2x x2 + m2y y 2 = E = x + y (say)
2m x y 2 2

Adopting a separation of variables for the x and y coordinates for W, i.e.,

writing W = X(x) + Y (y), we obtain straightforwardly from the above equa-
tion Z p Z q
W = 2mx m2 2x x2 dx + 2my m2 2y y 2 dy

In consequence the two momenta are

W p W q
px = = 2mx m2 2x x2 , py = = 2my m2 2y y 2
x y
The corresponding action variables are
I p I q
Jx = 2mx m2 2x x2 dx, Jy = 2my m2 2y y 2 dy
q q
2mx 2my
Substituting x = m2 2x sin , y = m2 2y sin we obtain for Jx and Jy
when integrated in the range (0, 2) the following forms:
2x 2y
Jx = , Jy =
x y

These imply
x Jx y J y
x = , y =
2 2
W W
Next, the angle variables being x = Jx and y = Jy can be evaluated
x y
to be and
2 2 ,
respectively, and correspond to the ordinary frequencies in
the x and y directions.

Example 9.14

Consider the Keplerian motion of a particle in a central field of force. The

kinetic and potential energies are
1
T = m(r2 + r2 2 + r2 sin2 2 ), V (r) =
2 r
where is a real constant. Then the Hamiltonian is given by
!
2 2
1 p p
H= p2r + 2 + 2 2
2m r r sin r
 HamiltonJacobi equation and action-angle variables 239
n
P
where we have employed the Legendre transformation H = pi qi L with
i=1
L = T V . The conjugate momenta are pr = mr, p = mr2 and p =
mr2 sin2 .
To obtain the HamiltonJacobi equation we make the replacements
S S S
pr = , p = p =
r
yielding for S the partial differential equation
1 S 2 1 S 1 S S
[( ) + 2 ( )2 + 2 2 ( )2 ] + =0
2m r r r sin r t
Taking for S the form

S(r, , , , t) = W (r, , , ) t
results in the corresponding equation for the characteristic function W
1 W 2 1 W 2 1 W 2
[( ) + 2( ) + 2 2 ( ) ] = = |E|
2m r r r sin r
where we have taken the energy |E| < 0 for a bound state.
Assuming separation of variables for the function W given by W = Wr (r)+
W () + W (), the above equation can be put in the form
dWr 2 1 dW 2 1 dW 2
( ) + 2( ) + 2 2 ( ) = 2m(|E| + )
dr r d r sin d r
where the partial derivatives have been replaced by ordinary derivatives.
By multiplying with r2 sin2 , the only -dependent term can be isolated
enabling us to express
dW
= = p
d
where is a constant.
Similarly carrying out separation of variables for the other two variables r
and , p and pr result from solving the corresponding differential equations.
These are
dW q
= p = 2 2 cosec2
d
and
r
dWr 2
= pr = 2m(|E| + ) 2
dr r r
and W can be obtained from the integral
240 Advanced Classical Mechanics

Z
W = (pr dr + p d + p d)

Of the three action variables, Jr , J and J , the last one J is the easiest
to evaluate. We straightforwardly find
I I I
dW
J = p d = = d = 2
d
Determination of Jr and J is relatively more complicated. First consider Jr .
It is given by
I r
2
I I
dWr
Jr = pr dr = = 2m(|E| + ) 2 dr
dr r r
If we take a partial derivative of Jr through the integral we can write it in
the way1
s
2m b
Z
Jr dr
= p
r |E| a (a r)(r b)
where a and b are the turning points where pr vanishes. From the form for pr
given above, we solve for the quadratic in r to obtain for a and b

2
a+b= , ab =
|E| 2m|E|
on comparing with the integrand of Jr . Defining = 12 (a + b) and = 12 (a b)
and making a replacement of r as r = + cos , the product (a r)(r b)
takes a simple form 2 sin2 making the integral easily tractable. We find
s
2m
Z r
Jr m 1
= ( + cos )d =
r |E| 0 2 |E| 32
which yields for Jr
s
2m
Jr = +
|E|
where is a constant.
can be determined by considering a circular orbit of radius a (and pr = 0)
when the Hamiltonian is given by

1 2 2
H= (p + 2 )
2m r r r
1 R. Hosley, Lecture notes on Hamiltonian dynamics (2014), University of Edinburgh,

unpublished.
 HamiltonJacobi equation and action-angle variables 241
2
Hamiltonian equation pr = H r = 0 gives the value of a = m using

the above form of H. Since the Hamiltonian gives the energy of the system,
2
substituting this value of a in H implies E = 12 m
2
. We are now in a position
to estimate by setting Jr = 0 since pr = 0. Using the just determined value
of E we find = . Hence we obtain Jr as
s
2m
Jr = 2 +
|E|
We refer the readers to the Goldstein book (see bibliography at the end) for
a more improved method, involving complex integrations, for the evaluation
of Jr .
Finally, let us turn to J which reads
I I I q
dW
J = p d = = 2 2 cosec2 d
d

One way to tackle the integral is to call the ratio = cos and rewrite J
as
I p
J = 1 cos2 cosec2 d

For the closed path we define the circuit as it is traced out by running from
say, 0 to 0 and back. The value of 0 is furnished by the equation sin 0 =
cos if we look at the vanishing condition of the integrand corresponding to
such a value of . In other words, 0 is determined from as 0 = 2 . In
this way, J is completely given by
Z 0 p
J = 4 1 cos2 cosec2 d
0
where the factor of 4 occurs because by symmetry it suffices to integrate over
the range (0, 0 ) only.
The rest is straightforward. We make a change of variable cos =
sin sin which implies that for = 2 , = 0 while for = 0, = 2 .
Note that = 0 implies = 2 so that the integrand in the expression for J
does not become singular.
In this way J acquires the form
Z
2 p
J = 4 1 cos2 cosec2 d
0

It is left as an exercise to evaluate the last integral (Hint: put t = tan ). The
final form of J is

J = 2( )
242 Advanced Classical Mechanics

It is interesting to observe that when we take the sum Jr + J + J all the

s cancel out and we are left with the following energy relation:

2m 2 2
E=
(Jr + J + J )2
where recall that we are considering bound state or periodic motion for which
E < 0. The corresponding frequencies r , and are the same and given
by the partial derivative of the respective J with the corresponding :

4m 2 2
r = = =
(Jr + J + J )3
which is completely symmetric in Jr , J , J .

9.7.1 Motion of a particle in a 2-dimensional rectangular

well
Let us now investigate2 the nature of the trajectories when the particle is
confined in an infinite rectangular well, with center at (0, 0) and vertices at
(a , b). The two-dimensional potential is given by:

0 |x| < a and |y| < b
V (x, y) = (9.81)
|x| > a and |y| > b

Employing a Type-II generating function, with the assumption that the Hamil-
tonian H does not depend on t explicitly, gives the following equation for the
characteristic function W :
" 2  2 #
1 W W
+ =E (9.82)
2m x y

Writing W in a separated form, namely, W = X(x) + Y (y), we obtain the

following differential equations:
 2  2
dX dY
= x2 , = y2 (9.83)
dx dy

1
x2 + y2 . This implies for the momenta


with E =
2m
px = x , py = y (9.84)
2 B. Bagchi, S. Mallik and C. Quesne, Infinite square well and periodic trajectories in

classical mechanics, preprint arXiv:physics/0207096 (2002); B. Bagchi and A. Sinha, A

classification of classical billiard trajectories, preprint arXiv:physics/09074892 (2009).
 HamiltonJacobi equation and action-angle variables 243

In the above the signs show the reversal in the direction of motion of the
particle each time it hits the barriers at x = a and y = b.

The action variables can now be calculated easily and turn out to be

Jx = 4ax (9.85)

Jy = 4by (9.86)
so that E takes the form
!
1 Jx2 Jy2
E= + (9.87)
32m a2 b2

Thus the natural frequencies of the system are obtained to be

E Jx E Jy
x = = , y = = (9.88)
Jx 16ma2 Jy 16mb2
which, with the help of (9.85) and (9.86) read equivalently
px py
x = , y = (9.89)
4ma 4mb
Thus the natural frequencies are functions of the particle velocity and the
dimensions of the well. Consequently, three types of trajectories are possible
for the particle trapped in the rectangular well, as discussed below.

9.8 Possible trajectories

In this section we shall discuss in detail three possible types of trajectories
of the trapped particle, namely,
1. Periodic trajectories
2. Open trajectories
3. Special trajectories when the billiard ball hits a corner

9.8.1 Periodic trajectories

To study periodic or closed trajectories we note that for the particle to
execute periodic motion, it must return to the starting point with its initial
momenta after a certain duration of time. This is possible only if

T = nx Tx = ny Ty (9.90)

where Tx and Ty represent the time in which the particle reaches the starting
point with its initial momenta in the x and y directions, respectively, T is the
244 Advanced Classical Mechanics

time period of the orbit, and nx , ny are integers. Thus, if the particle starts
from the origin at an angle to the x direction, where
py
tan = (9.91)
px

then for closed orbits tan must be rational,3 with the time period given by
(9.90). We can straightforwardly rearrange tan to write down

by bTx bny
tan = = = (9.92)
ax aTy anx

We now consider the following illustrative examples with the particle starting
from the origin in each case.

9.8.1.1 Some explicit examples for periodic trajectories

py b
Case 1 : ny = 1, nx = 4 , i.e., =
px 4a
The trajectories traced out by the particle are given by the following results
for the intervals of time specified:

3 We consider those cases where the linear momentum in the x and y directions, namely,

px , py are expressible in rational form, so that the action variables Jx , Jy as well as the
natural frequencies x , y are also rational.
 HamiltonJacobi equation and action-angle variables 245

px py
x= t y= t t16(r1) t t16r15
m m
px py b
x= (t t16r15 ) + a y= (t t16r15 ) + t16r15 t t16r14
m m 4
px py b
x= (t t16r14 ) y= (t t16r14 ) + t16r14 t t16r13
m m 2
px py 3b
x= (t t16r13 ) a y= (t t16r13 ) + t16r13 t t16r12
m m 4
px py
x= (t t16r12 ) y= (t t16r12 ) + b t16r12 t t16r11
m m
px py 3b
x= (t t16r11 ) + a y= (t t16r11 ) + t16r11 t t16r10
m m 4
px py b
x= (t t16r10 ) y= (t t16r10 ) + t16r10 t t16r9
m m 2
px py b
x= (t t16r9 ) a y= (t t16r9 ) + t16r9 t t16r8
m m 4
px py
x= (t t16r8 ) y= (t t16r8 ) t16r8 t t16r7
m m
px py b
x= (t t16r7 ) + a y= (t t16r7 ) t16r7 t t16r6
m m 4
px py b
x= (t t16r6 ) y= (t t16r6 ) t16r6 t t16r5
m m 2
px py 3b
x= (t t16r5 ) a y= (t t16r5 ) t16r5 t t16r4
m m 4
px py
x= (t t16r4 ) y= (t t16r4 ) b t16r4 t t16r3
m m
px py b
x= (t t16r3 ) a y= (t t16r3 ) t16r3 t t16r2
m m 4
px py b
x= (t t16r2 ) y= (t t16r2 ) t16r2 t t16r1
m m 2
px py b
x= (t t16r1 ) a y= (t t16r1 ) t16r1 t t16r
m m 4

(9.93)

ma
where r = 1, 2, and tn = (2n 1) , n = 1, 2,
px
Thus, in this case, the trajectories are periodic when the velocities of the
particle in the x and y directions are such that the time taken by the particle
246 Advanced Classical Mechanics

to cover the distance a in the x direction, is the same as the time it takes to
cover the distance b/4 in the y direction:

ma mb
t1 = =
px 4py

This gives the time period as T = 4Tx = Ty , where Tx = 4t1 and Ty = 16t1 .

2bp
Case 2 : ny = 2, nx = 3 , i.e., pxy = 3a
The particle can be shown to trace out the following trajectories:

px py
x= t y= t t12(r1) t t12r11
m m
px py 2b
x= (t t12r11 ) + a y= (t t12r11 ) + t12r11 t t12r10
m m 3
px a py
x= (t t12r10 ) + y= (t t12r10 ) + b t12r10 t t12r9
m 2 m
px py
x= (t t12r9 ) a y= (t t12r9 ) t12r9 t t12r8
m m
px a py
x= (t t12r8 ) + y= (t t12r8 ) b t12r8 t t12r7
m 2 m
px py 2b
x= (t t12r7 ) + a y= (t t12r7 ) t12r7 t t12r6
m m 3
px py
x= (t t12r6 ) y= (t t12r6 ) t12r6 t t12r5
m m
px py 2b
x= (t t12r5 ) a y= (t t12r5 ) + t12r5 t t12r4
m m 3
px a py
x= (t t12r4 ) y= (t t12r4 ) b t12r4 t t12r3
m 2 m
px py
x= (t t12r3 ) + a y= (t t12r3 ) t12r3 t t12r2
m m
px a py
x= (t t12r2 ) y= (t t12r2 ) b t12r2 t t12r1
m 2 m
px py 2b
x= (t t12r1 ) a y= (t t12r1 ) t12r1 t t12r
m m 3

(9.94)
ma
where r = 1, 2, and tn = (2n 1) , n = 1, 2,
px
 HamiltonJacobi equation and action-angle variables 247

It is easy to observe that in this case t2 = 3t1 /2, t3 = 3t1 , t4 = 9t1 /2, t5 = 5t1 ,
ma 2mb
etc. Thus, Tx = t5 t1 = 4t1 and Ty = t8 t2 = 6t1 , where t1 = = ,
px 3py
giving T = 12t1 = 3Tx = 2Ty .

9.8.2 Open trajectories

py
If the initial angle is such that is irrational, then the orbit is an open
px
one. This is due to the fact that the time periods in the x and y directions
are such that one cannot find integral values of nx , ny satisfying equation
(9.90). We have traced out such a trajectory for = 300 . Even after multiple
reflections from the perfectly elastic walls of the rectangular well, the orbit
does not close.

9.8.3 Special trajectories when the billiard ball hits a corner

We now address the interesting case of special non-periodic trajectories
when the particle hits one of the corners of the billiard table and gets pocketed.
For the ball to hit either the right or left wall, the distance travelled in the
x direction is (4n 1)a, where n is an integer. Similarly, to hit the top or
bottom wall the distance travelled in the y direction is (4m 1)b, where m
is also an integer. If the ball hits a corner, then these two conditions must be
satisfied simultaneously. Thus, if the particle hits a corner in time t, then

vy t (4m 1) b
= (9.95)
vx t (4n 1) a

and the condition for the billiard ball to get pocketed reduces to

py (4m 1) b
= (9.96)
px (4n 1) a

If the numerator has positive (negative) sign in (9.96), then the ball hits one
of the two corners where y is positive (negative). Similarly, if the denominator
has positive (negative) sign in (9.96), then the ball hits one of the two corners
where x is positive (negative). From equations (9.91), (9.92) and (9.96), it is
obvious that this occurs for odd integral values of both nx and ny . Based on
this we summarize in Table 9.1 the corner in which the ball will get pocketed.
Note that we assign the following numbers to the respective corners : (a, b) as
corner 1, (a, b) as corner 2, (a, b) as corner 3, and (a, b) as corner 4. It
is observed that the predicted corners are in fact the actual ones.
248 Advanced Classical Mechanics

Table 9.1 Corners Where the Ball is Pocketed.

ny /nx Sign (numerator, denominator) Corner

1/3 (, +) 2
1/5 (+, +) 1
1/7 (, +) 2
1/9 (+, +) 1
3/7 (, ) 3
5/7 (, +) 2
3/5 (+, ) 4
5/9 (+, +) 1
7/9 (+, ) 4

9.9 Summary
Since the choices of generalized coordinates in most cases are rather ar-
bitrary it is sometimes necessary to look for transformations that ease the
mathematical complications confronting a given physical problem. Canonical
transformations form one class of transformations that offer a simplified way
of looking at the physical properties of a mechanical system through equations
that can be represented in simple terms. Looking for invariance of Hamiltons
equations under a set of suitable transformations does indeed give the right
clue in this direction. For one thing we can equivalently think of those class
of transformations that preserve the invariance of Poisson brackets defined
in terms of new variables. For another, the symplectic character of canonical
transformations holds equivalently. Permuting the set of generalized coordi-
nates and the set of generalized or conjugate momenta one can define four dif-
ferent types of functions that generate these transformations. Such functions
are called generating functions. The extreme possibility of the transformed
variables becoming explicitly absent in the transformed Hamiltonian gives us
the HamiltonJacobi equation. This has many rich consequences the most
noteworthy being the formulation of action-angle variables. This chapter dis-
cussed through examples the underpinnings of canonical transformations and
the new directions they open up.
 HamiltonJacobi equation and action-angle variables 249

Exercises
1. Show that the following transformations of coordinates are canonical:

(i) Q = q cos 2p, P = q sin 2p

q 1 2
(ii) Q = tan1 ( ), P = (q + p2 )
p 2
sin p
(iii) Q = log( ), P = qcotp
q

(iv) Q = pq 2 , P = q 1

(v) Q = log(1 + q cos p), P = 2(1 + q cos p) q sin p
2. What is the canonical transformation generated by G1 = q 2 cotQ,
where is a constant?
3. Deduce a canonical transformation such that the Hamiltonian H of the
1
form H = p2 + (4q + 1)p + 4q 2 2q is coverted to K(Q, P ) = 2m P 2 mgQ
where m and g are constant quantities. Interpret K(q, P ).
4. Show that the set of transformations
r
2q p
Q= cos p, P = 2qk sin p
k
is canonical and the generating function G1 (q, Q) is
s
1 p 1 k
G1 (q, Q) = Q 2qk k 2 Q2 q cos (Q )
2 2q
5. Prove that the generating function facilitating transformation from
Cartesian to spherical coordinates through the usual relations

x = r sin cos , y = r sin sin , z = r cos

is given by the Type II form

G2 (p, Q) = (r sin cos px + r sin sin py + r cos pz )

6. Let a Hamiltonian be given by
1 2
H= (p + p2y ) + c2 (qx2 + qy2 )
2 x
If we transform to new variables defined by

p2x 1 2 2 1
Qx = qx2 + , Qy = c [(qx + qy2 ) + 2 (p2x + p2y )]
c2 2 c
250 Advanced Classical Mechanics

prove that the transformation is canonical and the transformed Hamiltonian

is given by simply Qy .
7. Show that the canonical transformation mapping the variables
(qx , px , qy , py ) to (Qx , Px , Qy , Py ) where the latter are given by

Qx = qx cos qy sin , Qy = qx sin + qy cos , Px = px cos py sin ,

Py = px sin + py cos

can be effected by the generating function G2 = (Px qx + Py qy ) cos + (Py qx

Px qy ) sin .
Using such a transformation, show that a coupled oscillator Hamiltonian

p2x + x2 qx2 p2y + y2 qy2

H= + + gqx qy
2 2
gets decoupled in a manner H K:

Px2 + 2x Q2x Py2 + 2y Q2y

K= +
2 2
where

1 2 q 1 2 q
2x = [x +y2 + (x2 y2 )2 + 4g 2 ], 2y = [x +y2 (x2 y2 )2 + 4g 2 ]
2 2
and tan(2) = 22g 2.
x y
8. The motion of a particle of mass m moving in a plane under a central
force is controlled by the Hamiltonian
1 2
H= (r + r2 2 ) + V (r)
2m
Determine Hamiltons characteristic function and solve the problem com-
pletely.
9. If a time-independent generating function G2 (q, P ) is given by
G2 (q, P ) = qP + aq 3 P + bqP 3 where  is a small quantity and a,b are
constants, obtain q and p in the forms

q = Q 3bQP 2 aQ3 , p = P + bP 3 + 3aQ2 P

keeping only up to O() terms. Use the above form for G2 , map the anharmonic
oscillator Hamiltonian H = 12 (p2 + 2 q 2 )+q 4 to K(Q, P ) = 12 (P 2 + 2 Q2 )+
c(P 2 + 2 Q2 )2 , where c is to be determined. Show that Q and P are in the
forms

Q = A cos(t + ), P = A sin(t + )
3A2
where A is a constant and = + 2 .
 HamiltonJacobi equation and action-angle variables 251

10. For a power law potential V (q) = q 2n , where > 0 and n is a positive
p2
integer, appearing in the Hamiltonian H = 2m +V (q), find the action variable
I in the form
1 nI n+1
2n 1 n+1
n
E = n+1 ( ) ( )
Jn 2m
where E is the energy and Jn is given by the integral
Z 1
1 12n
Jn = (1 t) 2 t 2n dt
0
References

1. Arnold, V.J., (1978) Mathematical Methods of Classical Mechanics,

Springer-Verlag, New York.
2. Arrowsmith, D.K. and Place, C.M., (1992) Dynamical Systems, Differ-
ential Equations, Maps and Chaotic Behavior, Chapman & Hall/CRC,
New York.
3. Berry, M.V., (1989) Principles of Cosmology and Gravitation, Cambridge
University Press, London.
4. Calkin, M.G.,(1996) Lagrangian and Hamiltonian Mechanics, World Sci-
entific Publishing Co. Pte. Ltd.
5. Corinaldesi, E., (1998) Classical Mechanics For Physics Graduate Stu-
dents, World Scientific Publishing Co. Pte. Ltd.
6. Dirac, P.A.M., (1964) Lectures on Quantum Mechanics, Belfer Graduate
School of Science, Yeshiva University.
7. Elsgatts, L., (1975) Differential Equations and the Calculus of Varia-
tions, Mir Publishers, Moscow.

8. Fasano, A. and Marmi, S., (2002) Analytical Mechanics, Oxford Univer-

sity Press, Oxford.
9. Fetter, A.L., and Walecka, J.D., (1980) Theoretical Mechanics of Parti-
cles and Continua, McGraw-Hill Companies, Inc., New York.

10. Fring, A., (2005) Lecture Notes on Dynamical Systems, City University,
London.
11. Gantmacher, F., (1970) Lectures in Analytical Mechanics, Mir Publish-
ers, Moscow (1975).

12. Goldstein, H., Poole, C. and Safko, J., (2000) Classical Mechanics,
Addison Wesley, San Francisco, CA.
13. Greiner, W., (2010) Classical Mechanics: Systems of Particles and
Hamiltonian Dynamics, Springer-Verlag, New York.
14. Gupta, S.N.,(1970) Classical Mechanics, Meenakshi Prakashan, Meerut.

253
254 References

15. Hand, L.N. and Finch, J.D., (1998) Analytical Mechanics, Cambridge
University Press, London.

16. Hilborn, R.C., (1994) Chaos and Nonlinear Dynamics: An Introduction

for Scientists and Engineers, Oxford University Press, Oxford.
17. Iro, H., (2002) A Modern Approach to Classical Mechanics, World Sci-
entific Publishing Co. Pte. Ltd.

18. Jones, D.S., Plank, M. and Sleeman, B.D., (2009) Differential Equations
and Mathematical Biology, CRC/Taylor & Francis Group, Boca Raton,
FL.
19. Kleppner D., and Kolenkov, R., (1973) An Introduction to Mechanics,
McGraw-Hill Companies, Inc., New York.

20. Landau, L.D., and Lifshitz, E.M., (1976) Mechanics, Pergamon Press,
Oxford.
21. Lass, H.,(1950) Vector and Tensor Analysis, McGraw-Hill Book Com-
pany, Inc., New York.

22. Loney, S.L., (1982) An Elementary on the Dynamics of a Particle and

of Rigid Bodies, Macmillan, Calcutta.
23. Pierre, K., (1994) Lectures Notes on the Introduction to the Theory of
Dynamical Systems, University of Brussels.

24. Rana, N.C., and Joag, P.S. (1991) Classical Mechanics, McGraw Hill
Education (India) Private Limited, New Delhi.
25. Raychaudhuri, A.K., Classical MechanicsA Course of Lectures, Oxford
University Press, Calcutta.
26. Sagdeev, R.Z., Usikov, D.A., and Zaslavsky, G.M., (1988) Nonlinear
Physics: From the Pendulum to Turbulence to Chaos, Harwood Aca-
demic Publishers, Pennsylvania.
27. Scheck, F., (1994) Mechanics: From Newtons Laws to Deterministic
Chaos, Speinger-Verlag, Heidelberg.

28. Simmons, G.F., (1974) Differential Equations, Tata-McGraw-Hill, New

York.
29. Sommerfeld, A., (1952) Lextures on Theoretical Physics-Mechanics, Aca-
demic Press, Inc., New York.
30. Spigel, M.R., (2006) Theoretical Mechanics, Tata-McGraw-Hill Educa-
tion Private Limited, New Delhi.
 References 255

31. Strogatz, S.H., (2007) Nonlinear Dynamics and Chaos, Levant Books,
Kolkata.

32. Symon, K.R., (1971) Mechanics, Addison-Wesley Publishing Company,

Inc., Philippines.
33. Whittaker, E.T. (1988) A Treatise on the Analytical Dynamics of Parti-
cles and Rigid Bodies, Cambridge University Press, London.
Index

and variations, 177 damped pendulum problem, 121

variation, 174 degree of freedom, 54
dissipative, 195
absolute space and time, 2 double pendulum, 83
action principle, 161 dynamical system, 117
action variable, 236
angle variable, 236 effective potential, 41, 46
attractor, 120, 145 elliptic trajectories, 122
autonomous dynamical system, 119 equation of the orbit, 41
autonomous system, 119, 141 equation of the separatrix, 124
equilibrium point, 120
bifurcations, 151 EulerLagranges equations, 164
bilateral constraints, 55 extended point transformation, 174
binary star, 68
bounded orbits, 43 fixed point, 120
brachistochrone problem, 180 flow, 119
forced harmonic oscillator, 17
canonical transformations, 215 forces of constraint, 55
Cauchy problem, 171 Foucaults pendulum, 192
centrifugal force, 188
characteristic equation, 125 Galilean group, 4
configuration, 54 Galilean principle of relativity, 5
conservation of angular momentum, Galilean transformations, 3
23 generating function, 221, 236
conservation of energy, 24 geometric constraint, 55
conservation of linear momentum, 22 gravitational constant, 192
conservative force, 11 group, 221
constant of motion, 93 gyroscopic, 195
constraint, 55 gyroscopic force, 195
Coriolis force, 188, 190
Coulomb potential, 45 Hamilton principal function, 230
coupled oscillators, 86 Hamiltons characteristic function,
critical damping, 16, 131 231, 233, 237
critical point, 120 Hamiltons equations of motion, 119
Hamiltons principle, 167
DAlemberts principle, 58, 59 HamiltonJacobi equation, 229, 231
DAlemberts solution, 171 Hamiltonian, 92
damped oscillator, 15, 70

257
258 Index

Hamiltonian canonical equations of Liouvilles theorem, 107, 217

motion, 94 Lipschitz condition, 119
heartbeat model, 136 LotkaVolterra model, 138
higher derivative classical systems, Lyapunov function, 141, 142
111
holonomic, 55 neutrally stable center, 127
homogeneity, 7 Newtons inverse square law of
Hopf bifurcation, 156 gravitation, 36
hyperbolic singularity, 123 Newtons three laws, 2
hyperbolic trajectories, 123 NewtonLaplace principle of
determinacy, 5
ignorable coordinates, 72 Noethers theorem, 200, 203
inertia, 2 non-inertial reference frames, 185
inertial, 2 nonholonomic, 56
inertial frame, 5 nonpotential forces, 194
inertial mass, 4 normal modes, 172
infinite rectangular well, 242
infinitesimal canonical open trajectories, 247
transformations, 228 operator approach, 205
infinitesimal rotation, 202 orbit, 119
invariance transformation, 165 orbit equation for the inverse square
invariance transformations, 200 law problem, 45
isochronicity, 21 oscillatory, 235
isochronous, 13 overdamped, 17
isocline, 120 overdamping, 131
isotropic, 7
isotropic oscillator, 48 PaisUhlenbeck oscillator, 112
parity, 207
Jacobi elliptic functions, 27 periodic trajectories, 243
Jacobian, 134 perturbative analysis, 28
phase space, 120
Keplers laws, 35 pitchfork bifurcation, 154
kinematical constraint, 55 plane pendulum, 66
kinetic energy of a holonomic planetary problem, 75, 99
system, 60 PoincareBendixson theorem, 150
point attractor, 120
Lagranges equations of motion, 62 Poisson bracket, 101, 215
Lagrangian, 31 Poisson theorem, 104
LaplaceRungeLenz (LRL) vector, power, 195
49 power law potentials, 42
Legendre transformation, 94 principle of equivalence, 38
liberation, 235 principle of least action, 176
limit cycle, 145 principle of linear superposition, 6
LindstedtPoincare method, 28 principle of shortest path, 167
linear system, 125 principle of stationary action, 163
Liouvilles class of Lagrangians, 76 properties of Poisson bracket, 102
 Index 259

quartic oscillator, 28 Type III canonical transformation,

225
repeller, 120 Type III generating function, 222
rheonomic, 55 Type IV canonical transformation,
rotating frame, 186 225
Routhian, 72 Type IV generating function, 222
saddle, 123 underdamping, 16, 131
saddle-node bifurcation, 152 uniform string, 169
scleronomic, 55 unilateral constraints, 55
separatrix, 124 unstable degenerate node, 129
simple harmonic motion, 12 unstable focus, 127
simple pendulum, 20, 26, 124 unstable node, 127
singular Lagrangians, 109 unstable saddle point, 129
sink, 120, 126 unstable spiral point, 127
small oscillations, 80 unstable star node, 129
source, 120, 126
spherical pendulum, 67 Van der Pol oscillator, 145
stability of the circular orbit, 43 virial, 209
stable degenerate node, 129 Virial theorem, 209
stable equilibrium, 123 virtual displacement, 59
stable focus or a stable spiral point,
127
stable node, 126
stable star node, 127
state space, 120
stationary point, 120
symmetries, 199
symplectic matrix, 219
symplectic property, 218

time-dependent HamiltonJacobi
equation, 230
time-independent HamiltonJacobi
equation, 233
time-reversal, 208
torque, 23
total derivative term, 65
trajectory, 119
transcritical bifurcation, 153
Type I canonical transformation, 223
Type I generating function, 222
Type II canonical transformation,
224
Type II generating function, 222