Modeling and Simulation in Science, Engineering, and Technology

Series Editor
Nicola Bellomo
Politecnico di Torino
Torino, Italy

Editorial Advisory Board

M. Avellaneda P. Koumoutsakos
Courant Institute of Mathematical Sciences Computational Science &
New York University Engineering Laboratory
New York, NY, USA ETH Zürich
Zürich, Switzerland

K.J. Bathe H.G. Othmer

Department of Mechanical Engineering Department of Mathematics
Massachusetts Institute of Technology University of Minnesota
Cambridge, MA, USA Minneapolis, MN, USA

A. Bertozzi L. Preziosi
Department of Mathematics Dipartimento di Matematica
University of California Los Angleles Politecnico di Torino
Los Angeles, CA, USA Torino, Italy

P. Degond K.R. Rajagopal

Mathématiques pour l’Industrie Department of Mechanical Engineering
et la Physique Texas A & M University
Université P. Sabatier Toulouse 3 College Station, TX, USA
Toulouse, France

A. Deutsch Y. Sone
Center for Information Services Department of Aeronautical Engineering
and High Performance Computing Kyoto University
Technische Universität Dresden Kyoto, Japan
Dresden, Germany

M.A. Herrero Garcia

Departamento de Matematica Aplicada
Universidad Complutense de Madrid
Madrid, Spain

For further volumes:

http://www.springer.com/series/4960
Antonio Romano

Classical Mechanics
with Mathematicar
Antonio Romano
Università degli Studi di Napoli
Dipartimento di Matematica e Applicazioni
Napoli, Italia

Please note that additional material for this book can be downloaded from
http://extras.springer.com

ISSN 2164-3679 ISSN 2164-3725 (electronic)

ISBN 978-0-8176-8351-1 ISBN 978-0-8176-8352-8 (eBook)
DOI 10.1007/978-0-8176-8352-8
Springer New York Heidelberg Dordrecht London
Library of Congress Control Number: 2012941619

Mathematics Subject Classification (2010): 70–01, 70E05, 70E17, 70H03, 70H05, 70H06, 70H09,
70H15, 70H20, 70H25, 70H30, 76–00

© Springer Science+Business Media New York 2012

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed. Exempted from this legal reservation are brief excerpts in connection
with reviews or scholarly analysis or material supplied specifically for the purpose of being entered
and executed on a computer system, for exclusive use by the purchaser of the work. Duplication of
this publication or parts thereof is permitted only under the provisions of the Copyright Law of the
Publisher’s location, in its current version, and permission for use must always be obtained from Springer.
Permissions for use may be obtained through RightsLink at the Copyright Clearance Center. Violations
are liable to prosecution under the respective Copyright Law.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
While the advice and information in this book are believed to be true and accurate at the date of
publication, neither the authors nor the editors nor the publisher can accept any legal responsibility for
any errors or omissions that may be made. The publisher makes no warranty, express or implied, with
respect to the material contained herein.

Printed on acid-free paper

Springer is part of Springer Science+Business Media

(www.birkhauser-science.com)
Preface

Books on classical mechanics can roughly be divided into two classes. Books of
the first class present the subject using differential calculus and analysis, whereas
books of the second class discuss the subject by resorting to the advanced language
of differential geometry. The former books, though more accessible to readers
with a standard background in calculus, do not allow for a deep understanding of
the geometric structures underlying modern mathematical modeling of mechanical
phenomena. On the other hand, books of the second class, requiring a wide
knowledge of differential geometry, are accessible to readers who are already
acquainted with the many aspects of mechanics but wish to understand the most
modern developments in the subject. The present book aims to bridge the gap
between these two classes.
Before discussing the contents of the book in detail, I wish to clarify why I
decided to follow a historical approach to presenting mechanics. Over the long
period (35 years) during which I taught mechanics to students in physics at the
University of Naples Federico II, I saw that students could grasp the fundamentals
of the subject if they could understand the historical process that led to the
comprehension of mechanical phenomena. This process is like a staircase: as you
climb the stairs, each successive step gives you an increasingly broader perspective.
To take in a wider horizon, you must go up to the next step. Then, I believe, students
needed to remain on a given step until they reached the limit of their perspective
and felt the urge to climb to the next step. Clearly, students would achieve the
broadest possible view if they started on the last step. However, this approach would
allow for, at most, a technical comprehension of the subject without a profound
understanding of the roots out of which grew the great tree of classical mechanics.
In the history of science, mechanics was the first to resort to mathematical models
in an attempt to describe the reality around us. Already Leonardo da Vinci (1452–
1519), conscious of this feature of mechanics, stated that mechanics was heaven
for mathematics. Later, Galileo Galilei (1564–1642), having discovered many of
the fundamental principles of mechanical theory, stated that Nature was written
in the language of geometry. The first almost complete picture of the mechanical
world is due to Isaac Newton (1642–1727), heir to Galileo and Johannes Kepler

v
vi Preface

(1571–1630). He expressed his gratitude to these two great minds stating: “If I
have seen farther it is by standing on the shoulders of giants.” In proposing his
fundamental laws of dynamics, Newton understood that the mathematics of his
time was not up to the task of solving the problems posed by his description of
the mechanical world. To make explicit some of the consequences of his laws, he
laid down, together with Gottfried Wilhelm Leibniz (1646–1716), the foundations
of infinitesimal analysis. However, believing that this new approach was too difficult
for his contemporaries to understand, he adopted the language of geometry in
writing his magnum opus, Philosophiae Naturalis Principia Mathematica.
Although Newton’s laws, in principle, describe the physical behavior of any
mechanical system, the description appears in such an implicit form that drawing it
out is a very difficult task. We could say that the history of mechanics from Newton
to our day is the history of the process of making explicit a part of the hidden content
of Newton’s laws. In this process of making things explicit, powerful mathematical
descriptions of mechanical systems have been discovered that can be applied to
many other branches of physics.
The first step in this process was made by Leonhard Euler (1707–1783), who
introduced many notations and definitions still in use today. After formulating the
fundamental balance equations of mechanics, he proposed a model of rigid bodies.
This model introduced an extension to Newton’s model, which was essentially
conceived with respect to material points, in particular for the solar system.
Further, Euler formulated the balance equations of perfect fluids, i.e., of particular
deformable and extended bodies, laying down the foundations of fluidodynamics.
The systematic treatment of a system S of rigid bodies subject to smooth
constraints is due to Joseph-Louis Lagrange (1736–1813), who in his famous
treatise Mécanique analytique reduced the analysis of the dynamical behavior of
such a system to determining a curve of <n whose parametric equations q h .t/
are solutions of the famous Lagrange equations. These second-order differential
equations possess the following advantages:
• They contain the lowest number of parameters needed to determine the position
of S , the so-called Lagrangian coordinates of S .
• They allow one to determine the motion of S without knowing the reactive forces
due to the presence of constraints.
With Lagrange, mechanics resolutely opted for the language of analysis.
William Rowan Hamilton (1805–1865) and Carl Gustav Jacob Jacobi (1804–
1851) proposed a new description of the dynamics of a system S of rigid bodies
with smooth constraints and acted upon by conservative forces. This description
is obtained by the Legendre map, which transforms the solutions of the Lagrange
equations into curves belonging to the phase space and satisfying Hamilton’s equa-
tions. Adopting this formalism, it is possible to determine some general properties
of any mechanical system. Finally, the Hamiltonian formalism is fundamental in
geometric optics, statistical mechanics, and quantum mechanics.
In the early twentieth century, the Hamiltonian formulation of mechanics
underwent a profound change brought about by a new geometric interpretation of its
Preface vii

structures. This new point of view gave rise to symplectic mechanics, which, using
the powerful instruments of differential geometry, shed new light on many aspects
of mechanics and allowed for the solution of some open problems. Today mechanics
is again written in the language of (differential) geometry.
This book surveys all the aforementioned models showing their field of appli-
cation and their limitations. It is divided into two parts. The first part contains an
introduction to linear algebra and differential geometry. In presenting this subject,
I preferred to put forward as evidence the concrete meaning of geometric concepts,
often sacrificing the formal aspects. Chapter 1 contains a brief introduction to vector
spaces and linear mappings. Chapters 2 and 3 deal with tensor and exterior algebras.
Euclidean and symplectic spaces are discussed in Chaps. 4 and 5. Chapters 6–9 con-
tain some fundamental concepts of differential geometry: manifolds, tensor fields,
one-parameter groups of diffeomorphisms, exterior differentiation and integration,
and affine connections. Finally, Chap. 10 presents some elements of dynamical
systems. Chapters 1, 2, 4, and 10 are sufficient for understanding Chaps. 11–16. The
remaining chapters of the first part are necessary for understanding the Lagrangian
and Hamiltonian formalisms (Chaps. 17–22) of the second part.
The second part, starting with Chap. 11, is devoted to mechanics. More precisely,
the kinematics of a material point and a rigid body are presented, respectively,
in Chaps. 11 and 12. Newton’s laws and some general theorems of dynamics are
discussed in Chap. 13, whereas the applications of these laws to a single material
point or to systems of material points are considered in Chap. 14. Further, Chaps. 15
and 16 are devoted to the dynamics of a rigid body and its applications. Lagrange’s
coordinate formalism on a configuration manifold is introduced in Chap. 17, which
also includes a geometric formulation of Maupertuis’ (Pierre-Louis Moreau de
Maupertuis, 1698–1759) principle, an analysis of the principal oscillations of a
mechanical system about a stable equilibrium position, and a transcription of the
Lagrange equations in the velocity space.
Some fundamental aspects of Hamiltonian mechanics are treated in Chap. 18.
First, it is shown that Legendre’s (Adrien-Marie Legendre, 1752–1833) transforma-
tion maps a phase portrait of the solutions of the Lagrange equations into a phase
portrait of the solutions of Hamilton’s equations. When the dynamics is independent
of time, the existence on the phase state of a symplectic structure characterizing the
canonical or symplectic coordinates is demonstrated. Then, the Hamilton–Jacobi
theory about these coordinates is presented in Chap. 19. A description of the
behavior of completely integrable Hamiltonian systems can be found in Chap. 20.
This chapter also contains an introduction to Hamiltonian perturbation theory whose
purpose is to determine the approximate behavior of mechanical systems that are
almost completely integrable systems. In particular, this theory shows that small
perturbations may generate large effects and that chaos lurks in the corner of many
mechanical systems.
An elementary introduction to Maxwell’s (James Clerk Maxwell, 1831–1879)
kinetic theory of perfect gases and to Gibbs’s (Josiah Willard Gibbs, 1839–1903)
formulation of the statistical mechanics of equilibrium can be found in Chap. 21.
In Chap. 22, the balance equations and the Lagrange equations are extended to
viii Preface

impulsive dynamics. Chapter 23 contains the basic elements of the dynamics of a

perfect fluid. This chapter appears in the book simply to show how we must modify
the axioms of mechanics of rigid bodies to take into account the deformability of
real bodies when they are acted upon by forces. Finally, an introduction to special
relativity and its four-dimensional formulation can be found in Chap. 24.
The book contains approximately 200 exercises. Further, it provides the names
of many notebooks, written using Mathematica, that are relevant to several of the
book’s chapters. These notebooks have the twofold aim of showing the possibilities
of this software and of helping the reader to manage some difficult problems of
dynamics.
The notebooks referring to the first part are Geometry.nb, Weierstrass.nb,
Phase2D.nb, and LinStab.nb. The other notebooks relate to the second part.
Contents

Part I Introduction to Linear Algebra and Differential

Geometry

1 Vector Space and Linear Maps . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 3

1.1 Definition of Vector Space.. . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 3
1.2 Dimension of a Vector Space .. . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 5
1.3 Basis Changes .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 7
1.4 Vector Subspaces.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 7
1.5 Algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 8
1.6 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 9
1.7 Linear Maps .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 11
1.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 13
2 Tensor Algebra .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 17
2.1 Linear Forms and Dual Vector Space . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 17
2.2 Biduality .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 19
2.3 Covariant 2-Tensors.. . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 20
2.4 .r; s/-Tensors .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 24
2.5 Tensor Algebra .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 25
2.6 Contraction and Contracted Multiplication .. . .. . . . . . . . . . . . . . . . . . . . 25
2.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 27
3 Skew-Symmetric Tensors and Exterior Algebra . . . .. . . . . . . . . . . . . . . . . . . . 29
3.1 Skew-Symmetric (0, 2)-Tensors .. . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 29
3.2 Skew-Symmetric .0; r/-Tensors.. . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 33
3.3 Exterior Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 36
3.4 Oriented Vector Spaces . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 37
3.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 39
4 Euclidean and Symplectic Vector Spaces . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 41
4.1 Representation Theorems for Symmetric
and Skew-Symmetric .0; 2/-Tensors .. . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 41

ix
x Contents

4.2 Degenerate and Nondegenerate .0; 2/-Tensors . . . . . . . . . . . . . . . . . . . . 44

4.3 Pseudo-Euclidean Vector Spaces . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 47
4.4 Euclidean Vector Spaces.. . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 48
4.5 Eigenvectors of Euclidean 2-Tensors . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 51
4.6 Orthogonal Transformations . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 53
4.7 Symplectic Vector Spaces . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 56
4.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 58
5 Duality and Euclidean Tensors . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 61
5.1 Duality .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 61
5.2 Euclidean Tensors .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 62
5.3 The Levi–Civita Tensor .. . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 63
5.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 65
6 Differentiable Manifolds .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 67
6.1 Differentiable Manifolds . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 67
6.2 Differentiable Functions and Curves on Manifolds . . . . . . . . . . . . . . . 71
6.3 Tangent Vector Space . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 72
6.4 Cotangent Vector Space . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 75
6.5 Differential and Codifferential of a Map. . . . . . .. . . . . . . . . . . . . . . . . . . . 78
6.6 Tangent and Cotangent Fiber Bundles . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 81
6.7 Riemannian Manifolds.. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 82
6.8 Geodesics of a Riemannian Manifold.. . . . . . . . .. . . . . . . . . . . . . . . . . . . . 84
6.9 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 88
7 One-Parameter Groups of Diffeomorphisms . . . . . . . .. . . . . . . . . . . . . . . . . . . . 95
7.1 Global and Local One-Parameter Groups . . . . .. . . . . . . . . . . . . . . . . . . . 95
7.2 Lie’s Derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 97
7.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 101
8 Exterior Derivative and Integration .. . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 105
8.1 Exterior Derivative .. . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 105
8.2 Closed and Exact Differential Forms . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 107
8.3 Properties of Exterior Derivative .. . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 109
8.4 An Introduction to the Integration of r-Forms . . . . . . . . . . . . . . . . . . . . 110
8.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 114
9 Absolute Differential Calculus . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 117
9.1 Preliminary Considerations.. . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 117
9.2 Affine Connection on Manifolds .. . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 120
9.3 Parallel Transport and Autoparallel Curves . . .. . . . . . . . . . . . . . . . . . . . 121
9.4 Covariant Differential of Tensor Fields . . . . . . . .. . . . . . . . . . . . . . . . . . . . 123
9.5 Torsion and Curvature Tensors . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 124
9.6 Riemannian Connection . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 128
9.7 Differential Operators on a Riemannian Manifold . . . . . . . . . . . . . . . . 130
9.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 132
Contents xi

10 An Overview of Dynamical Systems . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 135

10.1 Modeling and Dynamical Systems . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 135
10.2 General Definitions and Cauchy’s Problem . . .. . . . . . . . . . . . . . . . . . . . 138
10.3 Preliminary Considerations About Stability . . .. . . . . . . . . . . . . . . . . . . . 140
10.4 Definitions of Stability. . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 141
10.5 Analysis of Stability: The Direct Method.. . . . .. . . . . . . . . . . . . . . . . . . . 144
10.6 Poincare’s Perturbation Method .. . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 148
10.7 Introducing the Small Parameter .. . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 151
10.8 Weierstrass’ Qualitative Analysis . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 153
10.9 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 156

Part II Mechanics

11 Kinematics of a Point Particle .. . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 161

11.1 Space-Time Frames of Reference of Classical Kinematics .. . . . . . 161
11.2 Trajectory of a Point Particle .. . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 162
11.3 Velocity and Acceleration . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 163
11.4 Velocity and Acceleration in Plane Motions. . .. . . . . . . . . . . . . . . . . . . . 165
11.5 Circular Motion and Harmonic Motion .. . . . . . .. . . . . . . . . . . . . . . . . . . . 168
11.6 Compound Motions .. . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 170
11.7 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 174
12 Kinematics of Rigid Bodies . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 177
12.1 Change of the Frame of Reference .. . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 177
12.2 Velocity Field of a Rigid Motion .. . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 179
12.3 Translational and Rotational Motions.. . . . . . . . .. . . . . . . . . . . . . . . . . . . . 182
12.4 Spherical Rigid Motions .. . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 185
12.5 Rigid Planar Motions . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 188
12.6 Two Examples.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 188
12.7 Relative Motions .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 190
12.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 194
13 Principles of Dynamics .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 197
13.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 197
13.2 Principle of Inertia and Inertial Frames . . . . . . . .. . . . . . . . . . . . . . . . . . . . 198
13.3 Dynamical Interactions and Force Laws . . . . . . .. . . . . . . . . . . . . . . . . . . . 199
13.4 Newton’s Second Law . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 203
13.5 Mechanical Determinism and Galilean Relativity Principle .. . . . . 204
13.6 Balance Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 205
13.7 Kinetic Energy and König’s Theorem . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 208
13.8 Kinetic Energy Theorem . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 209
13.9 Dynamics in Noninertial Frames . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 210
13.10 Constrained Dynamics .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 211
13.11 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 213
xii Contents

14 Dynamics of a Material Point . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 215

14.1 Central Forces .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 215
14.2 Newtonian Forces .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 218
14.3 Two-Body Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 222
14.4 Kepler’s Laws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 224
14.5 Scattering by Newtonian Forces . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 226
14.6 Vertical Motion of a Heavy Particle in Air . . . .. . . . . . . . . . . . . . . . . . . . 227
14.7 Curvilinear Motion of a Heavy Particle in Air . . . . . . . . . . . . . . . . . . . . 230
14.8 Terrestrial Dynamics .. . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 234
14.9 Simple Pendulum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 237
14.10 Rotating Simple Pendulum .. . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 239
14.11 Foucault’s Pendulum.. . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 242
14.12 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 245
15 General Principles of Rigid Body Dynamics . . . . . . . .. . . . . . . . . . . . . . . . . . . . 247
15.1 Mass Density and Center of Mass. . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 247
15.2 Linear Momentum, Angular Momentum, and Kinetic
Energy of a Rigid Body.. . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 248
15.3 Tensor of Inertia.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 249
15.4 Some Properties of the Tensor of Inertia.. . . . . .. . . . . . . . . . . . . . . . . . . . 252
15.5 Ellipsoid of Inertia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 254
15.6 Active and Reactive Forces .. . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 256
15.7 Balance Equations for a Rigid Body: Theorem of
Kinetic Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 258
15.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 260
16 Dynamics of a Rigid Body . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 261
16.1 Rigid Body with a Smooth Fixed Axis . . . . . . . .. . . . . . . . . . . . . . . . . . . . 261
16.2 Compound Pendulum .. . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 263
16.3 Solid with a Fixed Point: Euler’s Equations . . .. . . . . . . . . . . . . . . . . . . . 264
16.4 Poinsot’s Motions .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 265
16.5 Uniform Rotations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 268
16.6 Poinsot’s Motions in a Gyroscope . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 269
16.7 Heavy Gyroscope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 271
16.8 Some Remarks on the Heavy Gyroscope . . . . . .. . . . . . . . . . . . . . . . . . . . 276
16.9 Torque of Terrestrial Fictitious Forces . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 280
16.10 Foucault’s Gyroscope and Gyroscope Compass.. . . . . . . . . . . . . . . . . . 282
16.11 Free Heavy Solid .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 285
17 Lagrangian Dynamics .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 287
17.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 287
17.2 Dynamics of Constrained Systems . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 288
17.3 Configuration Space . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 289
17.4 Virtual Velocities and Nonholonomic Constraints . . . . . . . . . . . . . . . . 290
17.5 Configuration Space-Time.. . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 293
17.6 Simple Example.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 294
Contents xiii

17.7 Principle of Virtual Power .. . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 296

17.8 Lagrange’s Equations .. . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 298
17.9 Conservative Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 301
17.10 First Integrals.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 304
17.11 Symmetries and First Integrals .. . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 306
17.12 Lagrange’s Equations for Nonholonomic Constraints . . . . . . . . . . . . 311
17.13 Small Oscillations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 313
17.14 Hamilton’s Principle .. . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 316
17.15 Geometric Formulations of Lagrangian Dynamics .. . . . . . . . . . . . . . . 321
17.16 Legendre’s Transformation .. . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 324
17.17 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 329
18 Hamiltonian Dynamics .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 337
18.1 Symplectic Structure of the Phase Space . . . . . .. . . . . . . . . . . . . . . . . . . . 337
18.2 Canonical Coordinates .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 339
18.3 Generating Functions of Canonical Transformations . . . . . . . . . . . . . 341
18.4 Isomorphism Between Vector Fields and Differential Forms .. . . . 343
18.5 Global and Local Hamiltonian Vector Fields . .. . . . . . . . . . . . . . . . . . . . 345
18.6 Poisson Brackets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 346
18.7 First Integrals and Symmetries .. . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 348
18.8 Poincaré’s Absolute and Relative Integral Invariants . . . . . . . . . . . . . 351
18.9 Two Fundamental Theorems . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 353
18.10 Volume Form on Invariant Submanifolds .. . . . .. . . . . . . . . . . . . . . . . . . . 356
18.11 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 359
19 The Hamilton-Jacobi Theory.. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 361
19.1 The Hamilton–Jacobi Equation . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 361
19.2 Separation Method .. . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 364
19.3 Examples of Separable Systems. . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 366
19.4 Hamilton’s Principal Function . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 369
20 Completely Integrable Systems . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 377
20.1 Arnold-Liouville’s Theorem.. . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 377
20.2 Angle-Action Variables .. . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 382
20.3 Fundamental Frequencies and Orbits . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 386
20.4 Angle-Action Variables of the Harmonic Oscillator . . . . . . . . . . . . . . 389
20.5 Angle-Action Variables for a Material Point
in Newtonian Fields. . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 390
20.6 Bohr–Sommerfeld Quantization Rules . . . . . . . .. . . . . . . . . . . . . . . . . . . . 394
20.7 A Sketch of the Hamiltonian Perturbation Theory . . . . . . . . . . . . . . . . 396
20.8 Overview of KAM Theorem . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 400
21 Elements of Statistical Mechanics of Equilibrium . .. . . . . . . . . . . . . . . . . . . . 403
21.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 403
21.2 Kinetic Theory of Gases . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 404
21.3 Boltzmann–Gibbs Distribution .. . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 408
21.4 Equipartition of Energy .. . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 412
xiv Contents

22 Impulsive Dynamics .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 415

22.1 Balance Equations of Impulsive Dynamics .. . .. . . . . . . . . . . . . . . . . . . . 415
22.2 Fundamental Problem of Impulsive Dynamics .. . . . . . . . . . . . . . . . . . . 417
22.3 Lagrangian Formulation of Impulsive Dynamics . . . . . . . . . . . . . . . . . 419
22.4 Analysis of the Lagrange Equations . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 421
22.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 422
23 Introduction to Fluid Mechanics . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 427
23.1 Kinematics of a Continuous System . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 427
23.2 Velocity Gradient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 429
23.3 Rigid, Irrotational, and Isochoric Motions .. . . .. . . . . . . . . . . . . . . . . . . . 430
23.4 Mass Conservation .. . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 432
23.5 Equations of Balance of a Perfect Fluid . . . . . . .. . . . . . . . . . . . . . . . . . . . 434
23.6 Statics of Fluids .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 436
23.7 Fundamental Theorems of Fluid Dynamics . . .. . . . . . . . . . . . . . . . . . . . 439
23.8 Boundary Value Problems for a Perfect Fluid .. . . . . . . . . . . . . . . . . . . . 444
23.9 Two-Dimensional Steady Flow of a Perfect Fluid.. . . . . . . . . . . . . . . . 445
23.10 D’Alembert’s Paradox and the Kutta–Joukowky Theorem . . . . . . . 453
23.11 Waves in Perfect Gases . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 456
24 An Introduction to Special Relativity . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 459
24.1 Galilean Relativity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 459
24.2 Optical Isotropy Principle . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 462
24.3 Simple Consequences of Lorentz’s Transformations.. . . . . . . . . . . . . 466
24.4 Relativistic Composition of Velocities and Accelerations . . . . . . . . 468
24.5 Principle of Relativity . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 470
24.6 Covariance of Maxwell’s Equations . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 472
24.7 Relativistic Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 474
24.8 Minkowski’s Space-Time .. . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 476
24.9 Physical Meaning of Minkowski’s Space-Time . . . . . . . . . . . . . . . . . . . 480
24.10 Four-Dimensional Equation of Motion . . . . . . . .. . . . . . . . . . . . . . . . . . . . 482
24.11 Tensor Formulation of Electromagnetism in Vacuum .. . . . . . . . . . . . 484

Appendix A First-Order PDE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 487

A.1 Monge’s Cone .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 487
A.2 Characteristic Strips . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 489
A.3 Cauchy’s Problem .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 492

Appendix B Fourier Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 495

References .. .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 499

Index . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 501
 Part I
Introduction to Linear Algebra and
Differential Geometry
Chapter 1
Vector Space and Linear Maps

Vectors are usually introduced with a comment that many physical quantities
(e.g., displacements, velocities, accelerations, forces, and torques) are conveniently
described by oriented segments characterized by intensity, direction, and versus.
Then, a vector space E is defined as the set of the oriented segments starting from
a given point O. Algebraic operations are introduced in the set E as the addition
x C y of two vectors x; y 2 E, the multiplication ax of a real number a by a vector
x, the scalar product x  y, and the cross or vector product x  y.
In this chapter, in agreement with the spirit of modern mathematics, vectors will
be defined as algebraic objects, i.e., as quantities that are not characterized by their
nature but by the algebraic operations that are defined on them. This approach,
which is followed in many fields of mathematics (for instance, in the theory of
numbers), leads us to attribute a more general meaning to a vector.1

1.1 Definition of Vector Space

Let < be the field of real numbers whose elements we denote by lowercase Latin
letters a, b; : : :, and let E be an arbitrary set whose elements we denote by bold
Latin letters x, y; : : :. In the set E we assign an internal composition operation,
called addition, denoted by C,
.x; y/ 2 E  E ! x C y 2 E; (1.1)
and satisfying the following properties:
x C .y C z/ D .x C y/ C z; (1.2)
x C y D y C x; (1.3)

1
The topics contained in the Chapters 1–9 can also be found in [2, 5, 7–9, 11, 13, 14, 16–18, 21, 52].

A. Romano, Classical Mechanics with Mathematica R

, Modeling and Simulation 3
in Science, Engineering and Technology, DOI 10.1007/978-0-8176-8352-8 1,
© Springer Science+Business Media New York 2012
4 1 Vector Space and Linear Maps

90 2 E W x C 0 D x; (1.4)
8x 2 E; 9.x/ 2 E W x C .x/ D 0; (1.5)
8x; y; z 2 E.
Besides the addition, we suppose that another composition law is defined by

.a; x/ 2 E ! ax 2 E; (1.6)
which associates to any pair .a; x/, with a 2 < and x 2 E, a new element ax 2 E
in such a way that the following properties are satisfied:

.ab/x D a.bx/; (1.7)

a.x C y/ D ax C ay; (1.8)
.a C b/x D ax C bx; (1.9)
1x D x; (1.10)

8a; b 2 < and 8x; y 2 E.

Definition 1.1. The set E, equipped with the above operations, is said to be a real
vector space or a vector space on <. The elements of E and < are respectively
called vectors and scalars. The vector x C y is the sum of the vectors x and y; the
vector ax is the product of the scalar a by the vector x. Finally, the vector 0 is said
to be the zero vector and x is the opposite vector of x.
Henceforth, the sum x C .y/ will be denoted by x  y.
Proposition 1.1. The zero vector is unique.
Proof. If there are two zero vectors 0 and 00 , then the result is

x C 0 D x; 8x 2 E;
so that, in particular, it is 00 C 0 D 00 . On the other hand, from the relation
x C 00 D x; 8x 2 E;

it follows that 0 C 00 D 0. From (1.3) we conclude that 00 D 0. 

Proposition 1.2. The opposite vector x of any x 2 E is uniquely determined.
Proof. If x0 and x00 are two opposite vectors of x, then it is

x0 C x D x00 C x D 0:
On the other hand, in view of (1.2), we have that

.x0 C x/ C x00 D x0 C .x C x00 /;

i.e., x0 D x00 . 
1.2 Dimension of a Vector Space 5

In a similar way, it is possible to prove the following propositions.

Proposition 1.3. 8x 2 E and a 2 <, it is
0x D 0; a0 D 0; .a/x D .ax/:

Proposition 1.4. 8x1 ; : : : ; xn 2 E and a 2 <, which yields

a.x1 C    C xn / D ax1 C    C xn :

1.2 Dimension of a Vector Space

Definition 1.2. Let E be a vector space and let x1 ; : : : ; xn be vectors of E. The

vectors x1 ; : : : ; xn are said to be linearly dependent if there exist nonzero real
numbers a1 ; : : : ; an such that

a1 x1 C    C an xn D 0: (1.11)

In contrast, if (1.11) implies a1 D    D an D 0, then we say that x1 ; : : : ; xn are

linearly independent or, equivalently, that they form a free system of order n.
It is evident that the vector 0 does not belong to a free system.
Definition 1.3. The vector space E is said to have a finite dimension n > 0
on < if
1. There exists at least a free system of vectors of order n;
2. Every system of vectors of order n C 1 is linearly dependent.
If E contains free systems of vectors of any order, then E is said to have infinite
dimension.
Throughout the book we only consider vector spaces with finite dimension.
Definition 1.4. Let En be a vector space with finite dimension n. Any free vector
system fe1 ; : : : ; en g of order n is a basis of En .
Remark 1.1. Henceforth, we denote by .ei / the vector set fe1 ; : : : ; en g; further, we
adopt the Einstein convention according to which a summation is understood over
a pair of upper and lower indices denoted by the same symbol. For instance, the
relation a1 x1 C    C an xn will be written in the compact form ai xi . The index i ,
which is summed over, is called a dummy index. In a computation, such an index
can be denoted by any Latin letter.
Theorem 1.1. A vector system .ei / is a basis of the vector space En if and only if
for any x 2 En it yields
x D x i ei ; (1.12)

where the coefficients x i are uniquely determined.

6 1 Vector Space and Linear Maps

Proof. If .ei / is a basis, then the system .x; .ei // is linearly dependent so that n C 1
scalars .a; ai / exist such that

ax C ai ei D 0: (1.13)

But a ¤ 0 since, otherwise, we would have ai ei D 0, with some scalars ai different

from zero, against the hypothesis that .ei / is a basis. Dividing (1.13) by a, we
obtain (1.12). To prove that the coefficients x i in (1.12) are uniquely determined,
we suppose that, besides the decomposition (1.12), we also have

x D x i ei :

Subtracting this decomposition from (1.12), we obtain the relation 0 D .x i  x i /ei ,

which implies x i D x i , since .ei / is a basis.
If (1.12) is supposed to hold 8x 2 En , then, for x D 0, we obtain

0 D x i ei :

Since the quantities xi are uniquely determined, this relation implies that x 1 D    D
x n D 0 and the vector system .ei / is free. To prove that .ei / is a basis, we must verify
that any vector system .u1 ; : : : ; un ; x/ of order n C 1 is linearly dependent. That is
true if .ui / is linearly dependent. By contrast, if .ui / is free, then from (1.12) we
derive the relation
u1 D ai ei ;
in which at least one of the scalars ai does not vanish since u1 ¤ 0. Supposing that
a1 ¤ 0, the preceding relation leads us to
n
1 1 X i
e1 D u1  a ei :
a1 a1 i D2

In view of this result and (1.12), we can state that any vector x 2 En can be
represented as a linear combination of the vectors .u1 ; e2 ; : : : ; en /. Consequently,
we can say that
u2 D b 1 u1 C b 2 e2 C    C b n en ;

where at least one of the scalars b 2 ; : : : ; b n does not vanish. If fact, if all these
coefficients were equal to zero, then we would have u2 D b 1 u1 , and the vector
system .u1 ; : : : ; un / would be linearly dependent. If b 2 ¤ 0, we can write that

e2 D c 1 u1 C c 2 u2 C c 3 e3 C    C en :

In view of (1.12), this relation implies that any vector x 2 En is a linear combination
of .u1 ; u2 ; e3 ; : : : ; en /. Iterating the procedure, we conclude that (1.12) and the
hypothesis that .ui / is free imply the relation x D x i ui , i.e., x  x i ui D 0. Then, the
vector system .u1 ; : : : ; un ; x/ is linearly dependent and the theorem is proved. 
1.4 Vector Subspaces 7

1.3 Basis Changes

Definition 1.5. The unique coefficients expressing any vector x as a linear combi-
nation of the basis .ei / are called the contravariant components of x in the basis
.ei / or relative to the basis .ei /.
Let .ei / and .e0i / be two bases of the n-dimensional vector space En . Then, x
admits the two representations

x D x i ei D x 0i e0i : (1.14)

Further, since .ei / and .e0i / are bases of En , any vector of the basis .e0i / can be
expressed as a linear combination of the vectors belonging to the basis .ei /, i.e.,
j
e0j D Aij ei ; ei D .A1 /i e0j ; (1.15)

j
where ..A1 /i / is the inverse matrix of .Aij /. Introducing (1.15) into (1.14), we
obtain the condition
j
x D x 0j e0j D .A1 /i x i e0j ;

which, given the uniqueness of the representation of a vector with respect to a basis,
implies the equations
j
x 0j D .A1 /i x i (1.16)

relating the contravariant components of the same vector evaluated in two different
bases. It is evident that the inverse formulae of (1.16) are

x i D Aij x 0j : (1.17)

Notice that the contravariant components of the vector x are transformed with
the inverse matrix of the basis change (1.14).

1.4 Vector Subspaces

Definition 1.6. A subset V  En is a vector subspace of En if

8x; y 2 V; 8a; b 2 < ) ax C by 2 V: (1.18)

It is evident that V is a vector space.

Let .e1 ; : : : ; em /, m  n, be a free vector system of En . It is plain to verify that
the totality V of the vectors

x D x 1 e1 C    C x m em ; (1.19)
8 1 Vector Space and Linear Maps

obtained on varying the coefficients x i in <, represents a vector subspace of En .

Such a subspace is said to be the subspace generated or spanned by the system
.e1 ; : : : ; em /. In particular, any vector space En is generated by any basis of it.
Definition 1.7. Let Up and Vq be two vector subspaces
T of En , having dimensions
p and q, p C q  n, respectively, and such that Up Vq D f0g. Then, we denote by

W D U p ˚ Vq (1.20)

the direct sum of the subspaces Up and Vq , that is, the vector set formed by all the
vectors u C v, where u 2 Up and v 2 Vq .
Theorem 1.2. The direct sum W D Up ˚ Vq of the vector subspaces Up and Vq is
a vector subspace of dimension m D p C q.
Proof. Let .u1 ; : : : ; up / and .v1 ; : : : ; vq / be bases of Up and Vq , respectively. By
Definition 1.7, we can state that none of these vectors belongs to both bases.
Moreover, any w 2 W can be written in the form w D u C v, where u and v
are unique vectors belonging to Up and Vq , respectively. Therefore, we have that
p q
X X
i
wD u ui C vi v i ;
i D1 i D1

where the contravariant components ui and vi are uniquely determined. In conclu-

sion, the vectors .u1 ; : : : ; up ; v1 ; : : : ; vq / form a basis of W . 

1.5 Algebras

Definition 1.8. Let E be a vector space. Let us assign an internal mapping .x; y/ 2
E  E ! xy 2 E, called a product, that for any choice of the scalars a and b and
any choice of the vectors x, y, and z verifies the following properties:

a.xy/ D .ax/y D x.ay/;

.ax C by/z D a.xz/ C b.yz/;
x.ay C bz/ D a.xy/ C b.xz/: (1.21)

The vector space E equipped with a product is an example of an algebra on E.

This algebra is commutative if
xy D yx: (1.22)
A vector e is the unit vector of the product if

xe D ex D x: (1.23)

It is simple to prove that (1.23) implies that e is unique.

1.6 Examples 9

Definition 1.9. We say that the preceding algebra is a Lie algebra if the product
• Is skew-symmetric, i.e.,
xy D yxI (1.24)
• Verifies Jacobi’s identity
x.yz/ C z.xy/ C y.zx/ D 0: (1.25)

1.6 Examples

In this section we present some interesting examples of vector spaces and algebras.
• The vector space of oriented segments. Let E3 be a three-dimensional Euclidean
!
space. We denote by AB an oriented segment starting from the point A 2 E3 and
ending at the point B 2 E3 . In the set of all oriented segments of E3 we introduce
the following equivalence relation R: two oriented segments are equivalent if
they are equipollent. Then we consider the set E D E3 =R and denote by x D
! !
ŒAB the equivalence class of an arbitrary oriented segment AB. Since all these
equivalence classes are in one-to-one correspondence with the oriented segments
!
starting from a fixed point O 2 E, we can introduce the notation x D ŒOA. If
!
y D ŒOB, then we introduce the following operations in E:
!
ax D ŒaOA; (1.26)
! !
x C y D ŒOA C OB; (1.27)
! !
where in (1.27) the oriented segment OAC OB is the diagonal of a parallelogram
! !
with sides OA and OB. It is plain to verify that the foregoing operations equip
E with the structure of a three-dimensional vector space since three unit vectors,
which are orthogonal to each other, form a basis of E.
It is also easy to verify that if we introduce into E the cross product x  y of
two vectors, the vector space E becomes a Lie algebra.
• The vector space of polynomials of degree  n. Denote by E the set of
polynomials P .x/ of degree r, where 0  r  n. Any polynomial can be written
as P .x/ D p0 C p1 x C    C pn x n , where some of its coefficients pi may vanish.
If, for any a 2 <, we define

aP .x/ D a.p0 C p1 x C    C pn x n / D ap0 C ap1 x C    C apn x n ;

P .x/ C Q.x/ D .p0 C p1 x C    C pn x n / C .q0 C q1 x C    C qn x n /
D .p0 C q0 / C .p1 C q1 /x C    C .pn C qn /x n ;
10 1 Vector Space and Linear Maps

then E becomes a vector space on <. Moreover, the n C 1 polynomials

P0 D 1; : : : ; Pn D x n (1.28)

are linearly independent since, from the linear combination

0 P0 .x/ C    C n Pn .x/ D 0

and the fundamental theorem of algebra, it follows that all the real numbers
0 ; : : : ; n vanish. On the other hand, any polynomial of degree n can be
expressed as a linear combination of the n C 1 polynomials (1.28). We can
conclude that E is an .n C 1/-dimensional vector space.
• The vector space of continuous functions on the interval Œa; b. Denote by
C 0 Œa; b the set of continuous functions on the closed and bounded interval
Œa; b. With the ordinary operations of multiplication of a function by a number
and the addition of two functions, C 0 Œa; b becomes a vector space. Since the
polynomials of any degree are continuous functions, the vector space of the
polynomial of degree n, where n is an arbitrary integer, is a vector subspace
of C 0 Œa; b. Consequently, E has infinite dimension since it contains free vector
systems of any order.
• The vector space of n  n matrices. Let E be the set of the n  n matrices
A D .aij /. The operations

aA D .aaij /; A C B D .aij C bij / (1.29)

equip E with the structure of a vector space. It is a very simple exercise to verify
that the n2 matrices
0 1 0 1
1 0  0 0 0  0
@           A ;::: ; @           A
0 0 0 0 0 0 0 1

form a basis of E that is an n2 -dimensional vector space. If we add to the

operations (1.29), consisting in the multiplication of rows by columns the two
matrices n  n
AB D .ai k bkj /; (1.30)
then E becomes a noncommutative algebra since

AB ¤ BA: (1.31)

Finally, with the operation

ŒA; B D AB  BA; (1.32)

E is a Lie algebra. In fact, operation (1.32) is skew-symmetric, and it is a simple

exercise to prove Jacobi’s identity (1.25) (Exercise 1).
1.7 Linear Maps 11

1.7 Linear Maps

Definition 1.10. Let En and Gm be two vector spaces on < with dimensions n and
m, respectively. We say that the map

F W En ! Gm (1.33)

is a linear map or a morphism if

F .ax C by/ D aF .x/ C bF .y/; (1.34)

8a; b 2 <, and 8x, y 2 En .

Henceforth, we denote by Lin.En ; Gm / the set of all linear maps of En into Gm .
Theorem 1.3. A linear map F 2 Lin.En ; Gm / is determined by the vectors F .ei / 2
Gm corresponding to the vectors of a basis .ei / of En .
Proof. In fact, 8x 2 En we have that

y D F .x/ D F .x i ei / D x i F .ei /: (1.35)


Since the vectors F .x/ belong to Gm , they can be represented as a linear
combination of the vectors of a basis .gi / of Gm , i.e.,

F .ei / D Fih gh ; (1.36)

where, for any fixed i , the scalars Fih denote the components of the vector F .ei /
relative to the basis .gh /. Considering the upper index of Fih as a row index and the
lower index as a column index, the numbers Fih become the elements of an m  n
matrix F with m rows and n columns. When the bases .ei / in En and .gh / in Gm are
chosen, such a matrix uniquely determines the linear map F since (1.35) and (1.36)
imply that
y D F .x/ D Fih x i gh : (1.37)
For this reason, F D .Fih / is called the matrix of the linear map F relative to the
bases .ei / and .gh /. Adopting the matrix notation, (1.37) becomes

Y D FX; (1.38)

where Y is the m  1 matrix formed with the components of y and X is the n  1

matrix formed with the components of x.
Definition 1.11. 8F 2 Lin.En ; Gm / the subset

Im.F / D fy 2 Gm j9x 2 En W y D F .x/g (1.39)

12 1 Vector Space and Linear Maps

is called the image of F , whereas the subset

ker.F / D fx 2 En jF .x/ D 0g (1.40)

is the kernel of F .
A linear map (1.33) is an epimorphism if Im.F /  G; further, F is called a
monomorphism if no two different elements of E are sent by F into the same
element of G. In particular, an epimorphism F W En ! En is called endomorphism
and the set of all endomorphisms is denoted by Lin.En /. The linear map (1.33) is an
isomorphism if it is a monomorphism and F .En / D Gm . In this case, there exists
the inverse linear map F 1 W Gm ! En . Finally, an isomorphism F W En ! En is
said to be an automorphism.
Theorem 1.4. Im.F / and ker.F / are vector subspaces of Gm and En , respectively.
Moreover, F is a monomorphism if and only if ker.F / D f0g.
Proof. If y1 , y2 2 Im.F /, then there are x1 , x2 2 En such that y1 D F .x1 / and
y2 D F .x2 /. Then, 8a; b 2 < the linearity of F implies that

ay1 C by2 D aF .x1 / C bF .x2 / D F .ax1 C bx2 /

and ay1 C by2 2 Im.F /. Similarly, if x1 , x2 2 ker.F /, then it follows that F .x1 / D
F .x2 / D 0, so that

0 D aF .x1 / C bF .x2 / D F .ax1 C bx2 /;

and we conclude that ax1 C bx2 2 ker.F /.
It remains to prove the second part of the theorem. Since F .0/ D 0, if the inverse
map F 1 of F exists, then the subspace F 1 .0/ reduces to the vector 0, that is,
ker.F / D 0. In contrast, supposing that this last condition is satisfied, if it is possible
to find two vectors x0 and x00 2 En such that F .x0 / D F .x00 /, then it is also true that
F .x0  x00 / D 0. But this condition implies that
x0  x00 D 0 ) x0  x00 2 Ker.F / ) x0 D x00 ;

and the theorem is proved. 

In particular, from the preceding theorem it follows that

F is an isomorphism , Im.F / D Gm ; ker.F / D 0: (1.41)

Theorem 1.5. Let .ei / be a basis of the vector space En and let .gh / be a basis
of the vector space Gm . For any F 2 Lin.En ; Gm /, the following statements are
equivalent:
(a) F is an isomorphism;
(b) m D n and the vectors F .ei / form a basis of Gm ;
(c) m D n and the matrix F of F relative to the pair of bases .ei / and .gh / is not
singular.
1.8 Exercises 13

Proof. (a)) (b). The condition i F .ei / D F .i ei / D 0 and (1.41) imply that
i ei D 0. Since .ei / is a basis of En , we have that i D 0, i D 1; : : : ; n, and
the vectors .F .ei // are independent. Moreover, if F is an isomorphism, 8y 2
Gm there exists one and only one x 2 Gm such that y D F .x/ D x i F .ei /.
Consequently, the vectors F .ei / form a basis of Gm and m D n.
(b)) (c). In view of (1.37), the relation i F .ei / D 0 becomes i Fih gh D 0. But
(b) implies that the linear system i Fih D 0 has only the solution i D 0,
i D 1; : : : ; n. Consequently, m D n and the matrix F is not singular.
(c)) (a). Equation (1.38) is the matrix form of the mapping F with respect to the
bases .ei / and .gh /. For any choice of the vector y 2 Gm , (1.38) is a linear system
on n equations in the n unknown x i . Owing to (c), this system admits one and
only one solution so that F is an isomorphism. 
We conclude this section with the following remark. Consider a basis change
j
g0h D Ghk gk ; e0i D Ei ej (1.42)

in the two vector spaces En and Gm . Then, the matrix F relative to the bases .eh / and
.gh / and the matrix F0 relative to the bases .e0h / and .g0h / are related by the following
equation:
F0 D G1 FE (1.43)

(Exercise 4).

1.8 Exercises

1. In the vector space En2 of the n  n matrices, prove that Jacobi’s identity is
verified.
We start by noting that

ŒA; ŒB; C D AŒB; C  ŒB; CA

D A.BC  CB/  .BC  CB/A
D ABC  ACB  BCA C CBA:

Jacobi’s identity is proved applying the foregoing identity to the three terms of
the expression ŒA; ŒB; C C ŒC; ŒA; B C ŒB; ŒC; A .
2. Let

e01 D cos 'e1 C sin 'e2 ;

e02 D  sin 'e1 C cos 'e2 (1.44)
14 1 Vector Space and Linear Maps

x
e2
e’2

e1

Fig. 1.1 Two bases in a plane

be a basis change in the vector space E2 (Fig. 1.1). Evaluate the components
.x 0i / in the basis .e0i / of the vector that has the components .1; 2/ in the
basis .ei /.
Hint: It is sufficient to verify that the inverse matrix of the basis change is
 
cos '  sin '
.A1 /ij D
sin ' cos '

and then to apply (1.16).

3. Let .ei /, i D 1; 2, be a basis of the two-dimensional vector space E2 and let
.gh /, h D 1; 2; 3, be a basis of the three-dimensional vector space G3 . Verify
that the ker and the Im of the linear mapping F W E2 ! G3 , whose matrix
relative to the bases .ei / and .gh / is
0 1
1 2
@0 1 A; ˛ 2 <;
˛˛

are given by the following subspaces:

ker.F / D f0g; Im.F / D fy 2 G3 W y D .y 1 ; y 2 ; ˛.y 1 C y 2 /g;

where .y h / are the contravariant components of any vector y 2 G3 .

4. Prove formula (1.43).
Let .ei /, .gh / be a pair of bases in the vector space En and denote by .e0i /,
0
.gh / a pair of bases in the vector spaces Gm . We have that, for any x 2 En ,
the vector y D F .x/ 2 Gm can be written [see (1.36), (1.37)] in the following
forms:

y D y h gh D Fih x i gh ;
y D y 0h g0h D Fi0h x i g0h :

From the foregoing representations of y we have that

Fih x i gh D Fi0h x 0i g0h :

1.8 Exercises 15

Taking into account the basis changes (1.42), this condition becomes

Fih Eli x 0l .G 1 /kh g0k D Fl0k x 0l g0k :

From the arbitrariness of x there follows (1.43).

5. Determine the image and the kernel of the following linear mapping:
0 1 0 10 11
y1 1 0 1 x
@ y 2 A D @ 2 1 0 A @ x2 A :
y3 1 a 2 x3

6. Determine the vectors .a; b; c/ of the vector space V generated by the vectors
u D .2; 1; 0/, v D .1; 1; 2/, and w D .0; 3; 4/.
Hint: We must find the real numbers x, y, and z such that

.a; b; c/ D x.2; 1; 0/Cy.1; 1; 2/Cz.0; 3; 4/ D .2xCy; xyC3z; 2y4z/;

i.e., we must solve the linear system

2x C y D a;
3y  6z D b;
0 D 2a  4b  3c:

Analyze this system.

7. Verify if the vectors .a; b; c/ generate a subspace W when they satisfy one of
the following conditions:
• a D 2b;
• ab D 0;
• a D b D c;
• a D b2.
8. Prove that the polynomials .1  x/3 , .1  x/2 , and .1  x/ generate the vector
space of polynomials of the third degree.
9. Let U be the subspace of <3 generated by the vectors .a; b; 0/, with a and
b arbitrary real numbers. Further, let W be the subspace of <3 generated
T .1; 2; 3/ and .1; 1; 1/. Determine vectors generating the
by the two vectors
intersection U W .
10. Determine the image and the kernel of the following linear map:

.x; y; z/ 2 <3 n ! .x C 2y  z; y C z; x C y  2z/ 2 <3 :

Chapter 2
Tensor Algebra

This chapter contains an introduction to tensor algebra. After defining covectors and
dual bases, the space of covariant two-tensor is introduced. Then, the results derived
for this space are extended to the general space of the .r; s/-tensors.

2.1 Linear Forms and Dual Vector Space

Definition 2.1. Let E be a vector space on <. The map

!WE!< (2.1)
is said to be a linear form, a 1-form, or a covector on E if
!.ax C by/ D a!.x/ C b!.y/; (2.2)
8a, b 2 < and 8x, y 2 E.1
The set E  of all linear forms on E becomes a vector space on < when we define
the sum of two linear forms !,  2 E  and the product of the scalar a 2 < with the
linear form ! in the following way
.! C  /.x/ D !.x/ C  .x/; .a!/.x/ D a !.x/; 8x 2 E: (2.3)

Theorem 2.1. Let En be a vector space with finite dimension n. Then, E  has the
same dimension n. Moreover, if .ei / is a basis of En , then the n covectors such that
 i .ej / D ıji 2 (2.4)

1
For the contents of Chaps. 2–9, see [7, 8, 10, 11, 13, 14].
2
Here ıji is the Kronecker symbol

0; i ¤ j;
ıji D
1; i D j:

A. Romano, Classical Mechanics with Mathematica R

, Modeling and Simulation 17
in Science, Engineering and Technology, DOI 10.1007/978-0-8176-8352-8 2,
© Springer Science+Business Media New York 2012
18 2 Tensor Algebra

define a basis of E  .
Proof. First, we remark that, owing to (2.4), the linear forms  i are defined over the
whole space En since, 8x D x i ei 2 En , we have that
 i .x/ D  i .x j ej / D x j  i .ej / D x i : (2.5)

To show that En is n-dimensional, it is sufficient to verify that any element of En
can be written as a unique linear combination of the covectors . i /. Owing to the
linearity of any ! 2 En , we have that
!.x/ D x i !.ei / D x i !i ; (2.6)
where we have introduced the notation

!i D !.ei /: (2.7)
On the other hand, (2.5) allows us to write (2.6) in the following form:

!.x/ D !i  i .x/; (2.8)

from which, owing to the arbitrariness of x 2 En , it follows that

! D !i  i : (2.9)
To prove the theorem, it remains to verify that the quantities !i in (2.9) are uniquely
determined. If another representation ! D !i0  i existed, then we would have 0 D
.!i  !i0 / i , i.e.,
.!i  !i0 / i .x/ D 0; 8x 2 En :

Finally, from (2.4) it follows that !i0 D !i . 

Remark 2.1. Note that (2.6) gives the value of the linear map ! when it is applied
to the vector x, whereas (2.9) supplies the linear map ! as a linear combination of
the n linear maps  i .
By (2.4), the dual basis  i of En is associated with the basis .ei / of En .
Consequently, to a basis change .ei / ! .e0i / in En expressed by (1.15) corresponds
a basis change . i / ! . 0i / in En . To determine this basis change, since it is

 0i .x/ D x 0i ;  i .x/ D x i ; (2.10)

and (1.16) holds, we have that

 0i .x/ D .A1 /ij x j D .A1 /ij  j .x/:

In view of the arbitrariness of the vector x 2 En , from the preceding relation we

obtain the desired transformation formulae of the dual bases:

 0i D .A1 /ij  j ;  i D Aij  0j : (2.11)

2.2 Biduality 19

The transformation formulae of the components ! i of the linear form !

corresponding to the basis change (2.11) are obtained recalling that, since ! is a
vector of En , its components are transformed with the inverse matrix of the dual-
basis change. Therefore, it is
j j
!i0 D Ai !j ; !i D .A1 /i !j0 : (2.12)

Remark 2.2. Bearing in mind the foregoing results, we can state that the com-
ponents of a covector relative to a dual basis are transformed according to the
covariance law (i.e., as the bases of the vector space En ). By contrast, the dual
bases are transformed according to the contravariance law (i.e., as the components
of a vector x 2 En ).
Remark 2.3. Since the vector spaces En and En have the same dimension, it is
possible to build an isomorphism between them. In fact, choosing a basis .ei / of
En and a basis . i / of En , anPisomorphism between En and En is obtained by
associating the covector ! D i x i  i with the vector x D x i ei 2 En . However,
owing to the different transformation character of the components of a vector and
a covector, the preceding isomorphism depends on the choice of the bases .ei /
and . i /. Later we will show that, when En is a Euclidean vector space, it is
possible to define an isomorphism between En and En that does not depend on
the aforementioned choice, i.e., it is intrinsic.

2.2 Biduality

We have already proved that En is itself a vector space. Consequently, it is possible
to consider its dual vector space En containing all the linear maps G W En ! <.
Moreover, to any basis . i / 2 En we can associate the dual basis .fi / 2 En
defined by the conditions [see (2.4)]
j
fi . j / D ıi ; (2.13)

so that any F 2 En admits a unique representation in this basis:

F D F i fi : (2.14)

At this point we can consider the idea of generating “ad libitum” vector spaces by
the duality definition. But this cannot happen since En and En are isomorphic, i.e.,
they can be identified. In fact, let us consider the linear map

x 2 En ! Fx 2 En

such that
Fx .!/ D !.x/; 8! 2 En : (2.15)
20 2 Tensor Algebra

To verify that (2.15) is an isomorphism, denote by .ei / a basis of En , . i / its dual

basis in En , and fi the dual basis of . i / in En . In these bases, (2.15) assumes
the form
F i !i D x i !i ; (2.16)

which, owing to the arbitrariness of !, implies that

F i D xi ; i D 1; : : : ; n: (2.17)

The correspondence (2.17) refers to the bases .ei /, . i /, and .fi /. To prove that
the isomorphism (2.17) does not depend on the basis, it is sufficient to note that the
basis change (1.15) in En determines the basis change (2.11) in En . But .fi / is the
dual basis of . i /, so that it is transformed according to the formulae
j
f0i D Ai fj :

Consequently, the components x i of x 2 En and the components F i of F 2 En

are transformed in the same way under a basis change.
We conclude by remarking that the foregoing considerations allow us to look at
a vector as a linear map on En . In other words, we can write (2.15) in the following
way:
x.!/ D !.x/: (2.18)

2.3 Covariant 2-Tensors

Definition 2.2. A bilinear map

T W En  En ! <

is called a covariant 2-tensor or a .0; 2/-tensor.

With the following standard definitions of addition of two covariant 2-tensors
and multiplication of a real number a by a covariant 2-tensor

.T1 C T2 /.x; y/ D T1 .x; y/ C T2 .x; y/;

.aT/.x; y/ D aT.x; y/;

the set T2 .En / of all covariant 2-tensors on En becomes a vector space.

Definition 2.3. The tensor product ! ˝  of !;  2 En is a covariant 2-tensor
such that
! ˝  .x; y/ D !.x/ .y/; 8x; y 2 En : (2.19)
2.3 Covariant 2-Tensors 21

Theorem 2.2. Let .ei / be a basis of En and let . i / be the dual basis in En . Then
. i ˝  j / is a basis of T2 .En /, which is a n2 -dimensional vector space.
Proof. Since T 2 T2 .En / is bilinear, we have that

T.x; y/ D T.x i ei ; y j ej / D x i y j T.ei ; ej /:

Introducing the components of the covariant 2-tensor T in the basis . i ˝  j /

Tij D T.ei ; ej /; (2.20)

the foregoing relation becomes

T.x; y/ D Tij x i y j : (2.21)

On the other hand, in view of (2.19) and (2.5), it also holds that

 i ˝  j .x; y/ D x i y j ; (2.22)

and (2.21) assumes the form

T.x; y/ D Tij  i ˝  j .x; y/:

Since this identity holds for any x; y 2 En , we conclude that the set . i ˝  j /
generates the whole vector space T2 .En /:

T D Tij  i ˝  j : (2.23)

Then, the covariant 2-tensors . i ˝  j / form a basis of T2 .En / if they are linearly
independent. To prove this statement, it is sufficient to note that from the arbitrary
linear combination
aij  i ˝  j D 0

we obtain
j
aij  i ˝  j .eh ; ek / D aij ıhi ık D 0;

so that ahk D 0 for any choice of the indices h and k. 

The basis change (1.15) in En determines the basis change (2.11) in En and a
basis change
j j
 0i ˝  0j D .A1 /ih .A1 /k  h ˝  k ;  i ˝  j D Aih Ak  0h ˝  0k (2.24)

in T2 .En /. On the other hand, it also holds that

T D Tij  i ˝  j D Tij0  0i ˝  0j ; (2.25)

22 2 Tensor Algebra

and taking into account (2.24) we obtain the following transformation formulae of
the components of a covariant 2-tensor T 2 T2 .En / under a basis change (2.24):

Tij0 D Ahi Akj Thk : (2.26)

Remark 2.4. If T D .Tij / is a matrix whose elements are the components of T, and
A D .Aij / is a matrix of the basis change (1.15), then the matrix form of (2.26) is

T0 D AT TA: (2.27)

Definition 2.4. A contravariant 2-tensor or a .2; 0/-tensor is a bilinear map

T D En  En ! <: (2.28)

It is evident that the set T 2 .En / of all contravariant 2-tensors becomes a vector
space by the introduction of the standard operations of addition of two contravariant
2-tensors and the product of a 2-tensor by a real number.
Definition 2.5. The tensor product of two vectors x; y 2 En is the contravariant
2-tensor
x ˝ y.!;  / D !.x/ .y/; 8!;  2 En : (2.29)
Theorem 2.3. Let .ei / be a basis of the vector space En . Then .ei ˝ ej / is a basis
of T 2 .En /, which is an n2 -dimensional vector space.
Proof. 8!;  2 En ,

T.!;  / D T.!i  i ; j  j / D !i j T. i ;  j /:

By introducing the components of the contravariant 2-tensor T relative to the basis

.ei ˝ ej /
T ij D T. i ;  j /; (2.30)
we can write
T.!;  / D T ij !i j : (2.31)
Since
ei ˝ ej .!;  / D !i j ; (2.32)
and !;  in (2.31) are arbitrary, we obtain the result
T D T ij ei ˝ ej ; (2.33)
which shows that .ei ˝ ej / generates the whole vector space T 2 .En /. Further, it is
a basis of T 2 .En / since, in view of (2.32), any linear combination

aij ei ˝ ej D 0

implies that aij D 0 for all the indices i; j D 1; : : : ; n. 

2.3 Covariant 2-Tensors 23

The basis change (1.15) in En determines the basis change

e0i ˝ e0j D Ahi Akj eh ˝ ek (2.34)

in T 2 .En /. The contravariant 2-tensor T can be represented in both bases by

T D T 0ij e0i ˝ e0j D T ij eh ˝ ek ; (2.35)

so that, taking into account (2.34), we derive the following transformation formulae
for the components of T:
j
T 0ij D .A1 /ih .A1 /k T hk : (2.36)

Exercise 2.1. Verify that the matrix form of (2.36) is [see (2.27)]

T0 D .A1 /T.A1 /T : (2.37)

Definition 2.6. A mixed 2-tensor or a .1; 1/-tensor is a bilinear map

T W En  En ! <: (2.38)

Once again, the set T11 .En / of all mixed 2-tensors becomes a vector space by the
introduction of the standard operations of addition of two mixed 2-tensors and the
product of a mixed 2-tensor by a real number.
Definition 2.7. The tensor product of a vector x and a covector  2 En is the
mixed 2-tensor

x ˝ !. ; y/ D x. /!.y/ D  .x/!.y/; 8 ; y 2 En : (2.39)

Theorem 2.4. Let .ei / be a basis of the vector space En and let . i / be the dual
basis in En . Then .ei ˝  j / is a basis of T11 .En /, which is an n2 -dimensional vector
space.
In view of this theorem we can write

T D Tji ei ˝  j ; (2.40)

where the components of the mixed 2-tensor are

Tji D T. i ; ej /: (2.41)

Moreover, in the basis change .ei ˝  j / ! .e0i ˝  0j / given by

j
ei ˝  j D Ahi .A1 /k eh ˝  k ; (2.42)

the components of T are transformed according to the formulae

Tj0i D .A1 /ih Akj Tkh : (2.43)

24 2 Tensor Algebra

Exercise 2.2. Verify that the matrix form of (2.43) is [see (2.27)]

T0 D .A1 /TA: (2.44)

2.4 .r; s/-Tensors

Definition 2.8. An .r; s/-tensor is a multilinear map

T W Enr  Ens ! <: (2.45)

It is quite clear how to transform the set Tsr .En / of all .r; s/-tensors (2.45) in a real
vector space.
p
Definition 2.9. Let T 2 Tsr .En / be an .r; s/-tensor and let L 2 Tq .En / be a .p; q/-
tensor. Then the tensor product of these two tensors is the .r C p; s C q/-tensor
rCp
T ˝ L 2 TsCq .En / given by

T ˝ L. 1 ; : : : ;  rCp ; x1 ; : : : ; xsCq /

D T. 1 ; : : : ;  r ; x1 ; : : : ; xs / L. rC1 ; : : : ;  rCp ; xrC1 ; : : : ; xsCq /: (2.46)

By imposing that the associative property holds, the tensor product can be
extended to any number of factors. Then we introduce the following definition of
the tensor product of vectors and covectors:

x1 ˝    ˝ xr ˝ !1 ˝    ˝ !s . 1 ; : : : ;  r ; u1 ; : : : ; us / D
x1 . 1 /    xr . r /!1 .u1 /    !s .us /: (2.47)

With the procedure revealed in the previous section it is possible to prove the
following theorem.
Theorem 2.5. The dimension of the vector space Tsr .En / of the .r C s/-tensors on
En is nrCs , and
e i1 ˝    ˝ e ir ˝  j 1 ˝    ˝  j s (2.48)

is a basis of it.
In (2.48), .ei / is a basis of En and . j / the dual basis in En . Instead of
Eqs. (2.40)–(2.43), we now obtain

T D Tji11j
ir
e ˝    ˝ e ir ˝  j 1 ˝    ˝  j s ;
s i1
(2.49)

j ;:::;j
Ti 11;:::;i s D T. j1 ; : : : ;  js ; ei1 ; : : : ; eir /; (2.50)
n
2.6 Contraction and Contracted Multiplication 25

e0i1 ˝    ˝ e0ir ˝  0j1 ˝    ˝  0js

j j
D Ahi11    Ahirr .A1 /k11    .A1 /kss eh1 ˝    ˝ ehr ˝  k1 ˝    ˝  ks ; (2.51)

Tj0i11j
ir
s
D .A1 /ih11    .A1 /ihrr Akj11    Akjss Tkh11k
hr
s
: (2.52)

2.5 Tensor Algebra

In the previous sections we defined the addition of tensors belonging to the same
tensor space. On the other hand, the tensor product of two tensors that might
belong to different tensor spaces defines a new tensor that belongs to another tensor
space. In conclusion, the tensor product is not an internal operation. However,
it is possible to introduce a suitable set that, equipped with the aforementioned
operations, becomes an algebra.
Let us consider the infinite direct sum (Sect. 1.4)

TEn D ˚r;s2N Tsr En (2.53)

whose elements are finite sequences fa; x; !; T; L; K; : : :g, where a 2 <, x 2 En ,

! 2 En , T 2 T02 .En /, K 2 T20 .En /, L 2 T11 .En /, etc. With the introduction of
this set, the multiplication by a scalar, the addition, and the tensor product become
internal operations and the set TEn , equipped with them, is called a tensor algebra.

2.6 Contraction and Contracted Multiplication

In the tensor algebra TEn we can introduce two other internal operations: the
contraction and the contracted product.
Theorem 2.6. Denote by .ei / a basis of the vector space En and by . i / the dual
basis in En . Then, for any pair of integers 1  h; k  n, the linear map

Ch;k W T D Tji11kj
hir
ei ˝    ˝ eir ˝  j1 ˝    ˝  js 2 Tsr .En /
s 1

! Ch;k .T/ D Tji11hj

hir
ei ˝    eih1 ˝ eihC1 ˝    ˝ eir ˝
s 1

 j1 ˝    ˝  ik1 ˝  ikC1 ˝    ˝  js 2 Ts1

r1
.En /; (2.54)

which is called a contraction, to any tensor T 2 Tsr .En / associates a tensor

r1
Ch;k .T/ 2 Ts1 .En /, which is obtained by equating the contravariant index h and
the covariant index k and summing over these indices.
Proof. To simplify the notations, we prove the theorem for a .2; 1/-tensor T D
ij hj
Th ei ˝ ej ˝  h . For this tensor C1;1 .T/ D Th ej , and it will be sufficient to prove
26 2 Tensor Algebra

hj
that, under a basis change, the quantities Th are transformed as vector components.
Since in a basis change we have that
0ij p
Th D .A1 /il .A1 /jm Ah Tplm ;

it also holds that

0hj p
Th D .A1 /hl .A1 /jm Ah Tplm D .A1 /jm Thhm ;

and the theorem is proved. 

The preceding theorem makes it possible to give the following definition.
Definition 2.10. A contracted multiplication is a map
rCp1
.T; L/ 2 Tsr .En /  Tqp .En / ! Ch;k .T ˝ L/ 2 TsCq1 .En /: (2.55)

ij
For instance, if T D Th ei ˝ ej ˝  h and L D Llm el ˝  m , then Ci;m .T ˝ L/ D
T hj Llh ej ˝ el .
Theorem 2.7. The nrCs quantities

Tji11j
ir
s
(2.56)

are the components of an .r Cs/-tensor T if and only if any contracted multiplication

of these quantities by the components of a .p; q/-tensor L, p  s, q  r generates
an .r  q; s  p/-tensor.
Proof. The definition of contracted multiplication implies that the condition is
necessary. For the sake of simplicity, we prove that the condition is sufficient
ij
considering the quantities Th and a .0; 1/-tensor L. In other words, we suppose
ij
that the quantities Th Li are transformed as the components of a .1; 1/-tensor. Then
we have that
0ij j
Th L0i D .A1 /l Am kl
h Tm Lk :

Since Lk D .A1 /rk L0r , the theorem is proved. 

Remark 2.5. Let T W x 2 En ! y 2 En be a linear map and denote by .ei / a basis
of En and by .Tji / the matrix of T relative to the basis .ei /. In terms of components,
the map T becomes
y i D Tji x j ; (2.57)

where y i are the components of the vector y D T.x/ in the basis .ei /. Owing to the
previous theorem, the quantities .Tji / are the components of a .1; 1/-tensor relative
to the basis .ei ˝  j /. It is easy to verify that any .1; 1/-tensor T determines a linear
map T W En ! En .
2.7 Exercises 27

2.7 Exercises

1. Determine all the linear maps that can be associated with an .r; s/-tensor, where
r C s  3.
2. Let .ei / be a basis of a vector space En , and let . i / be the dual basis. Verify
that a tensor that has the components .ıji / in the basis .ei ˝  j / has the same
components in any other basis.
3. Let .ei / be a basis of a two-dimensional vector space E2 , and denote by . i / the
dual basis. Determine the value of the covariant 2-tensor  1 ˝ 2  2 ˝ 1 when
it is applied to the pairs of vectors .e1 ; e2 /, .e2 ; e1 /, and .x; y/ and recognize
geometric meaning of each result.
4. The components of a .1; 1/-tensor T of the vector space E3 relative to a basis
.ei ˝  j / are given by the matrix
0 1
121
@2 1 2A:
121

Determine the vector corresponding to x D .1; 0; 1/ by the linear endomor-

phism determined by T.
5. In the basis .ei / of the vector space E3 , two vectors x and y have the components
.1; 0; 1/ and .2; 1; 0/, respectively. Determine the components of x ˝ y relative
to the basis .e0i ˝ e0j /, where

e01 D e1 C e3 ;
e02 D 2e1  e2 ;
e03 D e1 C e2 C e3 :

6. Given the .0; 2/-tensor Tij  i ˝  j of T2 .E2 /, where

 
12
.Tij / D ;
10

determine if there exists a new basis in which its components become

 
11
.Tij0 / D :
01

7. For which .1; 1/-tensor .Tji / of T11 .E3 / does the linear map F defined by the
matrix .Tji / satisfy the condition F .x/ D ax, 8x 2 E3 and 8a 2 <?
8. Given the .1; 1/-tensor Tji ei ˝  j of T2 .E2 /, verify that T11 C T22 and det.Tji /
are invariant with respect to a change of basis.
28 2 Tensor Algebra

9. Prove that if the components of a .0; 2/-tensor T satisfy either of the conditions

Tij D Tj i ; Tij D Tj i

in a given basis, then they satisfy the same conditions in any other basis.
10. Given .0; 2/-tensors that in the basis .ei /, i D 1; 2; 3, have the components
0 1
1 0 1
T1 D @ 0 1 2 A ;
1 2 1
0 1
0 1 1
T2 D @ 1 0 2 A ;
1 2 0
determine the covector !u depending on u such that

!u .v/ D T1 .u; v/;

!u .v/ D T2 .u; v/

8v. Further, find the vectors u such that

T1 .u; v/ D 0; T2 .u; v/ D 0; 8v:

Chapter 3
Skew-Symmetric Tensors and Exterior Algebra

3.1 Skew-Symmetric (0, 2)-Tensors

Definition 3.1. A tensor T 2 T2 .En / is skew-symmetric or alternating if

T.x; y/ D T.y; x/; (3.1)

8x; y 2 En . In particular, (3.1) implies T.x; x/ D 0.

Denote by x D x i ei and y D y i ei respectively the representations of x and y relative
to a basis .ei / of En . If . i / is the dual basis of .ei / and T D Tij  i ˝  j , then (3.1)
becomes
Tij x i y j D Tj i x i y j :

This equality is identically satisfied 8x; y 2 En if and only if the components Tij of
T verify the conditions

Tij D Tj i ; i ¤ j; Ti i D 0: (3.2)

Exercise 3.1. Prove that if conditions (3.2) hold in a basis, then they hold in any
basis.
If we denote by T the matrix of the components of T, then conditions (3.2)
assume the following matrix form:

T D TT : (3.3)

Definition 3.2. The exterior product of the covectors !;  2 En is the .0; 2/-
tensor such that
! ^  D ! ˝    ˝ !: (3.4)

A. Romano, Classical Mechanics with Mathematica R

, Modeling and Simulation 29
in Science, Engineering and Technology, DOI 10.1007/978-0-8176-8352-8 3,
© Springer Science+Business Media New York 2012
30 3 Skew-Symmetric Tensors and Exterior Algebra

Theorem 3.1. The exterior product (3.4) has the following properties:
• ! ^  is a skew-symmetric tensor;
• ! ^  D  ^ !;
• If  D a!, a 2 <, then ! ^  D 0.
Proof. From (3.4), 8x; y 2 En , we have that

! ^  .x; y/ D ! ˝  .x; y/   ˝ !.x; y/

D !.x/ .y/   .x/!.y/
D ! ^  .y; x/; (3.5)

and the skew symmetry of ! ^  is proved. Starting from (3.5) it is simple to verify
the other two properties. 
The set ƒ2 .En / of the skew-symmetric .0; 2/-tensors is a subspace of the vector
space T2 .En /. The dimension and the bases of ƒ2 .En / are determined by the
following theorem.
Theorem 3.2. Let .ei / be a basis of En and denote by . i / its dual basis in En .
 
n
The dimension of ƒ2 .En / is , and the vectors
2

. i ^  j /; i < j;

form a basis of ƒ2 .En /.

Proof. For any T 2 ƒ2 .En / we have that
X X
T D Tij  i ˝  j D Tij  i ˝  j C Tij  i ˝  j
i <j i >j
X X
D Tij  i ˝  j C Tj i  j ˝  i
i <j j >i
X X
D Tij  i ˝  j  Tij  j ˝  i :
i <j i <j

Therefore, we can write

T D T.ij /  i ^  j ; (3.6)

where
T.ij / D Tij ; i < j: (3.7)

Relation (3.6) shows that the set of the .0; 2/-tensors . i ^  j / 2 ƒ2 .En / generates
the whole subspace ƒ2 .En /. Consequently, to verify that they form a basis of
3.1 Skew-Symmetric (0, 2)-Tensors 31

ƒ2 .En /, it is sufficient to verify their linear independence. Now, from any linear
combination
a.ij /  i ^  j D 0; i < j;

we have the condition

a.ij /  i ^  j .eh ; ek / D 0;

which, in view of (3.5) and (2.4), implies a.ij / D 0, and the proof is complete. 
Definition 3.3. The quantities T.ij / , i < j , are called strict components of T
relative to the basis . i ^  j /.
To determine the transformation formulae of the strict components, we recall that
they are components of a .0; 2/-tensor. Consequently, in view of (2.26), we can write
0 h k
T.ij / D A.i Aj / Thk
X X
D Ah.i Akj / Thk C Ah.i Akj / Thk
h<k h>k
X
D .Ah.i Akj /  Ak.i Ahj / /T.hk/ ;
h<k

and the transformation formulae of the strict components of T 2 ƒ2 .En / are

ˇ ˇ
ˇ Ah Ah ˇ
0 ˇ i jˇ
T.ij / D ˇ k k ˇ T.hk/ : (3.8)
ˇ Ai Aj ˇ

Starting from the inverse formula of (2.26) we also have

ˇ ˇ
ˇ .A1 /h .A1 /h ˇ
ˇ j ˇ 0
T.ij / D ˇ 1 ki ˇT : (3.9)
ˇ .A /i .A1 /kj ˇ .hk/

The transformation formulae of the bases  i ^  j of ƒ2 .En / can be obtained by

noting that
0i 0j
0
T D T.ij / ^  D T.hk/  h ^  k : (3.10)

Introducing (3.8) and (3.9) into (3.10) we obtain

ˇ 1 i ˇ
ˇ .A / .A1 /i ˇ
 0i ^  0j D ˇˇ 1 jh 1 jk ˇˇ  h ^  k ; i < j; h < k; (3.11)
.A /h .A /k
ˇ i i ˇ
ˇA A ˇ
 i ^  j D ˇˇ jh jk ˇˇ  0h ^  0k ; i < j; h < k: (3.12)
Ah Ak
32 3 Skew-Symmetric Tensors and Exterior Algebra

Example 3.1. Let E2 be a two-dimensional vector space and denote by .e1 ; e2 / and
. 1 ;  2 / a basis of E2 and the dual basis of E2 , respectively. The dimension of
the vector space T2 .E2 / is 4, whereas the subspace ƒ2 .E2 / of the skew-symmetric
.0; 2/-tensors is one-dimensional. Consequently, any .0; 2/-tensor of ƒ2 .E2 / can be
written as follows:
T D T12  1 ^  2 :

The skew-symmetric basis  1 ^  2 has a remarkable geometric meaning. In fact,

8x; y 2 E2 ,

 1 ^  2 .x; y/ D  1 ˝  2 .x; y/   2 ˝  1 .x; y/

ˇ 1 2ˇ
ˇx x ˇ
D .x 1 y 2  x 2 y 1 / D ˇˇ 1 2 ˇˇ ; (3.13)
y y

and we can state that  1 ^  2 .x; y/ measures the area of the parallelogram
determined by the vectors x and y. In particular, from (3.13) we obtain

 1 ^  2 .e1 ; e2 / D 1;

and the parallelogram determined by the vectors e1 and e2 has a unit area. We note
that the area of the parallelogram formed by the vectors .e01 ; e02 / of another basis of
E2 , where

e01 D A11 e1 C A21 e2 ;

e02 D A12 e1 C A22 e2 ;

has the value

ˇ 1 2ˇ
ˇA A ˇ
 1 ^  2 .e01 ; e02 / D ˇˇ 11 12 ˇˇ D det A:
A2 A2

In conclusion, choosing a basis .e1 ; e2 / of E2 and the corresponding skew-

symmetric .0; 2/-tensor  1 ^  2 , we introduce a criterion to evaluate the areas of
parallelograms without resorting to a metric. This criterion does not depend on the
basis, provided that the basis changes satisfy the condition det A D 1.
Exercise 3.2. Verify that in a vector space E3 , related to the basis .ei /, a skew-
symmetric .0; 2/-tensor T has the following form:

T D T12  1 ^  2 C T13  1 ^  3 C T23  2 ^  3 ;

where . i / is the dual basis in E3 of .ei /. Prove that the skew-symmetric
.0; 2/-tensor
T D 1 ^ 2 C 1 ^ 3 C 2 ^ 3
3.2 Skew-Symmetric .0; r/-Tensors 33

associates to any pair of vectors x, y of E3 the sum of the areas of the projections
of the parallelogram formed by the vectors x, y onto the subspaces generated by
.e1 ; e2 /, .e1 ; e3 /, .e2 ; e3 /, respectively.
In the next section, the content of this section will be extended to .0; r/-tensors.

3.2 Skew-Symmetric .0; r/-Tensors

Definition 3.4. Let Sr D f1; 2; : : : ; rg be the set of the first r integer numbers. A
permutation of Sr is any one-to-one map
 W Sr ! Sr :

We denote by f.1/; .2/; : : : ; .r/g the set formed by the same numbers of Sr
placed in a different order. It is well known that the set …r of all the permutations
of Sr contains rŠ one-to-one maps. This set can be equipped with the structure of a
group by the usual composition of maps

;  2 …r !  ı  2 …r :

The identity of this group is a map that does not modify the position of the numbers
of Sr . Finally, the opposite of  is the inverse map  1 . Let i; j , i < j , be two
numbers of Sr . We say that the permutation  contains an inversion with respect to
Sr if
.i / > .j /:
A permutation  is said to be even or odd according to whether the total number
of inversions contained in S is even or odd. In the sequel, we denote by m./ the
total inversions of .
Definition 3.5. A tensor T 2 Tr .En /, r > 2, is skew-symmetric or alternating if

T.x1 ; : : : ; xr / D .1/m./ T.x.1/ ; : : : ; x.r/ / (3.14)

8x1 ; : : : ; xr 2 En and 8 2 …r .
The preceding definition implies that the value of T vanishes every time T is
evaluated on a set of vectors containing two equal vectors. In fact, it is sufficient
to consider a permutation that exchanges the position of these two vectors without
modifying the position of the others and then to apply (3.14). From this remark it
easily follows that the value of T vanishes if the vectors fx1 ; : : : ; xr g are linearly
dependent.
If the r vectors fx1 ; : : : ; xr g belong to a basis .ei /, then, in view of (2.50), we
express the skew symmetry of T in terms of its components:
Ti1 ir D .1/m./ T.i1 /.ir / : (3.15)
34 3 Skew-Symmetric Tensors and Exterior Algebra

In particular, this condition implies that all the components of T in which two
indices have the same value vanish. It is quite obvious that the set of all the skew-
symmetric tensors form a subspace ƒr .En / of Tr .En /.
Remark 3.1. Since r vectors, when r > n, cannot be linearly independent, we can
state that
ƒr .En / D f0g; n < r:

Definition 3.6. The exterior product of r covectors !1 ; : : : ; !r is the .0; r/-tensor

X
!1 ^    ^ !r D .1/m./!.1/ ˝    ˝ !.r/ : (3.16)
2…r

The tensor (3.16) is skew-symmetric since, in view of (2.47), we have that

X
!1 ^    ^ !r .x1 ; : : : ; xr / D .1/m./ !.1/ .x1 /    ˝ !.r/ .xr /
2…r
ˇ 1 ˇ
ˇ ! .x1 /    ! 1 .xr / ˇ
ˇ ˇ
D ˇˇ          ˇˇ : (3.17)
ˇ !r .x /    !r .x / ˇ
1 r

A permutation  of the vectors x1 ; : : : ; xr corresponds to a permutation of the

columns of the determinant in (3.17). This operation changes or does not change
the sign of the determinant according to whether the permutation is odd or even.
But this is just the property expressed by (3.14). Again recalling the properties of a
determinant, we can easily prove the following properties of the exterior product of
r covectors:
• It is a .0; r/-skew-symmetric tensor.
• It vanishes if one vector linearly depends on the others.
Now we extend Theorem 3.2 to skew-symmetric .0; r/-tensors.
i
Theorem 3.3. Let .ei / be a basis of the vector
 space En and denote by . / the
n
dual basis in En . Then, if r  n, then is the dimension of the vector space
r
ƒr .En / of the skew-symmetric .0; r/-tensors. Further, the skew-symmetric tensors

 i1 ^    ^  ir ; (3.18)

where i1 <    < ir is an arbitrary r-tuple of integer numbers chosen in the set of
indices f1; : : : ; ng, form a basis of ƒr .En /.
Proof. First, when r  n, any T 2 ƒr .En / can be written as follows:

T D Tj1 jr  j1 ˝    ˝  jr :
3.2 Skew-Symmetric .0; r/-Tensors 35

The summation on the right-hand side contains only terms in which all the indices
are different from each other since T is skew-symmetric. This circumstance makes
it possible to group all the terms as follows: first, we consider the terms obtained by
extracting arbitrarilyr different
 indices i1 <    < ir from the set f1; : : : ; ng. We
n
recall that there are different possible choices of these indices. Then, for each
r
choice, which is characterized by the indices i1 <    < ir , we consider all the terms
obtained by permutating in all the possible ways the indices i1 <    < ir . In view
of (3.15) and (3.16), we can write
X X
TD T.i1 /.ir /  .i1 / ˝    ˝  .ir /
i1 <<ir 2…r
X X
D Ti1 ir .1/m./  .i1 / ˝    ˝  .ir /
i1 <<ir 2…r
X
D Ti1 ir  i1 ^    ^  ir ;
i1 <<ir

that is,
T D T.i1 ir /  i1 ^    ^  ir : (3.19)

This result shows that the skew-symmetric .0; r/-tensors f i1 ^    ^  ir ; i1 <    ir g

generate the whole vector space ƒr .En /. We can easily verify that they are linearly
independent, so that they form a basis of ƒr .En /. 
The transformation formulae of the strict components T.i1 ir / of a skew-symmetric
tensor [see (3.8)–(3.12)] can be found by noting that

T.i0 1 ir / D Ahi11    Ahirr Th1 hr

X X .j / .j /
D .1/m./Ai1 1    Air r T.j1 /.jr / ;
j1 <<jr 2…r

where the indices j1 <    < jr are chosen in the set 1; : : : ; n in all possible ways.
Finally, we have that
ˇ ˇ
ˇ Aj1    Aj1 ˇ
ˇ i1 ir ˇ
ˇ ˇ
T.i0 1 ir / D ˇ          ˇ T.j1 jr / : (3.20)
ˇ jr jr ˇ
ˇ Ai    Ai ˇ
1 r

We can easily verify that the transformation formula of the bases of ƒr .En / [see
(3.11)] is
ˇ ˇ
ˇ .A1 /i1    .A1 /i1 ˇ
ˇ j1 jr ˇ
ˇ ˇ
 0i1 ^    ^  0ir D ˇ          ˇ  j1 ^    ^  jr : (3.21)
ˇ 1 ir ir ˇ
ˇ .A /j1    Ajr ˇ
36 3 Skew-Symmetric Tensors and Exterior Algebra

In particular, if T 2 ƒn .En /, then (3.20) and (3.21) give

0
T1n D det A T1n ; (3.22)
1
 01 ^    ^  0n D  1 ^    ^  n; (3.23)
det A

where A D det.Aij /.

3.3 Exterior Algebra

Definition 3.7. We define the exterior product of the skew-symmetric tensors

T D T.i1 ir /  i1 ^    ^  ir 2 ƒr .En /

and
L D L.j1 js /  j1 ^    ^  jr 2 ƒs .En /;

as the skew-symmetric tensor of ƒrCs .En / given by

X X
T^L D .1/m./ T.h1 /.hr / L.hrC1 /.hrCs /  h1 ^    ^  hrCs ;
h1 <<hrCs 2…rCs

(3.24)
where  is any permutation of the indices h1 <    < hrCs , in which .h1 / <
   < .hr /, .hrC1 / <    < .hrCs /, and m./ is the number of inversions of
.h1 /; : : : ; .hrCs /.
Example 3.2. The external product of the two skew-symmetric tensors

T D T12  1 ^  2 C T13  1 ^  3 C T23  2 ^  3 2 ƒ2 .E3 /;

! D !1  1 C !2  2 C !3  3 2 ƒ1 .E3 /

is the skew-symmetric tensor of ƒ3 .E3 / given by

T ^ ! D .T12 !3  T13 !2 C T23 !1 / 1 ^  2 ^  3 :

Exercise 3.3. Prove that the exterior product of T 2 ƒ2 .E5 / and L 2 ƒ2 .E5 / has
the component

T12 L45  T14 L25 C T15 L24 C T24 L15  T25 L14 C T45 L12

along the basis vector  1 ^  2 ^  4 ^  5 .

3.4 Oriented Vector Spaces 37

It is not difficult to verify that T^L is skew-symmetric and independent of the basis.
It can also be proved that
T ^ L D .1/rs L ^ T: (3.25)

We can define the exterior algebra as we did for the tensor algebra. First, we
introduce the notations
ƒ0 .En / D <; ƒ1 .En / D En :

Then we consider the set

ƒ.En / D ˚k2N ƒk .En /; (3.26)

whose elements are formed by finite sequences .a; !; T; : : :/, where a 2 <, ! 2
En , T 2 ƒ2 .En /, etc., and we recall that ƒr .En / D f0g for r > n. This set,
equipped with multiplication by a scalar, addition, and an exterior product, is the
exterior algebra over En .
It is evident that what we have proved for the skew-symmetric tensors T0s can be
repeated for the skew-symmetric tensors Tr0 . In this way, we can define the exterior
algebra ƒr .En /.

3.4 Oriented Vector Spaces

Let B be the set of the bases of a vector space En , and let

e0j D Aij ei (3.27)

be a basis change. Introduce in B the relation R such that

.e0i /R.ei / , A > 0; (3.28)

where A D det.Aij / ¤ 0.
Theorem 3.4. R is an equivalence relation partitioning B into two equivalence
classes.
j j
Proof. From the evident conditions ei D ıi ej , det.ıi / D 1 > 0, it follows that any
basis is in the relation R with itself and R is reflexive. If (3.28) holds, then we have
j
ei D .A1 /i e0j ;
j
where det..A1 /i / D 1=A > 0. Therefore, .ei / is in the relation R with .e0i / and
R is symmetric. It remains to prove that R is transitive. To this end, we consider a
third basis .e00i / such that
j
e00h D Bh e0j ; (3.29)
38 3 Skew-Symmetric Tensors and Exterior Algebra

j
where det.Bh / > 0. Then we also have

j j
e00h D Bh e0j D Bh Akj ek  Chk ek : (3.30)

j
Since det.Chk / D det.Bh / det.Akj / > 0, R is an equivalence relation. To verify
that R partitions B into two equivalence classes, we first note that these equivalence
classes are at least two since the two bases .ei / and .e0i /, such that
0 1
1 0  0
B 0 1  0 C
e0i D Cih eh ; .Cih / D B
@ 
C ; det.C h / D 1;
  A i

0 0  1

are not equivalent. If .e00i / is another arbitrary basis of En , then

j j
e00i D Ai e0j D Ai Cjh eh ;

so that the basis .e00i / is equivalent either to .e0i / or to .ei /. 

Definition 3.8. A vector space En is said to be oriented if one of the two
equivalence classes of R is chosen. In this case, the bases belonging to this class
are called positive, whereas the bases belonging to the other class are said to be
negative.
Let .ei / be a positive basis of the vector space En , and denote by . i / its dual
j
basis. If xi D xi ej , i D 1; : : : ; n, are n vectors of En , from (3.17) and (2.5) we
obtain that
0 1 1
x1    xn1
 1 ^    ^  n .x1 ; : : : ; xn / D det @          A : (3.31)
x1n    xnn

In particular, we have that

 1 ^    ^  n .e1 ; : : : ; en / D 1: (3.32)

Further, if .e0i / is another basis of En , related to .ei / by (3.27), then in view of (3.31),
we have that
 1 ^    ^  n .e01 ; : : : ; e0n / D det A: (3.33)

In conclusion, we have proved that (Example 3.1)

Theorem 3.5. The .0; n/-covector  1 ^    ^  n of the vector space ƒn .En /
associates to any n-tuple of vectors x1 ; : : : ; xn the volume of the parallelepiped
having these vectors as wedges.
3.5 Exercises 39

3.5 Exercises

1. Given the 1-forms

˛ D  1   2;
ˇ D  1   2 C  3;
 D  3;

the 2-form
 D  1 ^  3 C  2 ^  3;

and the 3-form

 D  1 ^  2 ^  3;

calculate the exterior products

˛ ^ ˇ; ˛ ^ ˇ ^  ; ˛ ^ ; ˛ ^ :

2. Evaluate the components of the forms of the preceding exercise under the basis
change

 01 D  1  2 2 ;
 02 D  1 C  3 ;
 03 D  3 :

3. Let .ei / be a basis of the vector space E3 and denote by . i / the dual basis. Given
the skew-symmetric tensors

T D T12  1 ^  2 C T13  1 ^  3 C T23  2 ^  3 ;

L D T123  1 ^  2 ^  3 ;

and the basis change

e01 D e1  e3 ;
e02 D e1 C 2e2 ;
e03 D e2  e3 ;

determine the components of the preceding tensors in the corresponding new

basis of ƒ2 .E3 / and ƒ3 .E3 /.
4. Determine the ratio between the volumes of the parallelepipeds formed by the
two foregoing bases.
5. Write arbitrary skew-symmetric tensors of ƒ2 .E4 / and ƒ2 .E5 /.
40 3 Skew-Symmetric Tensors and Exterior Algebra

6. Multiply a skew-symmetric tensor of ƒ2 .E4 / by a skew-symmetric tensor of

ƒ3 .E4 /.
7. Given the volume form  1 ^  2 ^  3 in a three-dimensional space E3 , determine
the volume of a parallelepiped whose edges are the vectors .1; 0; 2/, .1; 2; 1/,
and .1; 1; 0/.
8. Evaluate the volume of the parallelepiped of the previous exercise adopting the
volume form  01 ^  02 ^  03 , where

 01 D  1 C 2 2 ;
 02 D  2 C  3 ;
 03 D  1  2 3 :
Chapter 4
Euclidean and Symplectic Vector Spaces

4.1 Representation Theorems for Symmetric

and Skew-Symmetric .0; 2/-Tensors

In the preceding chapters we analyzed some properties of a finite-dimensional vector

space En . In this chapter we introduce into En two other operations: the scalar
product and the antiscalar product. A vector space equipped with the first operation
is called a Euclidean vector space, whereas when it is equipped with the second
operation, it is said to be a symplectic vector space. These operations allow us to
introduce into En many other geometric and algebraic concepts.
Definition 4.1. Let En be an n-dimensional vector space. A tensor T 2 T2 .En / is
said to be symmetric if

T.x; y/ D T.y; x/; 8x; y 2 En : (4.1)

It is easy to prove that (4.1) is equivalent to the condition

Tij D Tj i ; (4.2)

where Tij D T.ei ; ej / are the components of T in an arbitrary base .ei / of En .

Theorem 4.1. If T 2 T2 .En / is symmetric, then there exists a basis .ei / of En in
which the components of T are given by the matrix
0 1
Ir Ors Orp
@ Osr Is Osp A : (4.3)
Opr Ops Opp

In (4.3), Ir and Is are unit matrices of order r and s, respectively, where r C s  n.

Finally, p D n  .r C s/ and Oij is an i  j zero matrix.

A. Romano, Classical Mechanics with Mathematica R

, Modeling and Simulation 41
in Science, Engineering and Technology, DOI 10.1007/978-0-8176-8352-8 4,
© Springer Science+Business Media New York 2012
42 4 Euclidean and Symplectic Vector Spaces

Proof. If T D 0 or n D 1, then the theorem is trivial. In that case we suppose

T ¤ 0 and n > 1. If T ¤ 0, then there exists a pair .x ; y / of vectors such that
T.x ; y / ¤ 0. Now we prove the existence of at least one vector x1 for which
T.x1 ; x1 / ¤ 0. To this end it is sufficient to prove the existence of at least a real
value  such that T.x C y ; x C y / ¤ 0. In view of the symmetry of T we
obtain that

T.x C y ; x C y / D 2 T.x ; x / C 2T.x ; y / C T.y ; y /

 a2 C 2b C c; (4.4)

where a, b, and c are given real numbers. Now the trinomial in (4.4) vanishes
identically for any  if and only if a D b D c D 0. Since b D T.x ; y / ¤ 0, we
have provedp the existence of at least one vector x1 for which T.x1 ; x1 / ¤ 0. Putting
u1 D x1 = j˛j, where ˛ D T.x1 ; x1 /, it follows that T.u1 ; u1 / D ˙1, where the
sign is chosen according to the sign of ˛. We denote by U 1 the one-dimensional
subspace of En generated by u1 and by V 1 the following subspace of En :

V 1 D fv 2 En W T.u1 ; v/ D 0g:

It is En D U 1 ˚ V 1 . In fact, 8x 2 En the vector

T.u1 ; x/
vDx u1
T.u1 ; u1 /
verifies the condition
T.u1 ; x/
T.u1 ; v/ D T.u1 ; x/  T.u1 ; u1 / D 0;
T.u1 ; u1 /

so that v 2 V 1 . Introducing the notation a D T.u1 ; x/=T.u1 ; u1 /, we have that

x D v C au1 ; v 2 V 1 ; au1 2 U 1 : (4.5)

The decomposition (4.5) is unique. In fact, if there is another decomposition x D

v0 C bu1 , v0 2 V 1 , then we have v0  v 2 V 1 since V 1 is a vector subspace of En .
Further, v0  v D .a  b/u1 and v0  v 2 U 1 \ V 1 . As a consequence, we have that

T.v0  v; u1 / D .a  b/T.u1 ; u1 / D .a  b/˛:

Since ˛ ¤ 0, the preceding condition implies a D b and v0 D v. Now we consider

the restriction T1 of T over V 1 , which is a bilinear symmetric map. If T1 D 0
or n1 D dim.V 1 / D n  1 D 1, then the theorem is proved. If neither of these
conditions is verified, then we can again apply the foregoing procedure to T1 ; T2 ; : : :
until one of the conditions Ti D 0 or ni D 1 is verified. In this way we determine a
set fu1 ; : : : ; un g of independent vectors for which the following result is obtained:

T.ui ; ui / D ˙1; T.ui ; uj / D 0; i ¤ j:

4.1 Representation Theorems 43

By a convenient arrangement of these vectors, we obtain a basis .ei / in which the

representative matrix of T has the form of (4.3). 
i
The basis .ei / is called a canonical basis of En . If . / is the dual basis of .ei /,
then the symmetric tensor T assumes the following canonical form:
r
X rCs
X
i i i i
TD  ˝   ˝ : (4.6)
i D1 i DrC1

Theorem 4.2. If T 2 T2 .En / is skew-symmetric, then there exists a basis .ei / of En

in which the components of T are given by the matrix
0 1
Orr Ir Orp
@ Ir Orr Orp A : (4.7)
Orp Orr Orp

In (4.7), Ir is an r  r unit matrix, 2r  n, p D n  2r and Oij is an i  j zero

matrix.
Proof. The theorem is obvious if T D 0 or n D 1. If T ¤ 0 and n ¤ 1, then
there are at least two independent vectors u1 and u2 for which T.u1 ; u2 / ¤ 0. In
fact, if u2 D au1 , then it follows that T.u1 ; u2 / D aT.u1 ; u1 / D 0, in view of the
skew symmetry of T. Let U 1 denote the two-dimensional space generated by the
independent vectors u1 and u2 . From the foregoing considerations it follows that, in
this basis, the matrix representing the restriction of T onto U 1 has the form
 
0 1
:
1 0

Now we prove that En D U 1 ˚ V 1 , where

V 1 D fv 2 En ; T.v; u/ D 0; 8u 2 U 1 g

is a subspace of En . First, 8x 2 En we introduce the two vectors

u D T.x; u2 /u1  T.x; u1 /u2 ;

v D x  u:

Since u 2 U 1 and x D u C v, it remains to prove that v 2 V 1 and the vectors u, v

are uniquely determined. But for i D 1; 2 it is

T.v; ui / D T.x  u; ui / D T.x; ui /  T.u; ui /

D T.x; ui /  T.T.x; u2 /u1  T.x; u1 /u2 ; ui /
D T.x; ui /  T.x; u2 /T.u1 ; ui / C T.x; u1 /T.u2 ; ui / D 0;
44 4 Euclidean and Symplectic Vector Spaces

so that v 2 V 1 . Further, if there is another decomposition x D u0 C v0 , u0 2 U 1 , and

v0 2 V 1 , we have u0  u D v0  v. But v0  v 2 V 1 , and then

T.v0  v; ui / D T.u0  u; ui / D 0:

Since u0 u 2 U 1 , there are two real numbers a and b such that u0 u D au1 Cbu2 .
Consequently, from the foregoing result we obtain the condition

aT.u1 ; ui / C bT.u2 ; ui / D 0;

which implies a D b D 0 since T.u1 ; u2 / ¤ 0 and T.ui ; ui / D 0. The restriction

T1 of T onto V 1 is still a skew-symmetric tensor. Then, if n D 1 or T1 D 0, the
theorem is proved; otherwise we can repeat for T1 the foregoing reasoning. Iterating
the procedure, we determine a basis .ui / of En in which T is represented by the
matrix
0 1
0 1 0    0
B 1 0 0             C
B C
B C
B    0 1  C
B C:
B          1 0       C
B C
@       A
0 0     0

In the new basis,

ei D u2i 1 ; i D 1; : : : ; r;
erCi D u2i ; i D 1; : : : ; r;
ek D uk ; k D 2r C 1; : : : ; n;

the representative matrix of T becomes (4.7), and the theorem is proved. 

i
The basis .ei / is called a canonical basis of En . Let . / be the dual basis of .ei /.
Then the foregoing theorem supplies the following canonical form of the skew-
i j
symmetric tensor T in the basis . ˝  / of T2 .En /:
r
X i rCi rCi i
TD . ˝   ˝  /: (4.8)
i D1

4.2 Degenerate and Nondegenerate .0; 2/-Tensors

For any T 2 T2 .En / we consider the vector subspace of En

E0 D fx 2 En ; T.x; x/ D 0; 8y 2 En g:
4.2 Degenerate and Nondegenerate .0; 2/-Tensors 45

Definition 4.2. The tensor T 2 T2 .En / is said to be nondegenerate if E0 D f0g,

degenerate if dim.E0 / D k  1:
In components, the condition T.x; x/ D 0; 8x 2 En , is expressed by the system

Tij x i D 0; j D 1; : : : ; n; (4.9)

of n linear equations in n unknowns .x i /. It is well known that, denoting by T the

representative matrix of T in the basis .ei / of En , system (4.9) admits a zero solution
when the rank p of T is equal to n, and infinite solutions, which belong to a vector
subspace of En , when p < n. Consequently, dim.E0 / D n  p.
For a symmetric tensor T 2 T2 .En /, resorting to its canonical representation, we
can state that the rank p coincides with the number r C s of the elements of the
principal diagonal of matrix (4.3) that assume the values ˙1, whereas dim.E0 / is
equal to the number of zeros contained into the principal diagonal. For a symmetric
nondegenerate .0; 2/-tensor T there is no zero in the principal diagonal of (4.3).
Starting from (4.7), we conclude that for a skew-symmetric .0; 2/-tensor T, it results
that the rank .T/ D 2r and T is nondegenerate if and only if 2r D n. In particular,
if n is odd, there is no nondegenerate tensor.
Theorem 4.3. The numbers r and s appearing in the canonical representation (4.6)
of a symmetric tensor T 2 T2 .En / do not depend on the canonical basis of En .
Proof. Let .ei / be the canonical basis in which T is represented by the matrix
(4.3). The vector subspaces E C and E  of En generated by .e1 ; : : : ; er / and
.erC1 ; : : : ; erCs /, respectively, satisfy the condition

En D E C ˚ E  ˚ E0 : (4.10)

Moreover, for any choice of

r
X
xD x i ei 2 E C ;
i D1
rCs
X
yD y i ei 2 E  ;
i DrC1

we have that
r
X rCs
X
T.x; x/ D .x i /2 > 0; T.y; y/ D  .y i /2 < 0: (4.11)
i D1 i DrC1

In another canonical basis .e0i / of En , the representative matrix T of T has again

the form (4.3), with r 0 positive numbers and s 0 negative numbers in the principal
diagonal such that r 0 C s 0 D r C s D p. Denoting by E 0C and E 0 the vector
subspaces of En generated, respectively, by .e01 ; : : : ; e0r 0 / and .e0r 0 C1 ; : : : ; e0r 0 Cs 0 /,
again we have the decomposition
En D E 0C ˚ E 0 ˚ E0 ;
46 4 Euclidean and Symplectic Vector Spaces

and (4.11) gives

r 0 0 Cs 0
rX
X
T.x; x/ D .x 0i /2 > 0; T.y; y/ D  .y 0i /2 < 0:
i D1 i Dr 0 C1

We can easily prove that the intersection E C \.E 0 ˚E0 / D f0g. In fact, if x 2 E C
and x ¤ 0, then (4.11)1 is satisfied; further, any y 2 E C , y ¤ 0, verifies (4.11)2.
Consequently, the subspace E C ˚ E 0 ˚ E0 has a dimension r C n  r 0  n, so
that r  r 0 . Applying this reasoning to E 0C ˚ E  ˚ E0 , we obtain r D r 0 . 
Definition 4.3. The integer number r is called an index of T, whereas the difference
r  s is the signature of T.
Definition 4.4. Let T belong to T2 .En /. The map q W En ! < such that

q.x/ D T.x; x/

is said to be the quadratic form associated with T. Since q.x/ vanishes identically
when T is skew-symmetric, throughout the following sections we refer only to
symmetric tensors.
Definition 4.5. A symmetric tensor T 2 T2 .En / is said to be positive semidefinite
if 8x 2 En the following results are obtained:

q.x/  T.x; x/  0: (4.12)

If (4.12) assumes a zero value if and only if x D 0, then T is positive definite.

By adopting a canonical basis in En and resorting to Theorem 4.3, we conclude
that s D 0 when T is positive semidefinite. Further, T is positive definite if and only
if r D n. The character of positive definiteness of T does not depend on the basis
.ei / of En . In fact, in any basis (4.12) can be written as

Tij x i x j  0; 8.x i / 2 <n : (4.13)

It is well known that quadratic forms are positive semidefinite (positive definite) if
and only if all the principal minors T i , i D 1; : : : ; n, of the representative matrix
T D .Tij / of T with respect to the basis .ei / satisfy the following conditions:

T i  0; .T i > 0/; i D 1; : : : ; n: (4.14)

Theorem 4.4. Let T 2 T2 .En / be a symmetric and positive semidefinite .0; 2/-
tensor. Then, 8x; y 2 En Schwarz’s inequality
p p
jT.x; y/j  T.x; x/ T.y; y (4.15)
4.3 Pseudo-Euclidean Vector Spaces 47

and Minkowski’s inequality

p p p
T.x C y; x C y/  T.x; x/ C T.y; y/ (4.16)
hold.
Proof. Note that 8a 2 < and 8x; y 2 En the following results are obtained:

T.ax C y; ax C y/ D a2 T.x; x/ C 2aT.x; y/ C T.y; y/  0: (4.17)

Since the left-hand side of preceding inequality is a second-degree polynomial in

the variable a, we can state that its discriminant is not positive, that is,

jT.x; y/j2  T.x; x/T.y; y/  0; (4.18)

and (4.15) is proved. To prove (4.16), we start from the inequality

T.x C y; x C y/ D T.x; x/ C 2T.x; y/ C T.y; y/

 T.x; x/ C 2jT.x; y/j C T.y; y/;

which in view of (4.15) implies

p
T.x C y; x C y/  T.x; x/ C 2 T.x; x/T.y; y/ C T.y; y/
p p
D ΠT.x; x/ C T.x; x/2 ;

and (4.16) is proved. 

4.3 Pseudo-Euclidean Vector Spaces

Definition 4.6. A pair .En ; g/ of a vector space En and a symmetric nondegenerate

.0; 2/-tensor is called a pseudo-Euclidean vector space.
In such a space the tensor g introduces the scalar product x  y of two arbitrary
vectors x; y 2 En , which is a mapping

.x; y/ 2 En  En ! x  y 2 <;

where
x  y D g.x; y/: (4.19)

The scalar product has the following properties:

x  y D y  x;
x  .y C z/ D x  y C x  z;
48 4 Euclidean and Symplectic Vector Spaces

a.x  y/ D .ax/  y;
x  y D 0; 8y 2 En ) x D 0; (4.20)

8x; y; z 2 En . In fact, the first property follows from the symmetry of g; since g
is bilinear, the second and third properties hold; finally, the fourth property follows
from the positive-definite character of g.
A vector x is said to be a unit vector if

x  x D ˙1: (4.21)
Two vectors are orthogonal if
x  y D 0: (4.22)
In a basis .ei / of En the scalar product determines a symmetric nonsingular
matrix G D .gij / whose coefficients are

gij D ei  ej : (4.23)

In contrast, if a symmetric nonsingular matrix G D .gij / is given, then a scalar

product can be defined by the relation

x  y D gij x i y j (4.24)

in any basis .ei / of En . Theorem 4.1 guarantees the existence of canonical bases .ei /
in which the representative matrix G of the scalar product has the diagonal form
 
Ir O
GD ; (4.25)
O Is

where r C s D n. In these bases, which are called generalized orthonormal bases,

the scalar product assumes the following form:

r
X rCs
X
xyD xi yi  xi y i : (4.26)
i D1 i DrC1

4.4 Euclidean Vector Spaces

Definition 4.7. A Euclidean vector space is a pair .En ; g/, where En is a vector
space and g is a symmetric positive-definite .0; 2/-tensor.
In a Euclidean vector space the scalar product verifies the further condition

x  x  0; 8x 2 En ; (4.27)
4.4 Euclidean Vector Spaces 49

where the equality to zero holds if and only if x D 0. Property (4.27) makes it
possible to define the length or modulus of a vector as follows:
p
jxj D x  x: (4.28)

Also, in a Euclidean space the Schwarz and Minkowski inequalities (4.15) and
(4.16) hold. With the new notations, they assume the following form:

jx  yj  jxjjyj; (4.29)
jx C yj  jxj C jyj: (4.30)

Noting that (4.29) is equivalent to the condition

xy
1   1; (4.31)
jxjjyj

we can define the angle ' between two vectors x and y by the position
xy
cos ' D : (4.32)
jxjjyj

It is easy to verify that the vectors belonging to an orthonormal system are

independent; moreover, they form a basis if their number is equal to the dimension
of En . In any basis .ei / of En , the length of x 2 En and the angle between two
vectors x and y assume the form
q
jxj D gij x i x j ; (4.33)

gij x i y j
cos ' D p p : (4.34)
gij x i x j gij y i y j

Definition 4.8. Let .ei / be a basis of the Euclidean vector space En . We call
covariant components of the vector x relative to the basis .ei / the quantities

xi D x  ei D gij x j : (4.35)

When a basis .ei / is given, there is a one-to-one map between vectors and their
contravariant components. The same property holds for the covariant components.
In fact, it is sufficient to note that det.gij / ¤ 0 and refer to linear relations (4.35).
All the preceding formulae assume their simplest form relative to an orthonormal
basis .ei /. In fact, since in such a basis

ei  ej D ıij ; (4.36)
50 4 Euclidean and Symplectic Vector Spaces

we obtain also
n
X
xy D xi y j ; (4.37)
i D1
v
u n
uX
jxj D t .x i /2 ; (4.38)
i D1

xi D xi : (4.39)

After checking the advantage of the orthonormal bases, we understand the

importance of Schmidt’s orthonormalization procedure, which allows us to obtain
an orthonormal basis .ui / starting from any other basis .ei /. First, we set

u1 D e1 : (4.40)

Then, we search for a vector u2 such that

u2 D a21 u1 C e2 ; (4.41)
u1  u2 D 0: (4.42)

Introducing (4.42) into (4.43), we obtain the condition

a21 u1  u1 C u1  e2 D 0:

Since u1 ¤ 0, the preceding condition allows us to determine a21 and the vector u2
is not zero owing to the linear independence of e1 and e2 . Then, we search for a
vector u3 such that

u3 D a31 u1 C a22 u2 C u3
u1  u3 D 0;
u2  u3 D 0:

These relations imply the linear system

a31 u1  u1 C u1  e3 D 0;
a32 u2  u2 C u2  33 D 0;

which determines the unknowns a31 and a32 since the vectors u1 and u2 do not vanish.
Finally, the system .u1 ; u2 ; u3 / is orthogonal. After n steps, an orthogonal system
.u1 ; : : : ; un / is determined. Dividing each vector of this system by its length, we
obtain an orthonormal system.
4.5 Eigenvectors of Euclidean 2-Tensors 51

Definition 4.9. Let V be a vector subspace of En . Then, the set

V? D fx 2 En ; x  y D 0; 8y 2 V g ; (4.43)

containing all the vectors that are orthogonal to any vector of V , is said to be the
orthogonal complement of V .
Theorem 4.5. If En is a Euclidean vector space and V any vector subspace of En ,
then V? is a vector subspace of En ; in addition, the following results are obtained:

E n D V ˚ V? : (4.44)

Proof. If x1 ; x2 2 V? , a1 ; a2 2 <, and y 2 V , then we have that

.a1 x1 C a2 x2 /  y D a1 x1  y C a2 x2  y D 0;

and a1 x1 C a2 x2 2 V? . Further, we note that if .e1 ; : : : ; em / is a basis of V , then any

vector that is orthogonal to all the vectors of this basis is an element of V? . In fact,
when x  ei D 0, i D 1; : : : ; n, for any y 2 V we obtain that
m
X m
X
xyDx y i ei D y i x  ei D 0;
i D1 i D1

and then x 2 V? . Now, 8x 2 En we set

x0 D .x  e1 /e1 C    .x  em /em ;
x00 D x  x0 :

Since the vector x00 2 V? , the decomposition x D x0 C x00 is such that x0 2 V and
x00 2 V? . To prove that this decomposition is unique, we suppose that there is an-
other decomposition y0 Cy00 . Then, it must be that .y0 x0 /C.y00 x00 / D 0, where the
vector inside the first parentheses belongs to V , whereas the vector inside the other
parentheses belongs to V? . Finally, in a Euclidean space the sum of two orthogonal
vectors vanishes if and only if each of them vanishes and the theorem is proved.


4.5 Eigenvectors of Euclidean 2-Tensors

Definition 4.10. Let T be a .1; 1/-tensor of a Euclidean vector space En . We say

that the number  2 < and the vector x ¤ 0 are, respectively, an eigenvalue of T
and an eigenvector of T belonging to  if  and x satisfy the eigenvalue equation

T.x/ D x: (4.45)

52 4 Euclidean and Symplectic Vector Spaces

The property that T is linear implies that the set V of all the eigenvectors
belonging to the same eigenvalue  form a vector subspace of En . In fact, if a; b 2 <
and x, y 2 V , then we have that

T.ax C by/ D aT.x/ C bT.y/ D .ax C by/;

so that ax C by 2 V .
Definition 4.11. The dimension of the vector subspace associated with the eigen-
value  is called a geometric multiplicity of the eigenvalue ; in particular, an
eigenvalue with multiplicity 1 is also said to be simple. The set of all the eigenvalues
of T is called the spectrum of T. Finally, the eigenvalue problem relative to T
consists in determining the whole spectrum of T.
To find the eigenvalues of T, we start out by noting that in a basis .ei / of En , (4.45)
is written as  
Tji  ıji x j D 0; i D 1; : : : ; n: (4.46)

This is a homogeneous linear system of n equations in n unknowns x 1 ; : : : ; x n ,

which admits a solution different from zero if and only if

Pn ./  det.Tji  ıji / D 0: (4.47)

Now we show a fundamental property of the preceding equation: although the

components Tji of tensor T depend on the choice of the basis .ei /, the coefficients
of (4.47) do not depend on it. In fact, in the basis change
j
e0i D Ai ej ;

with the usual meaning of the symbols, we have that

Pn0 ./ D det.T0  I0 / D Œdet A1 .T  I/A

D det A1 det APn ./;

and then
Pn0 ./ D Pn ./: (4.48)

Since the polynomial Pn ./ does not depend on the basis .ei /, it is called a
characteristic polynomial of T. Denoting by Ii the coefficient of the power ni
and noting that I0 D .1/n , we can write Pn ./ as follows:

Pn ./ D .1/n n C I1 n1 C    C In : (4.49)

Remark 4.1. It is possible to verify that

Ii D .1/i Ji ; i D 1; : : : ; n; (4.50)
4.6 Orthogonal Transformations 53

where Ji is the sum of the all determinants of the principal minors of order i of
matrix T. In particular, I1 D T11 C    C Tnn and In D det T.
In conclusion, we have proved what follows.
Theorem 4.6. The eigenvalues of a .1; 1/-tensor T are the real roots of the
characteristic polynomial

Pn ./ D .1/n n C I1 n1 C    C In D 0: (4.51)

Definition 4.12. Equation (4.51) is the characteristic equation of the tensor T.

Further, the multiplicity of a root  of (4.51) is called the algebraic multiplicity
of the eigenvalue .
Let  be any real roots of (4.51), i.e., an eigenvalue of the spectrum of T.
Introducing  into (4.46), we obtain a linear homogeneous system whose solutions
form a subspace V of eigenvectors. The dimension of V is equal to k D n  p,
where p is the rank of the matrix T  I. In other words, we can find k independent
eigenvectors u1 ; : : : ; un , belonging to V , that form a basis of V . In particular, if
there exists a basis of En formed by eigenvectors of T belonging to the eigenvalues
1 ; : : : ; n , then the corresponding matrix T representative of T assumes the
following diagonal form:
0 1
1    0
T D @  A: (4.52)
0    n

The following theorem, whose proof we omit, is very useful in applications.

Theorem 4.7. Let T be a symmetric tensor of a Euclidean vector space En . Then,
all the eigenvalues of T are real and the dimension of the subspace V associated
with the eigenvalue  is equal to the multiplicity of . Further, eigenvectors
belonging to different eigenvalues are orthogonal to each other, and there exists
at least a basis of eigenvectors of T relative to which the matrix T, representative of
T, is diagonal.

4.6 Orthogonal Transformations

Definition 4.13. Let En be a Euclidean n-dimensional vector space. An endomor-

phism Q W En ! En is an orthogonal transformation if

Q.x/  Q.y/ D x  y; 8x:y 2 En : (4.53)

If the basis .ei / is orthonormal, then the n vectors Q.ei / are independent and
consequently form a basis of En . Therefore, Q is an isomorphism and the matrix
54 4 Euclidean and Symplectic Vector Spaces

Q, representative of Q in any basis .ei /, is not singular. The condition (4.53) can be
written in one of the following forms:

ghk Qik Qjk D gij ; (4.54)

QT GQ D G: (4.55)

In particular, relative to the orthonormal basis .ei /, in which G D I, the foregoing

relations can be written as follows:

QT Q D I , QT D Q1 : (4.56)

A matrix satisfying one of conditions (4.56) is said to be orthogonal.

Since the composite of two orthogonal transformations is still orthogonal, and
such are the identity transformation and the inverse transformation, the set of all
orthogonal transformations is a group O.n/, which is called an orthogonal group.
In view of (4.56) it follows that

det Q D ˙1: (4.57)

The orthogonal transformations of a three-dimensional Euclidean space E3 are

also called rotations; in particular, the rotations for which det Q D 1 are called
proper rotations. A group of rotations is denoted by O.3/, whereas the subgroup of
proper rotations is denoted by SO.3/. Finally, the orthogonal transformation I is
called the central inversion.
The following theorem is fundamental.
Theorem 4.8 (Euler). A rotation Q ¤ I of the three-dimensional vector space E3
always has the simple eigenvalue  D 1 whose corresponding eigenspace V1 is
one-dimensional and invariant under Q. Further, if the restriction Q? of Q to the
orthogonal complementary space V? is different from the identity, then  D 1 is the
only eigenvalue of Q.
Proof. Relative to an orthonormal basis .ei / of E3 the eigenvalue equation of Q is

.Qji  ıji /x j D 0;

where Q D .Qji / is an orthogonal matrix. On the other hand, the characteristic

equation is
P3 ./ D 3 C I1 2 C I2  C I3 D 0;

with [see (4.50)]

I1 D Q11 C Q22 C Q33 ; (4.58)

 2 2  1 1  1 1
Q2 Q3 Q1 Q3 Q1 Q2
I2 D    ; (4.59)
Q23 Q33 Q13 Q33 Q12 Q22
I3 D det Q D 1: (4.60)
4.6 Orthogonal Transformations 55

In view of (4.60), the characteristic equation can also be written as

3 C I1 2 C I2  C 1 D 0:

Denoting by Aij the cofactor of Qji , we have that

Aij
.Q1 /ij D ;
det Q

and recalling that Q is orthogonal, the preceding equation gives

j
.Q1 /ij D Aij D .QT /ij D Qi : (4.61)

This result shows that I2 D I1 and the characteristic equation becomes

3 C I1 2  I1  C 1 D 0;

so that  D 1 is a solution of the characteristic equation. This eigenvalue can have a

multiplicity of 3  p, where p is the rank of the matrix Q  I. A priori p can assume
the values 0; 1; 2. The value 0 must be excluded since it implies Q D I, against the
hypothesis of the theorem. If p D 1, then all the minors of order two vanish; in
particular, the following minors vanish:
 
Q11  1 Q21
D A33  Q11  Q22 C 1 D 0;
Q12 Q22  1
 
Q11  1 Q31
D A22  Q11  Q33 C 1 D 0;
Q13 Q33  1
 
Q22  1 Q32
D A11  Q22  Q33 C 1 D 0:
Q23 Q33  1

In view of (4.60), the preceding equations imply Q11 D Q22 D Q33 D 1. Taking into
account this result and the orthogonality conditions

.Q11 /2 C .Q12 /2 C .Q13 /2 D 1;

.Q21 /2 C .Q23 /2 C .Q23 /2 D 1;
.Q31 /2 C .Q32 /2 C .Q33 /2 D 1;

we obtain Qji D 0, i ¤ j , and again we have that Q D I, against the hypothesis.

Consequently, p D 2 and the eigenvalue  D 1 is simple.
Let V1 be the eigenspace belonging to the eigenvalue  D 1, and denote by
V? the two-dimensional vector space of all the vectors that are orthogonal to V1 .
56 4 Euclidean and Symplectic Vector Spaces

The vectors .u1 ; u2 ; u3 /, where .u1 ; u2 / is an orthonormal basis of V? and u3 a unit

vector of V1 , form a basis of E3 . Since Q is orthogonal, Q.v/ 2 V? , 8v 2 V? . In
particular, we have that
1 2
Q.u1 / D Q1 .u1 / C Q1 .u2 /;
1 2
Q.u2 / D Q2 .u1 / C Q2 .u2 /;
Q.u3 / D u3 ;

and the representative matrix of Q relative to the basis .u1 ; u2 ; u3 / gives

01 1
1
Q1 Q 2 0
B C
Q D @ Q21 Q22 0 A ;
0 0 1

whereas the orthogonality conditions become

1 2
.Q1 /2 C .Q1 /2 D 1;
1 2
.Q2 /2 C .Q2 /2 D 1;
1 1 2 2
.Q1 /.Q2 / C .Q1 /.Q2 / D 1:

These relations imply the existence of an angle ' 2 .0:2/ such that
0 1
cos ' sin ' 0
Q D @ sin ' cos ' 0 A : 
0 0 1

4.7 Symplectic Vector Spaces

Definition 4.14. A symplectic vector space is a pair .E2n ; /, where E2n is a vector
space with even dimension and  a skew-symmetric nondegenerate .0; 2/-tensor.
Remark 4.2. It is fundamental to require that the dimension of the vector space be
odd. In fact, owing to Theorem 4.2, in a vector space with even dimension, any
skew-symmetric .0; 2/-tensor is always degenerate.
Definition 4.15. In the symplectic vector space .E2n ; /, the antiscalar product is
the map .x; y/ 2 E2n  E2n ! Œx; y 2 < such that

Œx; y D .x; y/: (4.62)

4.7 Symplectic Vector Spaces 57

An antiscalar product has the following properties:

Œx; y D Œy; x; (4.63)

Œx; y C z D Œx; y C x; z; (4.64)
a Œx; y D Œax; y ; (4.65)
Œx; y D 0; 8y 2 E2n ) x D 0: (4.66)

In fact, the first property follows from the skew symmetry of . The second and
third properties follow from the bilinearity of . Finally, the fourth property is due
to the fact that  is nondegenerate.
We say that the vectors x and y are antiorthogonal if

Œx; y D 0: (4.67)

In view of (4.66), we can state that the only vector that is antiorthogonal to any other
vector is 0. Further, any vector is antiorthogonal to itself. In any basis .ei / of E2n ,
the representative matrix of  is skew-symmetric and the antiscalar product can be
written as
Œx; y D ij x i y j ; (4.68)

where x i and y i are, respectively, the components of x and x relative to .ei /. In a

canonical basis .ui / (Sect. 4.1), which is also called a symplectic basis,  is given
by the matrix
 
O I
D ; (4.69)
I O

where O is a zero nn matrix and I is a unit nn matrix. Moreover, in a symplectic
basis, in view of (4.68) and (4.69), we obtain
   
ui ; uj D unCi ; unCj ; (4.70)
 
ui ; unCj D ıij ; (4.71)
n
X
Œx; y D .x i y nCi  x nCi y i /: (4.72)
i D1

An automorphism S W E2n ! E2n of a symplectic vector space E2n is a

symplectic transformation if the antiscalar product does not change, that is,

ŒS.x; y/ D Œx; y ; 8x; y 2 E2n : (4.73)

The symplectic transformations correspond to the orthogonal transformations of a

Euclidean space. If we denote by S D .Skh / the matrix representative of S in any
basis .ei /, then condition (4.73) assumes the form

hk Sih Sjk D ij : (4.74)

58 4 Euclidean and Symplectic Vector Spaces

In matrix form this condition can be written as

ST S D : (4.75)

In a symplectic basis, when we adopt the notation

 
AB
SD ; (4.76)
CD

condition (4.75) can explicitly be written as

 T T     
A B O I AB O I
D :
CT DT I O CD I O

In conclusion, a transformation S is symplectic if and only if

AT B  CT A D O; (4.77)
T T
A D  C C D I; (4.78)
T T
B B  D A D O; (4.79)
T T
B D  D C D I: (4.80)

The set Sp.E2n ; / of the symplectic transformations of E2n is called a

symplectic transformation group of E2n . To verify that Sp.E2n ; / is a group,
we start out by noting that, 8S1 ; S2 2 Sp.E2n ; /,

ŒS1 .S2 .x//; S1 .S2 .y// D ŒS2 .x/; S2 .y/ D Œx; y

and S1 S2 2 Sp.E2n /. Further, from (4.75), since det  D 1, it follows that

.det S/2 D 1;

and there is the inverse automorphism S1 . Finally,

   
Œx; y D S1 .S.x//; S1 .S.y// D S1 .x/; S1 .y/ ;

and S1 2 Sp.E2n /.

4.8 Exercises

1. Let V be a subspace of the Euclidean vector space <4 generated by the vectors
.1; 0; 1; 3/ and .0; 1; 1; 2/. Determine the subspace V? of the vectors that are
orthogonal to V .
4.8 Exercises 59

2. Let E be the four-dimensional vector space of the 2  2 matrices with real

coefficients. If A and B are two arbitrary matrices of E, show that

A  B D tr.BT A/

is a scalar product and that

       
10 01 00 00
; ; ;
00 00 10 01

is an orthonormal basis of E. p p
3. Find an orthogonal 2 2 matrix whose first row is either .1= 5; 2= 5/ or .1; 2/.
4. Let En be a vector space, and denote by u a given vector of En . Prove that
•
† D fv 2 En ju  v D 0g
is an .n  1/-dimensional subspace of En ;
•
En D † ˚ ‚;
where ‚ D fv D au; a 2 <g;
• The orthogonal projection P† .v/ of v 2 En onto † is given by

P† .v/ D v  .v  u/u:

5. Let En be the n-dimensional vector space En of the polynomials P .x/ of

degree n in the interval .0; 1/ (Sect. 1.6). Prove that
Z 1
P .x/  Q.x/ D P .x/Q.x/dx; P .x/; Q.x/ 2 En
0

defines a Euclidean scalar product in En .

Chapter 5
Duality and Euclidean Tensors

5.1 Duality

In this section, we show that when En is a Euclidean vector space, there is an

isomorphism among the tensor spaces Tsr .En / for which r C s has a given value.
In other words, we show the existence of an isomorphism between En and En , of
isomorphisms between T02 , T11 , and T20 , and so on.
Theorem 5.1. Let .En ; g/ be a Euclidean vector space. Then, the map  W x 2
En ! !x 2 En , such that

!x .y/ D g.x; y/ D x  y; (5.1)

defines an isomorphism that is called a duality.

Proof. First, (5.1) defines a covector since g is linear with respect to y. Moreover,
the linearity of g with respect to x implies that the mapping  is linear. Finally,
ker  D fx 2 En ; .x/ D !x D 0g contains the vectors x such that

!x .y/ D g.x; y/ D x  y D 0; 8y 2 En :

But g is a nondegenerate 2-tensor and the preceding condition implies that x D 0.

Therefore, ker  D f0g, and  is an isomorphism.
Remark 5.1. Let .ei / be a basis of En and let . i / be its dual basis. The basis . ei /
of En corresponding to .ei / in the isomorphism  is not equal to . i /. In fact, it is
such that
 ei .y/ D g.ei ; y/ D g.ei ; ej /y j D gij y j D gij  j .y/;

i.e.,
 ei D gij  j : (5.2)


A. Romano, Classical Mechanics with Mathematica R

, Modeling and Simulation 61
in Science, Engineering and Technology, DOI 10.1007/978-0-8176-8352-8 5,
© Springer Science+Business Media New York 2012
62 5 Duality and Euclidean Tensors

In the bases .ei / and . i /, isomorphism (5.1) has the following representation:
!i D gij x j ; (5.3)

whereas the inverse map is

x i D g ij !j : (5.4)

In other words, isomorphism (5.1) associates with any vector x a covector whose
components in the dual basis are equal to the covariant components of x in the basis
.ei /. We can also state that (5.1) allows us to identify the vectors of En and the
covectors of En .
Definition 5.1. Any pair .x; !x / in which !x is given by (5.1) is said to be
a Euclidean vector. Further, the real numbers x i and !i satisfying (5.3) are
respectively called contravariant components and covariant components of the
Euclidean vector x.
Henceforth a Euclidean vector will be denoted by the first element of a pair
.x; !x /.

5.2 Euclidean Tensors

p
The isomorphism (5.1) can be extended to the tensor spaces Tsr .En / and Tq .En /,
for which r Cs D pCq. For the sake of simplicity, we prove the preceding statement
for the spaces T02 .En /, T11 .En /, and T20 .En /.
Consider the maps

T 2 T20 .En / ! T0 2 T11 .En / ! T00 2 T02 .En /

such that
T.x; y/ D T0 .x; !y / D T00 .!x ; !y /; 8x; y 2 En : (5.5)

These linear maps are isomorphisms since, adopting a basis in En and its dual basis
in En , they have the following coordinate representations:

Tj0i D g ih Thj ; T 00ij D g ih g j k Thk : (5.6)

Definition 5.2. The triad .T; T0 ; T00 / is called a Euclidean double tensor. Further,
the components Tij of T in the basis . i ˝  j /, the components Tji of T0 in the
basis .ei ˝  j /, and the components T ij of T00 in the basis .ei ˝ ej / are respectively
called covariant, mixed, and contravariant components of the Euclidean tensor
.T; T0 ; T00 /.
Henceforth we will denote a Euclidean tensor by the first element of a triad. It is
evident how to extend the preceding considerations to any tensor space Tsr .En /.
5.3 The Levi–Civita Tensor 63

In view of the foregoing results, we conclude that the tensor spaces T r .En / and
Tr .En / are isomorphic. We now verify that their subspaces ƒr En and ƒr En are
also isomorphic, and we determine the form of this isomorphism. For the sake of
simplicity, we refer to the case r D 2. If T 2 ƒr En , we have that
XX XX
T.ij / D gih gj k T hk C gih gj k T hk
i <j h<k i <j h>k
XX XX
D gih gj k T hk C gih gj k T kh
i <j h<k i <j h<k

D .gih gj k  gi k gj h /T .hk/ :

In conclusion, we have
 
gih gi k
T.ij / D det T .hk/ : (5.7)
gj h gj k
More generally, we could prove that
0 1
gi1 h1    gi1 hr
T.i1 :::ir / D det @          A T .h1 :::hr / : (5.8)
gir h1    gir hr
In particular, when r D n, (5.8) gives
T1n D det.g/T 1:::n : (5.9)

5.3 The Levi–Civita Tensor

Definition 5.3. Let En be a vector space. A .0; k/- pseudotensor density of

weight p is a multilinear map T W Enk ! < such that under a basis change
e0j D Aij ei (5.10)

of En , the components of T are transformed according to the law

Tj01 :::jk D sgn.A/jAjp Aij11    Aijkk Ti1 :::ik ; (5.11)
where
A D det.Aij / (5.12)

and p is a nonnegative rational number. In particular, if p D 0, then T is said to be

a pseudotensor.
Starting from the formula
gij0 D Ahi Akj ghk ;
64 5 Duality and Euclidean Tensors

and introducing the notations

g 0 D det.gij0 /; g D det.gij /; (5.13)

we obtain
g 0 D A2 g; jg 0 j D A2 jgj;

so that
p p
jg 0 j D ˙A jgj; (5.14)

p the C sign when A > 0 and the  sign when A < 0. This formula
where we take
shows that jgj is a pseudoscalar density of weight 1.
Introduce the skew-symmetric Levi–Civita symbol
8
< 0;
i1 :::in D 1; (5.15)
:
1;

where the value 0 corresponds to a permutation i1 ; : : : ; in with two or more indices

equal, the value 1 to an even permutation i1 ; : : : ; in , and the value 1 to an odd
permutation i1 ; : : : ; in . Before proceeding, we recall the following formula relative
to the development of a determinant A:

Aj1 :::jn D i1 :::in Aij11    Aijnn : (5.16)

Let En be a Euclidean vector space. We prove that the multilinear map , which
in a basis .ei / of En has the following components:
p
i1 :::in D gi1 :::in ; (5.17)

is a .0; n/-pseudotensor. In fact, in view of (5.14) and (5.16), under the basis change
(5.10), we have that
p p
0j1 :::jn D jg 0 jj1 :::jn D ˙A jgjj1 :::jn
p
D ˙Aij11    Aijnn gi1 :::in

D ˙Aij11    Aijnn i1 :::in ; (5.18)

and our statement is proved. In conclusion,  is a skew-symmetric tensor only under

congruent basis changes.
The skew-symmetric tensor  is an element of ƒn .En /. Since this space has
dimension 1 (Chap. 3), we can write
p
D jgj 1 ^    ^  n ; (5.19)
5.4 Exercises 65

where . i / is a basis of the dual vector space En . Taking into account what we have
just proved about the n-form  and recalling the results of Sect. 3.4, we can state
that  is a volume form that is invariant with respect to congruent basis changes.
We conclude this section with the following definition.
Definition 5.4. The Hodge star operator is a linear mapping

 W ƒk .En / ! ƒnk .En / (5.20)

such that
? T.i1 i.nk/ / D i1 i.nk/ j1 jk T .j1 jk / : (5.21)

The skew-symmetric tensor ?T is called the adjoint or the dual form of T.

5.4 Exercises

1. Verify that in an arbitrary base .ei / of the Euclidean vector space En , the
eigenvalue equation (4.47) relative to the Euclidean tensor T assumes the form

.Tij  gij /uj D 0; (5.22)

where gij D ei  ej .
2. Show that the components of the cross-product u  v of the vectors u and v of
the three-dimensional Euclidean space E3 are

.u  v/i D ij k uj vk : (5.23)

3. Using the Levi–Civita pseudotensor, evaluate the components of the following

vectors of the Euclidean space E3 :

u  .v  w/; .u  v/  .w  x/:
Chapter 6
Differentiable Manifolds

6.1 Differentiable Manifolds

Let U be an open set of <n . The real-valued function f W U ! < is said to be of

class C k .U / or a C k function in U , where k  0, if it is continuous with its partial
derivatives up to the order k. In particular, a C 0 function in U is a continuous one.
A map
f W .x 1 ; : : : ; x n / 2 U ! .y 1 ; : : : ; y m / 2 <m

is of class C k if any i th projection pr i ı f

y i D pr i ı f .x 1 ; : : : ; x n /  y i .x 1 ; : : : ; x n /

is a C k function.
A homeomorphism f W U ! V , where V is an open set of <n , is a continuous
map with its inverse. Finally, the map f is a diffeomorphism of class C k if both
f and f 1 are C k maps. A diffeomorphism is represented by an invertible system
of functions y i .x 1 ; : : : ; x n /, .x 1 ; : : : ; x n / 2 U , i D 1; : : : ; n, of class C k together
with the inverse functions. It is well known that the condition
 i
@y
det ¤0
@x j 0

at the point .x01 ; : : : ; x0n / 2 U is a sufficient condition for the invertibility of these
functions in a neighborhood of the point .x01 ; : : : ; x0n /.
A differentiable manifold can roughly be defined as an n-dimensional surface
embedded in <m , n < m. This approach to the analysis of differentiable manifolds
is more intuitive but not convenient for the following reasons. First, determining the
lowest-dimension n of the space <n in which we can embed the manifold is not an
easy task. For instance, the plane curves can be embedded in <2 , whereas the skew
curves can be embedded in <3 . Further, in this approach the geometric objects on the

A. Romano, Classical Mechanics with Mathematica R

, Modeling and Simulation 67
in Science, Engineering and Technology, DOI 10.1007/978-0-8176-8352-8 6,
© Springer Science+Business Media New York 2012
68 6 Differentiable Manifolds

X U

O ϕ xn

ϕ(U)

x1

Fig. 6.1 n-dimensional manifold

manifold are defined starting from the space <n in which they are embedded. It is
much more interesting to build the geometry of a manifold in an intrinsic way. This
approach makes it possible to answer questions like the following ones: Is it possible
to recognize if a manifold is a sphere staying on it? Is it possible to recognize the
geometric structure of the three-dimensional space in which we live by measures
that are necessarily internal to our space?
Definition 6.1. Let n be a positive integer number and denote by X a Hausdorff1
paracompact topological space.2 X is said to be an n-dimensional manifold if, 8x 2
X , there exist an open neighborhood U of x and a homeomorphism ' W U !
'.U /  <n . The pair .U; '/ is called a chart of domain U and coordinate map '.
Finally, the n numbers

.x 1 ; : : : ; x n / D '.x/ 2 '.U /
are the coordinates in the chart .U; '/ (Fig. 6.1).
Definition 6.2. An atlas of class C k on an n-dimensional manifold X is a
collection ˛ of charts on X satisfying the following conditions:
• The collection of the domains of the charts of ˛ is an open covering of X ;
• 8.U; '/; .V; / 2 ˛ the map
ı  1 W '.U \ V / ! .U \ V / (6.1)
is a C k diffeomorphism, called a coordinate transformation.
(See Fig. 6.2.)

1
A topological space X is a Hausdorff
T space if, 8x; y 2 X, x ¤ y, there are neighborhoods U ,
V of x, y, respectively, such that U V D ;.
2
A Hausdorff space is paracompact if every open covering contains a subcovering that is locally
finite.
6.1 Differentiable Manifolds 69

X
U

O ψ ϕ xn

ψ(U) ϕ(U )

x1
ψ ° ϕ−1

Fig. 6.2 Coordinate transformation

A chart .V; / is compatible with the atlas ˛ if, 8.U; '/ 2 ˛, map (6.1) is
of class C k . We call the collection of all the charts that are compatible with ˛ a
maximal atlas ˛ of X .
Definition 6.3. The pair Vn D .X; ˛/ is called an n-dimensional differentiable
manifold of class C k .
A difficult theorem of Whitney proves that any C k manifold Vn , k  1, becomes
an analytic manifold (i.e., the coordinate transformations between charts of an
atlas are analytic diffeomorphisms) by discarding a suitable collection of C k charts
belonging to the original maximal atlas. It is even more difficult to show that a C 0
manifold may fail to become a C 1 manifold.
Now we show how to obtain differentiable manifolds.
• Let U  <n be an open set, and denote by .u1 ; : : : ; un / a point of U . Let S be
the locus of the points .x 1 ; : : : ; x l / 2 <l , n < l, given by the set of C 1 functions

x 1 D x 1 .u1 ; : : : ; un /;
::::::::::::::::::::::::::::: (6.2)
x l D x l .u1 ; : : : ; un /;

whose Jacobian matrix

 
@x i
J D ;
@u˛

i D 1; : : : ; l, ˛ D 1; : : : ; n, has rank n at any point of U . In other words, S is

a regular n-dimensional surface of <l defined by the parametric equations (6.2).
In particular, for l D 3 and n D 1; 2, we obtain regular curves and surfaces
of <3 , respectively. We sketch the proof that all these regular surfaces are
70 6 Differentiable Manifolds

n-dimensional differentiable manifolds. First, S becomes a topological space

when it is equipped with the topology induced by <l . It is well known that the
open sets of this topology are obtained by intersecting the open sets of <l with
S . Further, suppose that at the point x0 2 S , which is the image of .u10 ; : : : ; un0 /,
the determinant of the minor
 i
@x
;
@u˛

i D 1; : : : ; n, ˛ D 1; : : : ; n, does not vanish. Then, the first n equations (6.2)

define a homeomorphism ' between a neighborhood U of x0 and a neighborhood
'.U / of .u10 ; : : : ; un0 /. We do not prove that all the coordinate transformations
among these charts are of class C k . Some examples of manifolds obtained by
this procedure are given in the exercises at the end of this chapter.
• Differentiable manifolds can also be obtained by the implicit representation of
C k n-dimensional surfaces of <l . Let S be such a surface implicitly defined by
the following system:

f1 .x 1 ; : : : ; x l / D 0;
::::::::::::::::::::::::::;
fm .x 1 ; : : : ; x l / D 0; (6.3)

where m < l, the functions f˛ , ˛ D 1; : : : ; m, are of class C k , and the Jacobian

matrix
 
@f˛
J D ; (6.4)
@x i

i D 1; : : : ; l, has rank equal to m. Again, S becomes a topological space with

the topology induced by <l . Further, let x0 D .x01 ; : : : ; x0n / be a point of S ,
and suppose that the determinant of the minor formed with the first m rows and
m columns of (6.4) does not vanish at x0 . Then, the m equations (6.3) can be
written as
x 1 D x 1 .x mC1 ; : : : ; x l /;
:::::::::::::::::::::::::::::::
x m D x m .x mC1 ; : : : ; x l /;

in a neighborhood V  <n , n D l  m, of .x01 ; : : : ; x0n /. The preceding equations

define a homeomorphism between V and the neighborhood

U D f.x 1 .x mC1 ; : : : ; x l /; : : : ; x m .x mC1 ; : : : ; x l /; x mC1 ; : : : ; x l /;

j.x mC1 ; : : : ; x l / 2 V g (6.5)

on S (see exercises at the end of the chapter).

6.2 Differentiable Functions and Curves on Manifolds 71

• A manifold can be obtained by the topological product of two manifolds. Let Vn

be a C k n-dimensional manifold, and let Wm be a C k m-dimensional manifold.
First, we equip Vn  Wm with the product topology. If ˛ D f.Ui ; 'i /gi 2I is a C k
atlas of Vn and ˇ D f.Vj ; j /gj 2J is a C k atlas of Wm , then it is easy to verify
that fUi  Vj ; .'i ; j /g.i;j /2I J is a C k atlas of Vn  Wm , which becomes a C k
.n C m/-dimensional manifold (see exercises at the end of the chapter).
• A manifold can be defined by a collection .Ui /i 2I of an open set of <n and a set
of diffeomorphisms among their parts (see exercises at the end of the chapter).

6.2 Differentiable Functions and Curves on Manifolds

Definition 6.4. Let Vn be an n-dimensional differentiable manifold, and denote by

˛ an atlas on Vn . We say that the real-valued function f W Vn ! < is a C k function
on Vn if the function
f ı ' 1 W '.U / ! < (6.6)

is a C k function 8.U; '/ 2 ˛.

Definition 6.5. A C h curve  on the C k manifold Vn , h  k, is a map  W Œa; b 
< ! Vn such that, 8.U; '/ 2 ˛, where ˛ is a C k atlas of Vn , the map

' ı  W Œa; b ! <n (6.7)

is a C h map (Fig. 6.3). The real-valued functions

x i .t/ D pri ı ' ı .t/; t 2 Œa; b;

are the parametric equations of  in the chart .U; '/. The curve  is closed if
.a/ D .b/.
Let .x i / be the coordinates defined by the chart .U; '/. If x0 D .x0i / 2 U is a
point of U , the i th coordinate curve at x0 is a curve with the following parametric
equations:

Vn U xn

ϕ(U )
x1
O t O
a b

Fig. 6.3 Curve on a manifold

72 6 Differentiable Manifolds

F
Vn U V
Wm

ym
O xn O

ϕ(U) ψ(V)

x1 y1

Fig. 6.4 Map between manifolds

x 1 D x01 ;
:::::::::::::;
x i D t;
::::::::::::::;
x n D x0n : (6.8)

Definition 6.6. Let Vn be an n-dimensional manifold with a C k atlas ˛, and let Wm

be an m-dimensional manifold with a C k atlas ˇ. The map

F W Vn ! W m

is of class C k if

ı F ı ' 1 W '.U /  <n ! .V /  <m ; (6.9)

is C k , 8.U; '/ 2 ˛ and 8.V; / 2 ˇ (Fig. 6.4).

If .x i /, i D 1; : : : ; n, are the coordinates relative to the chart .U; '/ and .y ˛ /,
˛ D 1; : : : ; m, the coordinates relative to .V; /, then map (6.9) is equivalent to a
system of m C k functions of n real variables:

y ˛ D y ˛ .x 1 ; : : : ; x n /; ˛ D 1; : : : ; m: (6.10)

6.3 Tangent Vector Space

We denote by  W Œa; b ! Vn a C k curve on a differentiable manifold Vn and by

F .x/ the <-vector space of the C k functions in a neighborhood of a point x D .t/.
6.3 Tangent Vector Space 73

Definition 6.7. The tangent vector to the curve  at x is the map

Xx W F .x/ ! < (6.11)

such that
 
d
Xx f D .f ı .t// : (6.12)
dt t

In other words, a tangent vector is defined as an operator that associates to any

C k function about the point x of the curve  the directional derivative along .t/ at
the point x. In the coordinates .x i / relative to the chart .U; '/ on Vn , we have that

f ı .t/ D f ı ' 1 ı ' ı .t/ D f ı ' 1 .x 1 .t/; : : : ; x n .t//;

where .x 1 .t/; : : : ; x n .t// are the parametric equations of  in the chart .U; '/. Then,
(6.12) gives
 
@ 1 dx i
Xx f D .f ı ' / : (6.13)
@x i '.x/ dt

To better understand the preceding definition, we consider a curve .t/ D

.x 1 .t/; x 2 .t/; x 3 .t// in the Euclidean three-dimensional space E3 . The directional
derivative of a C 1 function f .x 1 ; x 2 ; x 3 / along .t/ is given by
d
.f .x 1 .t/; x 2 .t/; x 3 .t// D .t r/x f  Xx f; (6.14)
dt
where t D .dx i .t/=dt/ is the tangent vector to  at the point x D .x i .t//. In other
words, by (6.14), a derivation operator corresponds to any vector t. It is evident
that, if the directional derivatives of three independent functions are given at x, then
(6.14) leads to a unique vector t. We note that our definition of tangent vector as
derivative operator does not requires an environment containing the manifold Vn .
Consider the set Tx Vn of all the maps (6.12) obtained upon varying the curve
.t/ at the point x 2 Vn . This set, equipped with the operations

.Xx C Yx /f D Xx f C Yx f;
.aXx /f D aXx f;

becomes an <-vector space that is called the tangent vector space to the manifold
Vn at point x.
Theorem 6.1. Let .x i / be a coordinate system relative to the chart .U; '/ of the
manifold Vn . Then, the relations
   
@ @ 1
f D .f ı ' / (6.15)
@x i x0 @x i '.x0 /

define n independent vectors tangent to the coordinate curves.

74 6 Differentiable Manifolds

Proof. Denoting by  i .t/ the i th coordinate curve crossing x0 , that is, the curve
with the following parametric equations,

x 1 D x01 ; : : : ; x i D t; : : : ; x n D x0n ;

the directional derivative along  i .t/ is

     j
d i @ 1 dx
.f ı  .t// D j
.f ı ' /
dt t0 @x '.x0 / dt t0
   
@ j @
j
.f ı ' 1 / ıi D .f ı ' 1 / ;
@x '.x0 / @x i '.x0 /

and relations (6.15) define n vectors tangent to the coordinate curves. Their linear
independence is proved applying the linear combination
 
j @
 D0
@x j x0

to the coordinate function x i D pr i ı ' and recalling (6.15). In fact, we obtain

 
@
j x i D j ıji D i D 0: 
@x j x0

From this result and (6.13) comes the following theorem.

Theorem 6.2. The tangent space Tx Vn is an n-dimensional <-vector space, and
the vectors .@=@x i /x form a basis of Tx Vn , which is called a holonomic basis or
natural basis relative to the coordinates .x i /. Therefore, 8Xx 2 Tx Vn the following
result is obtained:
 
i @
Xx D X ; (6.16)
@x i x

where the real numbers X i are the components of Xx relative to the natural basis.
It is fundamental to determine the transformation formulae of the natural bases
and the components of a tangent vector for a change .x i / ! .x 0j / of the local
coordinates. From (6.15) we obtain that
     
@ @x j @ j @
D  Ai ; (6.17)
@x 0i x @x 0i @x j x @x j x

and, consequently,
@x 0i j
X 0i D X  .A1 /ij X j : (6.18)
@x j
6.4 Cotangent Vector Space 75

6.4 Cotangent Vector Space

Definition 6.8. The dual vector space of Tx Vn (Sect. 2.1) is called the cotangent
vector space Tx Vn .
If Tx Vn is referred to as the natural basis .@=@x i /x relative to the coordinates .x i /,
the dual basis . ix / is characterized by the conditions

 ix .Xx / D Xxi ; (6.19)

where Xxi are the components of Xx relative to the basis .@=@x i /x .

Definition 6.9. If f 2 F .x/, the differential .df /x of f at the point x 2 Vn is the
linear map
.df /x W Tx Vn ! <; (6.20)

such that
.df /x Xx D Xx f; Xx 2 T x V n : (6.21)

In a natural basis relative to the coordinates .x i / of the chart .U; '/ of Vn , (6.21)
gives
.df /x Xx D ai Xxi ; (6.22)

where
 
@
ai D f: (6.23)
@x i x

In particular, for the differentials of the coordinate functions we obtain

 
i i @
.dx /x Xx D Xx x D Xxh x i D Xxh ıhi D Xxi : (6.24)
@x i x

Comparing (6.24) and (6.20), we can state the following theorem.

Theorem 6.3. The differentials .dx i /x of the coordinate functions form the dual
basis of the cotangent space Tx Vn . Consequently, any covector !x 2 Tx Vn can be
written as
!x D !i .dx i /x ; (6.25)

where
  
@
!i D !x : (6.26)
@x i x
76 6 Differentiable Manifolds

Owing to the results of Sect. 2.2, we can state that under the coordinate change
.x i / ! .x 0i /, the following transformation formulae of the dual bases and the
components of a covector hold [see (6.17)]:

@x 0i
.dx 0i /x D .dx j /x D .A1 /ij .dx j /x ; (6.27)
@x j
@x j j
!i0 D !j D Ai !j : (6.28)
@x 0i

Starting from Tx Vn and Tx Vn it is possible to build the whole tensor algebra
.Tsr /x Vn as well as the exterior algebra .^s /x Vn at any point x 2 Vn (Chap. 2). In
particular, the transformation formulae under a change of local coordinates .x i / !
.x 0i / of the components Tji11i
ir
s
of any .r; s/-tensor belonging to .Tsr /x Vn ,

@ @
T D Tji11i
ir
˝ ˝ i ˝ dx j1 ˝ ˝ dx js ; (6.29)
s
@x i1 @x 1
are
@x 0i1 @x 0ir @x k1 @x ks h1 hr
Tj0i11iis r D T : (6.30)
@x h1 @x hr @x 0j1 @x 0is k1 ks

The preceding definitions can be extended to the whole manifold. A C k vector

field is a map
X W x 2 V n ! Xx 2 T x V n : (6.31)
In local coordinates .x i /, map (6.31) assumes the form

@
X D X i .x 1 : : : x n / ; (6.32)
@x i

which differs from (6.16) since the components X i are C k functions of the
coordinates.
Similarly, a C k tensor field is a map

T W x 2 Vn ! Tx 2 .Tx /rs Vn : (6.33)

In local coordinates .x i /, map (6.33) assumes the form

@ @
T D Tji11j
ir
.x 1 ; : : : ; x n / ˝ i ˝ dx j1 ˝ ˝ dx js ; (6.34)
s
@x i1 @x r

where the components Tji11j

ir
s
.x 1 ; : : : ; x n / are C k functions of the coordinates. In
particular, a p-form is a map

 W x 2 Vn ! x 2 .^x /s Vn ; (6.35)
6.4 Cotangent Vector Space 77

which locally has the coordinate form

 D j1 js .x 1 ; : : : ; x n /dx j1 ^ ^ dx js : (6.36)

We conclude this section by introducing the Lie algebra of the vector field on a
manifold Vn .
Let Vn be a C 1 n-dimensional manifold and denote by F 1 Vn the <-vector
space of the C 1 functions on Vn and by 1 Vn the <-vector space of the C 1
vector fields on Vn . Then, to any C 1 vector field

X W x 2 V n ! Xx 2 T x V n

we can associate the linear map

X W f 2 F 1 Vn ! Xf 2 F 1 Vn (6.37)

such that
.Xf /.x/ D Xx f: (6.38)
It can be easily proved that map (6.37) verifies the following derivation property:

X.fg/ D gXf C f Xg; 8f; g 2 F 1 Vn : (6.39)

Definition 6.10. Let X; Y 2 1 Vn be two C 1 vector fields. The bracket of X and

Y is the C 1 vector field ŒX; Y such that

ŒX; Yf D .XY  YX/f; 8f 2 F Vn : (6.40)

From (6.38) and (6.32) we obtain the following coordinate form of (6.40):
 
@Y i @X i @
ŒX; Y D X j j  Y j j : (6.41)
@x @x @x i

It is not difficult to prove the following theorem.

Theorem 6.4. The bracket operation verifies the following properties:

ŒX; Y D ŒY; X;

aŒX; Y D ŒaX; Y D ŒX; aY;
ŒX; Y C Z D ŒX; Y C ŒX; Z ;
ŒX; ŒY; Z C ŒY; ŒZ; X C ŒZ; ŒX; Y D 0; (6.42)

8a 2 < and 8X; Y; Z 2 1 Vn .

This theorem proves that 1 Vn , equipped with the addition of vector fields, the
product of a real number by a vector field, and the bracket operation, is a Lie algebra.
78 6 Differentiable Manifolds

6.5 Differential and Codifferential of a Map

Definition 6.11. Let Vn and Wm be two C k manifolds with dimensions n and m,

respectively, and let .t/ be an arbitrary curve on Vn containing the point x 2 Vn
(Fig. 6.5). The differential at the point x 2 Vn of the C k map

F W Vn ! W m
is the linear map
Fx W Xx 2 Tx Vn ! YF .x/ 2 TF .x/Wm (6.43)

such that the image of the tangent vector Xx at x to the curve .t/ is the tangent
vector to the curve F ..t// at the point F .x/. Formally,
d
YF .x/ g D g ı F ı .t/; 8g 2 F .F .x//: (6.44)
dt

To find the coordinate representation of (6.44), we introduce a chart .U; '/ with
coordinates .x i /, i D 1; : : : ; n, in a neighborhood of x 2 Vn and a chart .V; /,
with coordinates .y ˛ /, ˛ D 1; : : : ; m, in neighborhood V of F .x/. We denote by
y ˛ D y ˛ .x 1 ; : : : ; x n / the coordinate form of the map F and by x i .t/ the parametric
equations of the curve .t/ in the coordinates .x i /. Then, (6.44) can be written as
follows:
@ dx i @y ˛ @ i @y
˛
@
Y ˛ .y ˇ .x i // g D g  X g; 8f 2 F F .x/:
@x ˛ dt @x i @y ˛ @x i @y ˛
In conclusion, the coordinate form of (6.43) is
   
i @ @y ˛ i @
Fx W Xx D X 2 T x Vn ! X 2 TF .x/ Wm : (6.45)
@x i x @x i @y ˛ F .x/
Starting from linear map (6.43), we can give the following definition.

Xx V
U F
x

O xn O

y
x

Fig. 6.5 Differential of a map F

6.5 Differential and Codifferential of a Map 79

Definition 6.12. The linear map

FF.x/ W  F .x/ 2 TF.x/Wm ! !x 2 Tx Vn ; (6.46)

between the dual spaces TF.x/Wm and Tx Vn , defined by the condition

!x .Xx / D  F .x/.Fx Xx /; 8Xx 2 Tx Vn ; (6.47)

is called codifferential of F W Vn ! Wm at F .x/.

Adopting the coordinates .x i / on Vn and .y ˛ / on Wm , we can write (6.47) as follows:

@y ˛ i
!i X i D ˛ X ;
@x i
and, taking into account the arbitrariness of Xx , we obtain

@y ˛
!i D ˛ : (6.48)
@x i
In conclusion, the coordinate form of (6.46) is

@y ˛ i
FF.x/ W ˛ dy ˛ 2 TF.x/ Wm ! ˛ dx 2 Tx Vn : (6.49)
@x i
Now we consider the extension of the differential F of F that is the new linear
map, denoted by the same symbol,

Fx W Tx 2 .T0r /x Vn ! TO F .x/ 2 .T0r /F .x/ Wm ; (6.50)

such that
Fx .X1 ˝ : : : ˝ Xr / D Fx X1 ˝ : : : ˝ Fx Xr ; (6.51)

8X1 ; : : : ; Xr 2 Tx Vn . It can be easily verified that map (6.50) transforms the .r; 0/-
tensor
@ @
T D T i1 ir i ˝ : : : ˝ i
@x 1 @x r
into the .r; 0/-tensor
˛1 ˛r
O D @y @y T i1 ir @ ˝ : : : ˝ @ :
T (6.52)
@x i1 @x ir @y ˛1 @y ˛r

Similarly, we can extend (6.46) by the linear map

FF.x/ W TF .x/ 2 .Ts0 /F .x/ Wm ! TO x 2 .Ts0 /x Vn (6.53)

80 6 Differentiable Manifolds

such that
FF.x/ . 1 ˝ ˝  s / D FF.x/  1 ˝ ˝ FF.x/  s (6.54)

8 1 ; : : : ;  s 2 TF.x/ . It is evident that FF.x/ maps the .0; s/-tensor

T D T˛1 ˛s dy ˛1 ˝ ˝ dy ˛s 2 .Ts0 /F .x/ Wm

into the .0; s/-tensor

@y ˛1 @y ˛s
TO D T˛ ˛ dx i1 ˝ ˝ dx is 2 .Ts0 /x Vn : (6.55)
@x i1 @x is 1 s
Finally, we can define the linear map
O 2 .ƒs /x Vn
Fx W F .x/ 2 .ƒs /x Wm !  (6.56)

such that
FF.x/. 1 ^ ^  s / D FF.x/  1 ^ ^ FF.x/  s (6.57)

81 ; : : : ; s 2 TF.x/ .
Again it is simple to verify that (6.56) maps

 D .˛1 ˛s / dy ˛1 ^ ^ dy ˛s 2 .ƒs /F .x/Wm

into
˛1 ˛s
O D @.y ; : : : ; y / .˛1 ˛s / dx i1 ^ ^ dx is 2 .ƒs /x Vn :
 (6.58)
@.x i1 ; : : : ; x is /

Remark 6.1. It is important to note that, in general, none of the preceding linear
maps can be extended over the entire manifolds Vn or Wm because F W Vn ! Wm
can be neither one-to-one nor onto. For instance, the vector field X W x 2 Vn !
Fx Xx is defined on F .Vn /, and it could assume more values at the same point if F
were not one-to-one.
If F W Vn ! Wm is a diffeomorphism, then n D m, and we can define an
isomorphism
Fx W .Tsr /x Vn ! .Tsr /F .x/ Wn (6.59)
such that
1
Fx .X1 ˝ ˝ Xr ˝ !1 ˝ ˝ !s / D Fx X1 ˝ ˝ Fx !s : (6.60)

Once again, it is simple to verify that the linearity of (6.59) implies that (6.59) maps
the tensor

@ @
T D Tji11j
ir
i
˝ ˝ i ˝ dx j1 ˝ ˝ dx js 2 .Tsr /x Vn
s
@x 1 @x r
6.6 Tangent and Cotangent Fiber Bundles 81

into the tensor

@y ˛1 @y ˛r @x j1 @x js i1 ir @
Fx T D T ˝
@x ii @x ir @y ˇ1 @y ˇs j1 js @y ˛1
@
˝ ˝ dy ˇ1 ˝ ˝ dy ˇs 2 .Tsr /F .x/ Wn : (6.61)
@y ˛r

In conclusion, if F W Vn ! Wm is a diffeomorphism, then n D m, and the

tensor and exterior algebras on Vn and Wn at the points x 2 Vn and F .x/ 2 Wn
are isomorphic. Further, upon varying x 2 Vn , Fx maps .r; s/-tensorial fields of Vn
onto .r; s/-tensorial fields of Wn .

6.6 Tangent and Cotangent Fiber Bundles

Given the C k manifold Vn , consider the set

T Vn D f.x; Xx /; x 2 Vn ; Xx 2 Tx Vn g : (6.62)
The map
W .x; Xx / 2 T Vn ! x 2 Vn (6.63)

is called a projection map, and the counterimage 1 .x/ D x  Tx Vn is called a

fiber on x.
If .U; '/ is a chart on Vn , then any x 2 U is determined by its coordinates .x i /.
Further, any vector Xx 2 Tx Vn , where x 2 U , is determined by its components
X i relative to the natural basis .@=@x i /x . In this way, we have defined a one-to-one
correspondence
 W U ! '.U /  <n ;
where
U D f.x; Xx /; x 2 U; Xx 2 Tx Vn g  T Vn :

T Vn becomes a Hausdorff topological space when we equip it with a topology

whose open sets have the form U  I n , with U an open set of Vn and I n an open set
of <n . It is not difficult to verify that the map  is a homeomorphism so that T Vn
is a topological manifold and .U  <n ; / a chart of T Vn . Collecting the domains
U of the charts of an atlas of Vn , we define an atlas of T Vn , which becomes a 2n-
dimensional topological manifold. We do not prove that, if Vn is a C k manifold,
then T Vn is a C k manifold.
Definition 6.13. The 2n-dimensional manifold T Vn is called a tangent fiber bun-
dle of Vn , and the preceding coordinates .x i ; X i / are called the natural coordinates
of T Vn .
82 6 Differentiable Manifolds

For instance, the tangent fiber bundle T S 1 of a circumference S 1 is the collection

of all pairs .x; Xx /, where x 2 S 1 and Xx is a tangent vector to S 1 at point x. The
fiber at x is the tangent straight line to S 1 at x. It is evident that T S 1 is diffeomorphic
to a cylinder S 1  <.
All that has been said about T Vn can be repeated starting from the set

T  Vn D f.x; !x /; x 2 Vn ; !x 2 Tx Vn g : (6.64)

The corresponding C k 2n-dimensional manifold is called the cotangent fiber

bundle of Vn . Natural coordinates in the open set U  <n , where U is an open
set of a chart of an atlas of Vn , are given by .x i ; !i /, with .x i / the coordinates of a
point x 2 U and .!i / the components of a covector !x 2 Tx Vn in the dual basis
.dx i /.

6.7 Riemannian Manifolds

Before giving the definition of a Riemannian manifold, we consider two examples.

First, let S 2 be the unit sphere embedded in the three-dimensional Euclidean space
E3 referred to Cartesian coordinates .x i / (Fig. 6.6). The parametric equations of S 2
are

x 1 D sin
cos ';
x 2 D sin
sin ';
x 3 D cos
:

x
O

x
Fig. 6.6 Sphere in E3
6.7 Riemannian Manifolds 83

We have already said that any regular surface in Euclidean space is a differen-
tiable manifold. Now we wish to equip a sphere with the metrics induced by the
metrics of E3 . This means that the square distance ds 2 between two points .x i / and
.x i C dx i / of S 2 is assumed to be equal to the Pythagorean distance in E3

3
X
ds 2 D .dx i /2 ; (6.65)
i D1

but expressed in terms of the coordinates .';

/ on S 2 . To this end, we differentiate
the parametric equations of the surface and obtain

ds 2 D sin2
d' 2 C d
2 ; (6.66)

where ' 2 Œ0; 2  and

2 Œ0; .
As a second example we consider the ellipsoid †2 in E3 with parametric
equations

x 1 D sin
cos ';
x 2 D sin
sin ';
x 3 D a cos
;

with axes 2, 2, and 2a. Again adopting the foregoing viewpoint, the square distance
between two points .x i / and .x i C dx i / of †2 becomes

ds 2 D sin2 d' 2 C .cos2

C a sin2
/d
2 : (6.67)

The sphere S 2 and the ellipsoid †2 are diffeomorphic so that they are essentially
the same manifold. In particular, the variables ' and
are local coordinates for both
surfaces. These two manifolds become different from each other when we introduce
a metric structure by the square distance ds 2 (Fig. 6.7).
Definition 6.14. A C k -differentiable n-dimensional manifold Vn is a Riemannian
manifold when on Vn a metric tensor, that is, a C k .0; 2/-symmetric, nondegenerate
tensor field g, is given.
Owing to the properties of g, we can introduce into any tangent space Tx Vn a
scalar product of two arbitrary tangent vectors Xx and Yx

Xx Yx D gx .Xx ; Yx /; (6.68)

and Tx Vn becomes a pseudo-Euclidean vector space. In other words, the whole

tensor algebra at any point of a Riemannian manifold becomes a pseudo-Euclidean
tensor algebra.
84 6 Differentiable Manifolds

x
O

Fig. 6.7 Ellipsoid in E3

In local coordinates .x i / relative to a chart .U; '/ of Vn , the tensor g assumes the
following representation:

g D gij dx i ˝ dx j ; gij D gj i ; (6.69)

where det.gij / ¤ 0, and the scalar product (6.68) becomes

Xx Yx D gij X i Y j : (6.70)

In view of the results of Chap. 4 regarding the symmetric .0; 2/-tensors, it is

always possible to find, about any point x 2 Vn , a coordinate system such that at x

@ @
gij D D ˙ıij : (6.71)
@x i @x j
The set f1; : : : ; 1; 1; : : : ; 1g is called the signature of the metric tensor and is
independent of the coordinates.
Finally, we define as the square distance ds 2 between two points .x i / and .x i C
i
dx / the quantity
ds 2 D gij dx i dx j : (6.72)

6.8 Geodesics of a Riemannian Manifold

Definition 6.15. Let Vn be a Riemannian manifold and suppose that the tensor gx
is positive definite at any point x 2 Vn . The length of a C 1 curve .t/ W Œa; b 
< ! Vn is the real number
6.8 Geodesics of a Riemannian Manifold 85

Z
l. / D ds: (6.73)


Denote by x i .t/, t 2 Œa; b, the parametric equations of .t/ in an arbitrary

system of coordinates .x i / whose domain contains the curve .t/. Then the length
l. / of .t/ assumes the form
Z bq
l. / D gij .x h /xP i xP j dt: (6.74)
a

It is evident that l. / depends neither on the choice of the coordinates nor on the
parameterization of the curve.
Let .xai / D .x i .a// and .xbi / D .x i .b// be the initial and final points of .t/, and
consider the one-parameter family  of curves

f i .s; t/ W .; /  Œa; b ! Vn ; (6.75)

• Including .t/ for s D 0:

f i .0; t/ D x i .t/I (6.76)

• In addition, they are such that any curve starts from .xai / and ends at .xbi /, i.e.,

f i .s; a/ D xai ; f i .s; b/ D xbi ; 8s 2 .; /: (6.77)

It is evident that the length of any curve of  is given by the integral

Z bq
I.s/ D gij .f h .s; t//fPi fPj dt: (6.78)
a

Definition 6.16. The curve .t/ between the points .xai / and .xbi / is a geodesic of
the metric g if the function I.s/ is stationary for s D 0 for any family  of curves
satisfying (6.76) and (6.77).
To determine the parametric equations of the geodesic between the points .xai /
and .xbi /, we start by analyzing the stationarity condition

dI
.0/ D 0: (6.79)
ds
Introducing the position
q
L.f h ; fPh / D gij fPi fPj ; (6.80)

and taking into account (6.78), we obtain that

Z !
dI b
@L @f h @L @fPh
.s/ D C dt:
ds a @f h @s @fPh @s
86 6 Differentiable Manifolds

Since
   
@L @fPh @ @L @f h @ @L @f h
D  ;
@fPh @s @t @fPh @s @t @fPh @s

we have that
Z b    b
dI @L @ @L @f h @L @f h
.s/ D  .0; t/dt C : (6.81)
ds a @f h @t @fPh @s @fPh @s a

From this formula, when we recall (6.76) and (6.77), we deduce that the condition
dI =ds.0/ D 0 is equivalent to requiring that the equation
Z b  
@L j j d @L j j @f h
.x ; x
P /  .x ; x
P / .0; t/dt D 0 (6.82)
a @x h dt @xP h @s

be satisfied for any choice of the functions @f h =@s.0; t/, h D 1; : : : ; n. It is possible

to prove that this happens if and only if the parametric equations x i .t/ of the curve
.t/ satisfy the Euler–Lagrange equations

@L j j d @L j j
h
.x ; xP /  .x ; xP / D 0; h D 1; : : : ; n; (6.83)
@x dt @xP h
and the boundary conditions

x h .a/ D xah ; x h .b/ D xbh ; h D 1; : : : ; n: (6.84)

Remark 6.2. In the Cauchy problem relative to (6.83), we must assign the initial
conditions x h .0/ D xah and xP h .0/ D X h , h D 1; : : : ; n. It is well known that, under
general hypotheses on the function L.x i ; xP i /, there is only one solution satisfying
the Euler–Lagrange equations and the initial data, that is, there is only one geodesic
starting from the point .xai / and having at this point the tangent vector .X i /. In
contrast, there is no general theorem about the boundary problem (6.83), (6.84), so
perhaps there is one and only one geodesic between the points .xah / and .xbh /, no
geodesic or infinite geodesics. In this last case we say that .xah / and .xbh / are focal
points.
Remark 6.3. A curve .t/ has been defined as a map  W t 2 Œa; b ! Vn .
Consequently, a change in the parameter into the parametric equations x h .t/ leads
to a different curve, although the locus of points is the same. On the other hand,
the presence of the square root under integral (6.74) implies that the value of the
length l. / is not modified by a change in the parameter t; in other words, l. / has
the same value for all curves determining the same set of points of Vn . This remark
implies that (6.83) cannot be independent. To show that only n1 of these equations
are independent, it is sufficient to note that from the identity
6.8 Geodesics of a Riemannian Manifold 87

n
X @L
LD xP i (6.85)
i D1
@xP i

it follows that
X n
@L @2 L
h
D xP i i h ; (6.86)
@x i D1
@xP @x

X n
@L @L @2 L
h
D h
C xP i i h ; (6.87)
@xP @xP i D1
@xP @xP

and then
n
X @2 L
xP i D 0: (6.88)
i D1
@xP i @xP h

From the foregoing conditions there follows the identity

n 
X 
@L d @L
i

i D1
@x dt @xP i
n
X n n
@L X i @2 L h X i @2 L h
D xP i  x
P x
P  xP xR ;
i D1
@x i i D1 @xP i @x h i D1
@xP i @xP h

which, in view of (6.86) and (6.87), becomes

X n  
@L d @L
 D 0: (6.89)
i D1
@x i dt @xP i

This relation shows that if the parametric equations x h .t/ satisfy the first n  1
Euler–Lagrange equations, then they satisfy the last one.
We use the arbitrariness of the parameter t to obtain a new form of the equations
of a geodesic. Setting
p
L D '; (6.90)

we can give (6.83) the following form:

 
1 @' d 1 @'
p  p D 0; h D 1; : : : ; n: (6.91)
2 ' @x h dt 2 ' @xP h

If we choose a parameter s for which the tangent vector .xP h / to the geodesic has
unit length, i.e., a parameter s such that

dx h
' D gij xP i xP j D 1; xP h D ; (6.92)
ds
88 6 Differentiable Manifolds

then the Euler–Lagrange equations assume the form

1 @ d
h
.gij xP i xP j /  .ghj xP j / D 0; h D 1; : : : ; n: (6.93)
2 @x ds

6.9 Exercises

1. Find an atlas for the sphere S 2 in three-dimensional Euclidean space starting

from the parametric equations

x 1 D sin
cos ';
x 2 D sin
sin ';
x 3 D cos
;

where 0  '  2 is the longitude and 0 

 the colatitude.
2. Find an atlas for the cylinder in three-dimensional Euclidean space starting
from its parametric equations

x 1 D cos ';
x 2 D sin ';
x 3 D z;

where 0  '  2 is the longitude and z 2 < is the ordinate along the axis of
the cylinder.
3. Find an atlas for the torus T 2 in three-dimensional Euclidean space starting
from its parametric equations

x 1 D sin
cos ';
x 2 D sin
sin ';
x 3 D cos
;

where 0  '  2 and 0 

 2 are shown in Fig. 6.8.
4. Find an atlas for the Moebius strip with parametric equations
 
1 '
x 1 D 1 C u sin cos ';
2 2
 
1 '
x 2 D 1 C u sin sin ';
2 2
'
x 3 D u sin ;
2

where 0  '  2 and u 2 Œ1; 1 are shown in Fig. 6.9.

6.9 Exercises 89

Fig. 6.8 Torus T 2 in E3

Fig. 6.9 Moebius strip

5. Find an atlas for a Klein bottle with parametric equations

 
1
x 1 D 6 cos u.1 C sin u/ C 4 1  cos u ;
2
 
1 '
x 2 D 1 C u sin sin ';
2 2
'
x 3 D u sin ;
2
where 0  '  2 and 0 
 2 (Fig. 6.10).
6. Let E3 be the three-dimensional space referred to Cartesian coordinates
.x 1 ; x 2 ; x 3 /. Find an atlas for the sphere and the cylinder starting from their
implicit equations
.x 1 /2 C .x 2 /2 C .x 3 /2  1 D 0;
.x 1 /2 C .x 2 /2  1 D 0:
90 6 Differentiable Manifolds

Fig. 6.10 Klein’s bottle

B A A’
B’

O x

Fig. 6.11 Atlas of a cylinder

7. The diffeomorphism

' W .x 1 ; x 2 / 2 .1 ; 1/.0; 1/ ! .x 01 D x 1 C ; x 02 D x 2 / 2 .1C ; 1/.0; 1/

overlaps the open subset .1  ; 1/  .0; 1/ of the square U D .0; 1/  .0; 1/

and the open subset .1 C ; 1/  .0; 1/ of the square V D .1; 2/  .0; 1/ in the
plane Ox 1 x 2 . Further, the diffeomorphism

W .x 1 ; x 2 / 2 .0; /.0; 1/ ! .x 01 D 2C x 1  ; x 02 D x 2 / 2 .2 ; 2/.0; 1/

overlaps the open subsets .0; /  .0; 1/ and .2  ; 2/  .0; 1/ of the same
squares (Fig. 6.11).
Show that U , V and , are an atlas for the cylinder S 1 .0; 1/. Essentially,
a cylinder is obtained from a rectangle by identifying the points of two opposite
sides.
6.9 Exercises 91

S r
x’

Fig. 6.12 Stereographic projection

8. The diffeomorphism

' W .x 1 ; x 2 / 2 .1 ; 1/.0; 1/ ! .x 01 D x 1 C ; x 02 D x 2 / 2 .1C ; 1/.0; 1/

overlaps the open subset .1  ; 1/  .0; 1/ of the square U D .0; 1/  .0; 1/

and the open subset .1 C ; 1/  .0; 1/ of the square V D .1; 2/  .0; 1/ in the
plane Ox 1 x 2 . Further, the diffeomorphism

W .x 1 ; x 2 / 2 .0; /.0; 1/ ! .x 01 D 2Cx 1  ; x 02 D 1x 2 / 2 .2 ; 2/.0; 1/

overlaps the open subsets .0; /  .0; 1/ and .2  ; 2/  .0; 1/ of the same
squares (Fig. 6.11). Show that U , V and , are an atlas for the Moebius strip.
Essentially, this surface is obtained by identifying the points of two opposite
sides after a torsion of 180ı.
9. Determine an atlas with four coordinate domains and the relative coordinate
transformations to obtain a torus. Show that the torus is obtained from a
rectangle by identifying the points of the two pairs of opposite sides.
10. This exercise shows how we can obtain a geographic chart using a map between
two manifolds. Let S 2  N be the unit sphere minus the North Pole. The
diffeomorphism F , which is called a stereographic projection, between S 2  N
and the tangent plane at the South Pole S , is shown in Fig. 6.12. F maps the
point x 2 S 2  N into the point x 0 2 corresponding to the intersection of
the straight line N x with . If we adopt spherical coordinates .';
/ on S 2 and
polar coordinates .'; r/ with its center at the South Pole on , the map F has
the following coordinate form (Fig. 6.13):
 

' D '; r D 2 tan C :

4 2
92 6 Differentiable Manifolds

/2
O

S x

x’
r

Fig. 6.13 Section of stereographic projection

S
x’
x
O
z

Fig. 6.14 Central projection

Determine how F transforms the coordinate curves, the tangent vectors, and
the covectors on S 2 . Analyze how it transforms the distances and the angles
between tangent vectors.
11. As another example of a geographic chart, we consider the diffeomorphism
between F W S 2  fS; N g ! C , where C is the cylinder in Fig. 6.14 and F
maps x 2 S 2 into the point of C belonging to the straight line Ox. Adopting
spherical coordinates .';
/ on S 2 and cylindrical coordinates .'; z/ on C , the
map F assumes the following coordinate form:

' D '; z D tan

:

Determine how F transforms the coordinate curves, the tangent vectors, the
covectors, the distances, and the angles between tangent vectors.
6.9 Exercises 93

12. We conclude the examples of geographic charts with Mercator’s projection

(1569). This representation is obtained considering a one-to-one map between
S 2  fS; N g and a cylinder C that, adopting the same coordinates of Fig. 6.14,
has the form
 

' D '; z D ln tan C :

4 2
Determine how F transforms the coordinate curves, the tangent vectors, the
covectors, the distances, and the angles between tangent vectors.
13. Determine the Riemannian metrics on a sphere, a cylinder, and a torus and find
the relative equations of geodesics.
Chapter 7
One-Parameter Groups of Diffeomorphisms

7.1 Global and Local One-Parameter Groups

Definition 7.1. A one-parameter global group of diffeomorphisms G on a mani-

fold Vn of class C k , k > 0, is a C k map

 W .t; x/ 2 <  Vn ! t .x/ 2 Vn (7.1)

such that
1. 8t 2 < the map t W x 2 Vn ! t .x/ 2 Vn is a C k diffeomorphism of Vn ;
2. 8t; s 2 < 8x 2 Vn , t Cs .x/ D t ı s .x/.
In particular, from the second property we have t .x/ D 0Ct .x/ D 0 ı t .x/,
so that
0 .x/ D x; 8x 2 Vn : (7.2)

Similarly, from x D 0 .x/ D t .x/ ı t .x/ we obtain that

t .x/ D .t /1 .x/; 8x 2 Vn : (7.3)

Definition 7.2. 8x0 2 Vn , the C k curve t .x0 / W < ! Vn on Vn is called the orbit
of G determined by x0 . In view of (7.2), the orbit contains x0 .
Theorem 7.1. Any point x0 2 Vn belongs to one and only one orbit, i.e., the orbits
determine a partition of Vn .
Proof. First, we show that if x2 2 t .x1 /, then the orbit determined by x2 coincides
with t .x1 / up to a change of the parameter. In other words, an orbit is determined
by any of its points. In fact, if x2 2 t .x1 /, then there exists a value t2 of t such
that x2 D t2 .x1 /. In view of property (7.3), x1 D t2 .x2 /. Consequently, the orbit
t .x1 / can also be written in the form t t2 .x2 /, which, up to the change t ! t  t2
of the parameter, gives the orbit determined by x2 . Now it remains to prove that if

A. Romano, Classical Mechanics with Mathematica R

, Modeling and Simulation 95
in Science, Engineering and Technology, DOI 10.1007/978-0-8176-8352-8 7,
© Springer Science+Business Media New York 2012
96 7 One-Parameter Groups of Diffeomorphisms

x2 does not belong to t .x1 /, then the orbits t .x1 / and t .x2 / do not intersect each
other. In fact, if there is a point x0 belonging to both orbits, then there exist two
values t1 ; t2 2 < such that

x0 D t1 .x1 /; x0 D t2 .x2 /:

In view of (7.3), x2 D t2 .x0 /, and then

x2 D t2 .x0 / D t2 .t1 .x1 // D t1 t2 .x1 /:

In conclusion, x2 belongs to the orbit determined by x1 , against the hypothesis. 

Theorem 7.2. Let Xx be the tangent vector to the orbit t .x/ at point x. The map
x 2 Vn ! Xx 2 Tx .Vn / defines a C k1 vector field X over Vn .
Proof. Let .U; x i / be a chart of Vn and denote by  i .t; x 1 ; : : : ; nn / the representa-
tion of the group in these coordinates. Then, 8x 2 U ,
 
@ i @
Xx D ;
@t t D0 @x i

and the theorem is proved. 

Definition 7.3. The vector field X is called the infinitesimal generator of the group
of diffeomorphisms.
Example 7.1. Let E2 be the Euclidean plane. It is easy to verify that the differential
map
 W <  E2 ! E2

such that
x 0  t .x/ D x C tu;

where u is a constant vector in E2 , is a one-parameter global group of diffeomor-

phisms of E2 , called the group of translations. Further, the orbit determined by the
point x0 is x 0 D x0 C tu. Finally, the constant vector field u D @t .x/=@t is the
infinitesimal generator of the one-parameter group.
Example 7.2. Let E3 be the Euclidean three-dimensional space referred to the
cylindrical coordinates r, ˛, and z. Then, the family of diffeomorphisms

r 0 D r;
˛ 0 D ˛ C t;
z0 D z

is a one-parameter group of diffeomorphisms, called the group of rotations about the

Oz-axis of the cylindrical coordinates. The orbits of the group are circumferences
having the center on the Oz-axis, and the infinitesimal generator is the vector field
7.2 Lie’s Derivative 97

@
XD :
@˛
Definition 7.4. Let U be an open region of Vn . A one-parameter local group of
diffeomorphisms is a differential map

 W .; /  U ! Vn ; (7.4)
where  > 0, such that
1. 8t 2 .; /, t W x 2 U ! t .x/ 2 t .U / is a diffeomorphism;
2. If t, s, t C s 2 .; /, then t Cs .x/ D t .s .x//.
Theorem 7.3. Let X be a differentiable vector field on Vn . For any x 2 Vn , there
exist an open region U  Vn , a real number  > 0, and a one-parameter local
group of diffeomorphisms  W .; /  U ! Vn whose infinitesimal generator is X.
Proof. Consider the system of ordinary differential equations (ODEs)

dx i
D X i .x 1 ; : : : ; x n /; i D 1; : : : ; n:
dt
If the vector field is differentiable, from known theorems of analysis, 8x 2 Vn there
exist a neighborhood U of x and an interval .; / such that one and only one
solution x i D .t; x0 /, t 2 .; /, of the preceding system exists satisfying the
initial data  i .0; x0 / D x0 , 8x0 2 U . For the uniqueness theorem, x i D .t; x0 / is
a diffeomorphism 8t 2 .; /. We omit the proof of the property .t C s; x0 / D
.t; .s; x0 //. 
Definition 7.5. The vector field X is said to be complete if it is an infinitesimal
generator of a global one-parameter group of diffeomorphisms.
Theorem 7.4. On a compact manifold Vn , every differential vector field is complete.
Proof. The preceding theorem allows one to state that, 8x 2 Vn , a map  W
..x/; .x//  Ux ! Vn exists verifying conditions 1 and 2 of Definition 7.4.
Since Vn is compact, the covering .Ux /x2Vn admits a finite subcovering .Uxi /i D1:::;s .
Setting by  D min f.x1 /; : : : ; .xs /g, the map  is defined on .; /  Vn . 

7.2 Lie’s Derivative

In Chap. 6, we proved that if the differentiable map F W Vn ! Vn is a

diffeomorphism, then its differential .F /x W Tx Vn ! TF .x/Vn is an isomorphism
that can be extended to the tensorial algebra and the exterior algebra at the point
x 2 Vn . In this section, we show that this result, which holds for any diffeomorphism
t of a one-parameter transformation group, makes it possible to introduce a
meaningful derivation operator on the manifold Vn .
98 7 One-Parameter Groups of Diffeomorphisms

Fig. 7.1 Lie’s derivative of a vector field

We denote by X a differentiable vector field on Vn , by F .Vn / the vector space

of the differentiable functions f W Vn ! <, and by t the one-parameter (local or
global) transformation group generated by X.
Definition 7.6. The Lie derivative of the function f with respect to the vector field
X is the map
LX W F .Vn / ! F .Vn /
such that
 
f .t .x//  f .x/ d
.LX f /x D lim D f .t .x/ : (7.5)
t >0 t dt sD0

Since X is tangent to the orbits of t , the Lie derivative of f at x is the directional

derivative of f along the vector Xx , that is,

LX f D Xf: (7.6)

Henceforth, we denote by ,  , and rs the F .Vn / modules of C 1 -vector fields,

1-forms, and .r; s/-tensor fields of Vn , respectively.
Definition 7.7. The Lie derivative on  with respect to the vector field X is the map
(Fig. 7.1)
LX W Y 2  ! LX Y 2 

such that, 8x 2 Vn ,
1
.LX Y/x D lim Œ.t / .t .x//Yt .x/  Yx : (7.7)
t !0 t

Definition 7.8. The Lie derivative on  with respect to the vector field X is the
map (Fig. 7.2)
LX W ! 2  ! LX ! 2 
7.2 Lie’s Derivative 99

Fig. 7.2 Lie’s derivative of a 1-form

such that, 8x 2 Vn ,
1
.LX !/x D lim Œ.t / .x/!t .x/  !x : (7.8)
t !0 t

It is evident how Lie’s derivative can be extended to the tensor fields of rs .
To find the coordinate expression of Lie’s derivative, we introduce a chart .U; x i /
on Vn and denote by y i D ti .x 1 ; : : : ; x n / the coordinate expression of the one-
parameter group of the diffeomorphisms t .x/ and by X i the components of the
infinitesimal generator of the group X. Then, the maps t .x/ and t .x/ in a
neighborhood of x are given by the expressions

y i D x i C X i .x/t C O.t/; (7.9)

i i i
x D y  X .y/t C O.t/; (7.10)

respectively. The coordinate expressions of the codifferential of (7.9) and of the

differential of (7.10) are given by
 i  i
 i @y i @X
..t / /j D j
D ıj C t C O.t/; (7.11)
@x x @x j x
 i  i
i @x i @X
..t / /j D j
D ıj  t C O.t/: (7.12)
@y y @y j y

From the preceding relations we have that

"  i  #  j 
i i @X j @Y h
..t / Yt .x/ / D ıj  t Y .x/ C X .x/t C O.t/:
@y j y @x h x
(7.13)
In view of (7.13), Lie’s derivative (7.7) assumes the following coordinate form:
 
@Y i @X i @
LX Y D X j j  Y j j : (7.14)
@x @x @x i
100 7 One-Parameter Groups of Diffeomorphisms

By comparing (7.14) and (6.41), we derive the important result

LX Y D ŒX; Y : (7.15)

Starting from (7.9), we derive the coordinate form of the codifferential .t .x//
  j     
j @X @!j
..t / !t .x/ /i D ıi C t !j .x/ C X h
.x/t CO.t/; (7.16)
@x i x @x h x

so that (7.8) becomes

 
@!i @X j
LX ! D X j j C !j dx i : (7.17)
@x @x i

It is not a difficult task to prove that for an arbitrary tensor T 2 rs , Lie’s
derivative is expressed by the following formula:
" r
@ i1 ir X i1 ik1 hikC1 ir @X ik
LX T D X h T  Tj1 js
@x h j1 js @x h
kD1
s
#
X i1 ir @X h @ @
C Tj1 jk1 hjkC1 js j i
˝˝ i
@x h @x 1 @x r
kD1

˝dx j1 ˝    ˝ dx js : (7.18)

In particular, for an s-form , the preceding formula gives

X  @ @X h
LX  D Xh  .j 1 j s / C  h.j 2 j s /
i1 <<is
@x h @x j1

@X h
C .j1 js1 /h dx j1 ^    ^ dx js : (7.19)
@x js

From the definition of Lie’s derivative there follows that LX is linear with respect
to X, and the derivation property with respect to the tensorial product and the
exterior product

LX .T ˝ S/ D LX T ˝ S C T ˝ LX S; (7.20)
LX .T ^ S/ D LX T ^ S C T ^ LX S: (7.21)

Definition 7.9. A tensor field T 2 rs Vn is invariant under a one-parameter group

t .x/ of diffeomorphisms if

Tt .x/ D .t T/t .x/ ; (7.22)

7.3 Exercises 101

where .t / denotes the extension of the differential .t / of the diffeomorphisms
of the group to the tensor algebra (Chap. 6).
In particular, a vector field Y and a 1-form ! are invariant under the group
t .x/ if

Yt .x/ D .t / .x/Yx : (7.23)

!t .x/ D .t / !x : (7.24)

We omit the proof of the following theorem.

Theorem 7.5. The tensor field T 2 rs Vn is invariant under the one-parameter
group t .x/ of diffeomorphisms if and only if

LX T D 0; (7.25)

where X is the infinitesimal generator of t .x/.

Definition 7.10. Let .U; x i / be a chart of Vn , and let

! D !.i1 is / dx i1 ^    dx is

be the coordinate representation of an s-form ! in the chart .U; x i /. The interior

product of ! 2 s Vn by the vector field X is the s  1-form iX !, which in the chart
.U; x i / of Vn has the following representation:

iX ! D X h !h.i1 is / dx i2 ^    dx is : (7.26)

Starting from (7.26), it is easy to prove the following theorem.

Theorem 7.6. The interior product has the following properties:
1. It is <-linear with respect to X;
2. If ! 2 s Vn and  2 p Vn , then

iX .! ^  / D iX ! ^  C .1/s ! ^ iX  I (7.27)

3. If ! 2  Vn , then
iX ! D !.X/:

7.3 Exercises

1. Determine the one-parameter global group acting on <2 whose infinitesimal

generator in the coordinates x; y is
@ @
XDx Cy :
@x @y
102 7 One-Parameter Groups of Diffeomorphisms

Hint: The parametric equations of the group’s orbits are solutions of the system
of ODEs
dx dy
D x; D y:
dt dt
The solutions are
x D x0 et ; y D y0 et ;

and it is evident that t .x0 ; y0 /D.x0 et ; y0 et / is the global group generated by X.

2. Determine the one-parameter local group acting on <2 whose infinitesimal
generator in the coordinates x; y is

@ @
XD C ey :
@x @y

Hint: The parametric equations of the group’s orbits are solutions of the system
of ODEs
dx dy
D 1; D ey :
dt dt
Consequently, the parametric equations of the orbits are

x.t/ D x0 C t; y.t/ D log.t C ey0 /;

where t>ey0 . The family of diffeomorphisms t .x0 ; y0 /D.x0 Ct; log.tCey0 //

defines a local group generated by X.
3. Determine the one-parameter global group acting on <2 whose infinitesimal
generator in the coordinates x; y is

@ @
X D y Cx :
@x @y

Hint: The parametric equations of the group’s orbits are solutions of the system
of ODEs
dx dy
D y; D x;
dt dt
i.e., they are given by the functions

x.t/ D x0 cos t  y0 sin t; y.t/ D x0 sin t C y0 cos t:

This family of diffeomorphisms of <2 is the global group generated by X.

4. Using the theory of linear differential equations, verify that the global group of
<2 generated by the vector field
7.3 Exercises 103

@ @
X D .x  y/ C .x C y/
@x @y

is
x.t/ D et .x0 cos t  y0 sin t/; y.t/ D et .y0 cos t C x0 sin t/:

Verify the result using the built-in function DSolve of Mathematica.

5. Given the one-parameter global group t .x/ of <2

x 0 D x C ˛t; y 0 D y C ˇt;

where x and y are Cartesian coordinates, determine the vector fields that are
invariant under the action of the group.
Hint: The infinitesimal generator of the group t .x/ is the vector field X D
.˛; ˇ/. The differential .t / of the diffeomorphisms t .x/ is represented by the
unit matrix. Further, in view of Theorem 7.5, we can determine the invariant
fields Y either by (7.23) or by (7.25). The last condition gives

@Y 1 @Y 1 dY 1
˛ Cˇ D D 0;
@x @y dt
@Y 2 @Y 2 dY 2
˛ Cˇ D D 0;
@x @y dt

whereas the former gives

Y 1 .x C ˛t; y C ˇt/ D Y 1 .x; y/;

Y 2 .x C ˛t; y C ˇt/ D Y 2 .x; y/:

Both the results show that the components of Y must be constant along the
group’s orbits.
6. On the unit sphere S referred to the spherical coordinates .'; /, consider the
one-parameter group G of diffeomorphisms

' 0 D ' C ˛t;  0 D  C ˇt;

where ˛ and ˇ are constant. Determine the orbits of G and the vector fields that
are invariant under the action of G.
Chapter 8
Exterior Derivative and Integration

8.1 Exterior Derivative

We denote by ^s Vn the set of differential r-forms on the manifold Vn . It is evident

that ^r Vn is both an <-vector space and an F Vn module. In the sequel, we use the
notation ^0 Vn D F Vn .
Definition 8.1. The exterior derivative is an <-linear map
d W ^r Vn ! ^rC1 Vn ; .r D 0; 1; : : : ; n/ (8.1)
such that
1. 8f 2 ^0 Vn the exterior derivative of f is the differential of f ;
2. 8! 2 ^r Vn , 8 2 ^s Vn , the map d is an antiderivation, i.e.,

d.! ^  / D d! ^  C .1/r ! ^ d I (8.2)

3. d2 D 0.
When the manifold Vn is paracompact (Chap. 6), it is possible to prove the
existence and uniqueness of the map d. Here, we limit ourselves to proving its local
existence in a chart .U; x i / of Vn .
In the domain U of the chart, any r-differential form can be written as

! D !.i1 ; ;ir / dx i1 ^    ^ dx ir :

If the exterior derivative exists, then, applying properties 2 and 3, we have that

d! D d!.i1 ; ;ir / dx i1 ^    ^ dx ir :

Finally, taking into account property 1, we obtain the coordinate form of the exterior
derivative

A. Romano, Classical Mechanics with Mathematica R

, Modeling and Simulation 105
in Science, Engineering and Technology, DOI 10.1007/978-0-8176-8352-8 8,
© Springer Science+Business Media New York 2012
106 8 Exterior Derivative and Integration

@!.i1 ; ;ir / h

d! D dx ^ dx i1 ^    ^ dx ir : (8.3)
@x h
It is plain to verify that (8.3) defines an exterior derivative in region U .
Before proceeding, we show that the preceding definition leads to familiar
concepts in a Euclidean space E3 . In fact, if f 2 F E3 , then (8.3) gives
@f
df D dx i : (8.4)
@x i
It is evident that the components of this differential form are covariant with respect
to a coordinate change. Denoting by .ei / the fields of the natural bases relative to
the coordinates .x i / and by gij the components of the scalar product, we can say
that the vector field
@f
rf D g ij j ei (8.5)
@x
defines the gradient of f .
Applying (8.3) to the differential form ! D !i dx i 2 ^1 E3 , we obtain

@wi j X @wi X @wi

d! D j
dx ^ dx i D j
dx j ^ dx i C j
dx j ^ dx i :
@x j <i
@x j >i
@x

The preceding relation can be written in the form

X  @wi @wj

d! D j
 i
dx j ^ dx i ; (8.6)
j <i
@x @x

and the quantities

@wi @wj
rij D j
 (8.7)
@x @x i
define the covariant components of a skew-symmetric tensor. The adjoint (Chap. 5)
of this tensor is the pseudovector r whose contravariant components are
1 ij k 1
ri D  rj k D p  ij k rj k ; (8.8)
2 2 jgj

that is,
 
1 @w3 @w2 @w1 @w3 @w2 @w1
rD p  3; 3  1; 1  2 : (8.9)
jgj @x 2 @x @x @x @x @x

In other words, the components of d! are proportional to the components of r  v,

where v is a vector field whose covariant components vi are equal to !i .
Finally, the exterior derivative of the 2-differential form

 D 12 dx 1 ^ dx 2 C 13 dx 1 ^ dx 3 C 23 dx 2 ^ dx 3

8.2 Closed and Exact Differential Forms 107

is
 
@23 @31 @12
d D C C dx 1 ^ dx 2 ^ dx 3 : (8.10)
@x 1 @x 2 @x 3

Introducing the pseudovector adjoint of 

1
ui D p  ij k j k ; (8.11)
2 jgj
i.e., a pseudovector vector with components

1 1 1
u1 D p 23 ; u2 D p 31 ; u3 D p 12 ; (8.12)
jgj jgj jgj

and defining the divergence of the vector field u (Chap. 9)

1 @ p
r uD p . jgj ui /; (8.13)
jgj @x i
relation (8.10) becomes
p
d D r  u jgjdx 1 ^ dx 2 ^ dx 3 ; (8.14)

where
p
jgjdx 1 ^ dx 2 ^ dx 3 (8.15)

is the volume differential 3-form of E3 (Chap. 5).

We can summarize the preceding results as follows. In the three-dimensional
Euclidean space E3 , we consider a function f W E3 ! <, a differential form ! D
!i dx i , and a 2-form . Further, we introduce the vector field v whose covariant
components vi are equal to the components !i of !, and the pseudovector field u,
which is the adjoint of the skew-symmetric tensor . Then, the following results
hold:

df D ! , rf D v; (8.16)
d! D  , r  v D u; (8.17)
d D 0 , r  u D 0: (8.18)

8.2 Closed and Exact Differential Forms

Definition 8.2. A differential r-form  2 ^r Vn is exact if a differential .r  1/-

form ! 2 ^r1 Vn exists such that

d! D : (8.19)
108 8 Exterior Derivative and Integration

Definition 8.3. A differential r-form  2 ^r Vn is closed if

d D 0: (8.20)

It is evident that the exact differential r-forms belong to the vector subspace
[see (1.39)]
Er D d.^r1 Vn / D Im.d/; (8.21)

where d W ^r1 Vn ! ^r Vn . In contrast, the closed differential r-forms belong to

the vector subspace [see (1.40)]

Cr D ker.d/; (8.22)

where d W ^r Vn ! ^rC1 Vn . In view of the third property of Definition 8.1, any

exact differential r-form is closed, and we have that

Er  Cr  ^r Vn : (8.23)

Definition 8.4. We call a cohomology relation the following equivalence relation

in the vector subspace Cr  ^r Vn :

8; 0 2 Cr ;   0 if   0 2 Er : (8.24)

This equivalence relation generates a partition of Cr into equivalence classes that

are called cohomology classes of order r. If the differential r-form  belongs to
the cohomology class Œ, then to Œ belong all the differential r-forms  C d!,
where ! 2 ^r1 Vn . Two closed differential r-forms belonging to the same class are
said to be cohomologs. In particular, Œ0 D Er .
The set Cr =Er of these equivalence classes becomes a vector space when it is
equipped with the following operations:

Œ1  C Œ2  D Œ1 C 2 ; aŒ D Œa;

81 ; 2 2 Cr and 8a 2 <.

Definition 8.5. The vector space Hr D Cr =Er is called the cohomology space
of order r. If we only refer to an addition, Cr =Er becomes the cohomology group of
order r. The integer number
br D dim.Hr / (8.25)
is the rth-Betti number of Vn . It is evident that br D 0 if and only if Hr reduces to
the cohomology class Œ0. In such a case, any closed form is also exact.
Definition 8.6. A subset W of a Euclidean space En is said to be star-shaped if
there exists a point x0 2 W such that, 8x 2 W , the points of the segment x0 C 
x!
0 xt,
t 2 Œ0; 1, belong to W .
8.3 Properties of Exterior Derivative 109

An open set U  Vn is smoothly contractible if there exists a point x0 2 U and

a differential map  W Œ0; 1  U ! U such that

.1; x/ D x; .0; x/ D x0 ; 8x 2 U:

For instance, a sphere of E3 is star-shaped and smoothly contractible to any point

of it. A region contained between two spheres with the same center and different
radii is not smoothly contractible.
We omit the proof of the following theorem.
Theorem 8.1 (Poincaré’s Lemma). If A is a star-shaped subset W of a Euclidean
space En or an open subset U of a manifold Vn , then

br D dim.A/ D 0;

that is, any closed form is exact.

8.3 Properties of Exterior Derivative

In this section, we prove some relations existing among the exterior derivative, the
differential of a map, the Lie derivative, and the inner product of a differential r-
form by a vector field.
In this regard the following theorem holds.
Theorem 8.2. For ˛ 2 ƒk Vn , f 2 F Vn , and X 2 Vn ,

if X ˛ D f iX ˛; (8.26)
iX df D LX f; (8.27)
LX ˛ D iX d˛ C d.iX ˛/; (8.28)
Lf X ˛ D f LX ˛ C df ^ iX ˛; (8.29)
LX .d˛/ D d.LX ˛/: (8.30)

Further, if F W Vn ! Wm is a C 1 map and ˛ 2 ƒk .Wm /; ˇ 2 ƒs .Wm /, then

F  .˛ ^ ˇ/ D F  .˛/ ^ F  .ˇ/; (8.31)

 
F .d˛/ D d.F .˛//: (8.32)

Proof. Equation (8.26) follows at once from (7.26). Further, iX df D X h @f =@x h ,

and, taking into account (7.6), the identity (8.27) is proved. To verify (8.28), we start
supposing that ˛ is a 1-form. Then, in view of (7.19), (7.26), (8.3), and (8.6), we
have
110 8 Exterior Derivative and Integration

 
j @˛i @X j
LX ˛ D X C ˛j dx i ;
@x j @x i
 
@˛i @˛j
iX d˛ D X j
 i dx i ;
@x j @x
@X j i @˛j
d.iX ˛/ D d.X j ˛j / D ˛j dx C X j i dx i ;
@x i @x
and (8.28) is proved for 1-forms. Since any r-form ˛ is a linear combination of
exterior products of 1-forms, and LX , iX , and d are linear maps, to prove the
theorem, it is sufficient to verify that (8.28) is satisfied for the exterior product
˛1 ^ ˛2 of two 1-forms. Now, owing to (7.2) and recalling that (8.28) has been
proved for 1-forms, we have that

LX .˛1 ^ ˛2 / D LX ˛1 ^ ˛2 C ˛1 ^ LX ˛2
D .iX d˛1 C d.iX ˛1 // ^ ˛2 C ˛1 ^ .iX d˛2 C d.iX ˛2 //
D .iX d˛1 / ^ ˛2 C d.iX ˛1 / ^ ˛2
C ˛1 ^ .iX d˛2 / C ˛1 ^ d.iX ˛2 /:

Similarly, in view of (8.2) and (7.27), we have also

iX d.˛1 ^ ˛2 / D iX .d˛1 ^ ˛2  ˛1 ^ d˛2 /

D iX d˛1 ^ ˛2 C d˛1 ^ iX ˛2  iX ˛1 ^ d˛2 C ˛1 ^ iX d˛2 I
d.iX .˛^ ˛2 // D d.iX ˛1 ^ ˛2  ˛1 ^ iX ˛2 /
D d.iX ˛1 / ^ ˛2 C iX ˛1 ^ d˛2  d˛1 ^ iX ˛2 C ˛1 ^ d.iX ˛2 /;

and (8.28) is proved. In the same way, we can prove (8.29) and (8.30). We omit the
proof of (8.31) and (8.32). 

8.4 An Introduction to the Integration of r-Forms

In Sect. 3.4, we showed that one of the two possible orientations of <r is determined
by choosing a basis .ei /. Denote by dx 1 ^    ^ dx r the volume r-form of <r .
Definition 8.7. The integral of the r-form

! D !.i1 ir / dx i1 ^    ^ dx ir

on the compact set U  <r is the number

Z Z
!D !.i1 ir / dx 1    dx r : (8.33)
U U
8.4 An Introduction to the Integration of r-Forms 111

Let U 0 be another compact region of <r , and denote by x 0 D x 0 .x 1 ; : : : ; x r / a

diffeomorphism of U onto U 0 . From well-known analytical results we have that
Z Z
J !.i1 ir / dx 1    dx r D !.i0 1 ir / dx 01    dx 0r ; (8.34)
U U0

where
 
@x 0i
J D det : (8.35)
@x j

Henceforth, we identify the compact set U with the r-cube defined as follows:
˚ 
U D .x 1 ; : : : ; x r / 2 <r ; ai  x i  ai C c i : (8.36)

We call faces of the r-cube U the .r  1/-cubes Ui  , i D 1; : : : ; r  1,  D 0; 1,

defined by the conditions
˚ 
Ui  D a1  x 1  a1 C c 1 ; : : : ; x i D ai C c i ; : : : ; an  x r  ar C c r : (8.37)

Starting from the .r  1/-cubes we can define the .r  2/-cubes, which are the faces
of the .r 1/-cubes, and so on. For instance, if r D 2, then the 2-cubes are rectangles
and the 1-cubes are the sides of the rectangles. Further, if r D 3, then a 3-cube is a
parallelepiped, the 2-cubes are its faces, and the 1-cubes are its edges.
Now we consider the vectors ni  , i D 1; : : : ; r,  D 0; 1, which in the basis .ei /,
chosen to determine an orientation of <r , have the components
j j
ni;0 D ıi ej ; ni;1 D ıi ej : (8.38)

In particular, if r D 2, then

n10 D e1 ; n11 D e1 ; n20 D e2 ; n21 D e2 :

It is evident that the vectors ni  are normal to the faces of the r-cube U and outward
oriented (Fig. 8.1 is relevant to the case r D 2).
.1/
On any face Ui  of the r-cube U , we consider a vector set †i  D .ni  ; vi  ; : : : ;
.r1/ .1/ .r1/
vi  /, where the vectors vi  ; : : : ; vi  are independent vectors belonging to Ui  ,
chosen in such a way that the bases †i  are congruent with the basis .ei /. When U
is equipped with these bases on its faces, we say that U is an oriented cube of <r
and we write .U; /, where  is the orientation determined by .ei /.
Definition 8.8. We define an oriented r-cube on a differentiable manifold Vn , a
tern C D .U; ; F /, where .U; / is an oriented r-cube of <r and F W U ! Vn is a
differentiable map.
Example 8.1. Given the 2-cube of <2 defined by the inequalities

0  r  1; 0   2
;
112 8 Exterior Derivative and Integration

Fig. 8.1 An oriented 2-cube

Fig. 8.2 An oriented 2-cube of E2

consider the map F W .r; / 2 U ! .x 1 ; x 2 / 2 E2 , where E2 is two-dimensional

Euclidean space such that (Fig. 8.2)

x 1 D r cos ; x 2 D r sin :

Example 8.2. Given the 2-cube of <2 defined by the inequalities

0  '  2
; 0  
=2;

consider the map F W .'; / 2 U ! .x 1 ; x 2 ; x 3 / 2 E3 , where E3 is three-

dimensional Euclidean space such that (Fig. 8.3)

x 1 D r cos sin '; x 2 D r sin sin '; x 3 D cos :

Definition 8.9. An r-chain C on a manifold Vn is a formal summation †i ai Ci of

oriented r-cubes Ci , where ai are relative integer numbers. Denoting by @Ci the
union of the faces of Ui , we call cochain @C of the chain C the formal summation
†i ai @Ci .
8.4 An Introduction to the Integration of r-Forms 113

Fig. 8.3 An oriented 2-cube of E3

Definition 8.10. Let ! 2 ƒr Vn be an r-form on a differentiable manifold Vn .

Denote by C D †ai Ci a chain of oriented r-cubes Ci D .Ui ; i ; Fi /. Then, the
integral of ! on the chain is given by
Z X Z X Z
!D ai !D ai F  .!/: (8.39)
C i Ci i Ui

Example 8.3. In three-dimensional Euclidean space E3 , referred to the Cartesian

coordinates .x 1 ; x 2 ; x 3 /, consider the oriented 1-cube C D .Œa; b; ; F /, where
Œa; b  <, F W t 2 Œa; b ! .x 1 .t/; x 2 .t/; x 3 .t// 2 E3 . If ! D !i dx i is a 1-form
of E3 , then
dx i
F  ! D !i
dt
and
Z Z b
dx i
!D !i dt
C a dt

is the ordinary curvilinear integral of ! along the curve  with parametric equations
x i .t/.
If s is the curvilinear abscissa along  and we denote by t D .dx i =ds/ the unit
tangent to  and by v a vector whose covariant components vi are equal to !i , then
the preceding integral assumes the familiar form
Z Z
!D v  tds: (8.40)
C 

Example 8.4. In three-dimensional Euclidean space E3 , referred to the Cartesian

coordinates .x 1 ; x 2 ; x 3 /, consider the oriented 2-cube C D .U; ; F /, where U is
a rectangle of <2 , F W .u1 ; u2 / 2 U ! .x 1 .u1 ; u2 /; x 2 .u1 ; u2 /; x 3 .u1 ; u2 // 2 E3 .
114 8 Exterior Derivative and Integration

If ! D !12 dx 1 ^ dx 2 C !13 dx 1 ^ dx 3 C !23 dx 2 ^ dx 3 is a 2-form of E3 , then [see

(6.58)]
@.x 1 ; x 2 / @.x 1 ; x 3 / @.x 2 ; x 3 /
F  ! D !12 C !13 C !23 ;
@.u1 ; u2 / @.u1 ; u2 / @.u1 ; u2 /
and
Z Z  
@.x 1 ; x 2 / @.x 1 ; x 3 / @.x 2 ; x 3 /
!D !12 C !13 C !23 du1 du2 (8.41)
C U @.u1 ; u2 / @.u1 ; u2 / @.u1 ; u2 /

is the integral of ! on the surface S with parametric equations x i .u1 ; u2 /.

In arbitrary curvilinear coordinates .x 1 ; x 2 ; x 3 / of E3 , we consider the adjoint
vector
1
u D p .!23 e1 C !31 e2 C !12 e3 /;
g

where .e1 ; e2 ; e3 / is the natural basis relative to the coordinates .x 1 ; x 2 ; x 3 /. In our

analysis it is proved that the Jacobian minors appearing in (8.40) coincide with the
components of the unit normal n to the surface S . Finally, the elementary area d of
p
S is equal to gdu1 du2 , and (8.41) gives the flux of u across the oriented surface S
Z Z
!D u  n d: (8.42)
C S

We conclude this section stating without the difficult proof the following
generalized Stokes theorem:
Theorem 8.3. If C is an oriented r-chain of the differentiable manifold Vn and @C
its cochain, then, 8! 2 ƒr1 Vn , we obtain
Z Z
d! D !: (8.43)
C @C

8.5 Exercises

1. Prove the equivalence between (8.3) and the following formula:

X X @!
.i2 /
.irC1 / i1
d! D .1/m.
/ dx ^    dx irC1 ; (8.44)
i1 <<irC1
2…rC1
@x
.i1 /

where ….r C 1/ is the set of all permutations of the indices i1 < i2 <    <
irC1 , for which i2 <    < irC1 , and m.
/ is the number of inversions in the
permutation f
.i1 /;
.i2 /; : : : ;
.irC1 /g.
8.5 Exercises 115

2. Evaluate by (8.44) the exterior derivative d of the 2-form

 D 12 dx 1 ^ dx 2 C 13 dx 1 ^ dx 3 C 23 dx 2 ^ dx 3

on a manifold V3 .
Hint: Since n D 3, the only choice i1 < i2 < i3 is 1; 2; 3. On the other hand, the
possible permutations
.1/;
.2/;
.3/ of these indices for which
.2/ <
.3/
are f1; 2; 3g, f2; 1; 3g, f3; 1; 2g. Therefore, (8.44) gives
 
@23 @13 @12
d D 1
 2
C dx 1 ^ dx 2 ^ dx 3 ;
@x @x @x 3

and we again obtain (8.10).

3. On a manifold V4 , apply (8.3) and (8.44) to the 2-form

 D 12 dx 1 ^ dx 2 C 13 dx 1 ^ dx 3 C 14 dx 1 ^ dx 4

23 dx 2 ^ dx 3 C 24 dx 2 ^ dx 4 C 34 dx 3 ^ dx 4 ;

and verify that they give the same result.

4. Adopting the volume form
p
gd'd

[see (8.16)] evaluate the area of the triangle and the parallelogram of Figs. 9.1
and 9.2, on the unit sphere referred to the spherical coordinates ', .
5. Integrate on the unit sphere the 1-form

xdx C dy C dz

of the three-dimensional Euclidean space E3 .

Chapter 9
Absolute Differential Calculus

9.1 Preliminary Considerations

In this chapter, we address the fundamental problem of extending the differential

calculus to manifolds. To understand the problem we are faced with, consider a C 1
vector field Y.t/ assigned along the curve x.t/ on the manifold Vn . We recall that
on an arbitrary manifold the components Y i .t/ of Y.t/ are evaluated with respect to
the local natural bases of local charts .U; x i /, U  Vn . Consequently, when we try
to define the derivative of Y along x.t/, we must compare the vector Y.t C t/ 2
Tx.t Ct /Vn , referred to the local basis ei .t C t/, with the vector Y.t/ 2 Tx.t / Vn ,
referred to the local basis ei .t/. Since we do not know how to relate the basis ei .t C
t/ to the basis ei .t/, we are not in a position to compare the two preceding vectors;
consequently, we cannot assign a meaning to the derivative of the vector field Y.t/
along the curve x.t/ of Vn .
To give a reasonable solution to the preceding problem, we start with some
elementary considerations in a Euclidean space En before giving an abstract
definition of an affine connection on a manifold. In a Euclidean space En it is always
possible to find rectilinear coordinates .y i /. We denote by .ui / the constant unit
vectors along the axes y i axes. Then we introduce an arbitrary local chart .U; x i / in
En and denote by .x; ei / the natural bases relative to the curvilinear coordinates x i .
If X.x/ D X i .x/ei is a C 1 vector field in region U , then the family  of the integral
curves of X is given by curves whose parametric equations x i .t/ in the chart .U; x i /
are a solution of the first-order differential system

dx i
D X i .x 1 ; : : : ; x n /; i D 1; : : : ; n: (9.1)
dt

It is evident that X.x i .t// is the tangent vector to the curve x i .t/ at any point. When
a point x0 .x0i / 2 U is given, there is one and only one (local) integral curve of X.x/
containing point x0 . We denote by x i .t; x0 / the general integral of (9.1).

A. Romano, Classical Mechanics with Mathematica R

, Modeling and Simulation 117
in Science, Engineering and Technology, DOI 10.1007/978-0-8176-8352-8 9,
© Springer Science+Business Media New York 2012
118 9 Absolute Differential Calculus

Let Y.x/ D Y i .x/ei be another C 1 vector field on region U . Then, the

directional derivative of Y.x/ along the integral curves  of the vector field X.x/ is

dY dY i dei
D ei C Y i
dt dt dt
 i  k
@Y h @eh dx
D C Y : (9.2)
@x k @x k dt

To attribute a meaning to the derivatives @ei =@x k that express the variation of the
basis vectors on varying the coordinates, we start by noting that if the functions
y i D y i .x 1 ; : : : ; x n /

define the coordinate transformation .x i / ! .y i /, then we also have that

@y m
ei D um :
@x i
Since the vectors ui are constant, the derivation in rectilinear coordinates is possible.
Then, from the preceding relations we have that

@ei @2 y m
D um
@x j @x i @x j
@2 y m @x k
D ek ;
@x i @x j @y m
and introducing the notations
@2 y m @x k
ijk D jki  ; (9.3)
@x i @x j @y m
we obtain
@ei
D ijk ek ; (9.4)
@x i
and we can give (9.2) the form
 
dY @Y i
D C kh Y X k ei :
i h
(9.5)
dt @x k
i
This formula is not satisfactory since the quantities kh are given in terms of the
i
rectilinear coordinates .y /, as is shown by (9.3), not in terms of the coordinates
.x i /. But this further problem can be solved easily because we are operating in a
Euclidean space. In fact, a scalar product

X  Y D gij X i Y j
9.1 Preliminary Considerations 119

between two arbitrary vectors X and Y of the same tangent space to En is defined.
In particular, ei  ej D gij . Consequently, we have that

@ei @ej @gij

 ej C ei  k D :
@x k @x @x k
In view of (9.4), the preceding relation becomes
@gij
ink gnj C kj
n
gni D : (9.6)
@x k
Cyclically permuting the indices, we obtain

@gj k
jni gnk C ink gnj D ; (9.7)
@x i
n @gki
kj gni C jni gnk D : (9.8)
@x j
Adding (9.6) and (9.7) and subtracting (9.8), we express the quantities ijn in terms
of the metric coefficients gij and their first derivatives in the coordinates .x i /:
 
1 nh @ghj @gj i @gih
ijn D g i
C h
 : (9.9)
2 @x @x @x j

It is an easy exercise to verify that the map

dY
rX W Y ! ;
dt
where dY=dt is given by (9.5), has the following properties:

rf XCgY D f rX C grX ; (9.10)

rX .f Y/ D .Xf /Y C f rX Y; (9.11)

for any choice of C 1 real functions f and g and the C 1 vector fields X and Y on En .
We conclude this section by noting that the expression (9.5) of the derivative of
Y.t/ along the curve x i .t/, where the quantities are given by (9.9), was obtained
under the following two conditions:
• There exists a rectilinear system of coordinates.
• In any tangent space to the manifold Vn , there is a scalar product.
In the following sections, we show how to extend the results of this section to
arbitrary manifolds.
120 9 Absolute Differential Calculus

9.2 Affine Connection on Manifolds

Let Vn be a C 1 n-dimensional manifold, and denote by F .Vn / the vector space of

the real C 1 functions on Vn . Then, the set .Vn / of the C 1 vector fields on Vn is a
F .Vn / module.
Definition 9.1. An affine connection on Vn is a map

rX W .Vn / ! .Vn / (9.12)

such that

rf XCgY D f rX C grX ; (9.13)

rX .f Y/ D .Xf /Y C f rX Y; (9.14)
8f; g 2 F .Vn / and X; Y 2 .Vn / [see (9.10) and (9.11)].
To find the coordinate representation of the map (9.12), we introduce a local chart
.U; x i / on Vn and consider two C 1 vector fields

X D X i ei ; Y D Y i ei ; (9.15)

where .ei / are the vector fields defining a natural basis of the tangent space at any
point of the open set U . From the properties (9.13) and (9.14) there follows
 i 
i @Y
rX Y D rX k ek .Y ei / D ei C Y rek .ei / X k :
i
(9.16)
@x k
h
Introducing the connection coefficients ki by the relations
h
rek .ei / D ki eh ; (9.17)

(9.16) becomes
 
@Y i i h
rX Y D C kh Y X k ei  rk Y i X k ei : (9.18)
@x k
Although (9.18) and (9.15) are formally identical, in (9.18) the n3 C 1 functions kh i
i
are arbitrary. In particular, they could not verify the symmetry conditions kh D
i i
hk . In any case, if we give an affine connection on Vn , then its coefficients kh
are determined in every coordinate domain. Conversely, a connection is defined in
i
every coordinate domain U by T giving the coefficients kh in U . If we introduce a
0i
new chart .V; x / such that U V ¤ ;, then the connection coefficients in the
new chart cannot be arbitrarily assigned. In fact, in the coordinate transformation
.x i / ! .x 0i /, the natural bases are transformed according to the rule

@x i
e0j D ei  Aij ei : (9.19)
@x 0j
9.3 Parallel Transport and Autoparallel Curves 121

Consequently, introducing the notation @f =@x i D f;i , we have that

rX Y D .Y;ik Ckh
i
Y h /X k ei D .Y;0lm Cmn 0l
Y 0n /X 0m e0l
 
D Akm ..A1 /li Y i /;k Cmn
0l
.A1 /nh Y h .A1 /m p r
p X Al er
h
D Akm .A1 /m r 1 l i k 1 m r
p Al .A /i Y ;k CAm .A /p Al .A /i;k Y
1 l i

i
C .A1 /nh Arl .A1 /m 0l
p mn Y
h
X p er :

But Akm .A1 /m k

p D ıp , and then the preceding formula gives

  k
rX Y D Y;ik C.Ail .A1 /lh C .A1 /m i 1 n 0l
k Al .A /h mn /Y
h
X ei :

Finally, we obtain the following transformation formulae for the connection coeffi-
cients:
i
kh D Ail .A1 /m 1 n 0l i 1 l
k .A /h mn C Al .A /h ;k ; (9.20)
whose inverse formulae are
0i
kh D .A1 /il Am n l 1 i l
k Ah mn C .A /l Ah ;k : (9.21)

These formulae show that the connection coefficients are not transformed like the
components of a .1; 2/-tensor unless the coordinate transformation is linear since,
in this case, Alh ;k D 0.
We do not prove that an affine connection can be assigned on any paracompact
manifold.

9.3 Parallel Transport and Autoparallel Curves

Let .U; x i / be a chart on a manifold Vn , equipped with an affine connection, and let
.x i .t// be the parametric equations of a curve  contained in region U . Finally, we
denote by
dx i
XD ei (9.22)
dt
the tangent vector to  evaluated in the natural bases .ei / along  .
Definition 9.2. The vector Y.t/ is said to be parallel along  if

dY
 rX Y D 0: (9.23)
dt
122 9 Absolute Differential Calculus

In view of (9.18), the coordinate form of (9.23) is

dY i i dx k
C kh Yh D 0; i D 1; : : : ; n: (9.24)
dt dt

This is a first-order differential system in the unknowns Y i .t/. If the functions

k
i
kh Y h dxdt are differentiable along  and the initial vector Y.x.t0 // is given by the
initial data Y i .x.t0 // D Y0i , then one and only one solution Y i .t; t0 ; Y0i / exists of
system (9.24). It is fundamental to remark that the vector Y.x i .t//, obtained by a
parallel transport of Y.x.t0 // along  , depends on  .
Definition 9.3. A vector field Y on a manifold Vn equipped with an affine
connection is uniform if
rX Y D 0 (9.25)
for any vector field X.
In view of (9.18), condition (9.25) can be expressed in the following coordinate
form:
rk Y i  Y i ;k Ckh
i
Y h D 0; i; k D 1; : : : ; n; (9.26)
i
in any domain U of a local chart .U; x /. When the connection is given, the
i
connection coefficients kh are known functions of the coordinates in region U .
Therefore, finding a uniform field Y in U requires the integration of system (9.25)
with respect to the unknowns Y i .x j /. In other words, we must solve a system of
n2 equations in n unknowns. It is well known that this is possible if and only
if suitable integrability conditions are satisfied. We analyze these conditions in
a subsequent section. In conclusion, on an arbitrary manifold Vn with an affine
connection uniform vector fields could not exist.
Definition 9.4. A curve .s/ on the manifold Vn , equipped with an affine connec-
tion, is said to be autoparallel if its tangent vector X.s/ satisfies the condition

rX X D 0; (9.27)

that is, if it is parallel along .s/.

In any chart .U; x i /, we have that X i .s/ D dx i =ds, and (9.27) can be written as
follows:
d2 x i k
i dx dx
h

2
C kh D 0; i D 1; : : : ; n: (9.28)
ds ds ds
This is a second-order differential system in the unknowns x i .s/. Consequently, if
a point x0 2 Vn is fixed together with a vector .X i .0// 2 Tx0 Vn , then one and only
one autoparallel curve exists that contains x0 and is tangent to .X i .0// at x0 .
Remark 9.1. We recall that a curve  of Vn is defined as a C 1 map  W s 2 Œa; b !
.s/ 2 Vn . This means that by changing the parameter, we obtain another curve,
9.4 Covariant Differential of Tensor Fields 123

even if the locus of its points does not change. As a consequence, we have that, if
.s/ is an autoparallel curve, .s.t//  .t/ could not be an autoparallel curve. To
identify the parameter changes that do not modify the autoparallelism of a curve,
we start by noting that

dx k dx k dt
D ;
ds dt ds
 2
d2 x k d dx k d2 x k dt dx k d2 t
D D C :
ds 2 ds ds dt 2 ds dt ds 2

Consequently, under the parameter change s D s.t/, system (9.28) becomes

 2 2
d2 x i k
i dx dx
h
ds d t dx i
2
C kh D  : (9.29)
dt dt dt dt ds 2 dt

This result allows us to state that the autoparallel character of the curve .s/ is not
modified by the parameter change s D s.t/ if and only if

t D as C b; (9.30)

where a and b are arbitrary constants. Parameters that do not modify the autoparal-
lelism of a curve are called canonical parameters.

9.4 Covariant Differential of Tensor Fields

We denote by rs .Vn / the set of .r; s/-tensor fields of class C 1 on the manifold Vn
equipped with an affine connection. In particular, we denote by .Vn / D 10 .Vn /
the set of C 1 vector fields on Vn . We remark that rs .Vn / is also a F .Vn / module.
Finally, we denote by FO the set of all the maps rs .Vn / ! rs .Vn /.
Theorem 9.1. On the manifold Vn there is only one <-linear map

rO W X 2 .Vn / ! rO X 2 FO (9.31)

with the following properties:

1. rO X f D Xf; 8f 2 F .Vn /;
2. rO X Y D rX Y; 8Y 2 .Vn /;
3. rO X .C T/ D C rO X T, where C is the contraction operator (Sect. 2.6) and T 2
rs .Vn /;
4. It verifies the derivation property

rO X .T ˝ S/ D .rO X T/ ˝ S C T ˝ .rO X S/: (9.32)

124 9 Absolute Differential Calculus

Proof. We prove the local existence of the preceding map in the domain of a chart
.U; x i /. We denote by .ei / the fields of the natural bases in U and by .dx i / the
differential forms of the dual bases. If ! 2 01 .Vn / and Y 2 .Vn /, then C.! ˝ Y/
is a function. From property 1 we obtain

rO X .C.! ˝ Y/ D X.C.! ˝ Y/:

On the other hand, in view of properties 2–4, we also have that

rO X .C.! ˝ Y// D C.rO X ! ˝ Y C ! ˝ rX Y/:

Expressing the preceding results in components, we can write that

X k .!i Y i /;k D .rO X !/i Y i C !i .Y i ;k Ckh

i
Y h /X k : (9.33)

Expanding the left-hand side of (9.33), simplifying the obtained relation and
recalling that it must hold for any vector field Y, we obtain the coordinate expression
of the covariant derivative of a 1-form:

.rO X !/i D .!;i ihk !h /X k : (9.34)

Starting from the coordinate expression of an arbitrary .r; s/-tensor T

T D Tji11i
ir
e ˝    ˝ eir ˝ dx j1 ˝    ˝ dx js
s i1

and applying properties 1–4, it is easy to verify that the coordinate expression of the
covariant differential of T is
h i
.rO X T/ij11i
is D X
r k
Tji11i
ir
s
;k C i1 hir
kh T j i
1 s
C  ir
kh T j
i1 h
i
1 s
  h
kj 1
T i1 ir
hi s
  h
kj s
T j
i1 ir
1 h :

(9.35)


Henceforth, we use the symbol rX instead of rO X .

9.5 Torsion and Curvature Tensors

Let f .x i / be a C 1 function in the chart .U; x i /. Then, owing to Schwarz’s theorem,

f;ij D f;j i :
9.5 Torsion and Curvature Tensors 125

However, the n2 quantities f;ij do not represent the components of a .0; 2/-tensor. In
contrast, the quantities ri f;j can be regarded as the components of a .0; 2/-tensor,
but we cannot change the derivation order since, in view of (9.34), we have that

ri f;j rj f;i D .jhi  ijh /f;h : (9.36)

The left-hand side gives the components of a .0; 2/-tensor and f;j are the compo-
nents of a covector; consequently, owing to the criteria given in Sect. 2.6, we can
state that the quantities
Sjhi D jhi  ijh (9.37)
are the components of a .1; 2/-tensor, which is called the torsion tensor of the affine
connection of Vn . This tensor field vanishes if and only if the connection coefficients
are symmetric with respect to the lower indices.
Consider a vector field X, which in the chart .U; x i / on the manifold Vn has
the coordinate representation X D X i ei , where .ei / is the natural basis of the
coordinates x i . Then, with simple calculations, it is possible to verify that

rj ri X k  ri rj X k D Rljk i X l  Sjl i rl X k ; (9.38)

where
Rljk i D ikl;j  jkl;i C jkh ihl  ih
k h
j l (9.39)
is the Riemann curvature tensor.
Now, we prove the following fundamental theorem.
Theorem 9.2. Let Vn be a manifold equipped with an affine connection. Then the
following statements are equivalent:
(a) There exists an atlas ˇ on Vn in each chart of which the connection coefficients
vanish;
(b) The torsion vanishes and, 8x; y 2 Vn , the parallel transport along any curve
connecting x and y does not depend on the curve;
(c) The torsion vanishes and there are n vector fields u1 ; : : : ; un that are indepen-
dent and uniform;
(d) The curvature and the torsion vanish.
T
Proof. (a)) (b). If .U; x i /; .U ; x i / 2 ˇ, then in the intersection U U of the
i i
domains, we obtain the results kh D  kh D 0. Consequently, in view of (9.20),
i
the coefficients Aj of transformation matrix (9.19) verify the conditions

Ail .A1 /lh ;k D 0;

that is,
.A1 /m i 1 l 1 l
i Al .A /h ;k D .A /h ;k D 0;
126 9 Absolute Differential Calculus

and we conclude that the elements Alh of the transformation matrix are constant
and the coordinate transformation x i ! x i is linear. Further, the parallel
transport (9.24) along any curve contained in a domain of a chart of the atlas
ˇ reduces to dX i =dt D 0. Let x; y be two arbitrary points of Vn . If both points
belong to the same domain U of a chart of ˇ, then the parallel transport of a
vector Yx D Yxi ei .x/ along a curve .t/  U , connecting x and y, is obtained
integrating the equation dY i =dt D 0 along .t/, and it is independent of .t/.
Suppose that x belongs
T to the chart .U; x i / of ˇ and y belongs to another chart
i
.U ; x /, with U U ¤ ;. Let 1 .t/ and 2 .t/ be two curves going from T x to y.
The parallel transport of Yx along 1 .t/ up to a point p1 D .t1 / 2 U U gives
the vector Yxi ei .p1 /, whereas the
T parallel transport of the same vector along 2 .t/
up to a point p2 D .t2 / 2 U U gives the vector Yxi ei .p2 / since this transport
for both the curves is expressed by the equation dY i =dt D 0. Now we regard p1
and p2 as points belonging to U . In the coordinates .x i /, the vectors Yxi ei .p1 / and
j j
Yxi ei .p2 / have, respectively, the representations Aij Yx ei .p1 / and Aij Yx ei .p2 /
T
since the quantities Aij are constant all over U U . To prove condition (b) we
j
must verify that the parallel transport of Aij Yx ei .p1 / along 1 .t/ from p1 to y
j
and of the vector Aij Yx ei .p2 / along 2 .t/ from p2 to y leads us to the same
i
vector Yy . But this is evident since we must integrate the equations dY =dt D 0
of the parallel transport along 1 .t/ and 2 .t/ with the same initial conditions.
(b)) (c). Let Y1 ; : : : ; Yn n be independent vectors at the arbitrary point x 2 Vn .
Since condition (b) holds, by a parallel transport of these vectors to any other
point of Vn , we obtain n vector fields u1 ; : : : ; un whose values at a point do not
depend on the curve along which the vectors are transported; in other words, they
verify the condition
rA ui D 0
for any vector field A tangent to the curve along which ui is transported. From
the arbitrariness of the curve, i.e., of A, it follows that the fields ui are uniform.
It remains to prove that they are independent at any point. In this regard, it is
sufficient to note that the parallel transport, when it is independent on the curves,
defines an isomorphism  W Tx .Vn / ! Ty .Vn / between the tangent spaces
corresponding to points x and y owing to the linearity of the parallel transport
equations and the existence and uniqueness theorems.
(c))(d). Since the torsion tensor vanishes, condition (9.38) becomes
rj ri Y k  ri rj Y k D Rljk i Y l :

Now we consider the vector field Y D ai ui , where the quantities a1 ; : : : ; an

are arbitrary constants and u1 ; : : : ; un are uniform vector fields existing owing to
property (c). For any field Y the preceding relation gives

Rljk i al D 0;

and the arbitrariness of the quantities al implies Rljk i D 0.

9.5 Torsion and Curvature Tensors 127

(d))(a). We must prove that, for any x 2 Vn , there exists a chart .U; x 0i /
such that in these coordinates the connection coefficients vanish. Let .V; x i / be
another chart containing x. Then, in view of (9.21), we must find a coordinate
transformation .x i / ! .x 0i / such that
0p q
j h D .A1 /pq Am n q 1 p
j Ah mn C .A /q Ah;j D 0;

q
where .Aij / D .@x i =@x 0j / and the connection coefficients mn are symmetric
with respect to m and n since the torsion tensor vanishes. Multiplying the
preceding equation by Alp , we obtain the system

Alh;j D mn
l
Am n
j Ah ; (9.40)
@x l
D Alj : (9.41)
@x 0j
To find the n2 Cn unknowns Alj .x/ and x 0i .x/ of the preceding n2 Cn3 equations,
the integrability conditions

@Alh @Alh
0j 0i
D ; (9.42)
@x @x @x 0i @x 0j
@2 x l @2 x l
D (9.43)
@x 0j x 0i @x 0i x 0j
must be satisfied. We write these conditions in the more compact form

Alh ;Œj i  D 0; x l ;Œj i  D AlŒj;i  D 0; (9.44)

where we have used the notation fŒij  D fij  fj i . The integrability conditions
(9.44)2 are satisfied since, from (9.40) and the hypothesis that the torsion tensor
vanishes, we obtain

AlŒi;j  D mn
l
Am n l m n
j Ai C mn Ai Aj
l
D mn Am n l n m n m l
j Ai C nm Ai Aj D Ai Aj Œn;m D 0:

Again for (9.40), the integrability conditions of (9.44)1 can be written as follows:

@  l m n @  l m n
Alh;Œj i  D  0i
mn Aj Ah C 0i mn Ai Ah
@x @x
l
@mn
D Am An  mn
l
Am n l m n
j;i Ah  mn Aj Ah;i
@x 0i j h
l
@mn
C Am An C mn
l
Am n l m n
i;j Ah C mn Ai Ah;j :
@x 0j i h
128 9 Absolute Differential Calculus

When we note that @=@x 0j D Akj @=@x k , take into account (9.40), and recall the
already proven result AlŒi;j  D 0, the preceding equations become
 
p
Alh;Œj i  D Am
i An k
A
h j  l
mn;k   l
k n;m C  l

kp mn
p
  l

mp k n

or, in equivalent form,

Alh;Œj i  D Am n k l
i Ah Aj Rnmk D 0; (9.45)

and the theorem is proved. 

9.6 Riemannian Connection

Theorem 9.3. A Riemannian manifold Vn admits one and only one affine connec-
tion r such that
• The torsion tensor vanishes and
• The scalar product is invariant under a parallel transport.
Proof. Let .U; x i / be a chart on Vn , and denote by kh
i
the connection coefficients
whose existence we must prove. If .t/ is any curve connecting two arbitrary points
x; y 2 U and X, Y two vector fields that are parallel along .t/, then we have

dX i i dx h
C kh Xk D 0; (9.46)
dt dt
dY i i dx h
C kh Yk D 0; (9.47)
dt dt

where x h .t/ are the parametric equations of .t/ in the chart .U; x i /. The torsion
tensor vanishes if and only if the connection coefficients are symmetric with respect
to the lower indices. Further, the scalar product is invariant under the parallel
transport if and only if the condition

d
.gij X i Y j / D 0 (9.48)
dt

is satisfied for any pair of vector fields X, Y that verify solutions of Eqs. (9.46) and
(9.47). Expanding (9.48) and taking into account (9.46) and (9.47), we obtain
 h 
dgij l dx dx h
 glj ih  gi l jl h X i Y j D 0:
dt dt dt
9.6 Riemannian Connection 129

Owing to the arbitrariness of X and Y and the curve x h .t/, the following conditions
must be verified:
l
gij;h D glj ih C gi l jl h : (9.49)
Cyclically permutating the indices, we obtain the other two equations

gj h;i D glh jl i C gj l hi

l
; (9.50)
l
ghi;j D gli hj C ghl ijl : (9.51)

Adding (9.49) and (9.50) and subtracting (9.51), we are led to the equations

1
ghl ijl D .ghi;j C gj h;i  gij;h /  Œj h; i : (9.52)
2
Finally, Eqs. (9.52) supply the following expressions of the connection coefficients:
1 hm
ijm D g .ghi;j C gj h;i  gij;h /  g mh Œj h; i : (9.53)
2
It is a simple exercise to verify that the connection defined by the preceding
coefficients is without torsion and leaves invariant the scalar product under a parallel
transport. 
Definition 9.5. The Riemannian connection is the connection whose coefficients
are given by (9.53).
Theorem 9.4 (Bianchi’s theorem). The metric tensor g is uniform, that is, its
covariant derivative vanishes:
rh gij D 0: (9.54)
Proof. It is sufficient to refer to (9.49) and recall the expression of the covariant
differential of a .0; 2/-tensor. 
Theorem 9.5. In a neighborhood of any point x0 of a Riemannian manifold, it is
possible to define a local chart .U ; x i / whose domain contains x0 such that
g ij .x0 / D i ıij ; g ij;h D 0; (9.55)

where D ˙1 according to the signature of Vn (Sect. 6.9).

Proof. Let .U; x i / be a chart in a neighborhood of x0 , and denote by .x0i / the
i
coordinates of x0 in this chart. If kh .x0i / are the Riemannian connection coefficients
evaluated at x0 , then the functions
1 i
x i D x i  x0i  kh .x0i /.x k  x0k /.x h  x0h / (9.56)
2
130 9 Absolute Differential Calculus

define a coordinate transformation in a neighborhood of x0 since .@x i =@x j .x0 // D

.ıij /. Further, it is
@x i i
.x0 / D kh .x0 /: (9.57)
@x h @x k
From the transformation formulae (9.21), when we take into account that Aij .x0 / D
ıji and recall (9.57), we obtain

i i i
 kh .x0 / D kh .x0 /  kh .x0 / D 0:

Since the connection is Riemannian, (9.50) holds, and then we have gij;h .x0 / D
0. Finally, by a linear transformation .x i / ! .x 0i /, which does not modify the
preceding results, we can reduce the metric coefficients at x0 to the form (9.55). 
The Riemannian manifold Vn is said to be locally Euclidean if in the neighbor-
hood of any point there is a system of coordinates .U; x i / in which gij D i ıij ,
8x 2 U .

9.7 Differential Operators on a Riemannian Manifold

Let f be a C 1 real function on the Riemannian manifold Vn . We call gradient of f

the vector field rf whose covariant components in a chart .U; x i / are

.rf /k D f;k : (9.58)

Consequently, in the natural basis .ei /, relative to the chart .U; x i /, we can write

rf D g ij f;j ei : (9.59)

Let X be a vector field on Vn . The skew-symmetric 2-tensor

.r  X/ij D rj Xi  ri Xj D Xi;j  Xj;i (9.60)

is called a curl tensor. From (9.59) and (9.60) it follows that

r  rf D 0: (9.61)

Finally, we call the function

r  X D ri X i (9.62)

the divergence of the vector field X.

9.7 Differential Operators on a Riemannian Manifold 131

To determine a useful form of (9.62), we start by noting that

r  X D X i ;i Cih
i
X h: (9.63)

On the other hand, in view of (9.49), we have that

j
g ij gij;h  ih
i
 j h D 0;

so that we can write

i 1 ij
ih D g gij;h : (9.64)
2
On the other hand, if we set g D det.gij /, it is well known that g;h D gg ij gij;h .
Consequently, (9.64) becomes

i 1 1 p
ih D g;h D p . jgj/;h (9.65)
2g jgj

and (9.63) assumes the final form

1 p
r  X D p . jgjX i /;i : (9.66)
jgj

The Laplace operator of the C 2 real function f is

4f D r  rf: (9.67)

In local coordinates .x i /, in view of (9.67) and (9.59), we can write the Laplace
operator in the form

1 p
4f D ri .g ij f;j / D g ij .fij  ijk f; k/ D p . jgjg ij f;j /;i : (9.68)
jgj

Any function f on Vn verifying the Laplace equation

4f D 0 (9.69)

is called a harmonic function. A system of coordinates .U; x i / is harmonic in the

metric gij if the coordinate functions x i , i D 1; : : : ; n, are harmonic. Taking into
account (9.68), we can state that the coordinates x i are harmonic if they satisfy the
equations
F i  4x i D g hk hk
i
D 0: (9.70)
132 9 Absolute Differential Calculus

S C
α

A
B

Fig. 9.1 Geodesic triangle

D
C

A
B

Fig. 9.2 Parallelogram on a sphere

9.8 Exercises

1. Let C be a cylinder in which we introduce the cylindrical coordinates ', z.

Determine the geodesics of the metrics on C and compare the parallel transport
of the vector .1; 0/ along the curve z D ', 0  ' 
=4, and along the curve
union of 0  ' 
=4; z D 0, and ' D
=4, 0  z 
=4.
2. Let S be a unit sphere in which we introduce the spherical coordinates ', . The
equator  of S has parametric equations .'; 0/, 0 < '  2
. Show that the
equator is a geodesic, and evaluate the parallel transport along  of the tangent
vector .0; 1/.
9.8 Exercises 133

3. On the unit sphere S consider the triangle ABC (Fig. 9.1) in which AC and BC
are arcs of meridians, AB an arc of the equator, and B a pole. Let ˛ be the angle
at the vertex C . Show that the three sides of the triangle are arcs of geodesics and
evaluate the parallel transport of the vector .1; 0/ along the sides of ABC from
A to A.
4. Using the notations of the preceding exercise, evaluate the parallel transport of
the vector .1; 0/ from A to A along the curve ABCA in Fig. 9.2,
where AD and BC are arcs of meridians and AB and CD are arcs of parallels.
Chapter 10
An Overview of Dynamical Systems

10.1 Modeling and Dynamical Systems

In previous chapters, some fundamental concepts of algebra and differential geom-

etry were presented. This chapter is devoted to an overview of dynamical systems
that play a fundamental role in building mathematical models of reality.1
Several interesting behaviors of physical, biological, economical, and chemical
systems can be described by ordinary differential equations (ODEs). Applied
scientists are interested in mathematical problems of models stated by ODEs.
The explicit form of their solutions can be found when the ODEs are linear, but
often nonlinearity represents an inner unavoidable feature of the model, and in
this case we cannot exhibit the explicit solutions. Mathematical methods must be
developed to tackle these difficulties. We usually resort to qualitative analysis, which
supplies important aspects of solutions such as their asymptotic behavior, stability
properties, the existence of limit cycles, and the possibility of bifurcation on varying
a parameter contained in the equations. When the applications require a quantitative
description of solutions, we can use known procedures that supply the approximate
time evolution of the dependent variable obtained by numerical integration, power
series, or expansion in one or more parameters.
In attempting to describe reality, people resort to models representing simplified
but useful mathematical descriptions of phenomena we are interested in. It is not
a simple task to identify the main characteristics of a phenomenon, describe it in
terms of mathematical variables, recognize the mathematical relations among these,
and finally verify if the expectations of the model agree with observation. Solving
these difficult problems is the main goal of the mathematical modeling of nature.
In this difficult process, the transcription of an aspect of reality is often
represented by a mathematical object called a (scalar or vector) ODE. To understand

1
The topics contained in this chapter can also be found in [5, 6, 9, 25, 28, 29, 35, 40, 55].

A. Romano, Classical Mechanics with Mathematica R

, Modeling and Simulation 135
in Science, Engineering and Technology, DOI 10.1007/978-0-8176-8352-8 10,
© Springer Science+Business Media New York 2012
136 10 An Overview of Dynamical Systems

why modeling leads us to this object, and what kinds of problems one finds at the
end of this process, we start with a very simple example containing the fundamental
ingredients of the problems we face.
Suppose that one wishes to describe the evolution of a population consisting of
people, animals, bacteria, radioactive atoms, etc. The importance of this problem is
clear: its resolution could allow decisive forecasts on the destiny of the examined
population. For example, the growth velocity of a population of infectious bacteria
in an organism could suggest the appropriate dosage of antibiotic. Similarly, if the
growth law of a population is known, the right amount of food for its survival can
be evaluated.
To formulate a growth model for a population, the factors that positively or
negatively influence the aforementioned process must first be identified and a
mathematical relation supplying the number N.t/ of living individuals at instant
t as a function of those factors must be formulated. Proceeding from the simplest
situation, all external influences on growth are neglected and the food source is
assumed unlimited. In this situation, it is quite reasonable to suppose that the
variation N.t/ of the number N.t/ in the time interval .t; t C t/ is proportional
both to N.t/ and t
N.t/ D ˛N.t/t;
where ˛ is a coefficient depending on the kind of population. Of course, this
coefficient is positive if the population increases and negative in the opposite case.
In the limit t ! 0, the number N.t/ satisfies the equation

dN.t/
D ˛N.t/;
dt

in which N.t/ is the unknown appearing in the equation with its first derivative.
Such a mathematical object is just an ODE of the first order in the unknown N.t/.
We underline that the hypotheses leading us to the previous equation are very
spontaneous, but this is not the case when forecasting the form of its solution2

N.t/ D C e˛t ;

where C is an arbitrary positive constant. Now the main reason why differential
equations frequently appear in modeling natural phenomena can be recognized: It is
much easier to formulate a reasonable relation between the unknown of a given
problem and its variations than to imagine the form of the function itself. The same
kind of population admits infinite evolutions depending on the initial datum or
condition N.0/ that assigns a value to C .
The previous model, where ˛ > 0, implies that the number of individuals goes
to infinity with time, and this is absurd. To improve the model, the habitat in which

2
This solution is obtainable at once by the method of variable separation. However, the reader can
easily verify that it is really a solution for any C .
10.1 Modeling and Dynamical Systems 137

the population lives is not supposed to support a population level greater than M ;
this means that when N.t/ reaches this value, the growth rate becomes zero. The
limit M is called the carrying capacity. To take into account this constraint, the
coefficient ˛ appearing in the preceding equation is assumed depending on the
quantity .M  N.t//=M . The simplest dependency on this variable is the direct
proportionality
 
N.t/
˛ Dˇ 1 ;
M
where ˇ > 0 is constant, and we arrive at the relation
 
N.t/
NP .t/ D ˇ 1  N.t/;
M

which is called the logistic equation. This nonlinear equation gives a more accurate
description of the evolution of a population since it introduces an upper bound
to its growth. Further, it represents the starting point of many models describing
the competition among species living in the same habitat or the modified prey–
predator model of Lotka–Volterra (see, for instance, [28, 29]). When the variables
are separated, the family of solutions

M
N.t/ D
1 C C eˇt

is easily obtained, where C is an arbitrary constant determined by assigning the

initial value N.0/. It is a very lucky circumstance that the solution of the logistic
equation can be exhibited. However, more frequently, one must resort to other
procedures if information about the unknown solutions is needed.3
Before we continue, some consequences of the previous considerations must be
underlined: (1) it is much easier to model by a differential equation than directly
finding the finite relation between the involved variables; (2) when solutions exist,
they are infinite in number, but one of them is assigned by giving the value of the
unknown at a certain instant; and (3) it might be impossible to find the closed form
of these solutions.
In all the examined cases, we were led to an ordinary first-order differential
equation, i.e., to an equation containing an unknown function and its first derivative.
In many other cases, we could be compelled to model our phenomenon by a system
of two or more higher-order ODEs with two or more unknowns. This means that we
are faced with a system of equations containing more unknowns and their higher-
order derivatives. In the chapters devoted to dynamics, we shall see that describing
the evolution of mechanical systems almost always leads to a system of differential
equations. In this regard, here we only recall that Galileo and Newton proved that the

3
For the application of differential equations to economy, see, for instance, [55].
138 10 An Overview of Dynamical Systems

wide variety of possible motions of material bodies confirm the three fundamental
principles of dynamics (Chap. 13). It is an everyday experience that a body may
fall in many different ways under the influence of its weight. Similarly, the planets
describe different orbits around the Sun and traverse them at different velocities.
Galileo and Newton discovered that a very simple relation of proportionality relates
the acceleration a of a material point P to the acting force F

ma D F;

where m is the mass of the body. If .x.t/; y.t/; z.t// are the coordinates in a
frame of reference Oxyz of the moving point P as a function of time, then the
components of the acceleration vector are .x.t/;
R y.t/;
R zR.t//. On the other hand, the
force depends on the position of P (that is, on .x.t/; y.t/; z.t//) as well as on its
velocity .x.t/;
P y.t/;
P zP.t//. Consequently, the fundamental equation of dynamics is
essentially a system of three ordinary second-order differential equations in the
unknowns .x.t/; y.t/; z.t//. By the auxiliary variables .u; v; w/ D .x; P y;
P zP/, this
reduces to a system of six first-order differential equations.

10.2 General Definitions and Cauchy’s Problem

Let <n be the vector space of the ordered n-tuples of real numbers x D .x1 ; : : : ; xn /
and kxk the Euclidean norm in <n . Moreover, let f W  ! <n be a map defined
on the open subset  of <n . We call a autonomous first-order (vector) differential
equation in the normal form the following equation

xP D f.x/: (10.1)

Essentially, (10.1) is an abbreviation of the system of differential equations

xP i D fi .x1 ; : : : ; xn /; i D 1; 2; : : : ; n:

The vector space <n , to which x belongs, is said to be the state space.
A smooth function x W I ! <n , where I is a nonempty interval of <, is a
solution of (10.1) if

x.t/ 2 ; xP .t/ D f.x.t// 8t 2 I:

The set x.t/; 8t 2 I is called an orbit, and it represents a curve in the state space.
The initial-value problem or Cauchy problem for (10.1) consists in searching
for the solution x.t; t0 ; x0 / of (10.1) that satisfies the following initial condition or
initial datum:
x0 D x.t0 /; x0 2 : (10.2)
10.2 General Definitions and Cauchy’s Problem 139

It is possible to prove that the solution of an autonomous equation depends on t  t0 .

For this reason, henceforth we denote by x.t; x0 / the solution of the Cauchy problem
(10.1) and (10.2). In a geometric language we search for a curve x.t; x0 / containing
x0 for t D t0 .
A solution x.t/ W I ! <n of (10.1) is a maximal solution if there is no other
solution y.t/ W J ! <n , where I  J such that x.t/ D y.t/; 8t 2 I . It is possible
to prove that the interval in which a maximal solution is defined is always open.
The function f.x/ is a Lipschitz function in  if it satisfies a Lipschitz condition
in  with respect to x, i.e., if a positive constant K exists such that
k f.x/  f.y/ k K k x  y k (10.3)

8x; y 2 . Moreover, we say that f.x/ satisfies a local Lipschitz condition in 

with respect to x if each point of  has a neighborhood A such that the restriction
of f to A \  is a Lipschitz function with respect to x. The fundamental theorem of
ODEs is as follows, which ensures the existence and uniqueness of the solution to
the Cauchy problem locally in time.4
Theorem 10.1 (Piccard–Lindel´’of). Let f W ! <n be a continuous and locally
Lipschitz function with respect to x. For any .t0 ; x0 / 2 <  , one and only one
maximal solution of (10.1) exists x.t/ W I.t0 ; x0 / ! <n such that t0 2 I.t0 ; x0 / and
x0 D x.t0 /. Moreover, if the notation
A D f.t; t0 ; x0 / W .t0 ; x0 / 2 <  ; t 2 I.t0 ; x0 /g
is introduced, the maximal solution x.t; x0 / W A ! <n of (10.1) is continuous.
The following examples explain Theorem 10.1.
p
Example 10.1. The right-hand side of the equation xP D x does not satisfy a
Lipschitz condition in any interval Œ0; a ; a > 0. It is easy to verify that the Cauchy
problem obtained by associating to the equation the initial datum x.0/ D 0 admits
the infinite solutions
8
ˆ
ˆ 0; a  t  a;
ˆ
ˆ
ˆ
<1 2
x D 4 .t  a/ ; a  t; 8a 2 <:
ˆ
ˆ
ˆ
ˆ1
:̂ .t C a/2 ; t  a;
4

Example 10.2. The right-hand side of the equation

1
xP D
x

4
For a proof, see, for instance, [40].
140 10 An Overview of Dynamical Systems

does not satisfy a Lipschitz condition in the interval .0; 1. However, the solution
p
xD x0 2 C 2t;

corresponding to the initial condition x0 > 0, is unique.

Another very important concept is that of the general integral of (10.1). This
is a function x.t; c/; c 2 <n (1) that is a solution of (10.1) for any choice of c
and (2) where for any initial datum .t0 ; x0 / only a choice c0 of c exists for which
x0 D x.t0 ; c0 /. We conclude this section by recalling that a first integral of (10.1) is
a smooth real function g.x/; .x/ 2  that is constant along any solution of (10.1).
Formally,

g.x.t// D const; 8t 2 I; (10.4)

n
for any x.t/ W I ! < such that xP D f.x.t//.
In geometrical language, we can say that a function g.x/ in the state space is a
first integral when an orbit starting from a point of the surface g.x/ D const of <n
lies completely on it. We can introduce the vector field X.x/ D f.x/. From (10.1),
the orbits at any point x are tangent to the field X.x/ or, equivalently, the orbits
are the integral curves of the field X.x/. This geometric interpretation will be used
throughout this chapter.
Finally, we call the phase portrait the family of the orbits of (10.1). The notebook
Phase2D supplies the phase portrait of (10.1) when n D 2 and Cartesian coordinates
are adopted in the plane. The notebook PolarPlot draws the phase portrait when
polar coordinates are used.

10.3 Preliminary Considerations About Stability

Let a real system S be modeled by an equation like (10.1). A particular evolution

x.t; x0 / of S is completely determined by assigning the Cauchy datum .t0 ; x0 /.
However, this datum is obtained by an experimental procedure and is therefore
affected by an error. If a “small” difference in the initial data leads to a new solution
that is “close” to the previous one, then x.t; x0 / is said to be stable in the sense
of Lyapunov (see, e.g., [9, 25, 28, 35]). To make precise the notion of Lyapunov
stability, one must attribute a meaning to terms like small and close. By considering
that x is a point of <n , the Euclidean norm for evaluating the difference between
two solutions or two initial data can be used. In particular, the stability property
can always be expressed for an equilibrium solution. In fact, by introducing the
new unknown x  x.t; x0 /, the analysis of stability of any solution can always
be reduced to an analysis of the equilibrium stability of the origin of a system,
which, in general, becomes nonautonomous. An equilibrium position that is not
stable is called unstable. In such a case, near the equilibrium there are initial data
10.4 Definitions of Stability 141

whose corresponding solutions go definitively away from the equilibrium. Finally,

the equilibrium is asymptotically stable if it is stable, and the solutions associated
to the initial data in a neighborhood of the equilibrium tend to the equilibrium
position when the independent variable goes to infinity. In this section, we refer
to the stability of the origin of an autonomous system.
Although the concrete meaning and the importance of the stability theory
are plain, at first sight one might think that, to check the equilibrium stability,
knowledge of all solutions whose initial data are close to equilibrium is required. If
this were true, it would be almost impossible to recognize this property because of
the difficulty of obtaining the closed form of solutions. In this chapter, the Lyapunov
direct method is described; it overcomes this difficulty by introducing a suitable
function (called a Lyapunov function) that verifies the solutions. To understand this
idea, we consider the system 
xP D y;
yP D x;
whose solutions are curves .x.t/; y.t// of the plane x; y. It is very easy to verify
that the function
1
V D .x 2 C y 2 /
2
is a first integral of the preceding system because along all the solutions it is VP D 0.
The level curves V .x; y/ D const are circles, so that the solutions starting near the
origin remain near it because they must belong to the circle to which the initial
datum belongs. This, in turn, means that the origin is stable; this property was
deduced without solving the system itself.
We conclude by remarking that the stability concept is much richer because it
includes many other aspects. For example, what happens to the solutions of (10.1)
if the function on the right-hand side is slightly changed? This is a very important
problem because the function f is a mathematical transcription of the system we
are describing. Further, f includes parameters that are the results of measures and
therefore are again affected by errors. The analysis of the behavior of the solution
on varying the form of f or of the parameters it includes is carried out by the total
stability and bifurcation theory (see, for instance, [6, 28, 29, 35]). However, this
subject will not be considered in this chapter.

10.4 Definitions of Stability

We suppose that the function f W D  <n ! <n of the autonomous system

(10.1) satisfies all the conditions that ensure the existence and uniqueness of
maximal solutions, namely, the vector function f satisfies a local Lipschitz condition
(Theorem 10.1). In the sequel, x D x.t; x0 /, t 2 I , denotes the maximal solution of
(10.1) corresponding to the initial datum x0 .
142 10 An Overview of Dynamical Systems

Any constant function x.t/ D x that is a solution of (10.1) is called an

equilibrium solution. Equivalently, x is a root of the equation f.x/ D 0. By a
suitable axis translation, it is always possible to reduce any equilibrium solution at
the origin. For this reason, in the sequel it is assumed that f.0/ D 0. Moreover, 
will denote the distance between the origin x D 0 and the boundary @D of D.
The origin is a stable equilibrium solution if a positive real number  exists such
that 8 20; Œ, 9ı./ 20; Œ for which

k x0 k< ı./ H)k x.t; x0 / k< ; 8t > 0: (10.5)

It is evident that ı./ < ; otherwise, condition (10.5) should not be verified at
the initial instant. It is important to note the following. First, the stability property
is equivalent to the continuity of solutions of (10.1) with respect to the initial
datum x0 uniformly in the unbounded time interval Œ0; 1. If (10.1) is interpreted
as a mathematical model of a real system S , its solutions represent the possible
evolutions of the variable x describing the state of S . To have S at the equilibrium
state x.t/ D 0,S must initially be put at 0. But this operation is physically
impossible because it is necessarily realized by procedures that introduce measure
errors, however accurate they may be. This means that the system is put at an initial
state x0 next to 0, and the corresponding solution could lead the state of the system
definitively far from the equilibrium. If this happened for some initial data, the
equilibrium state itself would not be physically observable. In contrast, if 0 is stable,
this situation is not confirmed. In fact, if the notation Uı D fx 2 <n Wk x k< ıg is
used, the stability notion can be formulated in the following terms. If any region
U  D is fixed around the origin, it is possible to find a whole neighborhood Uı 
U of initial conditions whose corresponding solutions are fully contained in U .
The equilibrium solution x D 0 is as follows:
1. Attractive if

.a/ x.t; x0 / exists 8t  0

9ı 20; Œ W 8x0 2 Uı H)
.b/ 8 > 0; 9T ./ Wk x.t; x0 / k< ; 8t  T ./I
(10.6)
2. Asymptotically stable if it is stable and

lim x.t; x0 / D 0I (10.7)

t !1

3. Unstable if it is not stable, i.e., if

9 20; Œ W 8ı 20; ; 9x0 2 Uı ; t  > 0 Wk x.t  ; x0 / k : (10.8)

The set 0 of initial data x0 for which (10.6) is satisfied is called the domain
of attraction. In particular, if 0 D D, x D 0 is globally asymptotically
stable. Figures 10.1–10.3 supply a rough representation of an attractive origin,
10.4 Definitions of Stability 143

U
x0 x
O

Fig. 10.1 Stable origin

Fig. 10.2 Attractive origin

U
x0 x
O

Fig. 10.3 Asymptotically stable origin

144 10 An Overview of Dynamical Systems

Fig. 10.4 Unstable origin

asymptotically stable origin, and unstable origin, respectively. In Fig. 10.1, the
origin is stable since there are two neighborhoods Uı and U of the origin, where
Uı  U , of the origin such that the orbits starting from any point in Uı are
always contained in U . In Fig. 10.2, the origin is attractive since the solutions
corresponding to the initial data belonging to 0 are contained in a neighborhood
of the origin after a suitable value of time. In Fig. 10.3, the origin is asymptotically
stable because the solutions corresponding to the initial data belonging to Uı are
contained in Uı and, moreover, tend to the origin when the variable t goes to
infinity. Finally, in Fig. 10.4, the origin is unstable because there are initial data
whose corresponding solutions go definitively away from the origin.

10.5 Analysis of Stability: The Direct Method

The previous stability definitions could generate the wrong idea that all solutions
of (10.1), corresponding to initial data in a neighborhood of 0, must be known
in order to recognize the stability of the equilibrium position. It is evident that
the solutions of (10.1) can be exhibited in only a few elementary cases (linear or
special systems) so that an alternative method is needed to check the stability of
the origin. In this section, it is shown that it is possible to recognize the stability
of the origin by analyzing the right-hand side of (10.1). This is realized in the
direct method of Lyapunov, which reduces the stability problem to that of the
existence of suitable functions (Lyapunov functions) having, with their derivatives,
determined the properties on the solutions of (10.5). This method is called direct
because, under the hypothesis that these functions are of class C 1 , the corresponding
derivatives on the solutions of (10.1) can be directly expressed by the right-hand
side of (10.1) without knowing the solutions themselves. The determination of a
10.5 Analysis of Stability: The Direct Method 145

Lyapunov function, in turn, is not easy. Therefore, the original idea was developed
in different directions to make easier the application of the method itself.
In this section, some stability or instability criteria related to the direct method
are discussed.
We extensively use the notation

VP .x/  f.x/  rV .x/; (10.9)

together with the remark that along the solutions of (10.1) VP .x/ coincides with the
total derivative of V .x.t; x0 // because

VP .x.t; x0 //  xP .t; x0 /  rV .x.t; x0 // D f.x.t; x0 //  rV .x.t; x0 //: (10.10)

This remark implies that the sign of VP .x.t; x0 // along the solutions can be
established without their preliminary knowledge if the sign of VP .x/ around the
origin is known.
Definition 10.1. Let V W U ! < be a continuous function such that V .0/ D 0.
V is said to be positive .negative/ definite on U if V .x/ > 0 .V .x/ < 0/ for x ¤ 0
or positive .negative/ semidefinite if V .x/  0 .V .x/  0/.
Theorem 10.2. Let V W U ! < be a class C 1 function that is positive definite
in a neighborhood U of the origin. If VP .x/ is negative semidefinite in U , then
the origin is stable. Moreover, if VP .x/ is negative definite in U , then the origin is
asymptotically stable.
Proof. 8 2 .0;    we denote a sphere with its center at the origin and radius
. Since U  U and in U  f0g the function V .x/ is positive definite and
continuous, there exists a positive minimum of the restriction of V .x/ on @U :

0 < V  min V .x/:

x2@U

Again for the continuity of V .x/ at the origin, there is a positive real quantity ı < 
such that
jxj < ı ) 0 < V .x/ < V :
Now we prove that any solution x.t; x0 / of (10.1), for which jx0 j < ı, remains for
any t > 0 in the sphere S . In fact, if this statement were false, there should be an
instant t  in which the orbit x.t; x0 / should cross @S . But initially it is V .x0 / < V ,
so that we should have V .x0 / < V  V .x.t  ; x0 // against the condition VP  0
along the orbit x.t; x0 /. 
The proof of the following theorem is omitted.
Theorem 10.3. Let V W U ! < be a class C 1 function in a neighborhood U
of the origin such that (1) V .0/ D 0 and (2) for any neighborhood U  U a point
146 10 An Overview of Dynamical Systems

Fig. 10.5 Asymptotically stable origin

x 2 U exists at which V .x/ > 0. If VP .x/ is positive definite in U , then the origin
is unstable.
It is possible to recognize the role of the hypotheses contained in Theorems (10.2)
and (10.3) by a simple geometric description.
In fact, let c be a trajectory starting from a point x0 of a level curve  of the
Lyapunov function. At this point, c has a tangent vector xP D f.x0 / that points
toward the internal region of  , and the angle between rV and xP is greater than
. Consequently, c meets the more internal level curves under the same conditions.
In this way, c approaches the origin (Fig. 10.5).
Similar considerations, applied to the case VP  0, allow us to conclude that the
trajectory does not leave the internal region to the level curve of V containing x0 :
However, the condition VP  0 does not exclude the possibility that the trajectory
will remain on a level curve so that it cannot approach the origin.
In particular, when VP D 0 along the solutions, one has the situation illustrated in
Fig. 10.6.
Before stating the next theorem of Barbashin–Krasovskii, a simple physical
example will be discussed. Let P be a harmonic oscillator of unit mass moving
on the Ox-axis subject to an elastic force kx, k > 0, and to a linear friction hx, P
h > 0. Its motion x.t/ is given by the linear system

xP D v;
vP D kx  hv:
It is quite evident from a physical point of view that the origin is an asymp-
totically stable equilibrium position. To verify this property of the origin, the total
10.5 Analysis of Stability: The Direct Method 147

Fig. 10.6 Stable origin

energy V .x; v/ is chosen as a Lyapunov function because it is positive definite and

decreases along any motion for the dissipation. In fact, from the total energy

1 2
V .x; v/ D .v C kx 2 /
2
we have
VP .x; v/ D hv2 :
However, VP .x; v/ vanishes not only at the origin but also along the x-axis. From
the previous theorems we can only conclude that the origin is stable. The following
theorems allow us to establish that the origin is asymptotically stable by using the
same Lyapunov function.
Theorem 10.4. (Barbashin–Krasovskii) Let V W P ! < be a class C 1 function
that is positive definite in a neighborhood P of the origin. If VP .x/ is negative
semidefinite in P and if x D 0 is the only solution of (10.1) for which VP .x/ D 0,
then the origin is asymptotically stable.
Theorem 10.5. (Krasovskii) Let V W P ! < be a class C 1 function such that
1. 8 2 .0;  /; 9x 2 P W V .x/ > 0I
2. VP .x/  0; 8x 2 P I and
3. x D 0 is the only solution of (5.1) for which VP .x/ D 0:
Then the origin is unstable.
Returning to the preceding example, we remark that the equation VP D hv2 D 0
implies v D 0 along the solutions. On the other hand, from the differential system it
148 10 An Overview of Dynamical Systems

follows that x D v D 0; then the hypotheses of Theorem 10.4 are satisfied and the
origin is asymptotically stable.
The preceding theorems make possible the analysis of the stability properties
of the equilibrium solution x D 0 given a Lyapunov function possessing suitable
property. We have already remarked that a first integral of (10.1) can be used as a
Lyapunov function. In some cases, polynomial Lyapunov functions can be easily
found (see exercises at end of chapter). However, in general, it is not an easy task to
determine a Lyapunov function. In this regard, the following theorem, which allows
us to state the stability of the equilibrium solution x D 0 by an analysis of the linear
part of (10.1) around 0, could be useful.5
More precisely, since f.0/ D 0, (10.1) can be written as follows:

xP D Ax C ˆ.x/; (10.11)

where A D .rf/xD0 is an n  n constant matrix and ˆ.x/ D o.k x k/ when x ! 0.

Then, it is possible to prove the following theorem.
Theorem 10.6. If all the eigenvalues of matrix A have negative real part, then
the solution x D 0 of (10.1) is asymptotically stable. If among the eigenvalues
of matrix A there is at least one whose real part is positive, the solution x D 0 of
(10.1) is unstable.
The notebook LinStab is based on this theorem.

10.6 Poincare’s Perturbation Method

In this section, we search for approximate solutions of (10.1). It should be possible

to search for such a solution as a Taylor expansion. This approach supplies a
polynomial of a fixed degree r, approximating the solution of (10.1) at least in
a neighborhood of the initial value x0 . A better approximation is obtained by
increasing the value r, that is, by considering more terms of the power expansion,
or values of the independent variable t closer to the initial value t D 0. In contrast,
Poincaré’s method tries to give an approximate solution in an extended interval
of the variable t, possibly for any t. On this logic, one is ready to accept a less
accurate solution, provided that it approximates the solution uniformly with respect
to t. In effect, Poincaré’s approach does not always give a solution with this
characteristic; more frequently, it gives an approximate solution whose degree of
accuracy is not uniform with respect to time. This is due, as we shall see in this
chapter, to the presence in Poincaré’s expansion of so-called secular terms that go
to infinity with t. This behavior of the approximate solution is not acceptable when

5
Readers will find in [35] many programs, written using Mathematica, that allow for the analysis
of many stability problems.
10.6 Poincare’s Perturbation Method 149

the solution we are searching for is periodic. However, in this case, by applying
another procedure (Lindstedt, Poincaré), a uniform expansion with respect to t can
be derived (see, for instance, [28, 29, 35]).
Poincaré’s method is applicable to a differential equation, which can be written
in the form
xP D f.x/ C F.x/; (10.12)
where  is a suitable “small” dimensionless parameter related to the physical system
modeled by the preceding differential equation.
The method can be developed, provided that the solution of the equation
xP D f.x/ C g.t/;

where g.t/ is a known function belonging to a suitable class, can be computed for
a given initial condition and if the perturbation term F is analytic in its argument.
When both conditions are fulfilled, it is possible to look for the solution in powers
of . More precisely, it can be proved that the solution of (10.12), under suitable
hypotheses on the functions f and F, can be written as a power series of the
parameter ,
x.t/ D x0 .t/ C x1 .t/ C  2 xn .t/ C    ;
that converges to the solution of (10.12) uniformly with respect to  but usually not
with respect to t. The general term xn .t/ of the preceding expansion is the solution
of the equation
xP n D f.xn/ C gn .t/;
where the function gn .t/ is completely determined by the previous terms x1 .t/,
: : : ; xn1 .t/.
The meaning of the attribute small applied to  requires an explanation. The
parameter  is small if  1 and the terms xP , f.x/, and F.x/ are all comparable
with unity. However, it is not possible to verify this property because usually there
is no knowledge of the solution. To overcome this difficulty, we can resort to a
dimensionless analysis of the problem modeled by (10.12), provided that we have
a sufficient knowledge of the phenomenon we are analyzing. We make clear this
procedure with some examples.
We analyze in more detail the case where the function f in (10.12) depends
linearly on x, i.e., we consider the following Cauchy problem:

xP D Ax C F.x; /;
(10.13)
x.t0 / D x;
where A is an n  n matrix with constant coefficients.
Poincaré proved that if F.x; / is an analytic function of its variables, then the
solution x.t; / of (10.13) is analytic with respect to ; consequently, it can be
expressed by an expansion

x.t; / D x0 .t/ C x1 .t/ C x2 .t/ 2 C    (10.14)

150 10 An Overview of Dynamical Systems

uniformly convergent with respect to  in a neighborhood of the origin. It is evident

that the functions xi .t/ must verify the initial data
8
ˆ
ˆ x0 .t0 / D x;
<
x1 .t0 / D 0;
(10.15)
ˆ x2 .t0 / D 0;
:̂
::::

To find the terms of the expansion (10.14), we expand F.x; / with respect to
 and then introduce (10.14) into (10.13)1. In this way, the following sequence of
Cauchy problems is derived, which determine x0 .t/, x1 .t/, x2 .t/, and so on:

xP 0 D Ax0 ;
(10.16)
x0 .t0 / D xN ;

xP 1 D Ax1 C F.x0 .t/; 0/;
(10.17)
x1 .t0 / D 0;


xP 2 D Ax2 C .F /.x0 .t/; 0/ C x1 .t/  .rx F/.x0 .t/; 0/;
(10.18)
x2 .t0 / D 0;

where .F /.x0 .t/; t; 0/ is the derivative of F with respect to  evaluated at .x0 .t/; 0/
and .rx F/.x0 .t/; 0/ denotes the gradient of F with respect to the variable x at
.x0 .t/; 0/.
It is very important to note that the problems (10.16), (10.17), : : : refer to the
same linear differential equation, which is homogeneous at the first step and nonho-
mogeneous at the next steps. However, the terms appearing at the i th step are known
when the previous Cauchy problems have been solved. Although the simplification
reached in solving the original Cauchy problem is clear, the calculations to write
and solve the different systems are very heavy and cumbersome. For these reasons,
the reader can use the notebook Poincare.
It is also very important to recall that expansion (10.14) is usually not uniform
with respect to time. As we shall see, in the series (10.14) some secular terms
 n t n sin nt,  n t n cos nt can appear. When this happens, one is compelled to accept
into (10.14) time values verifying the inequality t a=b, where a and b denote
the maximum values of the coefficients of, respectively, x0 .t/ and the secular terms
in x1 .t/, to be sure that the second term of the expansion (10.14) is small with
respect to the first one for time values t a=b.
10.7 Introducing the Small Parameter 151

10.7 Introducing the Small Parameter

To establish a model describing a certain system (e.g., physical, economical), we

must first identify a set of variables x D .x1 ; : : : ; xn / depending on an independent
variable t that should describe, in the framework of the mathematical model, the
physical state of the system we are dealing with.
Generally, independent and dependent variables of a model have dimensional
values related to the physical system. This may cause problems in comparing small
and large deviations of the variables. This problem can be overcome by putting all
the variables in a suitable dimensionless form, relating them to convenient reference
quantities. The criterion to choose these quantities must be such that the derived
dimensionless quantities have values close to one. Only after this analysis it is
possible to check whether the system has the form (10.13) and Poincaré’s method is
applicable.
This first step is a necessary step to apply Poincaré’s method and, in particular,
to verify whether (10.13) satisfies all the required conditions. Two mechanical
examples will be considered to explain this procedure in detail. Two other examples
will be considered in Chap. 14.
Example 10.3. If P is a material point of mass m subject to a nonlinear elastic force
hx  kx 3 and constrained to move on a straight line Ox, the Newtonian equation
governing its motion yields

mxR D hx  kx 3 : (10.19)

To put (10.19) in a dimensionless form, two reference quantities L and T are

introduced with, respectively, the dimensions of length and time
r
m
L D x0 ; T D ; (10.20)
h

where x0 denotes the abscissa of the initial position of P and T is equal to 2 times
the period of the oscillations one has when only the linear part hx of the elastic
force is acting. Defining the dimensionless quantities

x t
x D ; t D (10.21)
L T
and remarking that

L dx  L L d2 x  L
xP D D xP  ; xR D D 2 xR  ;
T dt  T T 2 dt 2 T
152 10 An Overview of Dynamical Systems

(10.19), in view of (10.20), (10.21), and the last relations, yields

kx02 3
xR  D x   x :
h
Since this equation is equivalent to the first-order system

xP  D y  ;
kx02 3
yP  D x   x ;
h
the dimensionless parameter
kx02
D
h
can be identified with the small parameter of Poincaré’s method, provided
that <<1.
Example 10.4. Now we take another mechanical example into account: a moving
point P under the action of its weight and a friction. The motion equation
is written as
vP D g  hvv; (10.22)
where g is the gravity acceleration, v D jvj, and h is a positive constant depending
on the medium in which P is moving as well as on its form. Refer the previous
vector equation to a frame Oxy whose origin O is at the initial position of P and
whose Ox-axis and Oy-axis are taken to be horizontal and vertical, respectively,
and such that the (vertical) plane Oxy contains the initial velocity of P . Since the
motion is planar, the whole trajectory is contained in the plane Oxy. To identify the
small parameter, we note that when friction is absent, the motion equation

vP D g

admits the solutions

vx .t/ D v0 cos ˛; (10.23)

vy .t/ D v0 sin ˛ C gt; (10.24)

where v0 D jv.0/j and ˛ is the angle between the initial velocity v.0/ and the
horizontal Ox-axis. Since the quantities vx .t/ and vy .t/, in the presence of friction,
are of the same order of magnitude as the previous ones, the height L at which a
heavy body arrives, before inverting its motion, and the time T to reach the soil
again can be taken as reference quantities of length and time, respectively. It is easy
to deduce from (10.23) and (10.24) the formulae

v20 2v0
LD ; T D
2g g
10.8 Weierstrass’ Qualitative Analysis 153

so that
L g
D :
T2 8
With the introduction of these reference quantities, (10.22) assumes the following
dimensionless form:
vP D 8k  hLvv; (10.25)
where k is the unit vector along the Oy-axis. If the factor

 D hL<<8;

then we have an equation to which the perturbation method is applicable. It is well

known that in the presence of nonlinear friction, the velocity of a heavy point tends
to a limit value. It has already been pointed out that Poincaré’s method cannot supply
correct results if the approximate solution is used in an extended time interval. This
implies that the method does not result in a behavior that exhibits a limit velocity. It
approximates the effective solution only if the motion lasts a sufficiently short time.

10.8 Weierstrass’ Qualitative Analysis

In this section, we consider the second-order scalar differential equation

xR D f .x/ (10.26)

in the unknown x.t/ since we will encounter it in many mechanical applications.

First, we notice that (10.26) is equivalent to the following first-order system:

xP D v; (10.27)
vP D f .x/; (10.28)

in the unknowns x.t/, v.t/. This pair of functions defines a curve  in the state space

whose structure depends on the meaning of the variable x. For instance, if x is the
angle ' varying on a circumference, then
is a cylinder. The equilibrium positions
of (10.27), (10.28) are the solutions of the system

f .x/ D 0; v D 0: (10.29)

The purpose of Weierstrass’s analysis consists in determining the qualitative

behavior of the phase portrait of (10.27), (10.28).
To localize the orbits of (10.27), (10.28), we note that

1 2
V .x; v/ D v C U.x/; (10.30)
2
154 10 An Overview of Dynamical Systems

where Z
U.x/ D  f .x/ dx (10.31)

is a first integral of (10.27), (10.28) since

VP D vPv C U 0 .x/xP D vPv  f .x/xP

vanishes along the solution of (10.27), (10.28).

Theorem 10.7 (Dirichlet). If U.x/ is a C 1 function with a minimum at the point
x  , then .x  ; 0/ is a stable equilibrium position of the system (10.27), (10.28).
Proof. If U is of class C 1 and has a minimum at x  , then we have U 0 .x  / D
f .x  / D 0 and .x  ; 0/ is an equilibrium position. Moreover, since U is defined
to within a constant, we can always suppose that U.x  / D 0. Consequently, we
have that V .x  ; 0/ D 0. Further, since U.x/ has a minimum at x  , there exists a
neighborhood of x  in which V .x; v/ D v2 =2 C U.x/ > 0. Finally, we have already
proved that VP D 0 along the solutions of (10.27), (10.28). In other words, V .x; v/
is a Lyapunov function that satisfies the hypotheses of Theorem 10.2. 
We omit the proof of the following theorem, which supplies an instability criterion.
Theorem 10.8 (Chetaiev). If the point x  is not a minimum of U.x/ and the
absence of the minimum can be determined by the derivatives U .k/ .x  / of U.x/
at x  , where k is a finite integer number, then .x  ; 0/ is an unstable equilibrium
position of the system (10.27), (10.28).
Now we show that (10.26) and (10.30) allow us to discover the behavior of the
solutions and the structure of the phase portrait applying a procedure proposed by
Weierstrass. In fact, from (10.30) we deduce that
p
v D
2.E  U.x//: (10.32)

This relation (recall that v D dx=dt) implies that

Z x
ds
t D
p : (10.33)
x0 2.E  U.x//

Equation (10.32) is the explicit form of level curves V D cost. It shows that these
curves are symmetric with respect to the Ox-axis. Moreover, in view of (10.27),
(10.28), when v D 0, we also have xP D 0. Consequently, the tangent straight line
to the level curves at the points in which they intersect Ox is orthogonal to Ox.
Finally, for any fixed value of E, the admissible values of x satisfy the condition
U.x/  E. Equation (10.33) gives the time needed to go from x0 to x.
If we assign an initial datum .x0 ; v0 /, where v0 > 0, then we must choose the
C sign in (10.32), (10.33) up to the instant at which U.x/ D E. Consequently, if
10.8 Weierstrass’ Qualitative Analysis 155

U.x/ < E, 8x > x0 , the function x.t; x0 ;p v0 / goes to infinity in a finite or infinite
time according to whether the function 1=. 2.E  U.x/// is integrable or not.
In contrast, if on the right-hand side of x0 there is a point x where U.x/ D E,
then we must consider the following two cases:

.a/ U 0 .x/ ¤ 0; .b/ U 0 .x/ D 0:

In case .a/, Taylor’s expansion of U.x/ at x is

U.x/ D E C U 0 .x/.x  x/ C O.jx  xj/:/ (10.34)

Therefore, the function under the integral in (10.33) is integrable in Œx0 ; x/, and
x.t; x0 ; v0 / reaches the point x in the finite time
Z x
ds
tD p : (10.35)
x0 2.E  U.x//

Further, U.x/ < E for x < x and U.x/ D E for x D x. Therefore, the function
U.x/ increases at the point x so that f .x/ D U 0 .x/ < 0. Finally, if we take
into account (10.28), then we also have vP .t ; x0 ; v0 / D x.t; R x0 ; v0 / < 0, whereas
(10.27) and (10.32) imply that v.t; x0 ; v0 / D x.t P ; x0 ; v0 / D 0. In conclusion,
x.t
P ; x0 ; v0 / < 0 for t < t and x.t ; x0 ; v0 / comes back toward x0 . In particular, if to
the left of x0 is another point x with the same characteristics of x, then the function
x.t; x0 ; v0 / is periodic with a period
Z x
ds
T D2 p : (10.36)
x 2.E  U.x//
In case .b/, Taylor’s expansion of U.x/ at x is
1
U.x/ D E C U 00 .x/.x  x/2 C O.jx  xj2 /; / (10.37)
2
p
and the approach time to x is infinity since the function 1=. 2.E  U.x/// is no
more integrable and x.t ; x0 ; v0 / tends asymptotically to x. We notice that, in this
case, x is an unstable equilibrium position since the conditions U.x/ < E for x < x,
U.x/ D E for x D x, and U 0 .x/ D 0, ensure that x is not a minimum.
The analysis we have just described can be carried out by the notebook
Weierstrass.
Example 10.5. Figure 10.7 shows the qualitative behavior of the potential energy
U.x/ and the corresponding phase portrait of (10.27), (10.28). For E < E1
there is no solution. For E D E1 the level curve reduces to the equilibrium
position .x1 ; 0/. For E1 < E < E  , the level curves are closed and the orbits
correspond to periodic motions. For E  < E < E2 , the level curves have two
connected components: the first of them corresponds to a periodic orbit, whereas
156 10 An Overview of Dynamical Systems

U (x)
E4

E3

E2

E*
E1
x
x1 x2 x3
v

Fig. 10.7 Phase portrait

the second one is an aperiodic orbit going to infinity. For E D E2 we have a level
curve with three connected components corresponding to a bounded open orbit, an
unstable equilibrium position, and an open orbit going to infinity. It is a simple
task to complete the analysis of the phase portrait. We conclude by noting that the
dashed lines represent orbits that tend to an unstable equilibrium position without
reaching it.

10.9 Exercises
1
 2 
1. Verify that V D 2 x C y 2 is a Lyapunov function for the system

xP D y C ax.x 2 C y 2 /;
yP D x C ay.x 2 C y 2 /;

and determine the stability property of the origin upon varying the constant a.
10.9 Exercises 157

1
 2 
2. Verify that V D 2 x C y 2 is a Lyapunov function for the system

xP D y C axy 2 ;
yP D x  yx 2 ;

and determine the stability property of the origin upon varying the constant a.
3. Determine the periodic orbits of the following system in polar coordinates
.r; '/:

rP D r.1  r/.2  r/;

'P D 2:

Are there equilibrium positions? Control the results using the notebook
PolarPhase.
4. Determine the equilibrium positions of the system

xP D 2xy;
yP D 1  3x 2  y 2 ;

and analyze their linear stability properties. Control the obtained results using
the notebook Linstab.
5. Determine the equilibrium positions of the system

xP D x C y;
yP D x C 2xy;

and analyze their linear stability properties. Control the obtained results using
the notebook Linstab.
6. Determine the equilibrium positions of the system

xP D x.x 2 C y 2  1/  y.x 2 C y 2 C 1/;

yP D y.x 2 C y 2  1/ C x.x 2 C y 2 C 1/;

and analyze their linear stability properties. Control the obtained results using
the notebook Linstab.
7. Determine an approximate solution of the system

xP D y;
yP D x C .1  x 2 /y

using Poincaré’s method, and control the obtained results using the notebook
Poincare.
158 10 An Overview of Dynamical Systems

8. Determine an approximate solution of the system

xP D y;
yP D x  x 3

using Poincaré’s method, and control the obtained results using the notebook
Poincare.
9. Determine an approximate solution of the system

xP D y C .x 2  y 2 /;
yP D x C y 3

using Poincaré’s method, and control the obtained results using the notebook
Poincare.
10. Determine the equilibrium configurations of the potential energy

1 2 1 4
U D x C x ;
2 4
their stability properties, and the phase portrait using a Weierstrass analysis.
Control the obtained results using the notebook Weierstrass.
11. Determine the equilibrium configurations of the potential energy

U.x/ D  cos.x/;

their stability properties, and the phase portrait using a Weierstrass analysis.
Control the obtained results using the notebook Weierstrass.
12. Determine the equilibrium configurations of the potential energy
2
U.x/ D ex cos.x/;

their stability properties, and the phase portrait using a Weierstrass analysis.
Control the obtained results using the notebook Weierstrass.
 Part II
Mechanics
Chapter 11
Kinematics of a Point Particle

Kinematics analyzes the trajectories, velocities, and accelerations of the points of a

moving body, the deformations of its volume elements, and the dependence of all
these quantities on the frame of reference. In many cases, when such an accurate
description of motion is too complex, it is convenient to substitute the real body
with an ideal body for which the analysis of the preceding characteristics is simpler,
provided that the kinematic description of the ideal body is sufficiently close to the
behavior of the real one. For instance, when the deformations undergone by a body
under the influence of the acting forces can be neglected, we adopt the rigid body
model, which is defined by the condition that the distances among its points do not
change during the motion. More particularly, if the body is contained in a sphere
whose radius is much smaller than the length of its position vectors relative to a
frame of reference, then the whole body is sufficiently localized by the position of
any one of its points. In this case, we adopt the model of a point particle.1

11.1 Space-Time Frames of Reference of Classical

Kinematics

To analyze physical phenomena, an observer adopts a body of reference S at whose

points he places identical clocks. For the present, we consider a clock as an arbitrary
device defining the local time, i.e., an increasing continuous variable t. The set R,
formed by S and the clocks placed in its points, is called a space-time frame of
reference.
By a space-time frame of reference R, we can introduce a time order into the
set of events happening at an arbitrary point A 2 S . In fact, it is sufficient to label
each event with the instant at which it takes place. Although we can introduce a

1
The contents of Chapters 11–17 can also be found in [4, 12, 20, 23, 24, 26, 32, 33, 45, 46, 49, 54].

A. Romano, Classical Mechanics with Mathematica R

, Modeling and Simulation 161
in Science, Engineering and Technology, DOI 10.1007/978-0-8176-8352-8 11,
© Springer Science+Business Media New York 2012
162 11 Kinematics of a Point Particle

chronology at any point A of S using the local time, at present we have no criterion
to compare the chronology of the events at A with the chronology of the events at
another point B of S . Only after the introduction of such a criterion can we state
when an event at A is simultaneous with an event at B or when an event at A happens
before or after an event at B. Already the measure of the distance d.A; B/ between
two points of S at a given instant requires the definition of simultaneity of events
happening at two different points.
We suppose that there is no problem in synchronizing two clocks at the same
point A of S . To introduce a definition of synchronism between two distant clocks,
we start by placing at the point A many identical and synchronous clocks. Then we
postulate that these clocks remain synchronous when they are placed at different
points of S . In other words, we assume the following axiom.
Axiom 11.1. The behavior of a clock does not depend on its motion. In particular,
it is not influenced by the transport.
This postulate allows us to define a universal time, i.e., a time that does not
depend on a point of S . In fact, it is sufficient to place synchronous clocks at A
and then to transport them to different points of S . Adopting this universal time,
measured by synchronous clocks distributed at different points of S , we can define
a rigid body as a body S for which the distance d.A; B/ between two arbitrary
points A and B of S is constant in time.
Another fundamental axiom of classical kinematics follows.
Axiom 11.2. There exist rigid bodies inside which the geometry of a three-
dimensional Euclidean space E3 holds.

11.2 Trajectory of a Point Particle

Let R D fS; tg be a frame of reference formed by the rigid body S and universal
time t. Henceforth, we will often identify a frame of reference with an orthonormal
frame .O; .ei //, i D 1; 2; 3, of the three-dimensional Euclidean space describing
the geometry of S (Axiom 11.2). In this section, we analyze the kinematic behavior
of a point particle P moving with respect to .O; .ei //.
The equation of motion of P is a function

r D r.t/; t 2 Œt1 ; t2 ; (11.1)

giving the position vector r.t/ of P relative to the orthonormal frame .O; .ei // at
any instant t (Fig. 11.1). If .xi / are the Cartesian coordinates of P in .O; .ei //, then
(11.1) can also be written in the following form:2

2
Henceforth we intend the summation from 1 to 3 on repeated indices. We use only covariant
indices since, in Cartesian orthogonal coordinates, there is no difference between covariant and
contravariant components.
11.3 Velocity and Acceleration 163

e r t

x
O
e
e

Fig. 11.1 Trajectory of a point particle

r D xi .t/ei ; t 2 Œt1 ; t2 ; (11.2)

where the functions xi .t/ should be of class C 2 Œt1 ; t2 .

The locus  of points (11.1) upon varying t in the interval Œt1 ; t2  is called the
trajectory of P . If we denote by s.t/ the value of the curvilinear abscissa on 
at instant t, then the trajectory and the equation of motion can respectively be
written as
r D rO .s/; (11.3)
r.t/ D rO .s.t//: (11.4)

11.3 Velocity and Acceleration

Let P be a point particle moving relative to the reference .O; .ei //, and let r.t/ be
the equation of motion of P . The vector velocity or simply velocity of P in .O; .ei //
is the vector function dr
rP D (11.5)
dt
and the scalar velocity is the derivative sP .t/.
Differentiating (11.4) with respect to time and recalling the first Frenet formula
dOr
D t;
ds
164 11 Kinematics of a Point Particle

where t is the unit vector tangent to the trajectory, we obtain the following relation
between the vector velocity and the scalar velocity:
rP D sP t: (11.6)

In other words, the vector velocity is directed along the tangent to the trajectory
and its length is equal to the absolute value of the scalar velocity. Finally, it has the
versus of t or the opposite versus depending on the sign of sP .
In the Cartesian frame of reference .O; .ei //, we can write r.t/ D xi .t/ei so that
the velocity has the following Cartesian expression:

rP D xP i ei : (11.7)

The vector acceleration, or simply the acceleration, of P is given by the vector

dPr d2 r
rR D D 2; (11.8)
dt dt
which, in view of (11.7), in Cartesian coordinates has the form

rR D xR i ei : (11.9)
Recalling the second Frenet formula

dt n
D ;
dt R
where n is the unit principal normal to the trajectory and R the radius of the
osculating circle, we obtain another important expression of acceleration differenti-
ating (11.6):
v2
rR D sR t C
n; .v D jPrj D jPs j/: (11.10)
R
The preceding formula shows that the acceleration
• Lies in the osculating plane of trajectory,
• Is directed along the tangent if the trajectory is a straight line, and
• Is normal to the trajectory when sP D 0 (uniform motion).
Henceforth, we will use the following definitions:

sR D scalar acceleration
sR t D tangential acceleration
v2
n D centripetal acceleration (11.11)
R
11.4 Velocity and Acceleration in Plane Motions 165

We say that the motion of P is accelerated if v (equivalently, v2 ) is an increasing

function of time and decelerated in the opposite case. Since

dv2 dPs 2
D D 2Ps sR ;
dt dt
the motion is accelerated if sP sR > 0 and decelerated if sP sR < 0.

11.4 Velocity and Acceleration in Plane Motions

Let P be a point moving on the plane ˛. Introducing in ˛ polar coordinates .r; '/,
the motion of P is represented by the equations

r D r.t/; ' D '.t/: (11.12)

The square distance ds 2 between the points .r; '/ and .r C dr; ' C d'/ is

ds 2 D dr 2 C r 2 d 2 ;
and the vectors
@ @
eO r D ; eO ' D
@r @'
of the holonomic base relative to the polar coordinates verify the conditions
eO r  eO r D 1; eO '  eO ' D r 2 ; eO r  eO ' D 0:
We denote by .P; er ; e' / the orthonormal frame relative to the polar coordinates at
any point P 2 ˛ (Fig. 11.2), where
1
er D eO r ; e' D eO ' :
r

v
e
er
P

r
j

O i

Fig. 11.2 Trajectory in polar coordinates

166 11 Kinematics of a Point Particle

Then we define the radial velocity vr and transverse velocity v' are the scalar
quantities
vr D rP  er ; v' D rP  e' ; (11.13)
that is, the components of vP relative to the base .er ; e' / at the point P 2  . To
determine the form of (11.13) in terms of (11.12), we note that, in polar coordinates,
the position vector r is written as
r D rer .r; '/:
Differentiating this expression with respect to time, we obtain
 
@er @er
rP D rP er C r rP C 'P :
@r @'

On the other hand, in a Cartesian frame of reference .O; i; j/ (Fig. 11.2), this
results in
er D cos 'i C sin 'j; e' D  sin 'i C cos 'j;
so that we have
@er @er
D 0; D e' ;
@r @' (11.14)
@e' @e'
D 0; D er ;
@r @'

and the velocity in polar coordinates is written as

rP D re
P r C r 'e
P ': (11.15)

In view of (1.13) and (1.15), the radial and transverse velocities are given by the
following formulae:
vr D rP ; v' D r ':
P (11.16)
Differentiating (11.15) with respect to time, we obtain the acceleration of P in polar
coordinates:
1 d 2
rR D .Rr  r 'P 2 /er C P ';
.r '/e (11.17)
r dt
so that its components along er and e' are

ar D a  er D rR  r 'P 2 ; (11.18)
1 d 2
a' D a  e' D P
.r '/: (11.19)
r dt
Now we introduce a new definition that plays an important role in describing the
motions of planets. Consider the area A spanned by the radius OP .t/, where P
belongs to the trajectory  , in the time interval .t; t C t/ (Fig. 11.3). Then the limit
11.4 Velocity and Acceleration in Plane Motions 167

P t+ t

A
r (t+ t )
Pt

r(t)
O t x

Fig. 11.3 Areal velocity

A
AP D lim
t !0 t

is the areal velocity of P . To obtain the explicit expression of AP in terms of (11.12),

we denote by rmin and rmax the minimum and the maximum of r.t/ in the interval
.t; t C t/, respectively. Since

1 2 1 2
r '  A  rmax ';
2 min 2
in the limit t ! 0, we obtain

1
AP D r 2 ':
P (11.20)
2
In some cases, the Cartesian form of the areal velocity is more convenient. It can be
obtained by considering the coordinate transformation from Cartesian coordinates
.x; y/ to polar ones
p
r D x2 C y 2 ;
y
' D arctan :
x
It is easy to verify that, in view of these formulae, (11.20) becomes

1
AP D .x yP  xy/:
P (11.21)
2
168 11 Kinematics of a Point Particle

11.5 Circular Motion and Harmonic Motion

Let P be a point moving on a circumference  with center O and radius r. Further,

let ' be the angle formed by the position vector r.t/ of P with the Ox-axis of an
arbitrary Cartesian system of coordinate .O; x; y/ (Fig. 11.4). Since the curvilinear
abscissa s is related to the angle ' and the radius r by the relation

s D r'.t/;
we have
sP D r '.t/;
P sR D r '.t/:
R
Introducing these results into (11.6) and (11.10), we obtain the expressions of
velocity and acceleration in a circular motion

rP D r 't;
P (11.22)
rR D r 't
R C r 'P 2 n; (11.23)

where t is the unit vector tangent to  and n is the unit vector orthogonal to  and
directed toward O. In particular, if P moves uniformly on  , then (11.23) gives

rR D ! 2 r; (11.24)

where we have introduced the notation

! D j'j:
P (11.25)

rt s

x
O
X

Fig. 11.4 Circular motion

11.5 Circular Motion and Harmonic Motion 169

Let P be a material point that uniformly moves counterclockwise ('P > 0) along
a circumference  with its center at O and radius r. We define harmonic motion
with its center at O along the straight line Ox the motion that is obtained projecting
the point P along any diameter Ox of  . In the Cartesian coordinates of Fig. 11.4,
we have that the motion of X is represented by the following equation:

x.t/ D r cos '.t/ D r cos.!t C '0 /: (11.26)

In (11.26), r is called the amplitude of the harmonic motion and ! its frequency.
Differentiating (11.26) twice, we obtain

P
x.t/ D r! sin.!t C '0 /; R
x.t/ D r! 2 cos.!t C '0 /;

so that in a harmonic motion the following relation between position and accelera-
tion holds:
R
x.t/ D ! 2 x.t/: (11.27)
It is simple to recognize that the homogeneous second-order linear differential
equation with constant coefficients (11.27) characterizes the harmonic motions, i.e.,
its solutions are only harmonic motions. In fact,  D i ! are the roots of the
characteristic equation of (11.27)

2 C ! 2 D 0;

and the general solution of (11.27) is

x.t/ D C1 ei !t C C2 ei !t ;

where C1 and C2 are arbitrary constants. Taking into account Euler’s formulae, the
preceding solution can also be written as

x.t/ D A cos !t C B sin !t;

where A and B are two arbitrary constants. Finally, we obtain (11.26) when we
write A and B in the form

A D r cos '0 ; B D r sin '0 ;

where r and '0 are still arbitrary constants.

170 11 Kinematics of a Point Particle

11.6 Compound Motions

Let P be a moving point with respect to a frame of reference .O; i; j; k/, where
i, j, and k are unit vectors along the .Ox; Oy; Oz/ axes of a Cartesian system of
coordinates. If the position vector r can be written in the form
X
n
r.t/ D ri .t/; (11.28)
i D1

then we say that the motion of P is a compound of the n motions ri .t/.

In this section, we analyze some interesting compound motions. First, we
consider the motions
r1 .t/ D v0 ti;
1
r1 .t/ D  gt 2 C z0 k;
2
where v0 , g, and z0 are arbitrary positive constants. Eliminating t from the preceding
equations, we at once see that the trajectory of the compound motion is a parabola
of the Oxz plane with the concavity downward.
Now we consider the motion obtained by compounding the following motions:

r1 .t/ D R cos.!t C '0 /i C R sin.!t C '0 /j;

r2 .t/ D zP0 tk:

The first equation represents a uniform motion along a circumference of radius R

and center O lying in the coordinate plane Oxy. The second equation is relative to
a uniform motion along the Oz-axis. The trajectory  of the compound motion is a
helix lying on a cylinder of radius R and axis Oz (Fig. 11.5). The helix is regular
since the unit tangent vector t D .!R sin.!t C '/; !R cos.!t C '/; zP0 / to  ,
where
q
 D ! 2 R2 C zP20 ;
forms a constant angle  with the directrices of the cylinder. In fact, it is
zP0
cos  D :


Finally, we consider the motion of a point P obtained by compounding two

harmonic motions along the Ox- and Oy-axes of a Cartesian system of coordinates:

r1 .t/ D x.t/i D r1 cos.!1 t C '1 /i;

r2 .t/ D y.t/j D r2 cos.!2 t C '2 /j: (11.29)

The compound motion of P may be very complex, as is proved by the following

theorem.
11.6 Compound Motions 171

y
O

Fig. 11.5 Helicoidal motion

Theorem 11.1. Let ABCD be the rectangle Œr1 ; r1   Œr2 ; r2  with vertices
A.r1 ; r2 /, B.r1 ; r2 /, C.r1 ; r2 /, and D.r1 ; r2 /. The compound motion of the
harmonic motions (11.29) has the following properties:
• If '1 D '2 , or '1 D '2   and !1 D !2 , then it is harmonic along one of the
diagonals of the rectangle ABCD;
• For all other values of '1 and '2 and !1 D !2 , the trajectory  is an ellipsis
tangent to the sides of the rectangle ABCD every time  touches one of them;
further, P moves along  with constant areal velocity, and the motion is periodic;
• If !1 ¤ !2 and the ratio !1 =!2 is a rational number, the motion is periodic.
Further, the trajectory may be either an open curve or a closed curve;
• If !1 ¤ !2 and the ratio !1 =!2 is an irrational number, then the motion is
aperiodic and the trajectory is everywhere dense on the rectangle ABCD (i.e.,
the trajectory intersects any neighborhood of any point in the rectangle).
Proof. For the sake of simplicity, we limit ourselves to proving only the first three
items. Introducing the notation ı D '2  '1 and supposing !1 D !2 D !, from
(11.29) we obtain
x
D cos.!t C '2  ı/ D cos.!t C '2 / cos ı C sin.!t C '2 / sin ı
r1
y
D cos ı C sin.!t C '2 / sin ı;
r2
172 11 Kinematics of a Point Particle

so that
x y
 cos ı D sin.!t C '2 / sin ı: (11.30)
r1 r2
On the other hand, in view of (11.29)2, we also have that
y
sin ı D cos.!t C '2 / sin ı: (11.31)
r2

Consequently, the trajectory of the compound motion is

x2 y2 2
2
C 2
 x y cos ı D sin2 ı: (11.32)
r1 r2 r1 r2

If '1 D '2 , then ı D 0, whereas if '1 D '2  , then ı D . In all these cases,
(11.32) gives
 
x y 2
 D 0; (11.33)
r1 r2
where the  sign corresponds to ı D 0, the C sign corresponds to ı D , and
the trajectory coincides with one of the diagonals of the rectangle ABCD. To prove
that the motion along the diagonal y D r2 x=r1 , for instance, is harmonic, we note
that the distance s of P from O along this diagonal is given by
p q
sD x 2 C y 2 D r12 C r22 cos.!t C '1 /:

When ı is different from 0 and , the trajectory (11.32) is an ellipsis. In this case,
the areal velocity A [see (11.21)]

1 1
AD P  x y/
.xy P D  !r1 r2 sin ı
2 2
is constant and the motion is periodic. Moreover, the ellipsis touches the sides
x D r1 of the rectangle ABCD when cos.!t C '1 / D 1, that is, when
!t C '1 D n, where n is an arbitrary relative integer. It is plain to verify that
the corresponding value of xP vanishes and the ellipsis is tangent to the rectangle any
time it touches the sides x D r1 .
We now determine the conditions under which the motion is periodic when
!1 ¤ !2 . The compound motion is periodic if and only if for any instant t there
exists a value T of time such that

cos.!1 .t C T / C '1 / D cos.!1 t C '1 /;

cos.!2 .t C T / C '2 / D cos.!2 t C '2 /:
11.6 Compound Motions 173

y
1.0

0.5

x
2 1 1 2

0.5

1.0

Fig. 11.6 r1 D 2, r2 D 1, !1 D !2 D 1, '1 D 0, '2 D =3

y
1.0

0.5

x
2 1 1 2

0.5

1.0

Fig. 11.7 r1 D 2, r2 D 1, !2 D 2, !2 D 1, '1 D 1, '2 D =2

The preceding conditions can equivalently be written as

!1 T D 2n;
!2 T D 2m;

where n and m are arbitrary relative integers. In conclusion, the motion is periodic
if and only if the ratio
!1 n
D (11.34)
!2 m
is a rational number. 
Figures 11.6–11.10 show different cases of compounding two harmonic motions
along orthogonal axes. They were obtained by using the notebook Composition.nb,
which allows one to draw the trajectory obtained by compounding two or three
motions.
174 11 Kinematics of a Point Particle

y
1.0

0.5

x
2 1 1 2

0.5

1.0

Fig. 11.8 r1 D 2, r2 D 1, !2 D 3, !2 D 1, '1 D 1, '2 D =2

y
1.0

0.5

x
2 1 1 2

0.5

1.0

Fig. 11.9 r1 D 2, r2 D 1, !2 D 3, !2 D 1, '1 D 1, '2 D =3

y
1.0

0.5

x
2 1 1 2

0.5

1.0
p
Fig. 11.10 r1 D 2, r2 D 1, !2 D 2, !2 D 1, '1 D 0, '2 D =3

11.7 Exercise

1. Let Oxy be a plane in which we adopt polar coordinates .r; '/, and let r D A'
be the trajectory of a point particle P , where A is a constant. Assuming that
'P D 'P0 is constant, determine the radial and transverse velocity and the radial
and transverse acceleration of P . Verify that the areal velocity is constant.
11.7 Exercise 175

2. In the same plane of the preceding exercise, a point particle P describes the
trajectory r D r0 e' , where r0 and 'P are constant. Determine the radial and
transverse velocity and acceleration.
3. In the plane Oxy, the equations of motion of a point particle P are
x D xP 0 t; y D y0  ˇ.1  e˛t /;

where xP 0 , y0 , ˛, and ˇ are positive constant. Determine the velocity and

acceleration of P , its trajectory  , and the unit vector tangent to  .
4. Let P be a point particle moving on a parabola  , and let y D x 2 be the
equation of  in the frame Oxy. Determine the velocity, acceleration, and the
curvilinear abscissa of P .
Hint: The curvature 1=R of a curve y D f .x/ is 1=R D jf 00 .x/j1=.1 C
f 0 .x/2 /3=2 .
5. Let Oxyz be a Cartesian frame of reference in space. A point particle P moves
on the parabola z D x 2 , which uniformly rotates with angular velocity ! about
the Oz-axis. Determine the velocity and acceleration of P .
6. The acceleration and velocity of a point particle P moving on the Ox-axis are
related by the equation
xR D ax;
P a > 0:
P
Determine the equation of motion of P , and evaluate the limit of x.t/ and x.t/
when t ! 1.
7. The acceleration and velocity of a point particle P moving on the Ox-axis are
related by the equation
xR D ajxj
P x;
P a > 0:
Prove that limt !1 xP D 0.
P and take into account that xP xR D
Hint: Multiply both sides of the equation by x,
1 dxP 2
2 dt .
8. Determine the velocity, acceleration, and trajectory of a point particle P whose
equations of motion in the frame Oxy are

x.t/ D sin !t; y.t/ D et :

Further, determine the instants at which the velocity is parallel to Oy.

9. Let P be a point particle moving on the ellipse  whose equation in polar
coordinates .r; '/ is
p
rD ;
1 C e cos '
where p is a constant and e the eccentricity. If P moves on  with constant
areal velocity, prove that the velocity square has a maximum at the perihelion
and a minimum at the aphelion.
176 11 Kinematics of a Point Particle

Hint: Evaluate rP taking into account that r 2 'P D c, where c is a constant. Then
recall that the radial velocity is rP and the transverse velocity is r '.
P
10. A particle P moves along a curve y D a sin bx, where a and b are constant.
Determine the acceleration of P assuming that the component of its velocity
along the Ox-axis is constant.
Chapter 12
Kinematics of Rigid Bodies

12.1 Change of the Frame of Reference

An important task of kinematics consists of comparing the measures of lengths and

time intervals carried out by two observers R and R0 moving each with respect to
the other with an arbitrary rigid motion. The solution of this problem allows us to
compare the measures of velocities, accelerations, etc., carried out by the observers’
comoving with R and R0 .
Let E D .P 0 ; t 0 / be an event happening at the point P 0 at instant t 0 for an
observer at rest in the frame of reference R. Let E D .P; t/ be the same event
evaluated from an observer at rest in R. The aforementioned problem of kinematics
can mathematically be formulated as follows:
Determine the functions

P D f .P 0 ; t 0 /; (12.1)
t D g.P 0 ; t 0 /; (12.2)

relating the space-time measures that R and R0 associate with the same event.
When we recall the postulate stating that the behavior of a clock is not influenced
by the transport, we can equip R and R0 with synchronized clocks and be sure that
they will remain synchronous. Therefore, (12.2) becomes t D t 0 . More generally,
even if R and R0 use clocks with the same behavior, they could choose a different
time unit and a different origin of time. In this case, instead of t D t 0 , we have

t D at 0 C b; (12.3)

where a and b are constants. We remark that (12.3) still implies that two simultane-
ous events for R are also simultaneous for R0 .
Let EA D .A0 ; t 0 / and EB D .B 0 ; t 0 / be two events happening in two different
points at the same instant for the observer R0 . The same events are also simultaneous

A. Romano, Classical Mechanics with Mathematica R

, Modeling and Simulation 177
in Science, Engineering and Technology, DOI 10.1007/978-0-8176-8352-8 12,
© Springer Science+Business Media New York 2012
178 12 Kinematics of Rigid Bodies

for R but happening at points A and B. To determine the function (12.1), classical
kinematics resorts to another axiom.
Axiom 12.1. At any instant, transformation (12.1) preserves the spatial distance d
between simultaneous events, that is,

d.A; B/ D d.f .A0 ; t 0 /; f .B 0 ; t 0 // D d.A0 ; B 0 /: (12.4)

In other words, Axiom 12.1 states that, at any instant, (12.1) is an isometry
between the three-dimensional Euclidean spaces E3 and E30 of observers R and R0 ,
respectively. Let us introduce Cartesian axes .O 0 ; .xi0 // in the frame R0 and
Cartesian axes .O; .xi // in R. If .xi0 / and .xi / are the coordinates that R0 and R,
respectively, associate with the points P 0 and P of the same event E D .P 0 ; t 0 / D
.P; t/, then from linear algebra we know that transformation (12.1) is expressed by
the formulae
xi D xi C Qij .t/xj0 ; (12.5)
where  is the place in which occurs the arbitrary event E D .; t/, evaluated by
observer R, and .Qij .t// is an arbitrary time-dependent orthogonal matrix, that is,
a matrix satisfying the conditions

Qij .QT /j h D ıih : (12.6)

Remark 12.1. When  D O, (12.5) reduces to the relation

xi D Qij .t/xj0 ;

which gives the transformation formulae of the components of the position vector
r0 in passing from the frame .O; .e0i // to the frame .O; .ei //. Consequently, the two
bases .ei / and .e0i / are related by the equations

e0i D .Q1 /ij ej D .QT /ij ej D Qj i ej ; (12.7)

so that the coefficients .QT /j i are the components of the vectors e0i with respect to
the base ei .
Remark 12.2. In many textbooks, the transformation formulae (12.5) are proved by
noting that from Fig. 12.1 we have that

rP D r C r0 : (12.8)

Therefore, referring this relation to the bases .ei / and .e0i /, we obtain

xi ei D xi ei C xj0 e0j :

12.2 Velocity Field of a Rigid Motion 179

x3

x’3 S
x’2
e’3 P
r’
e’2
Ω
rP
e3 e’1
rΩ

x’1 x2
O
e2

e1

x1

Fig. 12.1 Coordinates in a rigid motion

Now, representing the base .e0i / with respect to the base .ei / by the relations e0j D
Qj i ei , where .Qij / is an orthogonal matrix, we again obtain (12.5). However, this
approach hides a logical mistake. In fact, in both of the preceding relations, the
vector r0 is intended to be a vector belonging to both vector spaces relative to the two
observers R and R0 , that is, it is intended as the same vector. But this assumption is
just equivalent to the condition that the correspondence between the spaces relative
to the two observers is an isometry. In other words, we can accept (12.8) only after
proving (12.5).

12.2 Velocity Field of a Rigid Motion

In this section we evaluate the velocity field and the acceleration field of any rigid
body S moving relative to a rigid frame of reference R.
Let .O; .ei // be a Cartesian frame in R, and denote by .xi / the coordinates of any
point of S relative to .O; .ei //. Introduce a Cartesian frame .; .e0i // in the body
S (Fig. 12.1) with the origin at an arbitrary point  2 S . Owing to the rigidity of
S , any point P of S has coordinates .xi0 / relative to .; .e0i // that do not depend
on time. For the sake of brevity, we call R the space frame or the lab frame and
.; .e0i // the body frame.
180 12 Kinematics of Rigid Bodies

The relation between the coordinates .xi0 / and .xi / is expressed by (12.5). Since
we are now interested in the velocity field of S with respect to R and S is a rigid
body, we differentiate (12.5) with respect to time:

xP i D xP i C QP ij xj0 : (12.9)

On the other hand, in view of (12.5) and the orthogonality condition (12.6), we have
that
xj0 D .QT /j h .xh  xh /: (12.10)
Inserting the preceding expression into (12.9) yields the relation

xP i D xP i C Wih .xh  xh /; (12.11)

where
Wih D QP ij .QT /j h : (12.12)
In (12.11), xP i , xP i , and xh  xh are components of vectors; therefore, from the
tensorial criteria of Chap. 2, we can state that Wih are the components of a Euclidean
2-tensor W. This tensor is skew-symmetric since (12.6) implies the equality

QP ij .QT /j h D Qij .QP T /j h ; (12.13)

and consequently

Wih D Qij .QP T /j h D Qij QP hj D QP hj .QT /j i D Whi : (12.14)

Introducing the pseudovector (Chap. 5) ! such that

Wih D ij h !j ; (12.15)

where ij h is the Levi–Civita symbol, (12.9) becomes

xP i D xP i C ij h !j .xh  xh /: (12.16)

With the notations of Fig. 12.1, we can put (12.16) in the following vector form:
! !
rP P D rP  C !  P ; P D rP  r : (12.17)

Differentiating the preceding formula with respect to time and noting that

d ! d !
P D .rP  r / D rP P  rP  D !  P ;
dt dt
12.2 Velocity Field of a Rigid Motion 181

we derive the acceleration field of a rigid motion

! !
rR P D rR  C !
P  P C !  .!  P /: (12.18)

Exercise 12.1. Verify that (12.12) implies that

1
!i D  ij h Wj h : (12.19)
2
Remark 12.3. Equation (12.17) shows that the velocity field of a rigid motion is
completely determined by giving two vector functions of time, that is, the velocity
rP  .t/ of an arbitrary point  of the rigid body and the vector function !.t/.
We now prove the important Mozzi’s theorem, which supplies both an interesting
interpretation of (12.17) and the physical meaning of the vector !.t/.
Theorem 12.1. At any instant t at which !.t/ ¤ 0, the locus of points A such that
rP .t/ D .t/!.t/, where .t/ is a real function, is a straight line A.t/, parallel to
!.t/, which is called Mozzi’s axis. Moreover, all points belonging to A.t/ move
with the same velocity .t/.
Proof. The points A of S having the property stated by the theorem are character-
ized by the condition
rP A  ! D 0;
which, in view of (12.17), becomes
!
.Pr C !  P /  ! D 0:

When we take into account the vector identity .a  b/  c D .a  c/b  .b  c/a, the
preceding equation can also be written as
! !
rP   ! D .A  !/!  ! 2 A: (12.20)
!
In conclusion, the locus A.t/ coincides with the set of solutions A of (12.20).
Introducing the unit vector k D !=j!j, we have that
! ! !
A D .A  k/k C A? ;
!
where A?  k D 0, and (12.20) becomes
! ! !
rP   ! D .A  k/! 2 k  ! 2 Œ.A  k/k C A? :
182 12 Kinematics of Rigid Bodies

In the hypothesis ! ¤ 0, the preceding equation supplies

! rP   !
A? D  ;
!2
!
so that the requested vectors A are given by the formula
! rP   !
A D  C !; (12.21)
!2
where  is an arbitrary real number. The preceding formula proves that points A
belong to a straight line A.t/, parallel to !. Finally, we write (2.17) for any pair
!
P;  2 A.t/. Since in this case P is parallel to !, we obtain

rP P D rP  ;

and the theorem is proved. 

The existence of Mozzi’s axis allows us to write (12.17) in the following form:
!
rP P D  C !  AP ;  k !; A 2 A.t/: (12.22)

12.3 Translational and Rotational Motions

Definition 12.1. A rigid motion is said to be translational relative to the frame of

reference R if, at any instant t, the velocity field is uniform:

rP P D .t/: (12.23)

Theorem 12.2. The following statements are equivalent.

• A rigid motion is translational.
• !.t/ D 0, at any instant.
• The orthogonal matrix .Qij / in (12.5) does not depend on time.
Proof. The equivalence between the first two conditions is evident in view of
(12.22). The equivalence between the first and third conditions follows at once from
(12.12) and (12.15). 
Definition 12.2. A rigid motion is a rotational motion about axis a if during the
motion the points belonging to a are fixed. a is called the rotation axis.
To find the particular expression of the velocity field in a rotational motion about
the fixed axis a, we start by choosing the frames of coordinates as in Fig. 12.2.
Recalling the meaning of the orthogonal matrix .Qij / [see (12.7)] and looking at
Fig. 12.2, we at once see that
12.3 Translational and Rotational Motions 183

Fig. 12.2 Lab and body frames in a rotational motion

0 1
cos '  sin ' 0
.Qij / D @ sin ' cos ' 0 A :
0 0 1

Taking into account (12.7), (12.12), and (12.19), by simple calculations we obtain
the following results:
0 1
0 1 0
Wij D 'P @ 1 0 0 A ;
0 0 0

1
!i D  ij h Wj h D .0; 0; '/:
P
2
If the arbitrary point  in (12.17) is chosen on the fixed axis a and the preceding
expression of ! is taken into account, then we obtain the velocity field of a rotational
rigid motion with a fixed axis a:
!
rP P D 'e
P 3  P : (12.24)
184 12 Kinematics of Rigid Bodies

In this formula,  is an arbitrary point of a; in particular, identifying  with the

projection C of P on the axis a, we obtain
!
rP P D 'e
P 3  CP : (12.25)

Theorem 12.3. Let S be a rigid body rotating about the fixed axis a, and denote
by C the orthogonal projection of any point P 2 S on axis a and by P the plane
orthogonal to a and containing C . Then the following properties hold.

• Mozzi’s axis coincides at any instant with a.

• Any point P moves along a circumference  contained in the plane P with its
P
center at C . Further, P moves along  with velocity 'jCP j.

Proof. Mozzi’s axis A.t/ is a set of points whose velocities vanish or are parallel
to !. Since a is fixed, its points must belong to A.t/. Consequently, A.t/ coincides
with the fixed axis a. The second statement follows at once from both (12.25) and
the properties of the cross product. 
The vector !.t/ vanishes in a translational rigid motion, whereas in a rotational
motion about a fixed axis a, it is directed along a and its component along a gives
the angular velocity of all points of the rigid body.
Definition 12.3. A rigid motion is helicoidal if it is obtained by compounding a
rotational motion about axis a and a translational motion along a.
In view of (12.23) and (12.25), the velocity field in such a motion yields

!
rP P D  C 'k
P  OP ; (12.26)

where  is the sliding speed of a, k a unit vector along a, and 'P the angular velocity
about a.
Comparing the velocity field of a helicoidal motion (12.26) with (12.22), we
obtain the following equivalent form of Mozzi’s theorem.
Theorem 12.4. At any instant t, the velocity field of a rigid motion M coincides
with the velocity field of a helicoidal motion, called a helicoidal motion tangent to
M at instant t.
At this point, it is quite natural to call ! the instantaneous angular velocity of the
rigid motion.
12.4 Spherical Rigid Motions 185

12.4 Spherical Rigid Motions

Definition 12.4. A rigid motion with a fixed point O is said to be a spherical

motion.
Choosing   O in (12.5), we obtain

xi D Qij .t/xj0 ; (12.27)

so that a spherical motion is determined by giving the matrix function .Qij .t//.
Since this matrix is orthogonal, that is, it confirms the six conditions (12.6), we can
say that a spherical motion is determined assigning three independent coefficients
of the orthogonal matrix .Qij / as functions of time.
Further, the velocity field of a spherical motion is given by the equation
!
rP P D !.t/  OP ; (12.28)

in which both the value and direction of the angular velocity ! depend on time.
Instead of giving three independent coefficients of the orthogonal matrix .Qij /,
we can determine the orientation of the Cartesian axes .O; .xi0 // with respect to the
Cartesian axes .O; .xi // using three other independent parameters, called Euler’s
angles (Fig. 12.3). These angles can be defined when the planes Ox1 x2 and Ox10 x20
do not coincide. In fact, in this case, they intersect each other along a straight line,
called a line of nodes or nodal line, and we can define the angles , ', and  shown
in Fig. 12.3. The angle between the axis Ox1 and the line of nodes is called the
angle of precession, the angle ' between the line of nodes and the axis Ox10 is the

x3

x’3 x’2

e3
e’2
e’3
N
O x2
e’1
e1 n

x’1

line of nodes
x1

Fig. 12.3 Euler’s angles

186 12 Kinematics of Rigid Bodies

angle of proper rotation, and the angle between the axes Ox3 and Ox30 is called the
angle of nutation. To recognize the independence of these angles, it is sufficient to
verify that they define three independent rotations transforming the axes .Oxi / into
the axes .Ox 0i /. It is easy to understand that the following three rotations have this
property: a rotation through the angle about the axis Ox3 , a rotation through the
angle ' about the axis Ox30 , and a rotation through the angle  about the line of
nodes.
We now determine the fundamental Euler kinematical relations relating the an-
gular velocity with the Euler angles. First, we start with the following representation
of ! in the base .e0i /:
! D pe01 C qe02 C re03 : (12.29)
Owing to the independence of the Euler angles, we can write

! D P e3 C 'e P
P 03 C n: (12.30)

Denoting by N the unit vector in the plane .O; e01 ; e02 / and orthogonal to n and
resorting to Fig. 12.3, we can easily prove the following formulae:

N D sin 'e01 C cos 'e03 ;

e3 D cos e03 C sin N:

Combining these relations and recalling (12.30), we obtain

e3 D sin  sin 'e01 C sin  cos 'e02 C cos e03 ;

P  e0 D P sin  sin ' C P cos ':
p D !  e01 D P e3  e01 C n (12.31)
1

Applying a similar reasoning for evaluating q and r, we obtain the following

transformation formulae of the components of ! in going from the base .e3 ; e03 ; n/
to the base .e0i /:

p D P sin  sin ' C P cos ';

q D P sin  cos '  P sin ';
r D P cos  C ':
P

With tedious calculations, provided that  ¤ 0, it is possible to prove the following

inverse transformation formulae:

P D 1
.p sin ' C q cos '/; (12.32)
sin 
'P D .p sin ' C q cos '/ cot  C r; (12.33)
P D p cos '  q sin ': (12.34)
12.4 Spherical Rigid Motions 187

We conclude this section by evaluating the transformation formulae relating the

components .p; q; r/ of ! relative to the base .e0i / and the components .!1 ; !2 ; !3 /
relative to the base .ei /. Owing to (12.7), applying the matrix .QijT / to .ei /, we
obtain the base .e0i /. As we have already noted, we can bring the base .ei / to
coincide with the base .e0i / by applying the following three rotations: the rotation
through about Ox3 , the rotation through  about the line of nodes, and, finally,
the rotation through ' about D x30 . These rotations are, respectively, represented by
the following matrices:

0 1
cos sin 0
A D @  sin cos 0 A ;
0 0 1
0 1
1 0 0
B D @ 0 cos  sin 0 A ;
0  sin  cos 
0 1
cos ' sin ' 0
C D @  sin ' cos ' 0 A ;
0 0 1

so that

QijT D Cih Bhk Akj :

On the other hand, the components of a vector are transformed according to the
inverse matrix of .QijT /, that is, according to .Qij /, since .QijT / is orthogonal.
Consequently, the matrix giving the transformation .p; q; r/ ! .!i / is

Qij D ATih Bhk

T T
Ckj ;

and by simple calculations it is possible to show that

!1 D .cos ' cos  sin ' sin cos /p

.sin ' cos C cos ' sin cos /q C sin sin  r
!2 D .cos ' sin C sin ' cos cos /p
C. sin ' sin C cos ' cos cos /q  cos sin  r
!3 D sin ' sin  p C cos ' sin  q C cos  r: (12.35)
188 12 Kinematics of Rigid Bodies

12.5 Rigid Planar Motions

Definition 12.5. A rigid motion is a planar motion if the velocity field is parallel
to a fixed plane .
It is trivial that the rigidity of the planar motion implies that all points that belong
to the same straight line orthogonal to  have the same velocity. Consequently,
to characterize the velocity field of a planar motion, it is sufficient to analyze the
velocity field on an arbitrary plane parallel to . We must determine the restrictions
on the vectors  and ! of (12.22) due to the hypothesis of planar motion. The answer
to this question is given by the following theorem.
Theorem 12.5. In a planar motion, we have that
1. If .t/ ¤ 0, then necessarily !.t/ D 0 and .t/ is parallel to .
2. If !.t/ ¤ 0, then necessarily .t/ D 0 and !.t/ is orthogonal to .
Proof. At instant t, the vector  in (12.26) gives the common velocity of all points
belonging to Mozzi’s axis A.t/. Consequently, A.t/ must be parallel to . But also
!
!.t/  AP must be parallel to  for any choice of P , and that is possible if and
!
only if !.t/ D 0. Inversely, if !.t/ ¤ 0, then the term !.t/  AP is parallel to 
for any P if and only if !.t/ is orthogonal to . Further, .t/ must also be parallel
to both  and ! so that it vanishes. 
The preceding theorem states that, at any instant t, the velocity field of a planar
motion coincides either with that of a translational motion parallel to  or with the
velocity field of a rotational motion about an axis orthogonal to . In this last case,
since the points of this axis are at rest, it coincides with Mozzi’s axis A.t/. The
intersection point A of A.t/ with the plane  is called the instantaneous rotation
center. It is evident that, for any point P , A belongs to a straight line orthogonal to
the velocity rP P at point P .

12.6 Two Examples

In this section, we apply the results of the previous sections to two examples. Let
S be a disk and let a be an axis orthogonal to S containing the center C of S . We
suppose that S rotates with angular velocity ! D P k about a, where k is the unit
vector along a. Further, we suppose that C moves with constant angular velocity
'P on a circumference  having its center at O and radius R (Fig. 12.4). We take
the position vectors starting from O and denote by rA the unknown position vector
of the instantaneous rotation center A. Since we are considering a nontranslational
planar motion, Mozzi’s axis A.t/ is orthogonal to the plane  containing S and,
12.6 Two Examples 189

.
rC

C
R
A
rC

x
O

Fig. 12.4 Example of planar motion

consequently, is determined by the single point A at which it intersects . With our

notations, relation (12.22) reduces to the equation

rP C  P k
rA D rC  :
P2

Introducing the unit vector n orthogonal to  and directed toward O, the

preceding equation can also be written as

j'j
P
rA D rC ˙ R n;
P
j j

where we must choose the C sign when 'P and P have the same sign and the  sign
in the opposite case. In conclusion, during motion, the instantaneous rotation center
A describes a circumference with center at O and radius r less or greater than R,
depending on whether or not 'P and P have the same sign.
As a second example, we consider a disk S rotating with constant angular
velocity P about an axis a contained in the plane of S (Fig. 12.5). We also suppose
that the center C of S moves along a straight line Ox orthogonal to a with a velocity
rP C . To determine the evolution of Mozzi’s axis A.t/, we denote by k a unit vector
along a and by A the intersection point of A.t/ with the plane Oxy. Then from
(12.22) we obtain
rP C  P k jPrC j
rA D rC  D rC  j;
P2 j Pj
where j D k  i and i is the unite vector directed along Ox.
190 12 Kinematics of Rigid Bodies

k
z

y a
k
j
.
O rC x
i C

Fig. 12.5 Example of rigid motion

12.7 Relative Motions

Let R and R0 be two observers moving relative to each other. Let the relative
motion be rigid. If point P is moving relative to both R and R0 , what is the relation
between the velocities and the accelerations of P measured by both observers? In
this section, we answer this question.
We denote by .O; ei / and .; e0i / two systems of Cartesian coordinates fixed in R
and R0 , respectively. Further, we call the motion of P relative to R0 relative motion
and the motion of P relative to R absolute motion.
Let xi0 .t/ be the parametric equation of the moving point P in the frame .; xi0 /.
To know the coordinates of P relative to .O; xi /, we introduce the functions xi0 .t/
into (12.5):
xi .t/ D xi .t/ C Qij .t/xj0 .t/: (12.36)

Differentiating this relation with respect to time, we obtain

xP i .t/ D xP i .t/ C QP ij .t/xj0 .t/ C Qij .t/xP j0 .t/: (12.37)

Now, we must recognize the physical meaning of the different terms appearing in
(12.37). First, on the left-hand side are the components relative to the base .ei / of
the velocity rP P of P evaluated by observer R, that is, of the absolute velocity. The
first two terms on the right-hand side, which coincide with (12.9), are obtained by
differentiating (12.36) with respect to time and holding constant the components
xj0 .t/. In other words, these terms give the velocity v of the point at rest relative
to .; e0i / that, at instant t, is occupied by P . Since the motion of R0 relative to R
12.7 Relative Motions 191

is rigid, v is given by (12.17), i.e., by the first two terms on the right-hand side
of (12.37). Finally, the last term of (12.37) is the velocity rP 0P of P evaluated by
observer R0 . In conclusion, we can give (12.37) the following form:

rP P D v C rP 0P ; (12.38)

where
!
v D rP  C !  P : (12.39)
Equation (12.38) is the mathematical formulation of the classical principle of
velocity composition.
Let u be a vector varying in time with respect to both observers R and R0 . Due
to the relative motion of R0 relative to R, the dependence of u on time is expressed
by a function f.t/ in R and by a function g.t/ in R0 . We call

da u.t/
D Pf.t/
dt

the absolute derivative of u.t/ and

dr u.t/
D gP .t/
dt

the relative derivative of u.t/. Here, we show that (12.38) allows us to determine
the relation between these two derivatives. In fact, the following two representations
!
of the vector u.t/  AB hold in R and R0 , respectively (Fig. 12.6):

u.t/ D rB .t/  rA .t/; u.t/ D r0B .t/  r0A .t/; (12.40)

where rB .t/ and rA .t/ are the position vectors of points B and A relative to the
frame .O; ei /, and r0B .t/ and r0A .t/ are the position vectors of the same points in the
frame .; e0i /. It is evident that
da u.t/
D rP B .t/  rP A .t/;
dt
dr u.t/
D rP 0B .t/  rP 0A .t/:
dt
Using (12.38) and (12.39) in the preceding expressions, we obtain the fundamental
formula
da u.t/ dr u.t/
D C !  u.t/: (12.41)
dt dt
We are interested in determining the relation between the absolute acceleration
and the relative acceleration. We reach this result differentiating with respect to time
either (12.37) or (12.38). Here, we adopt the second approach, which is based on
(12.41).
192 12 Kinematics of Rigid Bodies

x3

x’2
x’3 B u
r’B
A

e’3 r’A

rB
e3 e’2
rA Ω x2
O
e2 e’1
e1
x’1

x1

Fig. 12.6 Absolute and relative derivatives

Differentiating (12.38) and taking into account (12.41), we obtain

da rP P da
rR P D D .Pr0P C v /
dt dt
0
dr rP P da v
D C !  rP 0P C
dt dt
da v
D rR 0P C !  rP 0P C : (12.42)
dt
On the other hand, from (12.39) we have that

da v !
D rR  C !
P  P C !  .PrP  rP  /
dt
!
P  P C !  .Pr0P C v P  rP  /
D rR  C !
! !
P  P C !  rP 0P C !  .!  P /:
D rR  C ! (12.43)

Collecting (12.42) and (12.43), we can write

rR P D rR 0P C a C ac ; (12.44)
12.7 Relative Motions 193

x2
x’2

.
rP
v x’1

e2
e’2 .
rP’
e’1 P
O x1
e1

Fig. 12.7 Example of relative motion

where
! !
a D rR  C !
P  P C !  .!  P /;
ac D 2!  rP 0P :

In view of (12.18), the term a gives the acceleration of that fixed point of .; e0i /,
which is occupied by P at instant t. The term ac is called Coriolis’ acceleration.
Example 12.1. Let P be a point moving on a straight line Ox1 that rotates
with constant angular velocity 'P0 about a fixed axis a orthogonal to Ox1 x2 and
containing O (Fig. 12.7). Verify that

rP 0 D xP 10 e01 ; v D x 0 'P0 e02 ;

rR D xR 10 e01 ; a D 'P02 x10 e01 ;

ac D 2'P 0 xP 10 e02 :
Later, we will use the result stated in the following theorem.
Theorem 12.6. Let P be a point uniformly moving relative to observer R. Then the
acceleration relative to another observer R0 vanishes if and only if the motion of R0
relative to R is translational and uniform.
Proof. If rR D 0 and the motion of R0 relative to R is translational and uniform, then
! D a D 0. Consequently, we also have that ac D 0 and rR 0 D 0. In contrast, if
rR D rR 0 D 0, then the condition

a C 2!  rP 0 D 0

must hold for any vector rP 0 , so that a D ! D 0, and the theorem is proved. 
194 12 Kinematics of Rigid Bodies

12.8 Exercises

1. Let !i be the components of the angular velocity in the lab frame .O; xi /.
Determine the evolution equations of the matrix Q D .Qij / relative to .O; xi /.
From (12.12) and (12.15) we obtain

P D WQ;
Q (12.45)

which in components is written as

QP ij D Wih Qhj D Qhj i kh !k : (12.46)

This is a system of nine ODEs in the nine unknowns .Qij .t//. However, we
can only accept solutions for which the matrix Q is orthogonal at any instant.
Now we prove that this property is confirmed by any solution of (12.46)
corresponding to initial data Q.0/ such that

Q.0/QT .0/ D I: (12.47)

In fact, (12.45) can equivalently be written as

P T D W D WT D QQ
QQ P T;

that is,
P T D 0 , d .QQT / D 0:
P T C QQ
QQ
dt
In other words, for any solution of (12.46), the matrix QQT is constant.
Therefore, if we assign the initial data satisfying (12.47), the corresponding
solution of (12.46) will be an orthogonal matrix.
2. Determine the time evolution of the matrix Q D .Qij / in the lab frame when
the angular velocity ! is given by the functions . P .t/; '.t/;
P P .t//.
When the initial values . 0 ; '0 ; 0 / of Euler’s angles are given, by integrat-
ing . P .t/; '.t/;
P P .t//, we obtain Euler’s angles . .t/; '.t/; .t//. Then, the
time evolution of the matrix Q D .Qij / follows from (12.32)–(12.35).
3. Let S be a rigid homogeneous bar whose end points A and B are constrained
to move, respectively, along the axes Ox and Oy of a Cartesian system of
coordinates. Prove that, during the motion of S , the instantaneous rotation
center C describes a circumference with its center at O and radius equal to the
length 2l of the bar (Fig. 12.8). Justify the result recalling that the instantaneous
rotation center belongs to the intersection of the straight lines orthogonal to the
velocities.
Hint: Point A has coordinates .2l sin '; 0/, where ' is the angle between S
and a downward oriented vertical straight line. Then, the velocity of A has
components .2l 'P cos '; 0/. In view of (12.21) and Theorem 12.4, we have that
12.8 Exercises 195

B
C

O l x
A

Fig. 12.8 A constrained bar

B
A

O G
B D x

Fig. 12.9 A rectangle with center moving along Ox

rP A  !
rC D rA  :
!2
Since ! D 'k, P where k is a unit vector orthogonal to the plane Oxy, we have
that rC D .2l sin '; 2l cos '/.
4. The center G of a rectangle ABCD (Fig. 12.9) moves along the Ox-axis of a
Cartesian system of coordinates. Further, ABCD rotates about an axis orthog-
onal to Oxy and containing G. Denoting by 2l the diagonal of the rectangle,
determine the velocity of point A and the trajectory of the instantaneous rotation
center.
!
Hint: rP A D xP G i C 'k
P  GA. Apply (12.21) to determine C .
5. A bar OC and a disk are constrained to move in the plane Oxy (Fig. 12.10).
The bar has point A moving on the Ox-axis and rotates about O. The disk
rotates about C . Determine the velocity of point P of the disk.
6. A uniform circular hoop of radius r rolls on the outer rim of a fixed wheel of
radius b; the hoop and the wheel are coplanar. If the angular velocity ! of the
hoop is constant, find
196 12 Kinematics of Rigid Bodies

y
P

O
x

Fig. 12.10 A bar and a disk moving in the plane Oxy

l l

O P
x

Fig. 12.11 System of two bars

• The velocity and acceleration of the center of the hoop;

• The acceleration of that point of the hoop which is at the greatest distance
from the center of the wheel.
7. A point A rotates about a fixed point with angular velocity !. A point B has
a uniform circular motion about A with angular velocity . What relation
connects  and ! if the acceleration of B is always directed toward O?
8. A wheel of radius r rolls without slipping along a straight line a. If the center of
the wheel has a uniform velocity v parallel to a, find at any instant the velocity
and the acceleration of the two points of the rim that are at the height h  r
above a.
9. Show that in a general motion of a rigid lamina in its plane, there is just one
point with zero acceleration.
10. Consider a system of two bars, as shown in Fig. 12.11. Determine the velocity
of point P in terms of the angular velocity 'P of the bar OA about O.
Chapter 13
Principles of Dynamics

13.1 Introduction

Dynamics has the aim of determining the motion starting from its causes. If we take
into account all the characteristics of real bodies, such as, for example, extension and
deformability, with the aim of reaching a more accurate description of the real world,
then we reach such a complex mathematical model that it is usually impossible
to extract concrete results to compare with experimental data. Consequently, it is
convenient to start with simplified models that necessarily neglect some aspects
of real bodies. These models will describe real behavior with sufficient accuracy,
provided that the neglected aspects have negligible effects on the motion.
In this chapter, we start with the Newtonian model in which the bodies are
modeled as material points or point masses. This model does not take into account
the extension and form of bodies, which are characterized only by their mass. It
is evident that such a model supplies a reasonable description of the real behavior
of bodies only when their extension can be neglected with respect to the extension
of the region in which the motion takes place. A famous example satisfying this
condition is supplied by the motion of the planets around the Sun. In fact, the
Newtonian model was created simply to describe the behavior of the planetary
system and, more generally, of celestial bodies such as comets, asteroids, and others.
The evolution of systems formed by material points is governed by the Newtonian
laws of dynamics, whose predictions have been confirmed by a large body of
experimental data.
In Chap. 15 we analyze the Eulerian model in which the extension of a body B
becomes a relevant characteristic of the model; the mass distribution of B plays a
crucial role but its deformations are neglected. In other words, the Eulerian model
analyzes the motion of extended rigid bodies. A model to describe a system of
material points or rigid bodies, free or subjected to constraints, is due to Lagrange
and will be analyzed in Chap. 17.

A. Romano, Classical Mechanics with Mathematica R

, Modeling and Simulation 197
in Science, Engineering and Technology, DOI 10.1007/978-0-8176-8352-8 13,
© Springer Science+Business Media New York 2012
198 13 Principles of Dynamics

13.2 Principle of Inertia and Inertial Frames

As was remarked in Chap. 11, the description of motion requires the introduction
of a frame of reference equipped with synchronized clocks. All frames of reference
having this characteristic are quite equivalent from a kinematical point of view. In
dynamics we are faced with a completely different situation since the choice of a
frame of reference deeply affects the connection between the motion and its causes.
Then it is fundamental to check the existence of a subset I of the whole class of
frames of reference adopted in kinematics, in which the description between motion
and causes admits the simplest description.
The criterion that helps us in determining set I is based on a fundamental
assumption: any action on a material point P is due to the presence of other bodies
B in the space around it; moreover, this action reduces to zero when the distances
between P and the bodies belonging to B go to infinity. This vague assumption on
the interaction among bodies makes reasonable the following definition: a material
point P is said to be isolated if it is so far from any other body that we can neglect
their action on P . It is natural to presume that in the class I the motion of an isolated
material point P is as simple as possible. In this respect, we have the following
principle.
Axiom 13.1. Principle of inertia (Newton’s First Law) – There is at least a frame
of reference I , called an inertial frame, in which any isolated material point is at
rest or is uniformly moving on a straight line.
An inertial frame I whose existence is stated by the preceding principle can be
localized using the following procedure. First, I is chosen in such a way that any
isolated material point moves on a straight line relative to I. On the straight line r
described by a particular isolated material point P we fix a point O and denote by x
the abscissa of P with respect to O. Then I is equipped with synchronized clocks
(Sect. 11.1) that associate with any position x of P the instant
x
tD ; (13.1)
v
where v is an arbitrary constant. It is evident that, with the time t, point P moves
uniformly on the straight line r with constant speed v. The variable t is defined up
to a linear transformation since we can change the origin of time by varying point O
and the unit of time by changing the value of v. The principle of inertia states that,
with this choice of time, the motion of any isolated material point is still rectilinear
and uniform. Such a time will be called a dynamical time.
In practice, we choose an arbitrary periodic device and then control whether
or not the time that it defines is dynamical. Usually, we adopt the Earth as a
fundamental clock, dividing its period of rotation into hours, primes, and seconds.
A more accurate measure of dynamical time is obtained by fractionizing the period
of revolution of the Earth about the Sun.
13.3 Dynamical Interactions and Force Laws 199

The existence of at least one inertial frame implies the existence of infinite
inertial frames, as is proved by the following theorem, which follows from
Theorem 12.5.
Theorem 13.1. Let I be an inertial frame equipped with the dynamical time t.
Then, a frame of reference I 0 is an inertial frame if and only if its motion relative
to I is translational and uniform and it is equipped with clocks measuring the time
t 0 D at C b, where a and b are arbitrary real constants.

13.3 Dynamical Interactions and Force Laws

The dynamics of a system of material points can be based on different sets of

axioms. These different axiomatic formulations, although they lead to the same
physical description of dynamical phenomena, differ from each other in the more or
less abstract principles on which they are based. Roughly, they can be partitioned
into two main classes according to whether the force is introduced as a primitive or
a derived concept. Here, we adopt an formulation of the first type for the following
two reasons. First, it is a faithful mathematical transcription of the fundamental
problem of dynamics that consists in finding the motion when its causes are given.
Second, many forces admit a static measure, whereas the axiomatic formulations
of the second class only lead to dynamical measures of forces. In any case, for
completeness, at the end of this section we present an axiomatic approach of the
second type proposed by Mach and Kirchhoff.
The following axiom describes the interaction between material points.
Axiom 13.2. Let S D fP1 ; P2 ; : : : ; Pn g be an isolated system of n material points.
In an inertial frame of reference I , the dynamical action on Pi 2 S , due to the
point Pj 2 S , is represented by a vector F.Pi ; Pj / that is applied at Pi and is
independent of the presence of the other points of S . Further, F.Pi ; Pj / depends on
the position vectors ri , rj and the speeds rP i , rP i relative to I of the material points
Pi and Pj , that is,
F.Pi ; Pj / D fij .ri ; rj ; rP i ; rP j /: (13.2)
The two vectors F.Pi ; Pj / and F.Pj ; Pi /, applied at Pi and Pj , respectively, are
directed along the straight line Pi Pj and confirm the condition

F.Pi ; Pj / D F.Pj ; Pi /: (13.3)

Finally, the action on Pi of the remaining material points Pie D S  Pi of S is

expressed by the following vector sum:
n
X
F.Pi ; Pie / D F.Pi ; Pj /; .j ¤ i /: (13.4)
j D1
200 13 Principles of Dynamics

The vector F.Pi ; Pj / is the force that Pj exerts on Pi , the function fij is called
a force law, condition (13.3) is the mathematical version of the action-reaction
principle (Newton’s third law), and, finally, (13.4) postulates the parallelogram
rule for forces.
The preceding axiom lays down the laws that hold for interactions in classical
mechanics. Before introducing the axiom that makes clear the connection between
these actions and the motion they produce, we must deal with another important
aspect of classical interactions. More precisely, we must define the behavior of
forces and force laws under a change of the frame of reference. It is important to
solve this problem for the following reasons. First, we can compare the dynamical
descriptions obtained by two inertial observers analyzing the same phenomenon.
Second, there are some interesting phenomena (for instance, dynamics relative to
the Earth) in which a noninertial frame is spontaneously adopted, so that we are
compelled to decide how the force law is modified by adopting this new frame of
reference.
To solve the preceding problem, we start by introducing some definitions.
Definition 13.1. A tensor is objective if it is invariant under a rigid frame change
R D .O; xi / ! R0 D .0 ; xi0 / given by

xi0 D xi .t/ C Qij .t/xj ; (13.5)

where .Qij .t// is an orthogonal matrix.

In particular, a scalar quantity is objective if it has the same value when it is
evaluated by R and R0 . A vector quantity v is objective if, denoting by .vi / and .v0i /
its components relative to R and R0 , respectively, the following results are obtained:

v0i D Qij vj : (13.6)

Similarly, a 2-tensor T is objective if

Tij0 D Qih Qj k Thk : (13.7)

Remark 13.1. To better understand the meaning of the preceding definition, we

note that the tensor quantity T is “a priori” relative to the frame of reference R in
which it is evaluated; more precisely, it belongs to the Euclidean space moving with
R. Consequently, in measuring the quantity T, the two observers R and R0 could
find two different tensors belonging to their own Euclidean spaces. For instance, a
measure of velocity of a material point moving relative to R and R0 leads to different
vectors: the absolute velocity and the relative velocity (Sect. 12.7). On the contrary,
the tensor T is objective if it is invariant under a change of frame or, equivalently, if
its components are modified according to (13.7). Bearing that in mind, we say that
T is objective if
T0 D T: (13.8)
13.3 Dynamical Interactions and Force Laws 201

Definition 13.2. Let

v D g.u1 ; : : : ; un / (13.9)
be a tensor quantity depending on the tensor variables .u1 ; : : : ; un /. The function g
is said to be objective if

v D g.u1 ; : : : ; un / D g.u01 ; : : : ; u0n / D v0 ; (13.10)

i.e., if it is invariant under a frame change.

Now we complete the description of classical interactions by adding the follow-
ing axiom.
Axiom 13.3. The force F.Pi ; Pj / acting between two material points Pi and Pj of
an isolated system S , as well as the force law (13.2), is objective under a change
from an inertial frame I to any rigid frame R0 .
The following theorem, which characterizes the force laws satisfying the preced-
ing axiom, is very important.
Theorem 13.2. The force law fij .ri ; rj ; rP i ; rP i / is objective if and only if

fij .ri ; rj ; rP i ; rP j / D 'ij .jrj; rP  k/k; (13.11)

where r
r D ri  rj ; rP D rP i  rP j ; kD ;
jrj
and 'ij is an objective scalar function.
Remark 13.2. The objectivity principle implies that the force fij .ri ; rj ; rP i ; rP i / acts
along the straight line joining points Pi and Pj , in accordance with the action and
reaction principle.
Proof. To simplify the notations, we omit the indices i and j and set ri D u,
rj D v. Further, we notice that the length of any vector r D u  v is invariant under
transformation (13.5). If (13.11) holds, then we have that
 
rP 0 r 0 rh0
fh0 .u0 ; v0 ; uP 0 ; vP 0 / D ' jr0 j; l 0 l
jr j jr0 j
!
.Qlk rPk C QP lk rk /Qlm rm Qhl rl
D ' jrj; :
jrj jrj

Since .Qij / is an orthogonal matrix and QP ij Qi l is skew-symmetric, we obtain

   
rP 0 r 0 rh0 rPm rm rl
' jr0 j; l 0 l D Q hl ' jrj; ;
jr j jr0 j jrj jrj
202 13 Principles of Dynamics

and the force law is objective. Conversely, if the force law is objective, i.e., if

fh .u0 ; v0 ; uP 0 ; vP 0 / D Qhk fk .u; v; u;

P vP /;

then we have that the condition

fh .Qkl .ul  xl /; Qkl .vl  xl /; Qkl .Pul  xP l /


C QP kl .ul  xl /; Qkl .Pvl  xP l / C QP kl .vl  xl /
D Qhm fm .uk ; vk ; uP k ; vP k / (13.12)

must hold for any possible choice of the quantities of Qkl , QP kl , xi , and xP i . Now
we choose
Qkl D ıkl ; QP kl D 0; xk D vk ; xP i D vP k :
This choice corresponds to taking a frame of reference R0 that, at the arbitrary
instant t, moves relative to I with a translational velocity equal to the velocity at
instant t of point Pj , which is at rest at the origin of R0 . For this choice, (13.12)
becomes
fh .rk ; 0; rPk ; 0/ D fh .uk ; vk ; uP k ; vP k /
so that f must depend on r D u  v and rP D uP  vP . Then, for any orthogonal matrix
.Qkl /, we have that

fh .Qkl rl ; Qkl rPl C QP kl rl / D Qhm fm .rk ; rPk /: (13.13)

Now we take Qkl D ıkl and suppose that QP kl is an arbitrary skew-symmetric

matrix. Then (Sect. 12.2) the equation QP kl rl D .!  r/k uniquely defines the vector
!. Moreover, decomposing rP to

rP D rP ? C .Pr  k/k;

where k D r=jrj and rP ?  k D 0, (13.13) can also be written as

fi .rk ; rPk / D fi .rk ; .Pr? /k C .!  r? /k C .Pr  k/k/:

Since this condition holds for any vector !, we have that

fi .rk ; rPk / D fi .rk ; .Pr  k/k/:

When we note that jrj and rP k are invariant under transformation (13.5), we conclude
that (13.13) assumes the final form

fi .rk ; .Pr  k/k/ D Qhi fh .Qkl rl ; .Prk  k/Qkl kl /;

13.4 Newton’s Second Law 203

so that f.r; .Pr  k/k/ depends isotropically on the variables r and .Pr  k/k. Finally,
(13.11) follows from known theorems of algebra about isotropic functions. 

13.4 Newton’s Second Law

The axioms of the preceding sections describe some fundamental aspects of the
classical interactions between material points. The following axiom describes the
connection existing between a force acting upon a material point and a correspond-
ing motion.
Axiom 13.4. Let S D P1 ; : : : ; Pn be an isolated system of n material points, and
.e/
let I be an inertial system of reference. Denote by Pi the set of all points of S
different from Pi . Then, to any material point Pi it is possible to associate a positive
number mi , the mass of Pi , which only depends on the material constitution of Pi .
Further, the acceleration rR i of Pi relative to I satisfies the equation [see (13.4)]

mi rR i D Fi .Pi ; Pie /: (13.14)

In particular, if we apply (13.14) to the pair Pi ; Pj 2 S and recall (13.3), we

obtain

mi rR i C mj rR j D 0: (13.15)

This result proves the following theorem.

Theorem 13.3. Let S D P1 ; : : : ; Pn be an isolated system of n material points,
and let I be an inertial system of reference. The accelerations rR i and rR j of the
points Pi ; Pj 2 S , produced by their mutual interaction, have the same direction
and opposite versus. Moreover, the ratio of their lengths is constant and equal to the
inverse ratio of their masses:

jRri j mj
D : (13.16)
jRrj j mi

This is precisely the result proved in the preceding theorem, which is postulated
in the formulation due to Mach and Kirchhoff. In this approach the force Fi .Pi ; Pj /,
which Pj exerts upon Pi , is defined as the product mi rR i . Postulating that the forces
satisfy the parallelogram rule, the action and reaction principle is derived. We
conclude by noting that (13.16) leads to an operative measure of mass by measures
of accelerations.
204 13 Principles of Dynamics

13.5 Mechanical Determinism and Galilean Relativity

Principle

In this section, we show that the Newtonian laws of dynamics allow one, at least in
principle, to determine the motion relative to an inertial frame I of material points
belonging to an isolated system S D fP1 ; : : : ; Pn g. In fact, applying (13.14) to any
material point Pi 2 S and taking into account (13.3), we obtain the system

mi rR i D fi .r1 ; : : : ; rn ; rP 1 ; : : : rP n /; i D 1; : : : ; n; (13.17)

where
n
X
fi .r1 ; : : : ; rn ; rP 1 ; : : : rP n / D fij .ri ; rj ; rP i ; rP j /
j ¤i;j D1

of n second-order ODEs in the n unknown functions r1 .t/; : : : ; rn .t/. Since the

system has a normal form, under suitable hypotheses about the functions fi ,
there exists one and only one solution of (13.17) satisfying the following initial
conditions (Cauchy’s problem):

ri .t0 / D r0i ; rP i .t0 / D rP 0i ; i D 1; : : : ; n: (13.18)

In other words, when the state of the system S is assigned by giving the position
and velocity of any point of S at the instant t0 , then the evolution of S is uniquely
determined for t  t0 . This circumstance, characteristic of classical physics, is
called mechanical determinism.
In our analysis it is proved that the explicit solution of the initial value prob-
lem (13.17) and (13.18) can only be found for some particular force laws. However,
we show throughout the book that we have at our disposal many mathematical
instruments to obtain meaningful information about the solution and its properties.
We are now in a position to state a fundamental principle of dynamics. Let I and
I 0 be two inertial frames, and let S D fP1 ; : : : ; Pn g be an isolated system moving
relative to them. Since the motion of I 0 relative to I is translational and uniform,
the acceleration of any point Pi 2 S is the same in both inertial frames [see (12.44)]

rR 0 D rR :

Further, the general axioms of dynamics state that the mass of Pi is invariant, and
the force Fi acting on it is objective, together with the force law. Consequently, in
the inertial frame I 0 , system (13.17) becomes

mi rR 0i D fi .r01 ; : : : ; r0n ; rP 01 ; : : : rP 0n /; i D 1; : : : ; n: (13.19)

This result proves the following theorem.

13.6 Balance Equations 205

Theorem 13.4 (Galilean relativity principle). The general laws of dynamics

preserve their form (are covariant) in any inertial frame.
The preceding theorem implies that, under the same conditions, the mechanical
evolution of an isolated system is the same for any inertial observer. More precisely,
since systems (13.17) and (13.19) have the same form, they will admit the same
solution when we choose the same initial conditions. In other words, two inertial
observers that repeat the same dynamical experiment will arrive at the same results.1

13.6 Balance Equations

Let Si D fP1 ; : : : ; Pn ; PnC1 ; : : : ; PnCm g be an isolated system of n C m material

points moving relative to an inertial frame I . We now propose to analyze the motion
of the points of the system S D fP1 ; : : : ; Pn g in the presence of the remaining
material points S e D fPnC1 ; : : : ; PnCm g. The motion of any point Pi 2 S is
governed by the equation
.i / .e/
mi rR i D Fi C Fi ; i D 1; : : : ; n; (13.20)
.i /
where Fi is the total force acting upon Pi , due to the other points belonging to
.e/
S .i / , and Fi is the total force due to the interaction of Pi with the material points
belonging to S .e/ . By adding all the equations (13.20) and taking into account the
action and reaction principle, according to which [see (13.3)]
n
X n
X n
X
.i /
Fi D Fij .Pi ; Pj /; (13.21)
i D1 i D1 j D1;j ¤i

we obtain
n
X n
X .e/
mi rR i D Fi : (13.22)
i D1 i D1

Introducing the linear momentum

n
X
QD mi rP i (13.23)
i D1

and the external total force

n
X .e/
R.e/ D Fi ; (13.24)
i D1

from (13.22) we derive the following theorem.

1
For a deeper analysis of the relativity principle, see Chap. 24.
206 13 Principles of Dynamics

Theorem 13.5. The time derivative of the linear momentum of the system S of
material points under the action of the external system S e is equal to the total
external force:
P D R.e/ :
Q (13.25)
Equation (13.25) is called a balance equation of linear momentum or first
cardinal equation of dynamics.
Define the center of mass G as the point whose position vector rG with respect
to the axes of the inertial frame I is given by the relation
n
X
mrG D m i ri ; (13.26)
i D1

Pn
where m D i D1 mi is the total mass of S . Then we have that

Q D mPrG ; (13.27)

and (13.25) can be written in the equivalent form

mRrG D R.e/ ; (13.28)

which makes the following statement.

Theorem 13.6. The center of mass of the system S of material points moves relative
to the inertial frame I as a point having the total mass of the system and subject to
the total force acting on S .
In particular, when the total force vanishes, from (13.25) and (13.28) we derive
that the linear momentum is constant and the center of mass moves uniformly with
respect to I .
Remark 13.3. The center of mass does not depend on the choice of the origin of
the axes introduced in the inertial frame I . In fact, choosing another origin O 0 and
evaluating the position vectors with respect to this new origin, we have that
n
X n
! X !
mr0G D mi r0i D mO 0 O C mi ri D mO 0 O C mrG :
i D1 i D1

Remark 13.4. It is important to note that (13.28) does not allow us to find the
motion of G since, in general, the total force depends on the position and velocity
of all material points of S . It is possible to determine the motion of G by (13.28)
only if the total force depends on the position and velocity of G.
Let ri be the position vector of Pi 2 S relative to the origin O of an
inertial frame I . Taking into account the action and reaction principle, after simple
calculations, from (13.20) we derive the condition
13.6 Balance Equations 207

n
X n
X .e/
ri  mi rR i D ri  F i : (13.29)
i D1 i D1

If we introduce the angular momentum with respect to the pole O

n
X
KO D ri  mi rP i (13.30)
i D1

and the total torque with respect to the pole O

n
X
.e/ .e/
MO D ri  F i ; (13.31)
i D1

then from (13.29) we obtain the following theorem.

Theorem 13.7. The time derivative of the angular momentum with respect to
pole O of the system S is equal to the total torque with respect to the same pole:

P O D M.e/ :
K (13.32)
O

Equation (13.32) is called a balance equation of angular momentum or second

cardinal equation of dynamics.
Remark 13.5. It is possible to extend (13.32) to the case where the angular
momentum and the torque are evaluated with respect to an arbitrary moving pole
A. In fact, we have that
n
X n
X
!
KA D AP i  mi rP i D .ri  rA /  mi rP i :
i D1 i D1

Differentiating with respect to time and recalling (13.25), (13.27), and (13.30),
instead of (13.32), we obtain

P A  rP A  mPrG D M.e/ :
K (13.33)
A

This generalization of the balance equation of angular momentum shows that (13.32)
holds either if A is fixed or if it coincides with the center of mass G.
Remark 13.6. The balance equations of dynamics do not determine the motion of
a system S since they do not contain the internal forces, which are fundamental
in determining the evolution of S . However, they will play a fundamental role in
describing the motion of a rigid system in which the internal interactions do not
affect its evolution.
208 13 Principles of Dynamics

13.7 Kinetic Energy and König’s Theorem

Let S D fP1 ; : : : ; Pn g be a system of n material points, and let R be an arbitrary

rigid frame of reference. Denoting by rP i the velocity of Pi 2 S , we define the kinetic
energy of S relative to R the (nonobjective) scalar quantity
n
1X
T D mi rPi2 ; rPi D jPri j: (13.34)
2 i D1

Introduce the frame of reference RG whose motion relative to R is translational with

velocity rP G . The motion of S relative to RG is called motion about the center of
mass. The following König’s theorem holds.
Theorem 13.8. If T 0 denotes the kinetic energy relative to the motion about the
center of mass, then the kinetic energy T of S relative to R is given by the formula

1
T D T 0 C mrPG2 : (13.35)
2

Proof. Denote by rP 0i the velocity of Pi relative to RG . Since the motion of RG

relative to R is translational with velocity rP G , we have that

rP i D rP 0i C rP G :

Consequently, we also have

n n
1X 1X
T D mi rPi2 D mi .Pr0i C rP G /2
2 i D1 2 i D1

X n
1
D T 0 C mrPG2 C rP G  mi rP 0i :
2 i D1

In the frame RG the center of mass is given by the formula

n
X
mr0G D mi r0i :
i D1

But in RG the center of mass is at rest; therefore, also

n
X
mi rP 0i D 0; (13.36)
i D1

and the theorem is proved. 

Another important result is stated by the following theorem.
13.8 Kinetic Energy Theorem 209

Theorem 13.9. The angular momentum K0 , relative to the motion about the center
of mass, is independent of the pole. Further, if KG is the angular momentum relative
to pole G in any frame of reference, it results that

KG D K0 : (13.37)

Proof. Let O and O 0 be two arbitrary points moving or at rest in RG . Then, taking
into account (13.36), we have
n
X n
X n
X
! 0! !
K0O 0 D O 0 Pi  mi rP 0i D O O  mi rP 0i C OPi  mi rP 0i D K0O :
i D1 i D1 i D1

To complete the proof of the theorem, we note that

n
X n
X
! !
KG D GPi  mi rP i D GPi  mi .Pr0i C rP G /
i D1 i D1
n
!
X !
D K0 C mi GPi  rP G ;
i D1

where, recalling the definition of center of mass, we obtain

n
X ! !
mi GPi D mGG D 0: 
i D1

13.8 Kinetic Energy Theorem

Let S D fP1 ; : : : ; Pn g be a system of material points. We denote by r1 .t/; : : : ; rn .t/

the trajectories of the points of S moving relative to the frame of reference R under
the action of the forces † D f.r1 ; F1 /; : : : ; .rn ; Fn /g. The work of the forces in the
time interval Œt0 ; t1  is given by
n Z
X t1
WŒt0 ;t1  D Fi  rP i dt: (13.38)
i D1 t0

From (13.20) we have the equation

n
X n 
X 
.i / .e/
mi rR i  rP i D Fi C Fi  rP i ;
i D1 i D1
210 13 Principles of Dynamics

which, taking into account (13.34), can be written as

n 
X 
.i / .e/
TP D Fi C Fi  rP i : (13.39)
i D1

Integrating (13.38) in the time interval Œt0 ; t1 , we obtain that the variation of kinetic
energy is equal to the work of internal and external forces acting upon the system S :
.i / .e/
T .t1 /  T .t0 / D WŒt0 ;t1  C WŒt0 ;t1  : (13.40)

The forces acting on S are said to be positional if they depend only on the
position vectors of the material point belonging to S and S .e/ (Sect. 13.6). In
this case, the work of these forces in the time interval Œt0 ; t1  only depends on the
trajectories of the points of S and S .e/ and is independent of their velocities. More
particularly, a system of positional forces † D f.r1 ; F1 /; : : : ; .r1 ; F1 /g acting upon a
system of material points fP1 ; : : : ; Pn g is conservative if there exists a C 1 function
U.r1 ; : : : ; rn /, called potential energy, such that

Fi D rri U: (13.41)

The work of conservative forces in the time interval Œt0 ; t1  assumes the form

WŒt0 ;t1  D .U.r1 .t1 /; : : : ; rn .t1 // C U.r1 .t0 /; : : : ; rn .t0 //; (13.42)

that is, the work depends not on the trajectories and velocities of the points of S but
only on their initial and final positions.
We conclude this section with the following theorem, which follows immediately
from (13.40) and (13.42).
Theorem 13.10 (Conservation of mechanical energy). If the forces acting on a
system S of material points are conservative and U is the corresponding potential
energy, during the motion the total energy is constant:

T C U D const: (13.43)

13.9 Dynamics in Noninertial Frames

In the preceding sections, the fundamental laws of dynamics were stated with
respect to inertial frames. However, in some cases it is necessary to analyze the
motion of a system S D fP1 ; : : : ; Pn g in an arbitrary frame of reference. For
instance, when we study the motion of a material point with respect to the Earth,
we adopt a terrestrial frame of reference that is not inertial.
13.10 Constrained Dynamics 211

Let I and R0 be, respectively, an inertial frame and an arbitrary rigid frame of
reference. To find the form that the Eqs. (13.17) assume in R0 , we first recall that
the masses and the force laws are invariant under an arbitrary change in the rigid
frame of reference. Then, taking into account (12.44), we can give (13.17) the form

mi rR 0i D fi .r01 ; : : : ; r0n ; rP 01 ; : : : ; rP 0n /  mi a i  2mi !  rP 0i ; (13.44)

where a i is the acceleration of point Pi due to the relative motion of R0 with respect
to I (Sect. 12.7). Relation (13.14) shows that, to describe the evolution of S in the
frame R0 , besides the forces fi , i D 1; : : : ; n, we must consider the forces mi a i
and 2mi !  rP 0i , which are called inertial or fictitious forces. More precisely, the
first of these is the drag force, whereas the second one is the Coriolis force. The
name we have given these forces is due to the fact that the observer R finds no
physical justification of their presence, that is, he cannot find bodies that, interacting
with S , can explain the existence of those forces. For these reasons, the forces due
to the presence of bodies are also called real forces.
We conclude by remarking that all the results we proved for inertial frames can be
extended to noninertial frames, provided we consider both real and fictitious forces.

13.10 Constrained Dynamics

In all the preceding sections, we considered systems S D fP1 ; : : : ; Pn g of material

points in which, a priori, there is no limitation to the possible positions of the
points Pi . Actually, among all the possible configurations, the points assume only
those that are compatible with the applied forces and the equations of motion.
However, in an application, it may happen that some obstacles or constraints
can limit the mobility of the points of S . It is evident that the presence of
constraints reduces the number of coordinates we must find by the equations of
motion to determine the trajectories of the points Pi . On the other hand, the
presence of constraints introduces as new unknowns the reactive forces that the
constraints exert on the points during motion. To understand why these forces are
unknown, we must remember that the constraints are due to the presence of suitable
rigid mechanical devices. The reactive forces are a consequence of very small
deformations undergone by these devices during motion. Since we have erased these
deformations assuming the rigidity of the obstacles, we cannot evaluate the reactive
forces and, consequently, cannot assign their force laws. For instance, a body B
on a table, provided it is not too heavy, deforms the table, which, to return to its
undeformed state, balances the weight of B with a reactive force. To determine this
force, one should resort to the theory of elasticity relating the reactive force to the
deformation of the table. However, such a problem is very difficult to solve, and
we prefer to assume that the table is rigid. This hypothesis makes it impossible to
determine the reactive forces that become unknowns.
212 13 Principles of Dynamics

Although we are not able to assign the reactive forces, it is possible by

experimentation to determine some of their important characteristics. In particular,
for smooth constraints, the reactive force ˆ acting on a material point Pi of S during
motion is orthogonal to the constraint at the point occupied by Pi .
At this point we must verify if the laws of dynamics are able to determine the
motion of S in the presence of constraints. In this chapter, we limit ourselves to
proving that the motion can still be determined in particular cases, deferring to a
later chapter the general analysis of this problem.
Let P be a material point constrained to move on the surface f .r/ D 0. If the
surface is smooth, then the reactive force acting on P can be written as ˆ D rf ,
where .r.t// is an unknown scalar function. Consequently, with the usual meaning
of the symbols, the equation of motion

mRr D F C rf; (13.45)

together with the equation of the surface, at least in principle, makes it possible for
us to determine the unknowns r.t/ and .r.t//.
Let P be a material point moving along the curve  obtained by intersecting the
surfaces
f1 .r/ D 0; f2 .r/ D 0 (13.46)
such that the rank of the Jacobian matrix of the functions f1 and f2 is equal to
2. If the surfaces are smooth, then the reactive force of the constraint is a linear
combination of two reactive forces orthogonal to each surface. Consequently, the
equation of motion can be written as

mRr D F C 1 rf1 C 2 rf2 : (13.47)

Equations (13.46) and (13.47) allow, at least in principle, to determine the unknowns
r.t/, 1 .r.t//, and 2 .r.t//.
If the curve  is given in parametric form

r D r.s/ (13.48)

as a function of the curvilinear abscissa s, then we have that [see (11.9)]

sP 2
rR D sR t C n; (13.49)
r
where t is the unit vector tangent to  , n the principal unit vector, and r the curvature
radius of  . The equation of motion
 
sP 2
m sR t C n D F C ˆ; (13.50)
r
13.11 Exercises 213

when it is projected along t and the condition ˆ  t D 0 is taken into account, gives
the scalar differential equation

mRs D Ft .s; sP /  F  t (13.51)

in the unknown s.t/.

13.11 Exercises

1. Let P1 and P2 be two point particles of masses m1 and m2 , respectively. Suppose

that P1 is constrained to moving along the Ox-axis and P2 along the Oy-axis
of a Cartesian system of coordinates Oxy under the action of the elastic force
!
k P1 P2 . If at the time t D 0 the positions and velocities of P1 and P2 are
x1 .0/ D x10 , x2 .0/ D x20 , xP 1 .0/ D xP 10 , and xP 2 .0/ D xP 20 , determine the trajectory
of the center of mass of the system P1 ; P2 .
2. At a certain instant t, a particle of mass m, moving freely in a vertical plane
under the action of gravity, is at a height h above the ground and has a speed v.
Determine when it strikes the ground using energy conservation.
3. A heavy particle rests on top of a smooth fixed sphere. If it is slightly displaced,
find the angular distance from the top at which it leaves the surface.
4. Show that, if an airplane of mass M in horizontal flight drops a bomb of mass m,
the airplane experiences an upward acceleration mg=M , where g is the gravity
acceleration.
5. Let S be a disk with a vertical fixed axis a, intersecting S at its center. A particle
P of mass m can move along a radius of S . If initially S rotates with angular
velocity !.0/ about a and P is at rest, determine at any instant t the relation
between the angular velocity !.t/ of S and the velocity v of P along the radius
of S .
6. A particle P moves along a vertical parabola y D x 2 under the action of its
weight starting from the point .2; 4/. Determine the velocity of P when it arrives
at the point .0; 0/ and evaluate where it arrives before inverting its motion.
7. Let P be a particle moving along the Ox-axis under the influence of the force
F D kx  hx. P Determine the work done by F in the time interval Œ0; t.
Chapter 14
Dynamics of a Material Point

14.1 Central Forces

Definition 14.1. A positional force is said to be central with center O if its force
law is
r
F D f .r/ ; r D jrj; (14.1)
r
where r is the position vector relative to O.
The elementary work dW of F in the elementary displacement dr is
r
dW D F  dr D f .r/  dr D f .r/dr:
r
Then, taking into account (13.41), we can state that a central force is conservative
and its potential energy is given by
Z
U D  f .r/dr: (14.2)

A first example of central force is the elastic force for which (14.1) and (14.2)
become
r 1
F D kr ; U D kr 2 ; (14.3)
r 2
respectively. Another important example of central force is the Newtonian force for
which
k r k
FD 2 ; U D : (14.4)
r r r
A Newtonian force is said to be attractive if k > 0, repulsive if k < 0.
The torque with respect to the pole O of any central force vanishes. Conse-
quently, the angular momentum KO is constant:
KO D KO .0/; (14.5)

A. Romano, Classical Mechanics with Mathematica R

, Modeling and Simulation 215
in Science, Engineering and Technology, DOI 10.1007/978-0-8176-8352-8 14,
© Springer Science+Business Media New York 2012
216 14 Dynamics of a Material Point

where KO .0/ is the initial value of KO . On the basis of this result we can prove the
following theorem.
Theorem 14.1. The trajectory of a material point P subject to a central force lies
in a plane containing the initial position and velocity of P and orthogonal to the
initial angular momentum KO .0/ (Laplace’s plane). Further, in this plane the areal
velocity of P is constant (Kepler’s second law).
Proof. Since KO is constant, we have that

KO D r  mPr D r.0/  mPr.0/ D KO .0/: (14.6)

Consequently, we can state that r and rP belong to the plane  orthogonal to KO

and containing r.0/ and rP .0/. Let Oxyz be a Cartesian system of coordinates such
that Oxy  . In these coordinates, we have that r D .x; y; 0/, rP D .x;
P y;
P 0/, and
KO .0/ D .0; 0; Kz.0//. Then (14.6) implies that

x yP  xy
P D Kz .0/=m; (14.7)

and the theorem is proved when we recall (11.21). t

u
The dynamic equation of a point mass P subject to a central force is
r
mRr D f .r/ : (14.8)
r

If we introduce polar coordinates .r; '/ with center O into the Laplace plane  and
we recall formulae (11.19), which give the components of the acceleration in polar
coordinates, from (14.8) we obtain the following system of ordinary differential
equations:

m.Rr  r 'P 2 / D f .r/; (14.9)

d  2 
r 'P D 0 (14.10)
dt

in the unknowns r.t/ and '.t/. Equation (14.10) again states that the areal velocity
is constant:
1 2 1 1
r 'P D Kz .0/  c: (14.11)
2 2m 2
Introducing (14.11) into (14.9), we obtain the following equation in the unknown
function r.t/:  
1 mc 2
rR D f .r/ C 3 : (14.12)
m r
Finally, as the central forces are conservative, the total energy [see (13.43)]

1 2
mPr C U.r/ D E (14.13)
2
14.1 Central Forces 217

assumes a constant value E depending on the initial data. In polar coordinates, we

have that rP 2 D rP 2 C r 2 'P 2 [see (11.15)] and taking into account (14.11), (14.13)
becomes  
1 c2
m rP C 2 C U.r/ D E:
2
(14.14)
2 r
In view of (14.11), (14.13) gives

2
rP 2 D .E  Ueff .r//; (14.15)
m
where we have introduced the effective potential

mc 2
Ueff .r/ D U.r/ C : (14.16)
2r 2
Equation (14.11) shows that, if the initial data are chosen in such a way that
c D 0, then '.t/ D '.0/ for any t > 0 and the trajectory lies on a straight line
containing O. When c ¤ 0, the function '.t/P has a constant sign, so that '.t/ is
always increasing or decreasing, according to the initial data. Consequently, there
exists the inverse function
t D t.'/; (14.17)
and the trajectory, instead of being expressed by the parametric equations r.t/, '.t/,
can be put in the polar form r.'/. To find a differential equation giving the function
r.'/, it is sufficient to note that the formulae
 
dr c dr d 1
rP D 'P D 2 D c ; (14.18)
d' r d' d' r
 
drP c drP c 2 d2 1
rR D 'P D 2 D 2 2 ; (14.19)
d' r d' r d' r

allow us to write (14.9) and (14.14), respectively, in the form (Binet)

 
d2 u 1 1
C u D  2 2f ; (14.20)
d' 2 mc u u
   
1 du 2 E 1 2 1 1
D 2
 u  2
U ; (14.21)
2 d' mc 2 mc u

where
1
uD :
r.'/
We highlight that (14.12) and (14.20) contain the constant quantity c. In other
words, these equations require a knowledge of the areal velocity or, equivalently,
218 14 Dynamics of a Material Point

of the angular momentum about an axis orthogonal to the plane in which the orbit
lies [see (14.11)] and containing the center of the force. This plane is obtained by
P
giving the initial data r.0/ and '.0/. We can always assume that the initial datum
'.0/ is equal to zero by a convenient choice of the polar axis. On the other hand, to
integrate (14.12), we must also assign the initial datum rP .0/. The data r.0/, '.0/, P
and rP .0/ are also necessary for integrating (14.20) since .dr=d'/.0/ D rP .0/='.0/ P
[see (14.18)].
Weierstrass’ analysis (Chap. 10) can be applied to (14.12) and (14.14) to
determine the qualitative behavior of the function r.t/ as well as the phase portrait
P In contrast, if we are interested in determining the qualitative
in the space .r; r/.
behavior of the function r.'/ and the corresponding phase portrait in the space
.r; dr=d'/, then we apply Weierstrass’s analysis to equations (14.20) and (14.21)
since (14.21) is a first integral of (14.20). We explicitly note that the preceding
phase portrait depends on the choice of the constant c. The different curves of the
phase portrait in the space .r; rP / are obtained by assigning the initial data r.0/, r.0/,
P
whereas the curve of the phase portrait in the space .r; dr=d'/ is determined by
the initial data .r.0/; .dr=d'/.0//. The notebook Weierstrass provides examples of
these phase portraits.
Regarding the central forces we recall the following important theorem.
Theorem 14.2 (Bernard). The elastic forces and the Newtonian forces are the only
central forces for which all bounded orbits are closed.

14.2 Newtonian Forces

Before analyzing the motion of a material point subject to a Newtonian force, we

remember one possible way to define a conic. In the plane  we fix a point O at a
distance D from a straight line a. Finally, we denote by OP and d the distances of
any point P 2  from O and a, respectively. Then, a conic is the locus  of the
points P such that the eccentricity of , defined by the ratio

OP
eD ;
d
is constant. The point O and the straight line a are respectively called the focus
and the directrix of the conic . Applying the preceding definition of a conic to
both cases of Fig. 14.1 and using polar coordinates in the plane , we obtain the
relations
r
eD ;
D  r cos.'  '0 /
r
eD ;
r cos.'  '0 /  D
14.2 Newtonian Forces 219

y y
a a d P
d
P

r r

O x O x

Fig. 14.1 Conic

which lead us to the following equations of a conic:

eD
rD ; (14.22)
1 C e cos.'  '0 /
eD
rD : (14.23)
1 C e cos.'  '0 /

For a material point subject to a Newtonian force (14.4), (14.20) yields

d2 u 1
CuD ; (14.24)
d' 2 p

where
mc 2
pD : (14.25)
k
The general integral of the linear equation (14.24), with constant coefficients, is

1
uD C R cos.'  '0 /;
p

where R > 0 and '0 are arbitrary constants depending on the initial data. Going
back to the variable r and introducing the quantity

e D jpjR; (14.26)

we obtain an equation of the orbit

p
rD ; if k > 0; (14.27)
1 C e cos.'  '0 /
p
rD ; if k < 0; (14.28)
1 C e cos.'  '0 /
220 14 Dynamics of a Material Point

p s
x
O

Fig. 14.2 Axis and parameter of the conic

for attractive .k > 0/ or repulsive .k < 0/ Newtonian forces. In view of (14.22)

and (14.23), we conclude that a material point under the influence of a Newtonian
force moves, with constant areal velocity, along a conic having a focus at the
center O.
To recognize the meaning of the constants p and '0 in (14.27) and (14.28), we
begin by noting that these relations do not depend on the sign of the difference
.'  '0 /. Consequently, the axis ' D '0 is a symmetry axis s of the conic. Further,
for ' D '0 ˙ =2, we have r D p, so that the parameter of the conic p is equal to
the length of the segment OQ, where Q is the intersection point between the conic
and the straight line orthogonal at point O to the axis s (Fig. 14.2). Finally, when
0  e < 1, and ' D '0 , r assumes its lowest value
p
rmin D ; (14.29)
1Ce
whereas its highest value, assumed for ' D '0 ˙ , is
p
rmax D : (14.30)
1e
The values (14.29) and (14.30) correspond to the perihelion and aphelion, respec-
tively. In conclusion, when 0  e < 1, we have that rmin  r  rmax , and the conic
is an ellipsis. In particular, it becomes a circumference when e D 0.
If e D 1, then we have
p
rmin D ; rmax D lim r D 1; (14.31)
2 '!'0 C

and the conic is a parabola.

Finally, when e > 1, the perihelion is still given by (14.29) but there exist two
values ' D '0 ˙ ˛ for which r D 1. The two straight lines with these angular
coefficients are asymptotes of the conic, which is a branch of a hyperbola with its
focus at O.
We can summarize the preceding considerations as follows: a material point
subject to an attractive Newtonian central force moves along a conic that may be
14.2 Newtonian Forces 221

an ellipse, a parabola, or a branch of a hyperbola. In this last case, the focus O is

internal to the branch of the hyperbola. If the force is repulsive, then the trajectory
is still a branch of the hyperbola and O is an external focus. In any case, during the
motion the areal velocity is constant.
It is important to relate the eccentricity of the conic to the total energy E. In view
of (14.14), the conservation of energy (14.18) assumes the form
"  2 #
1 2 d 1 1 k
mc C 2  D E:
2 d' r r r

Taking into account (14.25)–(14.27), after some calculations, we can prove the
result
k 2 k2
ED .e  1/ D .e 2  1/;
2p 2mc 2
which implies that r
2mc 2
eD 1C E: (14.32)
k2
This formula yields the following table:

Energy Eccentricity Orbit

E <0 e<1 Ellipse
ED0 eD1 Parabola
E >0 e>1 Hyperbola

relating energy and eccentricity.

We conclude this section with two important remarks.
Remark 14.1. After determining the orbit r.'/ of a material point subject to a
Newtonian force, we must know '.t/ to complete the analysis of the motion. We
limit ourselves to the case of closed orbits for which 0  e  1. Since the areal
velocity is constant, the function '.t/ is a solution of the equation [see (14.11) and
(14.27)]
c c
'P D 2 D 2 .1 C e cos.'  '0 //2 : (14.33)
r .'/ p
By changing the polar axis, we can always suppose '.0/ D '0 D 0. Then,
separating the variables in the preceding equation, we obtain the implicit form of
the solution
Z '
d' c
D 2 t: (14.34)
0 .1 C e cos '/
2 p
To find for elliptic orbits an approximate solution of (14.33), we note that if, in
particular, the orbit is a circumference (e D 1), then the angular velocity has the
constant value a  c=p 2 since r is constant. For elliptic orbits, which are similar to
222 14 Dynamics of a Material Point

circular orbits, it is reasonable to search for an approximate solution of (14.33) in

the form
'.t/ D at C e'1 .t/ C e 2 '2 .t/ C    : (14.35)
Introducing (14.35) into (14.33) we obtain

a C e 'P1 C e 2 'P2 C    D a.1 C e cos.a C e 'P1 C e 2 'P2 C    //2 : (14.36)

Applying Taylor’s expansion to the right-hand side of (14.36) and collecting the
terms that multiply the same power of the eccentricity, we obtain the following
sequence of equations:

'P0 D a; (14.37)
'P1 D 2a cos.at/; (14.38)
'P2 D a cos .at/  2a sin.at/'1 .t/;
2
(14.39)


Finally, solving the preceding equations with the initial data 'i .0/ D 0, i D
0; 1; 2; : : :, we obtain
 
3 5
'.t/ D at C 2e sin.at/ C e  at C sin.2at/ C    :
2
(14.40)
2 4

Remark 14.2. A material point P with mass m, subject to an attractive Newtonian

force, initially occupies the position .r0 ; '0 /. We want to determine for which initial
velocity v0 of P the orbit is a parabola or a hyperbola, i.e., is open. The lowest value
of these velocities is called the escape velocity. We have already shown that open
orbits are possible if and only if the total energy verifies the condition

1 2 k
ED mv   0:
2 0 r
Consequently, the escape velocity is given by
s
2k
v0 D : (14.41)
mr0

14.3 Two-Body Problem

Let S be an isolated system of two material points P1 and P2 subject to the action
of the mutual gravitational force. If m1 and m2 denote their masses, the motion of
S is described by the equations
14.3 Two-Body Problem 223

P1
G
r
P2
r1 rG

r2
O

Fig. 14.3 Two-body system

m1 m2
m1 rR 1 D h .r1  r2 /; (14.42)
jr1  r2 j3
m1 m2
m2 rR 2 D h .r2  r1 /; (14.43)
jr2  r1 j3

where the position vector ri , i D 1; 2, is relative to the origin O of an inertial frame

I (Fig. 14.3). Further, the position vector rG relative to O of the center of mass G
of S yields
.m1 C m2 /rG D m1 r1 C m2 r2 ; (14.44)
whereas the position vector r of P2 relative to P1 is given by

r D r2  r1 : (14.45)

Solving (14.44) and (14.45) with respect to r1 and r2 , we obtain

m2
r1 D rG C r; (14.46)
m1 C m2
m1
r2 D rG  r: (14.47)
m1 C m2

These equations show that the two-body problem can be solved by determining the
vector functions rG .t/ and r.t/. Since the system S is isolated, we can state that
the center of mass moves uniformly on a straight line; therefore, the function rG .t/
is known when the initial data r1 .0/, rP 1 .0/, r2 .0/, rP 2 .0/ are given [see (14.44)].
Regarding the function r.t/, we note that, dividing (14.42) by m1 and (14.42) by m2
and subtracting the second equation from the first one, we obtain the equation

m2 .m1 C m2 /
m2 rR D h r; (14.48)
jrj3
224 14 Dynamics of a Material Point

which can equivalently be written in the form

m1 m2
Rr D h r; (14.49)
jrj3

where we have introduced the reduced mass

m1 m2
D : (14.50)
m1 C m2

Both Eqs. (14.48) and (14.49) show that r.t/ can be determined by solving a
problem of a material point subject to a Newtonian force. In other words, if we
introduce the variables rG and r, the two-body problem is reduced to the problem
of a material point moving under the action of a Newtonian force.
Exercise 14.1. Let P1 x10 x20 x30 be a noninertial frame of reference with the origin at
the point P1 and axes parallel to the axes of an inertial frame Ox1 x2 x3 . Prove that
Newton’s equation relative to the material point P2 in the frame P1 x10 x20 x30 coincides
with (14.48), provided that we take into account the fictitious forces.

14.4 Kepler’s Laws

In this section, we study the motion of the planets about the Sun. First, we note
that the dimension of the planets is much smaller compared with their distances
from the Sun. Consequently, it is reasonable to model the planetary system with a
system of material points moving under the influence of the mutual gravitational
interaction. Such a problem is too complex to be solved. However, the solar mass
is much greater than the mass of any planet. This circumstance suggests adopting a
simplified model to analyze the motion of the planets. This model is justified by the
following considerations.
The motion of a planet Pi is determined by the gravitational action of the Sun
and the other planets. Consequently, in an inertial frame I , the motion of the planets
Pi of mass mi is a solution of the equation

M mi X mi mj
mi rR i D h r i  h rij ; (14.51)
jri j3 jrj3ij
j ¤i

where h is the gravitational constant, M is the solar mass, ri is the position vector
of the planet Pi with respect to the Sun, and rij D ri  rj is the position vector of
Pi relative to the planet Pj . Since the solar mass M  mi , for any planet, while the
distances jri j and jrij j are of the same order of magnitude, in a first approximation,
we can neglect the action of the other planets on Pi with respect to the prevailing
action of the Sun. In this approximation we are faced with a two-body problem: the
Sun and the single planet. Consequently, to describe the Sun–planet system, we can
14.4 Kepler’s Laws 225

resort to (14.48), which, owing to the condition M  mi , reduces to the equation

mi M
mi rRi D h ri ; (14.52)
jri j3

according to which, in a first approximation, the planet moves about the Sun as
a material point subject to a Newtonian force. We are in a position to prove the
following Kepler’s laws:
1. The orbit of a planet is an ellipse having a focus occupied by the center of the
Sun.
2. The orbit is described by a constant areal velocity.
3. The ratio between the square of the revolution period Ti and the cube of the
greatest semiaxis ai of the ellipse is given by

Ti2 4 2
3
D ; (14.53)
ai hM

i.e., it does not depend on the planet.

The first two laws are contained in the results we have already proved for the
motion of a point under the influence of a Newtonian force. To prove the third law,
we start by recalling that the area bounded by the ellipse described by the planet is
given by the formula ai bi , where bi is the smallest semiaxis of the ellipse. By the
second law, we have that [see (14.11)]

1
ai bi D ci Ti :
2

On the other hand, it can be proved that pi D bi2 =ai , so that

ci2 Ti2
pi D :
4 2 ai3

Comparing this result with (14.25), we obtain the formula

Ti2 4 2 mi
3
D
ai ki

which, recalling that for Newtonian forces ki D hmi M , implies the Kepler’s third
law.
The preceding considerations show that Kepler’s laws are not exact laws since
they are obtained by neglecting the mutual actions of the other planets on the motion
of a single planet. Taking into account this influence is a very difficult task since
it requires a more sophisticated formulation of the problem. This more accurate
226 14 Dynamics of a Material Point

analysis shows that the orbits of the planets are more complex since they are open
curves that, roughly, can be regarded as ellipses slowly rotating about the focus
occupied by the Sun.

14.5 Scattering by Newtonian Forces

Let P be a material point of mass m moving under the action of a repulsive

Newtonian force with center at O. If we suppose its energy E to be positive, the
trajectory of P will be a branch of a hyperbola  that does not contain the center
O. We denote by the angle between the two asymptotes of  and define as a
scattering angle the angle ˛ D   . This is the angle formed by the velocity
.i / .e/
v1 of P when it comes from infinity and the velocity v1 when P comes back to
infinity (Fig. 14.4). Finally, we call a shock parameter s the distance of O from the
.i /
asymptote corresponding to the velocity v1 .
In the polar coordinates .r; '/, the equation of  is
jpj
rD ;
e cos '  1
and it gives for the angle the value
1
cos D :
2 e
Further, we also have that
 
˛  1
sin D sin  D cos D
2 2 2 2 e

y
v (i )

s
r

O x

v (e)

Fig. 14.4 Scattering of a particle

14.6 Vertical Motion of a Heavy Particle in Air 227

and
˛ 1
D p
tan : (14.54)
2 e 1
2

On the other hand, the conservation of total energy and angular momentum is
expressed by the following formulae:

1 2 k 1  2
ED v  D m v.i1/ ; (14.55)
2 r 2
p
jKO j D lim jKO j D m s v.i1/ D s 2mE; (14.56)
r>1

which allow us to give (14.32) the equivalent form

s
 2
2sE
eD 1C : (14.57)
k

Using this relation and (14.56), we finally obtain that

k ˛
sD cot : (14.58)
2E 2
The preceding formula is important for the following reasons. Rutherford
proposed the nuclear atomic model according to which an atom is formed by a
small nucleus, containing the entire positive charge of the atom, and electrons
rotating along elliptic orbits with a focus occupied by the nucleus. This model was in
competition with the model proposed by Thomson in which the positive charge was
continuously distributed in a cloud having the dimension of the atom, whereas the
electrons were contained in fixed positions inside this cloud. Rutherford understood
that, to resolve the dispute regarding the correct model, it was sufficient to throw
charged particles (˛ particles) against the atoms contained in a thin gold leaf and to
evaluate the scattering angles of those particles. In fact, Thomson’s model, owing
to the assumed charge distribution inside the atom, implied small scattering angles.
In contrast, the concentrated positive charge of Rutherford’s model implied high
scattering angles, especially for small shock parameters. The experimental results
of Rutherford’s experience were in full agreement with Eq. (14.58), which, in
particular, implies a back scattering for very small shock parameters.

14.6 Vertical Motion of a Heavy Particle in Air

Let P be a mass point acted upon by its weight mg and the air resistance
v
F D R.v/ ; (14.59)
v
228 14 Dynamics of a Material Point

where m is the mass of P , g the gravitational acceleration, and v the length of

the velocity v. To suggest a reasonable form of the function R.v/, we recall that
a mass point is a schematic model of a small body B. When B is falling in the
air, the motion is influenced by the form of B, even if B is small. Consequently,
the air resistance must include something resembling those characteristics of B that
we discarded in adopting the model of a point particle. A good description of the
phenomenon we are considering can be obtained assuming that the air resistance
has the following form:
R.v/ D A˛f .v/; (14.60)
where  is the mass density of the air, A is the area of the projection of B on a plane
orthogonal to v, and ˛ is a form coefficient that depends on the profile of B. Finally,
f .v/ is an increasing function confirming the conditions

f .0/ D 0; lim f .v/ D 1: (14.61)

v!1

Taking into account the preceding considerations, we can state that the equation
governing the motion of P is
v
mPv D mg  R.v/ : (14.62)
v
Since in this section we limit our attention to falls along the vertical a, we begin
by proving that if the initial velocity v0 is directed along a, then the whole trajectory
lies on a vertical straight line. Let Oxyz be a frame of reference with the origin O
at the initial position of P and the axis Oz vertical and downward directed. In this
frame, the components along Ox and Oy of (14.62) are written as

R.v/
xR D  P
x; (14.63)
mv
R.v/
yR D  P
y: (14.64)
mv

Multiplying (14.63) by xP and (14.64) by y,

P we obtain

d 2 R.v/ 2
xP D 2 xP  0; (14.65)
dt mv
d 2 R.v/ 2
yP D 2 yP  0; (14.66)
dt mv

so that xP 2 .t/ and yP 2 .t/ are not increasing functions of time. Since initially x.0/
P D
P
y.0/ D 0, we can state that x.t/ P and y.t/
P vanish identically. But in addition, x.0/ D
y.0/ D 0, and then x.t/ D y.t/ D 0, for any value of time, and P moves vertically.
14.6 Vertical Motion of a Heavy Particle in Air 229

It remains to analyze the motion along the vertical axis Oz. To this end, we denote
by k the vertical downward-oriented unit vector and note that

v zP
 D  k D k; (14.67)
v jPzj

where we must take the  sign if zP > 0 (downward motion) and the C sign if zP < 0
(upward motion). Finally, to evaluate the vertical motion of P , we must integrate
the equation  
R.jPzj/
zR D mg 1 (14.68)
mg
with the initial conditions

z.0/ D 0; zP.0/ D zP0 : (14.69)

We must consider the following possibilities:

.a/ zP0 < 0; .b/ zP0  0: (14.70)

In case (a), we must take the C sign in (14.68) at least up to the instant t  for
which zP.t  / D 0. In the time interval Œ0; t  , we have that zR > 0, zP < 0, so that

dPz2
D 2PzzR < 0;
dt

and the motion is decelerated. Integrating (14.68) in the interval Œ0; t   we obtain
the formula Z 0
dPz
D gt  ; (14.71)
Pz0 R.jPzj/
1C
mg
and t  is finite since the function under the integral is bounded in the interval ŒPz0 ; 0.
In case (b), the motion is initially progressive, so that we take the  sign
in (14.68) at least up to an eventual instant t in which zP.t / D 0:
 
R.Pz/
zR D mg 1  : (14.72)
mg

On the other hand, it is evident that, under hypotheses (14.61) on f .v/, there is one
and only one value V such that

mg D R.V /: (14.73)

Consider the following three cases:

.i / zP.0/ D V; .i i / zP.0/ > V; .i i i / zP.0/ < V: (14.74)

230 14 Dynamics of a Material Point

It is plain to prove that zP.t/ D V is a solution of (14.72) verifying the initial

condition .i /. Owing to the uniqueness theorem, this is the only solution satisfying
datum .i /. In case .i i / we have that zP > V for any value of t. In fact, if there is an
instant t in which zP.t / D V , then, taking this instant as the initial time, we should
always have zP.t/ D V , against the hypothesis zP0 > V . Further, if zP.t/ > V for any t,
then from (14.72) we have zR.t/ < 0; consequently, zP.t/Rz.t/ < 0 and zP2 .t/ is always
decreasing. In conclusion, the motion is progressive and decelerated, and

lim zP.t/ D V: (14.75)

t !1

Reasoning in the same way, we conclude that in case .i i i / the motion is progressive
and accelerated, and (14.75) still holds.
As an application of the previous analysis, let us consider the case of a parachutist
of mass m for which we require a prefixed limit velocity V , i.e., an impact velocity
with the ground that does not cause damage to the parachutist. From (14.60) we
obtain the cross section A of the parachute in order to attain the requested limit
velocity:
mg
AD :
˛f .V /

14.7 Curvilinear Motion of a Heavy Particle in Air

In this section, a qualitative analysis of Eq. (14.62) is presented in the general case
in which the direction of the initial velocity v0 is arbitrary. Let Oxyz be a frame
of reference with the origin O at the initial position of P and an upward directed
vertical axis Oz. Further, we suppose that the coordinate plane Oxz contains v0 , and
Ox is directed in such a way as to have xP 0 > 0. Since in this frame yP0 D 0, and the
component of Eq. (14.62) along Oy is still (14.64), we have that y.t/ P D 0, at any
instant, and we can state that the trajectory lies in the vertical plane containing the
initial position and velocity.
P
From (14.63) and the initial condition x.0/ D xP 0 > 0 it follows that the function
jx.t/j
P is not increasing [see (14.65)]. Moreover, there is no instant t  such that
P  / D 0. In fact, if such an instant existed, we could consider the initial value
x.t
problem given by (14.63) and the initial datum x.t P  / D 0. Since the only solution
P
of this problem is given by x.t/ D 0, the motion should take place along the vertical
axis Oz, and this conclusion should be in contradiction with the condition x.0/ P D
xP 0 > 0. In conclusion, x.t/
P is a positive nonincreasing function for any t  0
so that

P
lim x.t/ D xP 1 < xP 0 ; (14.76)
t !1

R
lim x.t/ D 0: (14.77)
t !1
14.7 Curvilinear Motion of a Heavy Particle in Air 231

/2

/2

Fig. 14.5 Behavior of function ˛.t /

In the limit t ! 1, (14.63) and (14.77) lead us to the result

P
lim x.t/ D 0: (14.78)
t !1

We have already proved that the trajectory  of the material point P is contained
in the vertical plane Oxz. To discover important properties of  , we introduce the
unit vector t D v=v tangent to  and denote by ˛ the angle that t forms with the
Ox-axis. Then, projecting (14.62) onto the Ox-axis and along t, we have the system

d
m .v cos ˛/ D R.v/ cos ˛; (14.79)
dt
mPv D mg sin ˛  R.v/ (14.80)

in the unknowns ˛.t/ and v.t/. Introducing into (14.79) the value of vP deduced
from (14.80), we obtain the equation

d g
.cos ˛/ D sin 2˛; (14.81)
dt 2v
which can also be written as
g cos ˛
P
˛.t/ D : (14.82)
v

P
Since we refer to the case x.0/ > 0, at the initial instant t D 0 the angle ˛
satisfies the condition  
 < ˛0 < :
2 2
Then from (14.81) we have that
P
˛.0/ < 0: (14.83)
Consequently, we can state that ˛.t/, starting from any initial value belonging to the
interval .=2; =2/, decreases at least up to an eventual instant t  in which the
right-hand side of (14.82) vanishes, that is, when (Fig. 14.5)
232 14 Dynamics of a Material Point

O x

Fig. 14.6 Trajectories in air


˛.t  / D :
2

The value t  is infinite since, in the opposite case, if we adopt the initial datum
P  / D v.t  / cos ˛.t  / D 0, the corresponding motion should always take place
x.t
along a vertical line, against the hypothesis that for t D 0 the velocity has a
component along the Ox-axis.
In conclusion, we can state that the trajectory is downward concave [˛.t/
decreases] and has a vertical asymptote (Fig. 14.6).
Finally, we want to prove that, if v.t/ is a regular function of t at infinity, then

lim v.t/ D V; (14.84)

t !1

where V is the unique solution of (14.73). In fact, the regularity of the positive
function v.t/ implies one of the following possibilities:

lim v.t/ D C1; lim v.t/ D v1 < C1: (14.85)

t !1 t !1

In the first case, since v.t/ is positive, either of the following conditions holds:

lim vP .t/ D C1; lim vP .t/ D a > 0:

t !1 t !1

On the other hand, when limt !1 v.t/ D C1, from (14.80) we obtain limt !1 vP
.t/ D 1, and then (14.85)1 is false.
14.7 Curvilinear Motion of a Heavy Particle in Air 233

When (14.85)2 is verified, we have

P D 0;
lim v.t/
t !1

so that, by (14.80), we prove (14.84).

We wish to conclude this section with some considerations about the mathemati-
cal problem we are faced with. Equations (14.82) and (14.80) can also be written in
the form
g cos ˛
P
˛.t/ D ; (14.86)
v
dv R.v/
P
˛.t/ D g sin ˛  ; (14.87)
d˛ m
from which we deduce the equation

dv vR.v/
D v tan ˛ C (14.88)
d˛ mg cos ˛

in the unknown v.˛/. Its solution,

v D F .˛; ˛0 ; v0 /; (14.89)

for given initial data ˛0 and v0 , defines a curve in the plane .˛; v/. When this function
is known, it is possible to integrate the equation of motion. In fact, in view of (14.86),
we have that
dx g cos ˛ dx
v cos ˛ D xP D ˛P D ;
d˛ v d˛
dz g cos ˛ dz
v sin ˛ D zP D ˛P D :
d˛ v d˛
In turn, this system implies

dx v2 F 2 .˛; ˛0 ; v0 /
D D ;
d˛ g g
dz v2 F 2 .˛; ˛0 ; v0 /
D  tan ˛ D  tan ˛;
d˛ g g

so that
Z ˛
1
xD F 2 .; ˛0 ; v0 /d  G.˛; ˛0 ; v0 /; (14.90)
g ˛0
Z ˛
1
zD F 2 .; ˛0 ; v0 / tan ˛ d  H.˛; ˛0 ; v0 /: (14.91)
g ˛0
234 14 Dynamics of a Material Point

These relations, which give the parametric equations of the trajectory for any choice
of the initial data, represent the solution of the fundamental ballistic problem since
they make it possible to determine the initial angle ˛0 and the initial velocity v0 to
hit a given target.

14.8 Terrestrial Dynamics

By terrestrial dynamics we mean the dynamics relative to a frame of reference

Re D .O 0 ; x1 ; x2 ; x3 / at rest with respect to the Earth. Before writing the equation
of motion of a material point in the frame Re D .O 0 ; x1 ; x2 ; x3 /, we must determine
how Re D .O 0 ; x1 ; x2 ; x3 / moves relative to an inertial frame I D .O1 ; 2 ; 3 /.
We take the frame I D .O1 ; 2 ; 3 / with its origin at the center O of the Sun, axes
oriented toward fixed stars, and containing the terrestrial orbit in the coordinate
plane O1 ; 2 (Fig. 14.7). It is well known that the motion of the Earth relative to I
is very complex. However, we reach a sufficiently accurate description of terrestrial
dynamics supposing that the Earth describes an elliptic orbit about the Sun while it
rotates uniformly about the terrestrial axis a.
Then we consider the frame Re0 D .O 0 ; 10 ; 20 ; 30 / with its origin at the center O 0
of the Earth, axes O30  a, and the other axes with fixed orientation relative to
the axes of I . It is evident that the motion of Re0 relative to I is translational but
not uniform. Finally, we consider a frame Re D .O 0 ; x1 ; x2 ; x3 / with Ox3  a
and the other two axes at rest with respect to the Earth. Because the dynamical
phenomena we want to describe take place in time intervals small compared with
the solar year, we can assume that the rigid motion of Re0 relative to I is translational
and uniform. In other words, Re0 can be supposed to be inertial, at least for time
intervals much shorter than the solar year. Finally, the motion of Re relative to Re0 is
a uniform rotation about the terrestrial axis a. Now we are in a position to determine
the fictitious forces acting on a material point P moving relative to the frame Re :
!
 ma D m!T2 QP ; mac D 2m!T  rP 0 ; (14.92)

R’e
’3 x3

O
x2
O’
’2
Re
’ x1

Fig. 14.7 Terrestrial frame

14.8 Terrestrial Dynamics 235

where m is the mass of P , !T the terrestrial angular velocity, Q the projection of

P onto the rotation axis a, and rP 0 the velocity of P relative to Re . Consequently, the
equation governing the motion of P relative to the frame Re is
!
mRr0 D F C mA C m!T2 QP  2m!T  rP 0 ; (14.93)

where
MT m 0
mA D h r (14.94)
jrj03
is the gravitational force acting on P due to the Earth and F is the nongravitational
force.
We define the weight p of P as the opposite of the force Fe we need to apply to
P , so that it remains at rest in the terrestrial frame Re . From (14.93) we have that
!
Fe C mA C m!T2 QP D 0;

and then the weight of P is given by

 !
p D m A C !T2 QP  mg; (14.95)

where g is the gravitational acceleration. The preceding equation shows that the
weight of P is obtained by adding the gravitational force mA, which is exerted
!
on P by the Earth, and the centrifugal force m!T2 QP . Supposing that the Earth
is spherical and recalling that the first force is always directed toward the center
O 0 of the Earth, whereas the second force is orthogonal to the rotation axis of
the Earth, we can state that the weight reaches its minimum at the equator and its
maximum at the poles. Further, at the poles and at the equator, the direction of p
coincides with the vertical to the surface of the Earth. Finally, we remark that the
direction of g does not depend on the mass of P . This fundamental property follows
from the identification of the masses m appearing in (14.94) and in the centrifugal
!
force m!T2 QP . If we define as gravitational mass mg the mass appearing in the
gravitational force (14.94) and as inertial mass the mass mi appearing in Newton’s
equation of motion, instead of (14.95), we obtain
mi 2 !
gDAC ! QP : (14.96)
mg T

In other words, if we distinguish the two masses respectively appearing in the

universal attraction law and in Newton’s law of motion, then, due to the presence
of the ratio mi =mg , the acceleration g depends on the nature of the body. However,
ERotvRos and Zeeman showed, with very high experimental accuracy, that g does not
depend on the ratio mi =mg . Consequently, if we adopt the same unit of measure for
236 14 Dynamics of a Material Point

north
y

z
k

west O x east

south

Fig. 14.8 Terrestrial frame

mi and mg , we can assume that mi =mg D 1. This result plays a fundamental role
in general relativity.
With the introduction of the weight (14.95) and in the absence of other forces,
(14.93) becomes
rR 0 D g  2m!T  rP 0 : (14.97)
To consider a first application of (14.97), we choose a frame of reference Re as
in Fig. 14.8 and write (14.97) in a convenient nondimensional form.
Let us introduce a reference length L and a reference time interval T . Since the
rotation of the Earth produces much smaller effects on the falls of heavy bodies
than does the weight, we can take L and T in such a way that L=T 2 D g. On the
other hand, we cannot use the same time interval T as a measure of ! since a
turn of the Earth about its axis requires 24 h, whereas a heavy body falls in a few
minutes or seconds depending on the throw height. Consequently, we introduce a
new reference time interval, TT . If we use the same letters to denote dimensional
and nondimensional quantities, (14.97) can also be written as

rR 0 D k C 2 !T  rP 0 ; (14.98)

where k is the unit vector along Oz (Fig. 14.8) and is the nondimensional
parameter
T
D :
TT
For a falling time T D 60 s and T D 86; 400 s (24 h) we have ' 0:0007. In other
words, we are faced with a problem to which we can apply Poincaré’s method. Since
in the frame Re k D .0; 0; 1/ and !T D .0; !T cos ; !T sin  /, (14.98) gives the
system
14.9 Simple Pendulum 237

xR D 2 !T .Pz cos   yP sin  /;

yR D 2 !T xP sin ;
zR D 1 C 2 !T xP cos ;

which is equivalent to the first-order system

xP D u;
yP D v;
zP D w;
uP D 2 !T .w cos   v sin  /;
vP D 2 !T u sin ;
wP D 1 C 2 !T u cos :

The first-order Poincaré expansion of the solution of the preceding system,

obtained using the notebook Poincare and referred to by x.t/, y.t/, and z.t/, is
1 3
xD t ! cos ;
3
y D 0;
t2
z D 1 :
2
This approximate solution shows that the rotation of the Earth, while a heavy body
is falling, gives rise to an eastward deviation.

14.9 Simple Pendulum

We call a heavy particle P with mass m, constrained to moving along a smooth

circumference  lying in a vertical plane  (Fig. 14.9), a simple pendulum .
The equation of motion of P is

mRr D mg C ˆ; (14.99)

where g is the gravitational acceleration and ˆ the reactive force exerted by the
constraint. Projecting (14.99) onto the unit vector t tangent to  and recalling that the
curvilinear abscissa s D l', where l is the radius of  , and ˆ  t D 0 [see (13.51)],
we obtain a second-order differential equation
g
'R D  sin ' (14.100)
l
238 14 Dynamics of a Material Point

mg

Fig. 14.9 Simple pendulum

E
E

E
E

Fig. 14.10 Phase portrait of simple pendulum

in the unknown '.t/. This equation admits one and only one solution when the
P are given. Since there is no solution in a finite form of the
initial data '.0/ and '.0/
preceding equation, we resort to a qualitative analysis of (14.100). Due to the form
of (14.100), we can resort to the method discussed in Sect. 10.8 to obtain a phase
portrait (Fig. 14.10).
The conservation of the total mechanical energy is written as

1 2 2
ml 'P  mgl cos ' D E;
2
14.10 Rotating Simple Pendulum 239

so that we have
2
'P 2 D .E C mgl cos '/: (14.101)
ml 2
Since the potential energy U.'/ D mgl cos ' is periodic, it is sufficient to analyze
the motion for ' 2 Œ; .
The motion is possible if E C mgl cos '  0. When E D mgl, point P
occupies the position ' D 0 with velocity equal to zero [see (14.101)]. This position
is a stable equilibrium position since the potential energy has a minimum at ' D 0.
For mgl < E < mgl the motion is periodic between the two simple zeros of the
equation
E C mgl cos ' D 0;

and the corresponding period of the motion is given by the formula

Z '2 .E/
d'
T D2 r ; (14.102)
'1 .E/ 2
.E C mgl cos '/
ml 2

which is easily obtained by (14.101). For E D mgl we have again an equilibrium

position ' D  that is unstable since in this position the potential energy has a
maximum. In view of (14.101), the level curve corresponding to this value of the
energy is
g
'P 2 D 2 .1 C cos '/: (14.103)
l
This level curve contains three trajectories. One of them corresponds to the already
mentioned unstable equilibrium position. Since in this position the first derivative
of potential energy vanishes, any motion with initial condition  < '.0/ <  and
velocity such that the total energy assumes the value E D mgl tends to the position
˙ without reaching it. This circumstance is highlighted by the dashed part of the
level curve.

14.10 Rotating Simple Pendulum

In this section we consider a simple pendulum P moving along a smooth vertical

circumference  that uniformly rotates about a vertical diameter a with angular
velocity !. We study the motion of P relative to a frame of reference R D C xyz,
having its origin at the center C of  , C z  a, and the plane Cyz containing 
(Fig. 14.11). Since R is not an inertial frame, the equation governing the motion
of P is
!
mRr0 D mg C m! 2 QP  2m!  rP 0 C ˆ; (14.104)
240 14 Dynamics of a Material Point

z a

Q t y
x
P
mg

Fig. 14.11 Rotating simple pendulum

where Q is the orthogonal projection of P on a, t the unit vector tangent to  , and

ˆ the reactive force satisfying the condition ˆ  t D 0. Denoting by l the radius
of  , recalling the relation s D l', where s is the curvilinear abscissa on  , and
projecting (14.104) along t, we obtain the second-order differential equation
 g
m'R D sin ' ! 2 cos '  : (14.105)
l
There is no closed solution of this equation, and then we again resort to the
qualitative analysis of Sect. 10.8.
Multiplying (14.105) by ', P we obtain
 
2 m! 2 l 2
'P 2 D E C mgl cos '  cos2 ' : (14.106)
ml 2 2

To determine the behavior of the solutions of (14.105), we analyze the potential

m! 2 l 2
U.'/ D mgl cos ' C cos2 ' (14.107)
2

in the interval Œ; . First, we have

 
!2l
U.0/ D ml g ; (14.108)
2
14.10 Rotating Simple Pendulum 241

E
E

E
E

Fig. 14.12 Phase portrait for .g=! 2 l/ > 1

and

U 0 .'/ D ml sin '.g  ! 2 l cos '/: (14.109)

Therefore, if
g
> 1;
!2l
then U 0 .'/ D 0 if and only if ' D 0; ˙. In this hypothesis, we can state that
the sign of U 0 .'/ [see (14.109)] depends on the sign of sin '. Consequently, U.'/
decreases for ' < 0, increases for ' > 0, and has a minimum when ' D 0 and a
maximum for ' D ˙ (Fig. 14.12).
In contrast, if
g
< 1;
!2l
 
then U 0 .'/ D 0 when ' D 0; ˙; ˙ arccos !g2 l . Moreover, ' D 0; ˙ are
maxima of U.'/, whereas the angles ˙ arccos.g=! 2 l/ correspond to minima. The
relative phase portrait is shown in Fig. 14.13. This analysis shows that the phase
portrait undergoes a profound change when the ratio g=! 2 l changes in a small
neighborhood of 1. For this reason it is called a bifurcation value.
242 14 Dynamics of a Material Point

Fig. 14.13 Phase portrait for .g=! 2 l/ < 1

14.11 Foucault’s Pendulum

Let P be a heavy point moving on the surface S of a smooth sphere with radius l
whose center C is fixed with respect to a terrestrial frame of reference R. We wish
to study the small oscillations of this spherical pendulum relative to a terrestrial
observer R D Oxyz. The axis Oz is chosen to coincide with the vertical of the place
containing the center C of S , and the axes Ox and Oy are horizontal (Figs. 14.14
and 14.15). The equation governing the motion of P relative to a terrestrial observer
R D Oxyz is
rR D g  2!T  rP C ˆ; (14.110)
where !T is the angular velocity of the Earth and ˆ is the reactive force exerted on
P by the spherical constraint S with the equation

f .x; y; z/  x 2 C y 2 C .z  l/2  l 2 D 0: (14.111)

The sphere S is smooth so that ˆ D

rf , where
is an unknown function
of .x; y; z/. Since also !T D .!T sin ˛; 0; !T cos ˛/, the components of (14.110)
along the axes of R are

xR D 2!T yP cos ˛ C 2x
; (14.112)
yR D 2!T zP sin ˛  2!T xP cos ˛ C 2y
; (14.113)
zR D g  2!T yP sin ˛  2.l  z/
: (14.114)
14.11 Foucault’s Pendulum 243

Fig. 14.14 Spherical pendulum

y
z
C l

O x

Fig. 14.15 Spherical pendulum and terrestrial frame

We are interested in the small oscillations of the spherical pendulum about the
position ' D 0. This circumstance simplifies the study of Eqs. (14.112)–(14.114),
as we can see by a nondimensional analysis. Introduce as the length of reference
the radius l of the sphere S , and denote by ' the angle corresponding to the largest
oscillation; then we have that (Fig. 14.16)

x; y ' l'; z ' l' 2 : (14.115)

244 14 Dynamics of a Material Point

z
C

l
ϕ

S 2 y
}lϕ
lϕ mg

Fig. 14.16 Order of magnitude of x, y, and z

Now we introduce the two reference times

s
l
T D ; TT ; (14.116)
g

where T is approximately equal to the oscillation period of a simple pendulum and

TT is equal to the number of seconds in a day. Writing (14.114) in nondimensional
form and using the same symbols for nondimensional and dimensional quantities,
we have
l' 2 l'
2
zR D g  2 !T yP sin ˛  2l
C l' 2 z
;
T T TT
that is,
g

' : (14.117)
2l
Operating in the same way with (14.114) and taking into account (14.117), we
obtain the approximate system
g
xR D 2!T yP cos ˛  x; (14.118)
l
g
yR D 2!T xP cos ˛  y; (14.119)
l
which in vector form is written as
g
rR ? D  r?  2!  rP ? ; (14.120)
l
where r? is the vector obtained projecting the position vector r of P onto the
horizontal plane Oxy and ! D .0; 0; !T cos ˛/.
14.12 Exercises 245

To solve (14.120), we introduce a new frame of reference R0 D .Ox 0 y 0 z0 / with

the same origin of R, Oz0  Oz, and rotating about Oz with angular velocity ! .
In the transformation R ! R0 we have [see (12.38), (12.39), and (12.44)]

r D r0 ;
rP D rP 0  !  r0 ;
rR D rR 0 C a  2!  rP 0 D rR 0  ! 2 r0  2!  rP 0 :

Introducing the preceding relations into (14.120) and noting that

2!  .!  r0 / D 2! 2 r0 C 2.!  r0 /! D 2! 2 r0

since !  r0 D 0, (14.120) becomes

g 
rR 0 D  C ! 2 r0 : (14.121)
l

Thisqequation shows that in the frame R0 , the pendulum oscillates with a period
2= gl C ! 2 , whereas its plane of oscillation rotates with angular velocity !
about the vertical axis Oz.

14.12 Exercises

1. Neglecting the resistance of the air, prove that the velocity of a point particle
projected upward in a direction inclined at 60 to the horizontal is half of its
initial velocity when it arrives at its greatest height.
2. Neglecting the resistance of the air, find the greatest distance that a stone can
be thrown with initial velocity v0 .
3. A simple pendulum executes small oscillations in a resisting medium with
damping h'. P Determine the qualitative behavior of the motion and the loss
of energy after n oscillations.
4. Supposing the Earth to be spherical, with what velocity must a projectile be
fired from the Earth’s surface for its trajectory to be an ellipse with major axis
a?
5. A bomb is dropped from an airplane flying horizontally at height h and with
speed v. Assuming a linear law of resistance mgkv, show that, if k is small,
the fall time is approximately
s s !
2h 1 2h
1 C gk :
g 6 g
246 14 Dynamics of a Material Point

6. A particle moves in a plane under the influence of the central force

b c
f .r/ D 2
C 4;
r r

where a and b are positive constants and r is the distance from the center.
Analyze qualitatively the motion, prove the existence of a circular orbit, and
determine its radius.
7. Repeat the analysis required in the preceding exercise for a particle moving
under the action of the central force

f .r/ D kr 2 er :
2

8. A point moves under the action of a Newtonian attractive force. Resorting to

the stability theorems of Chap. 10 relative to the equilibrium stability, verify
that the elliptic orbits are stable. Are the motions along these orbits stable?
9. A particle moves in a plane attracted to a fixed center by a central force

k
f .r/ D  ;
r3

where k is a positive constant. Find the equation of the orbits.

10. Let  be a vertical parabola with equation y D x 2 . Determine the motion of a
particle P of mass m constrained to moving smoothly on  under the influence
!
of its weight and of an elastic force F D k QP , where k is a positive constant
and Q is the orthogonal projection of P onto the axis of the parabola.
Chapter 15
General Principles of Rigid Body Dynamics

15.1 Mass Density and Center of Mass

To improve the dynamical description of real bodies, we resort to a model of rigid

bodies that takes into account both the extension and the mass distribution in the
spatial region occupied by the body.
Henceforth we denote by R D .Ox1 x2 x3 / and R0 D .x10 x20 x30 / the lab frame
and the body frame, respectively. Further, we denote by ei and e0i , i D 1; 2; 3, the
unit vector directed along the orthogonal axes of R and R0 , respectively. Finally, C
is the fixed region occupied by the rigid body B in the body frame R0 , and r0 is the
position vector of any point of C relative to the origin  of R0 .
We suppose that there exists a positive and Riemann-integrable function .r0 / W
C ! RC such that the integral
Z
m.c/ D .r0 / dc (15.1)
c

gives the mass of the arbitrary measurable region c  C . The function .r0 /, which
is called the mass density of the body B, could exhibit finite discontinuity across
some surfaces contained in B.
We call the point G 2 C whose position vector rG relative to R0 is defined by
the formula Z
m.C /rG D .r0 /r0 dc (15.2)
C
the center of mass.
Exercise 15.1. Prove that the center of mass is a characteristic point of the rigid
body B, i.e., it is independent of the origin  of the body frame.

A. Romano, Classical Mechanics with Mathematica R

, Modeling and Simulation 247
in Science, Engineering and Technology, DOI 10.1007/978-0-8176-8352-8 15,
© Springer Science+Business Media New York 2012
248 15 General Principles of Rigid Body Dynamics

15.2 Linear Momentum, Angular Momentum, and Kinetic

Energy of a Rigid Body

The velocity field, which in a rigid motion has the form (12.17)

rP D rP  .t/ C !.t/  .r  r / D rP  .t/ C !.t/  r0 ;

can be considered as depending on the time t and the position vector r in the lab
frame or as a function of t and the position vector r0 in the body frame. In all of the
following formulae, it is supposed to be a function of t, r0 .
In this section, we extend to a rigid body the results of Sects. 13.6 and 13.7
relative to systems of material points. To this end, besides the lab and body frames
R and R0 , we need to introduce a third frame of reference RG D .Gx 1 x 2 x 3 / having
a translational motion relative to R with velocity rP G . As in Sect. 13.7, we call the
motion of B relative to RG the motion about the center of mass. Any quantity a
related to RG will be denoted by a.
We call linear momentum the vector
Z
QD Pr dc (15.3)
C

and angular momentum with respect to the pole O the vector

Z
!
KO D OP  rP dc: (15.4)
C

Since the region C does not depend on time owing to the rigidity of B,
differentiating (15.2) with respect to time, we obtain

Q D mPrG : (15.5)

We can comment on this result as follows. The momentum of B is equal to the

momentum of a mass point having the total mass m of B and moving with the
velocity of the center of mass rG .
Since in the frame RG the center of mass is at rest, we have that

Q D 0: (15.6)

We now prove the following important properties of KO .

Theorem 15.1. The angular momentum with respect to G in the frame R is equal
to the angular momentum with respect to G in the frame RG ; that is,

KG D KG : (15.7)
15.3 Tensor of Inertia 249

The angular momentum relative to RG does not depend on the pole; that is,

KO D KG (15.8)

for any pole O.

Proof. Since the motion of RG relative to R is translational, it is rP D rP C rP G .
Consequently, we have that
Z
!
KO D OP  .rP C rP G / dc
C
Z 
!
D KO C OP dc  rP G ;
C

and, taking into account (15.2), the following formula is proved:

!
KO D KO C mOG  rP G : (15.9)

Choosing O  G in this formula, we obtain (15.8). Further, the definition of angular

momentum allows us to write that
Z Z
! ! P !
KO D .OG C GP /  r dc D OG  rP dc C KG :
C C

In view of (15.6), the integral on the right-hand side of the preceding equation
vanishes and the theorem is proved. t
u
We conclude this section defining the kinetic energy of the body B as follows:
Z
1
T D Pr2 dc: (15.10)
2 C

It is trivial to extend (13.35) to the case of a rigid body so that

1
T D T C mPr2G : (15.11)
2

15.3 Tensor of Inertia

The quantities we introduced in the preceding section refer to any possible rigid
motion, but we have not yet used the explicit expression of the rigid velocity
field. In this section, we determine useful expressions of angular momentum and
kinetic energy that explicitly take into account the relation existing in a rigid motion
between the angular velocity and the velocity of any point.
250 15 General Principles of Rigid Body Dynamics

(1) Let B be a rigid body with a fixed point O in the frame of reference R. It is
always possible to take O coinciding both with the origin of the lab frame R and the
!
origin  of the body frame R0 . With this choice of the frames of reference, OP D r
and the velocity field is written as rP D !  r. Consequently, (15.4) gives
Z
KO D r  .!  r/ dc
C
Z
D .jrj2 !  .r  !/r/ dc:
C

In conclusion, we can state that

KO D IO  !; (15.12)

where the 2-tensor Z

IO D .jrj2 1  r ˝ r/ dc (15.13)
C
is said to be the tensor of inertia of the body B relative to point O. Formula (15.12)
shows that KO depends linearly on !. The components IOi i of IO are called
moments of inertia with respect to the axes ei , whereas the components IOij , i ¤ j ,
are the products of inertia. It is plain to verify that
Z Z
IO11 D .x2 C x3 /dc; IO22 D
2 2
.x12 C x32 /dc; (15.14)
C C
Z Z
IO33 D .x12 C x22 /dc; IOij D xi xj dc; .i ¤ j /; (15.15)
C C

so that IO is symmetric.
When a rigid motion with a fixed point O and its angular velocity ! are given, it
is possible to evaluate not only the angular momentum KO by (15.12) but also the
kinetic energy. In fact, introducing the velocity rP D !  r into (15.10), we obtain
Z Z
1 1
T D !  r  rP dc D r  rP  ! dc:
2 C 2 C

If we note that the angular velocity does not depend on the point of C and we take
into account (15.4) and (15.12), then the preceding relation can also be written as
1 1
T D KO  ! D !  IO !: (15.16)
2 2
The kinetic energy is a positive quantity (T  0) that vanishes if and only if the
velocity rP D !  r vanishes identically. In turn, this condition is equivalent to
requiring that ! D 0. In conclusion, in view of (15.16), we can state that the tensor
of inertia is symmetric and positive definite.
15.3 Tensor of Inertia 251

The quantities KO and ! in relations (15.12) and (15.16) are relative to the
lab frame R. However, since they are vector quantities, we can evaluate their
components in any frame of reference. Consistent advantages are obtained if (15.12)
and (15.16) are projected in the body frame R0 . In fact, with respect to the basis .e0i /
of the unit vectors directed along the axes of R0 , we have

0 1 0 0 0
KOi D Iij0 !ij0 ; T D I !! ; (15.17)
2 ij i j
where the components of the tensor of inertia are
Z
Iij0 D .r 02 ıij  xi0 xj0 / dc: (15.18)
C

In the body frame R0 the components Iij0 are independent of time since they only
depend on the geometry and mass distribution of the body B. On the other hand,
we have already noted that the 2-tensor IO is symmetric and positive definite.
Consequently, the eigenvalues A, B, and C of IO are positive, and there exists at
least an orthonormal basis .e0i / formed by eigenvectors of IO . These vectors are at
rest in R0 , i.e., relative to the body B. Any frame of reference that is formed by an
arbitrary point O and axes parallel to eigenvectors of IO is called a principal frame
of inertia relative to O, whereas the eigenvalues A, B, and C of IO are the principal
moments of inertia relative to O. In particular, if O is the center of mass of S , then
we substitute the attribute principal with central.
In a principal frame of inertia, the tensor of inertia is represented by the following
diagonal matrix:
0 1
A 0 0
@ 0 B 0 A; (15.19)
0 0 C
so that (15.12) and (15.16) become

KO D Ape01 C Bqe02 C C re03 ; (15.20)

1 2 
T D Ap C Bq 2 C C r 2 ; (15.21)
2
where the eigenvalues A, B, and C are constant.
We now determine the expressions of KO nd T when the solid B has a fixed axis
a or is free from any constraint.
(a) In the first case, provided that we choose the lab frame R and the body frame
R0 as in Fig. 12.2, we have that ! D 'eP 03 . Consequently, for any O 2 a, (15.18)
gives
0 0 0 0 0 0 1 0 2
KO D '.IP 13 e1 C I23 e2 C I33 e3 /; T D I33 'P : (15.22)
2
252 15 General Principles of Rigid Body Dynamics

In particular, if e03 is an eigenvector of IO belonging to the eigenvalue C , then

IO e03 D C e03

or, equivalently,
0 0 0 0
10 1 0 1
I11 I12 I13 0 0
@I0 I0 I0 A@0A D C @0A:
12 22 23
0 0
I13 I23 C 1 1
This equation implies that
0 0
I13 D I23 D 0;
and finally we can write (15.22) in the following simple form:

P 03 ;
KO D C 'e (15.23)

which holds if the rotation axis is a principal axis of inertia.

(b) Let B be a rigid body freely moving relative to the lab frame R D .O; ei /. Then,
in view of (15.4) and (15.7), the angular momentum KO relative to R is
!
KO D KG C OG  mPrG ; (15.24)
where KG is the angular momentum of S relative to the motion about the mass
center, i.e., relative to the frame RG D .G; ei /. Consequently, in the motion of
S with respect to RG the center of mass is at rest and the angular momentum KG
has the form (15.20), provided that we denote by A, B, and C the eigenvalues
of IG and we choose the basis .e0i / moving with S as a central base of inertia.
Bearing in mind the preceding remarks, we can write
!
KO D Ape01 C Bqe02 C C re03 C OG  mPrG ; (15.25)
1 2  1
T D Ap C Bq 2 C C r 2 C mPr2G : (15.26)
2 2

15.4 Some Properties of the Tensor of Inertia

At first sight, it seems that we have to evaluate the tensor of inertia IO every time
we change the fixed point O. In this section we show that there exists a simple
relation between the tensor IG and the tensor of inertia IO relative to any point O.
Further, the evaluation of IG becomes simpler when the rigid body B possesses some
symmetries.
Denoting the position vector r of any point P of B in the lab frame R by rG , the
position vector of the center of mass of B in R, and by r0 the position vector of P
0
in the body frame RO , we have that r D rG C r0 , and relation (15.13) can also be
written as follows:
15.4 Some Properties of the Tensor of Inertia 253

Z
IO D Œ.jrG j2 C jrj02 C 2rG  r0 /1  .rG C r0 / ˝ .rG C r0 / dc:
C

Since Z Z
IG D .jrG j2 1  r0 ˝ r0 / dc; r0 dc;
C C
we obtain the following formula:

IO D IG C m.jrG j2 1  rG ˝ rG /; (15.27)

which relates IG to the tensor of inertia IO relative to any point O.

Formula (15.27) allows us to prove the following theorem.
Theorem 15.2. Choose frames of reference .O; xi / and .G; x i / in R and RG ,
respectively, in such a way that the corresponding axes are parallel. The momenta
of inertia relative to the axes are modified according to the formula

IOi i D I Gi i C mıi2 ; (15.28)

where ıi is the distance between Oxi and Gx i ; moreover, the products of inertia
are modified according to the rule

IOij D I Gij  mxGi xGj ; i ¤ j: (15.29)

If the center of mass belongs to the axis Oxi , then

IOij D I Gij ; i ¤ j: (15.30)

Finally, if .G; x i / is a principal frame of inertia, then any frame .O; xi /, where
O 2 x i and the axes are parallel to the corresponding axes of .G; x i /, is a principal
frame of inertia.
Proof. Relations (15.28) and (15.29) follow at once from (15.27). Equation (15.30)
follows from (15.29) when we note that two coordinates of G vanish when one of
the axes Oxi contains G. Finally, (15.30) follows from (15.29) since, when .G; x i /
is a central frame of inertia, I Gij D 0 for i ¤ j . t
u
To determine IG , we arbitrarily fix a frame of reference .G; x i / and we evaluate
the components of IG in this frame by (15.14) and (15.15). Then we solve the
eigenvalue equation
I Gij uj D ui D ıij uj
to find the eigenvalues and eigenvectors of IG , i.e., the central frames of inertia.
The following theorem shows that the last problem can be simplified if the body
B exhibits some symmetries with respect to the center of mass.
254 15 General Principles of Rigid Body Dynamics

Theorem 15.3. If the plane  is a symmetry plane of B, then  is a principal plane

of inertia for any point of it.
Proof. Let .O; xi / be a frame of reference whose coordinate plane Ox1 x2 coincides
with . Since  is a symmetry plane of B, for any point r D .x1 ; x2 ; x3 / of B there
is a symmetric point r0 D .x1 ; x2 ; x3 / such that .r/ D .r0 /. Consequently,
Z Z
IO13 D x1 x3 dc D 0; IO23 D x2 x3 dc D 0;
C C

and the matrix of the components of IO becomes

0 1
I11 I12 0
@ I12 I22 0 A :
0 0 I33

It is a simple exercise to verify that an eigenvector of IO is orthogonal to  so that

the other eigenvectors belong to . t
u
The preceding theorem yields the following result.
Theorem 15.4. If the body B has two orthogonal planes of symmetry 1 and 2 ,
then they intersect along a straight line r, which is a principal axis of inertia for
any O 2 r. Further, a principal frame of inertia relative to O is obtained by
intersecting 1 and 2 with a third plane 3 that is orthogonal to the first two planes
and contains O.

15.5 Ellipsoid of Inertia

We define the moment of inertia of a rigid body B relative to the straight line a by
the quantity Z
Ia D ı 2 .r/ dc; (15.31)
C

where ı denotes the distance of an arbitrary point r of B from the straight line a
(Fig. 15.1). Denoting by u a unit vector directed along a, we have that

ı 2 D jrj2  jr  uj2 D jrjıij ui uj  xi xj ui uj

and recalling (15.18), relation (15.31) can also be written as follows:

Ia D u  IO u: (15.32)
15.5 Ellipsoid of Inertia 255

u
O
r

x x
a

Fig. 15.1 Ellipsoid of inertia

Finally, introducing the notation

u
Dp ; (15.33)
Ia

(15.32) assumes the form

  IO  D IOij i j D 1: (15.34)

The positive-definite character of the tensor of inertia IO implies that (15.33) defines
an ellipsoid E, which is called the ellipsoid of inertia relative to O. In particular,
if O  G, then E is called the central ellipsoid of inertia. In view of (15.32), the
moment of inertia of B relative to any axis a containing the center O of the ellipsoid
is given by the formula
1
Ia D : (15.35)
jj2
In other words, the ellipsoid of inertia relative to O allows one to determine
geometrically the moment of inertia of B with respect to any axis containing O.
Recalling the theorems we have proved regarding the dependence on O of the tensor
IO , we understand that knowledge of the central ellipsoid of inertia is sufficient to
determine the ellipsoid of inertia relative to any other point O. If the unit vectors
.ei / along the axes of the frame of reference .O; xi / are eigenvectors of IO , then
(15.35) becomes
Ax12 C Bx22 C C x32 D 1; (15.36)
where A, B, and C are the corresponding eigenvalues. When A D B, the ellipsoid
E is an ellipsoid of revolution about Ox3 ; finally, it reduces to a sphere when A D
B D C.
256 15 General Principles of Rigid Body Dynamics

We conclude this section by proving an important relation existing between the

ellipsoid of inertia IO relative to O and the angular momentum KO . If we introduce
the notation ! D j!ju, where u is a unit vector, then we can write [see (15.12)]
p u
KO D IO  ! D j!jIO u D Ia j!jIO p ;
Ia

where a is the axis containing O and parallel to !. In view of (15.33), the vector
 D puI is the position vector of the intersection point P of the ellipsoid E with
a
axis a; therefore, we have that
p
KO D Ia j!jIO :

On the other hand, from the equation f ./    IO   1 D 0 of E [see (15.34)]

we obtain .rf /P D 2IO , and the preceding equation assumes the following final
form:
1p
KO D Ia j!j.rf /P : (15.37)
2
Since .rf /P is orthogonal to E at point P , we can state that the angular momentum
KO is orthogonal to the plane tangent to the ellipsoid of inertia E at point P in which
the axis parallel to the angular velocity ! intersects the ellipsoid.

15.6 Active and Reactive Forces

The forces acting on a rigid body B can be divided into two classes: active forces
and reactive forces. A force belongs to the first class if we can assign a priori the
force law, i.e., the way in which the force depends on the position of B, its velocity
field, and time. These forces can be concentrated forces if they are applied at a
point of the body B and mass forces if they act on the whole volume occupied by B.
Consequently, the total force and the total torque relative to a pole O of the active
force can be written as
Z X
p
.a/
R .a/
D b dc C Fi ; (15.38)
C i D1
Z X ! p
.a/ ! .a/
MO D OP  b dc C OP i  Fi ; (15.39)
C i D1

.a/
where  is the mass density of B, b is the force per unit mass, and Fi , i D 1; : : : ; p,
are the concentrated forces acting at the points P1 ; : : : ; Pn of B.
The reactive forces are due to the contact between B and the external obstacles
(constraints). In turn, these forces are determined by the very small deformations
15.6 Active and Reactive Forces 257

that both B and the constraints undergo in very small regions around the contact
surfaces. The hypothesis that both B and the constraints are rigid eliminates the
preceding small deformations so that it is impossible to determine the force law of
the reactive forces. We suppose that the reactive forces can be distributed on the
surfaces 1 ; : : : ; m of B as well as on very small areas around the points Q1 ; : : : Qs
of contact between B and the constraints. Then we assume that the reactive forces
acting on these small areas can be described by a total force ˆ i and a couple with a
moment  i , which is called rolling friction. As a consequence of these assumptions,
we have that the total reactive force and torque can be written as follows:
m Z
X X
s
R .r/
D t d C ˆi ; (15.40)
i D1 i i D1
m Z
X X ! s
.r/ !
MO D OP  t dc C .OQi  ˆ i C  i /; (15.41)
1 i i D1

where t is the force per unit surface.

The vectors ˆ i and  i are not arbitrary since, although their force laws cannot
be assigned, they must satisfy some general phenomenological conditions. In this
regard, it is useful to distinguish between contact with friction and frictionless con-
tacts. In the latter case, starting from experimental results the following definition is
induced.
Definition 15.1. A constraint acting on a rigid body B is called ideal or smooth if
at the contact point Qi of B with the constraint surface †,  i D 0 and the reactive
force ˆ i exerted by † is normal to the boundary and directed toward the interior of
B. Moreover, t is orthogonal to † at any contact point.
If the constraints are rough, then friction is present and the preceding conditions
must be replaced by more complex ones. Moreover, they depend on whether B is
moving or at rest relative to the constraints. The conditions that are satisfied by the
reactive forces are called Coulomb’s friction laws.
For the sake of simplicity, we refer these empirical laws to the case of concen-
trated reactive forces since its extension to surface reactive forces is obvious. It
is convenient to distinguish in ˆ a component ˆ ? normal to the contact surface
†, which is always present and corresponds to the frictionless situation, and a
component ˆ k tangential to †. With similar meanings, we set  D  ? C  k ,
where the components  ? and  k are respectively called spin and rolling spin.
All the aforementioned reactive forces are modeled by the following empirical
laws.
Definition 15.2. Rough constraints in static conditions are such that
258 15 General Principles of Rigid Body Dynamics

jˆ k j  s jˆ ? j; (15.42)
j k j  hs jˆ ? j; (15.43)
j ? j  ks jˆ? j; (15.44)

where the constants 0 < s ; hs ; ks < 1, depending on the contact between the body
and the constraints, are called static friction coefficients.
Definition 15.3. Rough constraints in dynamic conditions are such that

rP
jˆ k j D  d jˆ ? j ; (15.45)
jPrj
!k
j k j D  hd jˆ ? j ; (15.46)
j!k j
!?
j ? j D  kd jˆ ? j ; (15.47)
j!? j

where rP is the relative velocity of the contact point of B with respect to the con-
straint, and !k and !? denote respectively the tangential and normal components
of the relative angular velocity ! of B with respect to the constraint. The coefficients
0 < d < s < 1, hd , and kd have the same meaning as in the static case but are
now called the dynamic friction coefficients.

15.7 Balance Equations for a Rigid Body: Theorem of

Kinetic Energy

In the preceding section, the characteristics of the forces acting on a rigid body
B were given. In this section, we introduce the balance equations governing its
dynamics.
It is reasonable to assume that, owing to the hypothesis of rigidity, the internal
forces, acting between parts of B, do not influence the motion. Consequently, they
must not appear in the dynamic equations describing the effect of the acting forces
on the motion of B. Now, the equations possessing this characteristic are the balance
equations of linear and angular momentum, (13.28) and (13.32). Although these
equations were proved for a system of mass points, it is quite natural to require
their validity for a rigid body that is a collection of the elementary masses  dc.
Therefore, we postulate that the dynamical behavior of a rigid body B is governed
by the following balance equations:

mrRG D R.a/ C R.r/ ; (15.48)

KPO D MO C MO ;
.a/ .r/
(15.49)
15.7 Balance Equations for a Rigid Body: Theorem of Kinetic Energy 259

where m is the total mass of B and G its center of mass, O coincides with G or with
an arbitrary fixed pole, and the angular momentum is expressed by (15.25). In the
next chapter, we prove that the preceding balance equations allow us to determine
the motion of a rigid body B when it has a fixed axis or a fixed point or is free of
constraints.
We conclude this section by proving an important theorem that holds during the
motion of any rigid body regardless of the constraints to which it is subjected.
Let us suppose that B moves relative to the frame of reference .O; xi / under the
action of the forces F1 ; : : : ; Fs applied at the points P1 ; : : : ; Ps of B. Then, taking
into account that the motion of B is rigid, the power W of these forces is

X
s X
s
!
W D Fi  rP i D Fi  .Pr C !  P i /
i D1 i D1
! !
X
s X
s
!
D Fi  rP  C Pi  Fi  !;
i D1 i D1

where  is an arbitrary point of B and ! is the angular velocity of the rigid motion.
Introducing the total force R and torque MO acting on B, we can put the preceding
relation in the form
W D R  rP  C MO  !;P (15.50)
which shows that the power of the forces acting on B depends only on their total
force and torque.
Denoting by T the kinetic energy of B, we prove the following formula, which
expresses the theorem of kinetic energy:

TP D W: (15.51)

First, we choose O  G in (15.50). Then, by adding the scalar product of (15.48)

by rP  and the scalar product of (15.49) by ! and taking into account (15.50), we
obtain  
d 1 P G  ! D W .a/ C W .r/ :
mjPrjG C K
2
(15.52)
dt 2
On the other hand, in a central frame of inertia .G; e0i /, ! D pe01 C qe02 C re03 , and,
in view of (15.25), we have that

P G D Ape
K P 01 C B qe
P 02 C C rP e03 C !  KG :

Recalling (15.26), we finally obtain

P G  ! D Ap pP C Bq qP C C r rP D dT ;
K
dt
and (15.51) is proved.
260 15 General Principles of Rigid Body Dynamics

15.8 Exercises

1. A cannon resting on a rough horizontal plane is fired, and the muzzle velocity of
the projectile with respect to the cannon is v. If m1 is the mass of the cannon and
m2 the mass of the projectile, the mass of the powder being negligible, and is
the coefficient of friction, show that the distance of recoil of the cannon is
 2
m2 v 1
;
m1 C m2 2 g

where g is the gravity acceleration.

2. Two men, each of mass m2 , are standing at the center of a homogeneous
horizontal beam of mass m1 that is rotating with uniform angular velocity !
about a vertical axis through its center. If the two men move to the ends of the
beam and !1 is then the angular velocity, show that
m1
!1 D !:
m1 C 6m2

3. If a shell at rest explodes and breaks into two fragments, show that the two
fragments move in opposite directions along the same straight line with speeds
that are inversely proportional to their masses.
4. Prove that, if is the coefficient of friction, the couple necessary to set into
rotation a right circular cylinder of radius r and weight mg that is standing on its
base on a rough horizontal plane is 23 mgr .
5. A uniform disk of radius a that is spinning with angular speed ! about a vertical
axis is placed upon a horizontal table. If the coefficient of the friction between
the surface of the disk and the plane is , determine when the disk stops.
6. A heavy bar’s end points A and B are moving, respectively, along the axes Ox
and Oy of a vertical plane . Neglecting the friction of the constraints, determine
the equations of motion of the bar and the reactive forces at A and B using the
balance equations.
Chapter 16
Dynamics of a Rigid Body

16.1 Rigid Body with a Smooth Fixed Axis

Consider a rigid body B with a fixed axis that should be smooth. Figure 16.1 shows
two devices forming a fixed axis r: two spherical hinges and a spherical hinge
coupled with a cylindrical one. If the contact surfaces between the moving parts
of these devices are smooth, then the reactive forces are orthogonal to them and, at
least in the devices shown, their application straight lines intersect the fixed axis r.
.r/
Consequently, the component M3 of the total reactive force vanishes:

.r/
M3 D 0: (16.1)

Henceforth we will assume this condition as being characteristic of any smooth

device that gives rise to a fixed axis.
The lab and body frames are chosen as shown in Fig. 16.2. It is evident that the
motion of B is determined by a knowledge of the function '.t/, which gives the
angle between the fixed plane Ox1 x3 and the plane Ox10 x30 moving with the body.
To find an equation containing the function '.t/, we note that, since the axis
r could not be a principal axis of inertia, we cannot assume that the body frame
is a principal frame of inertia. Therefore, the angular momentum KO is given by
(15.22). Introducing this expression into the balance of angular momentum (15.49),
projecting this equation on r, and taking into account (16.1), we obtain that

.a/
I33 'R D M3 ; (16.2)

where I33 is the moment of inertia of B relative to the axis r. It is possible to

determine, at least in principle, the function '.t/ by (16.2), provided that we recall
that active forces depend on the position of B, its velocity field, and time. But the

A. Romano, Classical Mechanics with Mathematica R

, Modeling and Simulation 261
in Science, Engineering and Technology, DOI 10.1007/978-0-8176-8352-8 16,
© Springer Science+Business Media New York 2012
262 16 Dynamics of a Rigid Body

Fig. 16.1 Devices for a fixed axis

Fig. 16.2 Solid rotating about a fixed axis

position of B is determined by the angle ', and the velocity field is given by the
!
relation rP D 'e
P 3  r, where r D OP . Consequently, (16.2) assumes the final form

.a/
I33 'R D M3 .'; ';
P t/; (16.3)

which determines one and only one motion satisfying the initial conditions (t D 0)

'.0/ D '0 ; P
'.0/ D 'P0 : (16.4)
16.2 Compound Pendulum 263

16.2 Compound Pendulum

A compound pendulum is a heavy rigid body B rotating about a smooth horizontal

axis Ox3 , which is called the suspension axis (Fig. 16.3). Let O be an orthogonal
projection of the center of mass G of B on the suspension axis, and let ' be the
angle between OG and the vertical straight line containing O. Finally, we denote
by h the length of OG, by m the mass of B, and by g the gravity acceleration.
Since the active force reduces to the weight, which is equivalent to the single force
mg applied at the center of mass, we have that the component of the active torque
becomes M3 D mgh sin ', and (16.3) assumes the following form:

g
'R D  sin '; (16.5)
l

where

I33
lD (16.6)
mh

is called the reduced length of the pendulum. Equation (16.6) shows that a
compound pendulum moves as a simple pendulum whose length is equal to the
reduced length of the compound pendulum.

x h

mg

x’
x

Fig. 16.3 Compound pendulum

264 16 Dynamics of a Rigid Body

16.3 Solid with a Fixed Point: Euler’s Equations

Let B be a rigid body constrained to moving around a smooth fixed point O. If the
constraint is realized with a spherical smooth hinge, then the reactive forces are
orthogonal to the contact spherical surface. Consequently, their application straight
lines contain point O and the total reactive torque relative to O vanishes:

.r/
MO D 0: (16.7)

We use the preceding condition as a definition of a smooth fixed point, regardless of

the device realizing the fixed point O.
The position of B is determined by the Euler angles , ', and  formed by the
body frame .O; e0i / with the lab frame .O; ei /. Consequently, we need to find three
differential equations that can determine these variables as functions of time. In view
of (16.7), the balance of angular momentum assumes the form

P O D M.a/ :
K (16.8)
O

.a/
The torque MO of the active forces must depend on the position of B, its velocity
!
field, and time. Since the velocity of any point r D OP is given by the formula
rP D !  r, where ! is the angular velocity of the rigid motion, we can state that

.a/ .a/
MO D MO . ; '; ; p; q; r; t/; (16.9)

where p, q, and r are the components of ! in the body frame. On the other hand,
we have proved that the angular momentum relative to the lab frame of a solid with
a fixed point, when it is projected onto the body frame, is given by (15.20). The time
derivative of KO , in view of (12.42), is given by

P O D Ape
K P 01 C B qe
P 02 C C rP e03 C !  KO : (16.10)

.a/ P O into the balance equation

Introducing expressions (16.9) and (16.10) of MO and K
of angular momentum, we obtain the following equation for a solid with a smooth
fixed point:

P 01 C B qe
P 02 C C rP e03 C !  KO D MO . ; '; ; p; q; r; t/:
.a/
Ape (16.11)

Finally, projecting this equation along the axes of the body frame (which are
principal axes of inertia relative to the fixed point O), we obtain the Euler equations
16.4 Poinsot’s Motions 265

ApP  .B  C /qr D MOx1 . ; '; ; p; q; r; t/; (16.12)

B qP  .C  A/rp D MOx2 . ; '; ; p; q; r; t/; (16.13)
C rP  .A  B/pq D MOx31 . ; '; ; p; q; r; t/: (16.14)

These three equations, since they contain the six unknowns .t/, '.t/, .t/, p.t/,
q.t/, and r.t/, cannot determine the motion of B. However, if we take into account
the Euler kinematic relations (12.32)–(12.34),

P D 1
.p sin ' C q cos '/; (16.15)
sin 
'P D .p sin ' C q cos '/ cot  C r; (16.16)
P D p cos '  q sin '; (16.17)

we obtain a system that, under general hypotheses of the regularity of the functions
appearing on the right-hand side of (16.12)–(16.14), admits one and only one
solution satisfying the initial data

.0/ D 0; '.0/ D '0 ; .0/ D 0 ; (16.18)

p.0/ D p0 ; q.0/ D q0 ; r.0/ D r0 ; (16.19)

which correspond to giving the initial position and velocity field of B.

The nonlinear equations (16.12)–(16.17) exhibit a very complex form. It is
utopian to search for explicit solutions of them, even in simple cases. In the
following sections, we analyze some simple but interesting situations in which we
do not succeed in finding explicit solutions, but, by a qualitative analysis, we obtain
meaningful information about the behavior of the motion.

16.4 Poinsot’s Motions

We call free rotations or Poinsot’s motions all motions of a solid B with a smooth
.a/
fixed point O in the absence of the total active torque MO :

.a/
MO D 0: (16.20)

In other words, these motions correspond to the solutions of Euler’s equations when
their right-hand sides vanish. Also in this case, it is not possible to find the closed
form of solutions. However, Poinsot (1851) proposed a very interesting geometric
qualitative analysis of free rotations. As a result of this analysis, we are able to
determine the configurations that B assumes during the motion, although we are
not able to give the instant at which B occupies one of them.
266 16 Dynamics of a Rigid Body

The preceding analysis is based on the following conservation laws of angular

momentum and kinetic energy:

KO D KO .0/; T D T .0/; (16.21)

which follow at once from (15.49) and (15.51) and hypothesis (16.20).
Let EO be the ellipsoid of inertia of B relative to point O, and denote by r.t/ the
instantaneous rotation axis, that is, the straight line parallel to the angular velocity
!  j!ju and containing O. Finally, we denote by P .t/ the intersection point of
r.t/ and E (Fig. 16.5).
We have proved [see (15.37)] that the angular momentum KO is orthogonal to the
plane  tangent to the ellipsoid of inertia at the point P .t/. Since, in view of (16.21),
KO is a constant vector, we can state that during motion this tangent plane remains
parallel to itself. Now we prove that the conservation of kinetic energy implies that
the distance h of  from O is constant during the motion. First, we note that

! KO
h D OP  : (16.22)
jKO j

Further, the point P .t/ 2 E confirms the equation of the ellipsoid so that, in view
of (15.33), we can write

u KO !  KO 2T
hD p  D p D p : (16.23)
Ir jKO j j!j Ir jKO j j!j Ir jKO j

On the other hand, we have

2T D !  IO  ! D j!j2 u  IO  u D j!j2 Ir ;

and (16.23) can be written in the form

p
2T .0/
hD ; (16.24)
jKO .0/j

which shows that h is constant.

In conclusion, during the motion of B, the plane  tangent to the ellipsoid of
inertia E at point P , where P is the intersection of the instantaneous axis of
rotation with E, does not change its position. To understand the consequences of this
result on the motion of B, we assign the initial angular velocity !.0/. This vector
determines the initial values of KO and TO and, consequently, owing to (16.24),
the fixed position of the plane  relative to O and the initial position P .0/ of P .t/.
16.4 Poinsot’s Motions 267

r(t)
E

KO
O
polhode
h

P( t) herpholode
fixed plane

Fig. 16.4 Polhode and herpolhode

T
Since P .t/ 2 E r.t/, its velocity is equal to zero. Finally, we can state that during
a Poinsot motion of B about O, the ellipsoid of inertia E rolls, without slipping, on
the fixed plane .
During motion, the point P .t/ describes a curve on E, which is called a polhode,
and a curve on , known as a herpolhode (Fig. 16.4). It is possible to prove that
the polhode is a closed curve on E, which may reduce to a point. In contrast, the
herpolhode could be an open curve. More precisely, consider the arc  described by
P .t/ during a complete turn on the ellipsoid E, and let ˛ be the angle subtended by
 having the vertex in the orthogonal projection A of O on . If ˛ D 2.m=n/,
where m and n are integers, then the herpolhode is closed; otherwise it is open. It
is also possible to prove that the herpolhode is contained in an annulus  whose
center is A, is always concave toward A, and, when it is open, is everywhere dense
in . In particular, if the initial angular velocity !.0/ is parallel to a principal axis
of inertia, then the angular momentum KO is always parallel to !.0/ and h is equal
to the half-length of the axis of E. Consequently, the polhode and the herpolhode
reduce to a point. If the ellipsoid E has a symmetry of revolution around an axis
a, then the polhode and the herpolhode become two circles. Finally, if E reduces
to a sphere, the polhode and the herpolhode reduce to a point and the motion is
always a uniform rotation. With the notebook MotiPoin it is possible to simulate all
the preceding cases. Figure 16.5 shows a herpolhode obtained with this notebook
corresponding to the data A D 1, B D 1:5, C D 0:5, r0 D 3, and .0/ D =4 in an
interval of motion of 15 s.
268 16 Dynamics of a Rigid Body

0.6

0.4

0.2

0.6 0.4 0.2 0.2 0.4 0.6

0.2

0.4

0.6

Fig. 16.5 Herpolhode

16.5 Uniform Rotations

Let B be a solid with a smooth fixed point O. In this section, we prove that, among
all the possible Poinsot motions, there are uniform rotations about certain fixed axes
containing O. First we note that the condition ! D const: must be considered in
the lab frame in which we are searching for uniform rotations. However, in view of
(12.42), we have
da ! dr ! dr !
D C!!D : (16.25)
dt dt dt
In other words, if a uniform rotation is possible in the lab frame, then this rotation
is also uniform in the body frame, i.e., the components p, q, and r of ! along the
axes of a principal frame of inertia in the body are constant. In view of Euler’s
equations (16.12)–(16.14), these components must satisfy the following algebraic
system:

.B  C /qr D 0; (16.26)
.C  A/rp D 0; (16.27)
.A  B/pq D 0: (16.28)

To find the solutions of this system, we distinguish three cases:

16.6 Poinsot’s Motions in a Gyroscope 269

• If the eigenvalues A, B, and C differ from each other, then system (16.26)–
(16.28) becomes
qr D pr D pq D 0;
and its solutions are

fp D p0 ; q D r D 0g; fp D r D 0; q D q0 g; fp D q D 0; r D r0 g:

• If A D B ¤ C , then system (16.26)–(16.28) becomes

pr D qr D 0;

and its solutions are

fp D p0 ; q D q0 ; r D 0g; fp D q D 0; r D r0 g:

• If A D B D C , then the solutions are given by arbitrary values of p, q, and r.

We note that in the first case, there is a single body frame formed by principal axes
of inertia; in the second case, any body frame formed by the eigenvectors belonging
to the axis Ox30 and any pair of orthogonal axes that are also orthogonal to Ox30
form a principal frame of inertia. Finally, if A D B D C , then any lab frame is a
principal frame. We can summarize the preceding results as follows: if a uniform
rotation is realized, then its axis of rotation coincides necessarily with a principal
axis of inertia of the tensor of inertia IO .
One of these rotations can be obtained by assigning the initial angular velocity
!.0/ directed along a principal axis of inertia. Thus, we have just proved that a
uniform rotation !.t/ D !.0/ is possible about this axis. On the other hand, we
know that there is only one solution satisfying the assigned initial data. Therefore,
this uniform rotation is the only possible motion satisfying the initial data.
We conclude this section by recalling a fundamental result relative to uniform
rotations about principal axes of inertia. If the eigenvalues A, B, C of the tensor
of inertia IO relative to the smooth fixed point O are assumed to satisfy the
condition A < B < C , then the uniform rotations about the principal axis
of inertia corresponding to A and C are stable, whereas the uniform rotations
about the principal axis corresponding to the eigenvalue B is unstable. The axes
corresponding to A and C are said to be principal steady axes.

16.6 Poinsot’s Motions in a Gyroscope

A gyroscope B is a rigid body whose central tensor of inertia IG has a double

eigenvalue .A D B/, which differs from the remaining simple eigenvalue C .
The one-dimensional eigenspace a corresponding to the eigenvalue C is called a
gyroscopic axis. Figure 16.6 shows a gyroscope whose fixed point O is obtained
270 16 Dynamics of a Rigid Body

Fig. 16.6 Gyroscope with its suspension device

with a complex structure (cardan joint) that allows for the complete mobility of B
about O.
From the properties we proved about the tensor of inertia in Chap. 15 we know
that a body frame whose axes are principal axes of inertia can be obtained by adding
to the axis Ox30 D a any arbitrary pair of axes Ox10 and Ox20 that are orthogonal
to each other and to Ox30 . Moreover, any other frame .x100 ; x200 ; x300 /, where  2 a
and Ox10 is parallel to x100 , Ox20 is parallel to x200 , and Ox30  x100 , is again a
principal frame of inertia. Finally, since A D B for the central tensor of inertia, the
corresponding ellipsoid of inertia E has a symmetry of revolution about a, and these
properties hold for all the tensors of inertia and corresponding ellipsoids relative to
any other point of a.
After recalling these properties, we are now in a position to evaluate Poinsot’s
motions of a gyroscope with a smooth fixed point O belonging to the gyroscopic
axis a. Since we have A D B and MO D 0, from the third Euler equation (16.14)
we derive the following conservation law:

r D r0 ; (16.29)

which states that in a free rotation of a gyroscope the component of the angular
velocity along the gyroscopic axis is constant. Further, the angular momentum is
constant. Collecting these results, we can write that
 
KO D A pe01 C qe02 C C r0 e03 D KO .0/; (16.30)
16.7 Heavy Gyroscope 271

where the unit vectors e0i are directed along the principal axes of the body frame. By
the preceding relation, we can write the angular velocity ! D pe01 C qe02 C re03 in
the following form:
! D !1 C !2 ; (16.31)
where  
C 1
!1 D 1  r0 e03 ; !2 D KO : (16.32)
A A
The preceding relations show that any Poinsot motion of a gyroscope is a spherical
motion obtained by composing a uniform rotation about a fixed axis containing O
and parallel to KO , and a uniform rotation about the gyroscopic axis.

16.7 Heavy Gyroscope

In this section we consider a gyroscope B with a fixed smooth point O, O ¤ G,

of its gyroscopic axis that is acted upon by its weight p (Fig. 16.7). The balance of
angular momentum yields
KP O D hG e03  p; (16.33)
where hG is the abscissa of the center of mass G along the gyroscopic axis Ox30 .
We now prove that the three Euler equations can be substituted, to great advantage,
by three conservation laws.

x
x’

x’

G
h
G

e’
p x
e’ e’

n
x’

Fig. 16.7 Heavy gyroscope

272 16 Dynamics of a Rigid Body

First, since A D B and the component Mx30 D hG e03  e03 D 0, the third Euler
equation still implies the first conservation law r D r0 .
Second, the total energy is constant. In fact, in view of (15.50), we have that the
power of the reactive forces vanishes,

.r/
W .r/ D R.r/  rP 0 C MO  ! D 0;

.r/
since rP 0 D 0 and MO D 0. On the other hand, the power W .a/ of the active forces
is given by
d d
p  rP G D .p  rG / D  .phG cos /:
dt dt
Taking into account the theorem of kinetic energy (15.51), we obtain conservation
of the total energy E:
T C phG cos  D E; (16.34)
which, by expression (15.21) of the kinetic energy of a solid with a fixed point and
the conditions A D B and r D r0 , assumes the form
 
A p 2 C q 2 C C r02 C 2phG cos  D 2E: (16.35)

Finally, the component M3 of the active torque along the vertical axis Ox3
vanishes since M3 D hG e03  p  e3 D 0. Consequently, the component KO  e3
of the angular momentum along the vertical axis Ox3 is constant. If we denote by
i the components of e3 in the body frame, this last conservation law can be written
as follows:
A.p1 C q2 / C C r0 3 D K3 .0/: (16.36)
The algebraic equations (16.35) and (16.36) in the unknowns p and q can replace
Euler’s equations to great advantage. However, (16.36) contains the components i
of e3 along the axes of the body frame. In terms of Euler’s angles, these components
are given by the following relations:

1 D sin  sin '; (16.37)

2 D sin  cos '; (16.38)
3 D cos : (16.39)

The presence of Euler’s angles in these relations requires the use of the Euler
kinematic relations (Sect. 12.4):

p D P sin  sin ' C P cos '; (16.40)

q D P sin  cos '  P sin '; (16.41)
r0 D P cos  C ':
P (16.42)
16.7 Heavy Gyroscope 273

In conclusion, the motion of a heavy gyroscope B can be determined by solving the

system of eight equations (16.35)–(16.42) in the unknowns p.t/, q.t/, 1 .t/, 2 .t/,
3 .t/, .t/, '.t/, and .t/.
From (16.40) and (16.41) we obtain that

p 2 C q 2 D P 2 sin2  C P 2 : (16.43)

Introducing this equation into (16.35) and relations (16.37)–(16.41) into (16.36) and
taking into account (16.42), we can write the following system:

P 2 sin2  C P 2 D ˛  a cos ; (16.44)

P sin  D ˇ  br0 cos ;
2
(16.45)
P cos  C 'P D r0 ; (16.46)

where

2E  C r02 2phG
˛D ; aD ; (16.47)
A A
K3 .0/ C
ˇD ; bD > 0: (16.48)
A A

From (16.46) we obtain

P D ˇ  br0 cos  ; (16.49)
sin2 
and substituting this result into (16.44) we arrive at the system

.ˇ  br0 cos /2 C P 2 sin2  D .˛  a cos / sin2 ; (16.50)

P D ˇ  br0 cos  ; (16.51)

sin2 
'P D r0  P cos : (16.52)

The first of the preceding equations can be written in a more convenient form by
introducing the new variable u D cos . In fact, since

uP D P sin ; sin2  D 1  u2 ; (16.53)

system (16.50)–(16.52) assumes the final form

274 16 Dynamics of a Rigid Body

f(u)

- u u
u

Fig. 16.8 Behavior of function f .u/

uP 2 D .˛  au/.1  u2 /  .ˇ  br0 u/2  f .u/; (16.54)

P D ˇ  br0 u ; (16.55)
1  u2
ˇ  br0 u
'P D r0  u: (16.56)
1  u2
A complete integration of the preceding system is impossible without resorting
to numerical methods. However, a qualitative analysis of the solution can be
developed.
Consider the unit sphere † (Fig. 16.7) with its center at O and the curve  that
the gyroscopic axis Ox30 draws on † during the motion. When we note that both
axes Ox3 and Ox30 are orthogonal to the line of nodes, identified in Fig. 16.7 by
the unit vector n, we deduce that the angle between the plane Ox3 x30 and the fixed
axis Ox1 results in ı D  =2. Consequently, in spherical coordinates, the curve
 can be described in terms of the angles ı and . Remembering that u D cos ,
(16.54) and (16.55) represent a system of differential equations in the unknown
parametric equations of the curve , i.e., in the functions ı.t/ and .t/. Finally,
(16.56) describes how the gyroscope rotates about its axis.
We do not prove that the relations existing among the quantities ˛, ˇ, a, and b
imply that the three roots of the equation f .u/ D 0 are real. In view of (16.54) and
the meaning of u, we must have

f .u/  0; 1  u  1: (16.57)

On the other hand, since a > 0, we also have (Fig. 16.8)

lim f .u/ D 1; lim f .u/ D 1I (16.58)

u!1 u!1

f .1/ D .ˇ C br02 /; f .1/ D .ˇ  br02 /: (16.59)

16.7 Heavy Gyroscope 275

The preceding conditions imply that the two acceptable roots u1 and u2 of the
equation f .u/ D 0 belong to the interval Œ1; 1, that is,

 1  u2  u1  1: (16.60)

Equivalently, since u D cos , we can state that during the motion of the gyroscope
the angle  satisfies the condition

1    2 : (16.61)

This shows that the curve  described on the unit sphere † is always contained
between the parallel  D 1 and  D 2 . In particular, if the initial data are such
that 1 D 2 , then  reduces to the parallel  D 1 D 2 and, in view of (16.55)
and (16.56), both P and 'P are constant. In other words, if 1 D 2 , the gyroscopic
axis moves around the vertical with constant slope 1 , constant precession velocity
P , and proper angular velocity '.P
If 1 ¤ 2 , then the motion of the gyroscopic axis is called nutation. Denoting

ˇ
uD ;  D arcsin u; (16.62)
br0

we have the following three possibilities:

•  … Œ1 ; 2 ;
then (16.55) shows that P always has the same sign. Moreover, for  D 1 and
 D 2 , P D 0 and  is tangent to the aforementioned parallels.
•  2 .1 ; 2 /;
then P D 0 when  D  and the curve  is perpendicular to the parallel  D
 . Moreover, P has a definite sign when  2 .1 ;  / and an opposite but still
definite sign if  2 .; 2 /. This means that .t/ increases in the first interval
and decreases in the second one or vice versa. Moreover, for  D 1 and  D 2 ,
we have P D 0 and the curve  is tangent to the parallels  D 1 and  D 2 .
•  D ;
for  D 2 both P and P vanish, but P has a constant sign at any other point. It
is possible to prove that  exhibits cusps when  D  D 1 .

We omit proving that the case 2 D  is impossible.

The previously listed cases are shown in Figs. 16.9–16.11, which were obtained
by the notebook GyroPes and refer to a gyroscope for which p D hG D 1. Their
captions contain the relative initial data.
276 16 Dynamics of a Rigid Body

Fig. 16.9 A D B D 1, C D 0:5, p.0/ D 0:3, q0 D 0:5, r.0/ D 3:5, .0/ D =4

Fig. 16.10 A D B D 1, C D 1:2, p.0/ D 0:3, q0 D 0:5, r.0/ D 3:5, .0/ D =4

16.8 Some Remarks on the Heavy Gyroscope

In this section, we analyze in detail the motion of a gyroscope corresponding to the

following initial conditions:

0 < 0 < ; p0 D q0 D 0; r0 > 0: (16.63)

In other words, an initial rotation r0 is impressed about the gyroscopic axis that
forms an angle 0 with the vertical. With the preceding initial conditions, taking
into account (16.35), the constants (16.47) and (16.48) become
16.8 Some Remarks on the Heavy Gyroscope 277

Fig. 16.11 A D B D 1, C D 1:2, p.0/ D q0 D 0, r.0/ D 2, .0/ D =4

2E  C r02 2P hG
˛D D cos 0 D au0 ; (16.64)
A A
ˇ D br0 cos 0 D br0 u0 : (16.65)

Introducing these values of ˛ and ˇ into (16.54), we obtain

 
uP 2 D .u0  u/ a.1  u2 /  b 2 r02 .u0  u/  f .u/: (16.66)

Since the polynomial f .u/ admits the root

u1 D u0 ) cos 1 D cos 0 ; (16.67)

the motion of the gyroscope corresponds to case .c/ of the previous section. The
other solution u2 in the interval Œ1; 1 is the root of the equation

a.1  u22 / D b 2 r02 .u0  u2 /; (16.68)

from which, in view of (16.67), we derive the following expression for the difference
of the two roots:
a.1  u22 /
u1  u2 D : (16.69)
b 2 r02
The preceding difference is positive since u1 ; u2 2 Œ1; 1, so that we have u2 < u1 .
But during the motion, u1  u  u2 , and consequently u1  u  u1  u2 . Bearing in
mind these remarks, (16.55) supplies
u1  u u1  u2
0  P D br0  br0 :
1u 2 1  u2
278 16 Dynamics of a Rigid Body

Fig. 16.12 A D B D 1, C D 0:5, p.0/ D 0:3, q0 D 0:5, r.0/ D 3:5, .0/ D =4

Recalling (16.69), from the preceding condition we obtain that

a
0 P ' : (16.70)
br0

The results (16.69) and (16.70) illustrate the following points.

• The nutation angle is proportional to 1=r02 ; in particular,  ! 0 when r0 ! 1.
• The velocity of precession is proportional to 1=r0.
We do not prove that jjP is proportional to r0 .
Finally, the balance of energy (16.35) can be written as

A.p 2 C q 2 / C C r02 C 2phG u D C r02 C 2phG u0 ;

and taking into account (16.69), we attain the result

a.1  u22 /
A.p 2 C q 2 / D 2phG .u0  u/ < 2phG .u1  u2 /  :
b 2 r02

This inequality allows us to state that

 
1
KO D C r0 e03 C O : (16.71)
r0

The preceding remarks prove that, for the initial conditions (16.63) and high
values of r0 , the motion of a gyroscope is approximatively a slow regular precession
around a vertical. Since jj P increases with r0 , during the low precession, the
gyroscopic axis vibrates with high frequency (Fig. 16.12).
16.8 Some Remarks on the Heavy Gyroscope 279

This result can also be obtained supposing that the motion of the heavy
gyroscope, corresponding to the initial data (16.63) and high values of r0 , is
described by the equation [see (16.71)]

C ro eP 03 D hG e03  p: (16.72)

In fact, denoting by ! the angular velocity of the gyroscope and recalling (12.41),
the preceding equation can also be written as
 
hG
! p  e03 D 0;
C r0

so that
hG
! D e03 C p;
C r0
where the approximate value of is

hG
D r0 C p  e03 ' r0 :
C r0

Finally, we obtain that the angular velocity is given by

hG
! D r0 e03  p: (16.73)
C r0

In conclusion, substituting the balance equation of momentum with (16.72), we de-

termine with sufficient approximation the position of the gyroscopic axis. However,
the velocity field is not well approximated since the fast vibrations of the gyroscopic
axis are neglected.
We say that for any solid B with a smooth fixed point O, the principle of the
gyroscopic effect holds if, substituting the balance of angular momentum with the
equation
C r0 eP 03 D MO ; (16.74)
we determine, with sufficient approximation, the configurations that B assumes
during its motion. The preceding considerations prove that such a principle holds
for a heavy gyroscope.
Many references on this subject can be found in [34] where many interesting
cases are analyzed in which the principle of the gyroscopic effect holds, at least
when the initial rotation is impressed about a steady principal axis of inertia
(Sect. 16.5) and the condition
MO  e03 D 0
is satisfied. The reader interested in analyzing the behavior of any body under the
preceding conditions can resort to the notebook solid.
280 16 Dynamics of a Rigid Body

16.9 Torque of Terrestrial Fictitious Forces

In the previous sections, we considered the motion of a solid B with a smooth fixed
point when the torque MO relative to O of the acting forces vanishes or is equal
to the torque of the weight. It is evident that the lab frame Ox1 x2 x3 , in which we
analyze the motion of B, is at rest with respect to the Earth, i.e., Ox1 x2 x3 is a
terrestrial frame. This implies that, in the torque of the acting forces, we should
also include the torque of the fictitious forces due to the noninertial character of the
terrestrial frame.
In this section, we evaluate the torque of these forces with respect to the center
of mass G of a gyroscope B. Let Ox1 x2 x3 be the lab frame, which is at rest relative
to the Earth. If we analyze the motion of B during a time interval which is much
shorter than a month, we can suppose that the motion of the lab frame with respect
to an inertial frame is a uniform rotation with angular velocity !T . Then, if C is the
region occupied by B and
the mass density of B, then the torque relative to G of
the fictitious forces due to the Earth’s rotation is
Z
.t / ! !
MG D !T 2

GP  QP dc; (16.75)
C

where Q is the orthogonal projection of the point P 2 B on the rotation axis a of

!
the Earth. Denoting by Q the orthogonal projection of G on a, we have QP D
! ! ! ! !
QQ C Q G C GP . But, 8P 2 C , we have that jQQj  jQ Gj, and (16.75)
yields Z
.t / ! !
MG D !T2 Q G 
GP dc D 0; (16.76)
C
owing to the definition of center of mass. It remains to evaluate the torque relative
to G of Coriolis’ forces:
Z
.c/ !
MG D 2
GP  .!T  rP P / dc: (16.77)
C

For the next applications, it will be sufficient to consider the center of mass at rest in
!
a lab frame. In this case, rP P D !  GP and (16.77) can also be written as follows:
Z
.c/ ! h ! i
MG D 2
GP  !T  .!  GP / dc: (16.78)
C

On the other hand, this results in

 !  ! !
!T  !  GP D !T  GP !  .!T  !/GP ;
16.9 Torque of Terrestrial Fictitious Forces 281

! h  ! i  !

so that
!
GP  !T  !  GP D  !T  GP !  GP :

Taking into account this result, (16.78) becomes

Z 
.c/ ! !
MG D 2! 
!T  GP GP dc: (16.79)
C

Denoting by !T0 i the components of !T in a central frame of inertia Gx10 x20 x30 , we
have that
Z  Z
! !   

!T  GP GP dcD
!T0 1 x10 C!T0 2 x20 C!T0 3 x30 x10 e01 Cx20 e02 Cx30 e03 dc:
C C

Since the body frame Gx10 x20 x30 is a central frame of inertia, the products of inertia
vanish and the preceding relation reduces to the following one:
Z  Z Z
! !

!T  GP GP dc D !T0 1 e01
x102 dc C !T0 2 e02
x202 dc
C C C
Z
C!T0 3 e03
x302 dc: (16.80)
C

Since B is a gyroscope, we have

Z Z Z Z
   
AD
x202 C x302 dc D
x102 C x302 dc D B )
x202 dc D
x102 dc:
C C C C

On the other hand, we also have

Z Z
 
C D
x102 C x202 dc D 2
x102 dc;
C C

so that
Z Z
C

x102 dc D
x202 dc D : (16.81)
C C 2
Z
C

x302 dc D A  : (16.82)
C 2

In view of (16.81) and (16.82), Eq. (16.80) assumes the final form
Z   
! !   C
2
!T  GP GP dc D C !T0 1 e01 C !T0 2 e02 C A  !T0 3 e03 : (16.83)
C 2
282 16 Dynamics of a Rigid Body

Introducing this expression into (16.79), we finally obtain

   
C
.!  e03 /e03 :
.c/
MG D !  C !T C A  (16.84)
2

16.10 Foucault’s Gyroscope and Gyroscope Compass

In this section, we show two interesting applications of formulae (16.76) and

(16.84).
A Foucault gyroscope B is a heavy gyroscope whose center of mass G is fixed
with respect to the Earth. Let x1 x2 x3 be an inertial frame of reference with the
origin at the center of the Sun and the axes oriented toward fixed stars. Consider
another frame Gx1 x2 x3 whose axes are parallel to the axes of the inertial frame
x1 x2 x3 . The frame Gx1 x2 x3 is not an inertial frame since its origin is at rest with
respect to the Earth. The forces acting on B are the gravitational attraction of the
Sun and the fictitious forces. The latter is applied at G so that its torque relative to
G vanishes. Further, the fictitious forces reduce to the drag forces since the motion
of the frame Gx1 x2 x3 relative to x1 x2 x3 is translational. In view of (16.76), the
torque of these forces vanishes. Therefore, the balance equation of momentum in
the frame Gx1 x2 x3 is
P G D 0;
K (16.85)
and the motion of B reduces to a regular precession. Consequently, if initially the
angular velocity ! is directed along the gyroscopic axis a, then the motion of
B reduces to a uniform rotation about a. In other words, the gyroscopic axis is
constantly directed toward a fixed point of the celestial sphere. An observer in a
frame of reference G xO 1 xO 2 xO 3 at rest relative to the Earth will see the gyroscopic axis
moving with the terrestrial angular velocity !T .
The gyroscopic compass is a heavy gyroscope B having a center of mass at rest
in a terrestrial frame Gx1 x2 x3 and the gyroscopic axis a constrained to move on a
fixed plane . Suppose that an initial rotation is applied to B about the axis a. We
prove that this axis remains at rest on  if and only if it coincides with the projection
of the terrestrial axis. Consequently, if  is horizontal (Fig. 16.13), then B serves as
a declination compass since a is directed toward the pole; if  is a meridian plane,
then B gives the terrestrial axial tilt.
Let G1 2 3 be a terrestrial frame with axes oriented as in Fig. 16.13, and let
Gx1 x2 x3 be a central frame of inertia with an axis Gx3  a. If , ', and 
denote the Euler angles, the condition that a lies on  implies that D 0 since
the nodal line coincides with the vertical axis Gx1 . Bearing this in mind, we can
write (12.30) as
! D 'e P 1;
P 03 C e (16.86)
16.10 Foucault’s Gyroscope and Gyroscope Compass 283

x
G

Fig. 16.13 Gyroscopic compass

where e03 is the unit vector along the gyroscopic axis Gx3 and e1 is the unit vector
along the lab axis Gx1 . Moreover (Sect. 12.4)

p D P cos '; q D P sin '; r D ';

P (16.87)

and the angular momentum of B is given by

P
KG D A.cos 'e01  sin 'e02 / C C 'e
P 03 : (16.88)

Finally, the balance equation of angular momentum becomes

P G D M.c/ C M.v/ ;
K (16.89)
G G

.c/ .v/
where MG and MG denote the torques of fictitious forces and reactive forces,
respectively. It is possible to prove (see next chapter) that, if the constraints are
.v/
smooth, then MG is orthogonal to the plane G1 x3 , that is, to the unit vectors e1
and e03 . Consequently, if we consider the projections of (16.89) along these vectors,
then we obtain two differential equations without reactive forces. Now, since e1 is
fixed and e01  e1 D cos ', e02  e1  sin ', we have that

P G  e1 D d .KG  e1 / D A:
K R (16.90)
dt
On the other hand, in view of (16.86) and (16.88), we also have

 
P G  e03 D d KG  e03  KG  !  e03
K
dt
d  
D KG  e03 C KG  ! D C ':
R (16.91)
dt
284 16 Dynamics of a Rigid Body

It remains to evaluate the components of MG along e1 and e03 . First, from (16.84)
.c/

we obtain
MG  e03 D C !  !T  e03 :
.c/
(16.92)
On the other hand, the terrestrial angular velocity is contained in the meridian plane
G1 3 (Fig. 16.13), and we can write

!T D !T cos  e1  !T sin  e3 ; (16.93)

where  is the colatitude of G. In view of (16.86), we give (16.92) the following

form:

MG  e03 D C 'e
.c/ P 1  .!T cos  e1  !T sin  e3 /  e0
P 03 C e 3

P T sin  e2  e0 D C !T sin  P sin :

D C ! (16.94)
3

We explicitly remark that (16.94) shows that a fundamental sufficient condition to

apply the gyroscopic effect is not satisfied (see end of Sect. 16.9).
Taking into account (16.84), (16.86), and (16.94), we deduce the following
.c/
expression of MG  e1 :

.c/
MG  e1 D C !T sin  'P sin : (16.95)

Collecting the preceding results we obtain the equations of motion

'R D !T sin  P sin ; (16.96)

AR D C !T sin  'P sin ; (16.97)

to which we associate the following initial data:

P
.0/ D 0 ; .0/ D 0; '.0/ D 0; '.0/
P D r0 : (16.98)

Since !T and sin  are constant and P sin  D d.cos /=dt, when we recall
(16.98), from (16.96) we derive the following result:
 
!T
'P D r0 C !T sin .cos   cos 0 / D r0 1 C sin .cos   cos 0 / : (16.99)
r0

Introducing (16.99) into (16.97), we obtain

 
!T
AR D C !T sin  r0 1 C sin .cos   cos 0 / sin : (16.100)
r0
16.11 Free Heavy Solid 285

When we suppose that r0 !T , (16.99) and (16.100) become

'P D r0 ;
AR D C !T sin  r0 sin :

In conclusion, the gyroscopic compass rotates uniformly about its axis, which
oscillates around the initial position. For the presence of friction, after a few
oscillations, the axis stops in the position  D 0, i.e., it is directed along the
meridian.

16.11 Free Heavy Solid

We conclude this chapter considering a free solid B for which we can write the
balance equations of momentum and angular momentum in the form

mRrG D R.a/ ; (16.101)

PG D
K
.a/
MG ; (16.102)

.a/
where m is the mass of B and R.a/ , MG denote the active total force and total torque
relative to the center of mass G of the active forces acting on B.
We denote by Ox1 x2 x3 a lab frame and by Gx 1 x 2 x 3 a frame of reference with
its origin at G and axes parallel to the corresponding axes of the lab frame. Finally,
Gx10 x20 x30 is the body frame, which is a principal frame of inertia. In view of (15.7)
and taking into account that the motion of Gx 1 x 2 x 3 relative to the lab frame is
translational, we have

KG D KG ; P G D da KG D dr KG ;
K (16.103)
dt dt

where dr =dt denote a derivative in the frame Gx 1 x 2 x 3 . In other words, we can

substitute (16.101) and (16.102) with others, as follows:

mRrG D R.a/ ; (16.104)

dr KG .a/
D MG : (16.105)
dt
We remark that in the frame Gx 1 x 2 x 3 the center of mass G is fixed so that the
components of (16.105) along the body frame are the Euler equations. If we add to
(16.104) and (16.105) Euler’s kinematic relations, then we obtain a system of nine
scalar differential equations whose unknowns are Euler angles .t/, '.t/, .t/, the
components p.t/, q.t/, and r.t/ of the angular velocity in the body frame, and the
coordinates of the center of mass x1G , x2G , and x3G . This conclusion is also due to
286 16 Dynamics of a Rigid Body

the fact that the acting forces, since they depend on the position of B, its velocity
field, and time, are functions of the aforementioned variables.
We explicitly remark that the presence of all the unknowns in all the equations
make it impossible to analyze the motion of the center of mass independently of the
motion about the center of mass. This is possible, for instance, when B is only acted
upon by its weight mg. In such a case, (16.104) and (16.105) become

rR G D g.a/ ; (16.106)
dr KG
D 0: (16.107)
dt
These equations imply that the center of mass G moves along a parabola, whereas
the motion about the center of mass is a free rotation (Sect. 16.4).
Chapter 17
Lagrangian Dynamics

17.1 Introduction

In previous chapters, we analyzed some fundamental aspects both of Newton’s

model, describing a system of material points, and Euler’s model, referring to a
single, free or constrained, rigid body. If we attempt to apply the latter model to a
system S of N constrained rigid bodies, we face great difficulties. In fact, it is not
an easy task either to express analytically the constraints to which S is subject or to
formulate mathematically the restrictions on the reactive forces exerted by smooth
constraints.
D’Alembert and Lagrange proposed an elegant and efficient method to analyze
free or constrained mechanical systems. This method exhibits the advantage of
supplying differential equations of motion that
• Do not contain unknown reactive forces and
• Contain the essential functions determining the motion of S .
These equations, called Lagrange’s equations, reduce any dynamical problem to
the integration of a system of ordinary differential equations. This reduction of
mechanics to analysis, of which Lagrange was very proud, justifies the name
analytical mechanics of this branch of dynamics. It must again be noticed that the
effective integration of Lagrange’s equation is impossible in almost all cases, so
that we are compelled to resort to other strategies to obtain meaningful information
about unknown solutions. One of the most important of these approaches is given
by Noether’s theorem, relating symmetries of mechanical systems to conservation
laws.
Another important aspect of the Lagrangian formulation of dynamics is repre-
sented by the fact that Lagrange’s equations coincide with the equations defining the
extremals of a functional. This remark suggests an integral formulation of motion
(Hamilton’s principle) opening new perspectives on the development of dynamics.
Hamilton proposed a set of equations that, though equivalent to the Lagrange
equations, are more convenient in describing the behavior of dynamical systems.

A. Romano, Classical Mechanics with Mathematica R

, Modeling and Simulation 287
in Science, Engineering and Technology, DOI 10.1007/978-0-8176-8352-8 17,
© Springer Science+Business Media New York 2012
288 17 Lagrangian Dynamics

In this new formulation of dynamics, it is possible to prove very important

theorems, shedding a new light on the behavior of dynamical systems. Further,
the Hamiltonian formalism has been applied to statistical mechanics and quantum
mechanics.
At the beginning of the last century, a geometric formulation of dynamics made
it possible to prove many meaningful theorems. Finally, using this new approach
we can formulate the laws of dynamics in an intrinsic way, i.e., independently of
Lagrangian coordinates.

17.2 Dynamics of Constrained Systems

In this section, we formulate the fundamental problem of dynamics of a system S

of N rigid bodies B1 ; : : : ; Bn , subject to constraints and acted upon by active forces.
We can apply the balance laws (15.48), (15.49) of linear momentum and angular
momentum to each body of S :

.a/ .r/
mRrGi D Ri C Ri ; (17.1)
P Gi D
K
.a/
MGi C
.r/
MGi ; (17.2)

where i D 1; : : : ; N . Choosing a central frame of inertia Gi x10 x20 x30 as a body

frame for the body Bi , we can state that the motion of any body is known when
we determine the three Euler angles i , 'i , i that Gi x10 x20 x30 forms with the axes
of the lab frame as well as the three coordinates xGi 1 , xGi 2 , xGi 3 of the center of
.r/
mass of Bi relative to the lab frame. Since the six components of the vectors Ri
.r/
and MGi are unknowns, the 6N differential equations, obtained by projecting (17.1)
and (17.2) along the axes of the lab frame, contain 12N unknowns. In other words,
the model we are proposing seems unable to describe the motion of S . However, we
must not give up hope since we have not yet considered the conditions due to the
.r/ .r/
presence of the constraints or the restrictions on the vectors Ri and MGi , deriving
from the hypothesis that the constraints are smooth. For instance, it is evident that
the constraints, reducing the mobility of the system S , lead to some relations among
the 6N parameters necessary to determine the configuration of S . On the other
hand, we have not yet taken into account that the reactive forces are orthogonal to
the contact surfaces between bodies and constraints.
D’Alembert and Lagrange proved that system (17.1), (17.2), together with the
restrictions on the mobility of S and the reactive forces due to the constraints, leads
to a system of equations that completely determine the motion of S . In the following
sections, we present this approach and analyze some important consequences of it.
17.3 Configuration Space 289

17.3 Configuration Space

Let S be a system of N rigid bodies B1 ; : : : ; Bn . We define a constraint as any

restriction a priori imposed on the mobility of S , that is, both on the positions of the
bodies of S and their velocity fields. To describe a constraint mathematically, we
introduce the system SL obtained by eliminating the constraints of S . Since a con-
figuration of SL is determined assigning the 6N numbers  D .1 ; : : : ; 6N / given
by xGi 1 , xGi 2 , xGi 3 ; i , 'i , i , i D 1; : : : ; N , we can state that the configurations
of SL are in a one-to-one correspondence with the points of the manifold V6N D
<3N  O.3/, where O.3/ is the space of the orthogonal matrices (Sect. 12.4).1 The
introduction of this 6N manifold supplies a geometric representation of the motion
of S . In fact, while S is moving in the three-dimensional Euclidean space E3 , the
point  describes a curve  in V6N , which is called the dynamical trajectory of S .
In the absence of constraints, this curve can invade any region of V6N and its tangent
vector t, with components xP Gi 1 , xP Gi 2 , xP Gi 3 ; P i , 'Pi , Pi , i D 1; : : : ; N , can have any
direction for the arbitrariness of the velocity field of S .
Adopting this geometric interpretation, we say that S is subject to position
constraints independent of time or to holonomic constraints independent of time,
if the dynamical trajectory  , corresponding to any possible motion of S , lies on a
submanifold Vn of V6N , where n < 6N . The manifold Vn is called the configuration
space of S , and any set .q 1 ; : : : ; q n / of local coordinates on Vn is called a set of
Lagrangian coordinates for S . Finally, the circumstance that Vn is n-dimensional
is also expressed by saying that the system S has n degrees of freedom. Recalling
the contents of Chap. 6, we can say that the immersion map i W Vn ! V6N can be
expressed locally in implicit form by a system

f1 ./ D 0;
: : : : : : : : : : ::
fm ./ D 0; (17.3)

where the functions f1 ; : : : ; fn are of class C 1 and satisfy the condition

 
@fi
rank Dm
@j

in an open region  of V6N . In fact, under these hypotheses, system (17.3) defines in
 a submanifold of class C 1 and dimension n D 6N m. Equivalently, if q 1 ; : : : ; q n
are arbitrary local coordinates on Vn , i.e., arbitrary Lagrangian coordinates of S , this
manifold can be represented in the parametric form

1
More precisely, V6N is an open n-dimensional submanifold of <3N  O.3/ since we must exclude
those values of xGi 1 , xGi 2 , xGi 3 , and i , 'i , i , i D 1; : : : ; N , for which parts of two or more bodies
of S occupy the same region of the three-dimensional space.
290 17 Lagrangian Dynamics

 D .q 1 ; : : : ; q n /: (17.4)

We conclude this section by exhibiting the configuration spaces of some dynam-

ical systems with time-independent holonomic constraints.
• Let S be a system formed by a single material point P constrained to remain on a
curve  . Then the system SL reduces to P freely moving in the three-dimensional
space E3 and Vn coincides with  . The immersion map i W  ! E3 can be written
in implicit form

f1 .x1 ; x2 ; x3 / D 0;
f2 .x1 ; x2 ; x3 / D 0;

or in parametric form
r D r.s/;
where x1 , x2 , and x3 are the Cartesian coordinates of P , r D .x1 ; x2 ; x3 /
the position vector of P , f1 D 0 and f2 D 0 the implicit equations of two
surfaces intersecting each other along  , and s the curvilinear abscissa along  .
A Lagrangian coordinate is s, and S has one degree of freedom.
• Let S be a rigid body with a fixed axis a. The manifold VL associated with SL
is <  O.3/. The configuration space V1 is a circumference C . In fact, with the
notations of Fig. 16.2, the configurations of S are given assigning the angle '
between the fixed plane Ox1 x3 and the body plane Ox10 x30 . In turn, this angle
defines a point belonging to C . ' is a Lagrangian coordinate of S that has one
degree of freedom.
• When S is a rigid body with a fixed point, we have VL D <3  O.3/ and
V3 D O.3/. Euler’s angles are Lagrangian coordinates, and S has three degrees
of freedom.
• Let S be a double pendulum. Then the configuration space is the torus Vn D
S 1  S 1.
• The bar AB of Fig. 17.1 is contained in a plane Oxy, has point A constrained
to move on the straight line Ox, and can rotate about an axis containing A and
orthogonal to Oxy. The configuration space is a cylinder <  S 1 , and x and '
are Lagrangian coordinates. The system has two degrees of freedom.
Other examples can be found in the exercises at the end of the chapter.

17.4 Virtual Velocities and Nonholonomic Constraints

The existence of a one-to-one map between the configurations of a system S of N

rigid bodies B1 ; : : : ; Bn and the points of the configuration space Vn is equivalent to
stating that the position vector r of any point P of Bi , i D 1; : : : ; n, is determined
17.4 Virtual Velocities and Nonholonomic Constraints 291

x A x
O
x
r’
r

Fig. 17.1 Constrained bar

by giving the values of the Lagrangian coordinates q 1 ; : : : ; q n . In other words, there

exists a function
r D r.r0 ; q 1 ; : : : ; q n /; (17.5)
where r0 is the constant position vector of P 2 Bi in the body frame of Bi .
For instance, in the example of Fig. 17.1, the position vector r can be written as
follows:
r D xe1 C jr0 j.cos 'e1 C sin 'e2 /  r.r0 ; x; '/;
where e1 and e2 are the unit vectors along the Ox- and Oy-axes, respectively. It
could be very difficult to determine the explicit form of the function (17.5). In any
case, we are interested not in its form but in its existence.
The sequence of configurations that S assumes during its effective motion
are represented by the dynamical curve  of the configuration space Vn . If
q 1 .t/; : : : ; q n .t/ are the parametric equations of this curve, then the spatial trajectory
of the point r0 2 S is given by

r.t/ D r.r0 ; q 1 .t/; : : : ; q n .t//: (17.6)

We recall that the position vector r0 is constant since it is evaluated in a body frame.
Therefore, differentiating (17.6) with respect to time, we obtain the velocity of
the point r0 in terms of the Lagrangian coordinates and the Lagrangian velocities
qP 1 ; : : : ; qP n
@r
rP D h qP h : (17.7)
@q
We remark that .qP 1 ; : : : ; qP n / is tangent to the dynamical trajectory  , whereas the
velocity vector rP is tangent to the trajectory of the point r0 in three-dimensional
space. These vectors are related to each other as it is shown by (17.7).
We define virtual velocity v of P 2 S as any possible velocity compatible with
the constraints of S . It is evident that a virtual velocity is obtained by (17.7) by
292 17 Lagrangian Dynamics

substituting the tangent vector .qP 1 ; : : : ; qP n / with an arbitrary tangent vector w D

.w1 ; : : : ; wn /
@r
v D h wh : (17.8)
@q
We say that the system S , in addition to the holonomic constraints, is subject to
nonholonomic constraints if, at any instant, its velocity field is a priori subject to
some restrictions. To give a mathematical definition of nonholonomic constraints,
we consider the tangent space Tq .Vn / at the point q  .q 1 ; : : : ; q n / of Vn . In the
absence of nonholonomic constraints, by (17.8), to any vector w D .w1 ; : : : ; wn / 2
Tq .Vn / corresponds a possible velocity v. We say that S is subject to nonholonomic
constraints if the possible Lagrangian velocities .w1 ; : : : ; wn / belong to an r-
dimensional subspace Uq .Vn / of Tq .Vn /, where r < n. We define Uq .Vn / at any
point q 2 Vn by r equations

!i .w/ D aij .q 1 ; : : : ; q n /dq j .w/

D aij .q 1 ; : : : ; q n /wj D 0; (17.9)

where !i D aij .q 1 ; : : : ; q n /dq j , i D 1; : : : ; r, are r differential forms on Vn . The

differential forms !i must be nonintegrable. In fact, if they are integrable, there exist
r functions fi .q 1 ; : : : ; q n / such that (17.9) are equivalent to the system

fi .q 1 ; : : : ; q n / D ci ; i D 1; : : : ; r;

where c1 ; : : : ; cr are constants, and then the constraints (17.9) are holonomic.
Example 17.1. Let S be a disk that rolls without slipping on a horizontal plane
Ox1 x2 . If we suppose that the plane containing S remains vertical during motion,
the configuration space of S is <2  S 2 ; moreover, the coordinates x1 and x2 of the
contact point A and the angles ' and  in Fig. 17.2 are Lagrangian coordinates for
S . Finally, the angular velocity ! of S about a vertical axis containing C is

! D 'e P 3;
P 03 C e (17.10)

and the condition of rolling without slipping becomes

!
rP A D rP C C !  CA D 0:
!
Noting that rP C D xP 1 e1 C xP 2 e2 , CA D ae3 , e3 D sin e1  cos e2 , and taking
into account (17.10), the condition of rolling without slipping leads to the system

xP 1 C a'P cos  D 0;
xP 2 C a'P sin  D 0;
17.5 Configuration Space-Time 293

x’

e x’

C
e’
x’ O x

x A

Fig. 17.2 Disk rolling on a smooth surface

which can also be written in terms of differential forms:

dx1 C a cos d' D 0;

dx2 C a sin d' D 0:

It is evident that these differential forms are nonintegrable, so that they define a
nonholonomic constraint.

17.5 Configuration Space-Time

We say that the system S of N rigid bodies B1 ; : : : ; BN , i D 1; : : : ; N , is subject to

constraints depending on time if there exists a submanifold Vn of V6N such that the
position vector r of a point r0 2 S is given by the relation [see (17.5)]

r D r.r0 ; q 1 ; : : : ; q n ; t/: (17.11)

The manifold Vn  < is called the configuration space-time. In these conditions,

the spatial trajectory of r0 is given by
 
r.t/ D r r0 ; q 1 .t/; : : : ; q n .t/; t ; (17.12)

and its velocity yields

@r h @r
rP D qP C : (17.13)
@q h @t
294 17 Lagrangian Dynamics

.
r x’

w
x

Fig. 17.3 Moving constraint

For holonomic constraints depending on time we define virtual velocity at the

instant t as a possible velocity that is compatible with the constraints at that instant.
Consequently, the virtual velocity is still given by (17.8).
Example 17.2. Let S be a material point constrained to move on the axis Ox10 .
Suppose that this axis rotates about a fixed axis a, orthogonal at O to Ox10 , with
known angular velocity 'P (Fig. 17.3). rP and w denote the effective velocity and a
possible virtual velocity, respectively.

Example 17.3. Let S be a rotating simple pendulum (Sect. 14.10). In this case,
V6N is the three-dimensional Euclidean space and V1 the rotating circumference on
which the pendulum is constrained to remain.

17.6 Simple Example

Before introducing the main ideas of the new approach due to D’Alembert and
Lagrange, we start with a simple example that contains all the ingredients of this
theory.
Let S be a material point constrained to move on a smooth surface † having
parametric equations
r D r.q 1 ; q 2 /; (17.14)
where q 1 and q 2 are surface parameters. It is evident that the configuration space of
S is † and q 1 and q 2 are possible Lagrangian coordinates. If we denote by ˆ the
reactive force, by F.r; rP / the active force acting on S , and by m the mass of S , then
the motion of S on † is governed by the fundamental equation of dynamics

mRr D F.r; rP / C ˆ: (17.15)

Let q 1 .t/ and q 2 .t/ be the parametric equations of the trajectory  of S . Starting
from the position occupied by S on  at a given instant t, we consider an arbitrary
17.6 Simple Example 295

x
F
e
w
e

Fig. 17.4 Point moving on a smooth surface

virtual velocity w (Fig. 17.4). From (17.15) we obtain that the virtual power of the
force of inertia mRr and active force vanishes,

.mRr C F/  w D 0; 8w 2 Tr .†/; (17.16)

since the hypothesis that † is smooth implies ˆw D 0. In (17.16) Tr .†/ denotes the
tangent space to † at the point r occupied by S at the considered instant. Introducing
the expression (17.8) of the virtual velocity into (17.16), we obtain the condition

@r h
.mRr C F/  w D 0; 8.w1 ; w2 /: (17.17)
@q h

On the other hand, the vectors eh D @r=@q h are tangent to the coordinate curves,
and therefore rRh D rR  eh and Fh D F  eh denote the (covariant) components of rR and
F along the vector eh . Finally, from (17.17), taking into account the arbitrariness of
.w1 ; w2 /, we derive the following two scalar equations:

mrRh D Fh ; h D 1; 2: (17.18)

Since from (17.14) we obtain

@r h
rP D qP ; (17.19)
@q h
@r h @2 r
rR D h
qR C h k qP h qP k ; (17.20)
@q @q q

we can say that Fh are functions of q h and qP h , whereas rRh depend on qRh , qP h , and qh .
In other words, Eqs. (17.18) are a system of two second-order differential equations
296 17 Lagrangian Dynamics

in the unknown functions q 1 .t/ and q 2 .t/ that completely determine the motion of
S on the surface †.
We can summarize the preceding results as follows.
1. Equation (17.15) implies (17.17).
2. Equation (17.17) does not contain the unknown reactive force.
3. Equation (17.17) leads to equations that contain the unknowns determining the
motion of S .
In the following sections we prove that the foregoing procedure can be extended
to an arbitrary system S of N constrained rigid bodies.

17.7 Principle of Virtual Power

Let S be a system of N constrained rigid bodies B1 ; : : : ; BN , and denote by Gi the

center of mass of Bi . Finally, let Ci be the fixed region occupied by Bi in its body
frame Gi x10 x20 x30 . The motion of S is governed by the balance equation of linear
momentum (17.1) and the balance equations of angular momentum (17.2), which in
this more general situation replace (17.15) in the example considered in the previous
section. Following the logic of Sect. 17.6, we must evaluate the virtual power of the
inertial, active, and reactive forces and show that the virtual power vanishes for any
virtual velocity field v. To simplify the notations, we suppose that the active forces
on Bi are distributed on the volume Ci , whereas the reactive forces act on the surface
@Ci . Further, we denote by Rr and F the inertia forces and the active forces per
unit volume. Finally, ˆ is the reactive force per unit surface. Then the virtual powers
of these forces can be written respectively as

N Z
X
P .m/ D  Rr  v dc; (17.21)
i D1 Ci

N Z
X
P .a/ D F  v dc; (17.22)
i D1 Ci

N Z
X
P .r/
D ˆ  v dc: (17.23)
i D1 @Ci

We now prove the following theorem.

Theorem 17.1 (D’Alembert, Lagrange). The equations of balance (17.1) and
(17.2) imply the condition

P .m/ C P .a/ C P .r/ D 0; 8v; (17.24)

17.7 Principle of Virtual Power 297

where v is an arbitrary virtual velocity field, i.e., compatible with the constraints to
which the system is subject.
Proof. The virtual velocity field v is compatible with the constraints. In particular,
it must be rigid for each body Bi of S . Consequently, the restriction of the velocity
field v to the region Ci has the form
!
vi D vGi C !i  Gi P ; (17.25)

where !i is the virtual angular velocity of Bi . Introducing (17.25) into (17.22), we

obtain the equation

X
N Z X
N Z
!
P .a/ D vGi  F dc C !i  Gi P  F dc
i D1 Ci i D1 Ci

which can also be written in the form

X
N
.a/
X
N
.a/
P .a/ D vGi  Ri C !i  MGi : (17.26)
i D1 i D1

Similarly, we have

X
N
.r/
X
N
.r/
P .r/ D vGi  Ri C !i  MGi (17.27)
i D1 i D1

and
X
N Z X
N Z
!
P .m/ D  vGi  Rr dc  !i  Gi P  rR dc:
i D1 Ci i D1 Ci

This last relation assumes the final form

X
N X
N
P .m/
D vGi Pi 
Q P Gi
!i  K (17.28)
i D1 i D1

since
Z Z
d P i;
Rr dc D Pr dc D Q
Ci dt Ci
Z Z
! d ! P Gi ;
Gi P  rR dc D Gi P  rP dc  rP Gi  Qi D K
Ci dt Ci
298 17 Lagrangian Dynamics

and rP Gi  Qi D rP Gi  mPrGi D 0. Collecting (17.26)–(17.28), we obtain

N h
X  
P .m/ C P .a/ C P .r/ D P i C R.a/ C R.r/
vGi  Q i i
i D1
 i
P Gi C M.a/ C M.r/ ;
C !i  K Gi Gi

and the theorem is proved. t

u
By this theorem we can state that property 1, proved at the end of Sect. 17.6, holds
for any constrained system S of rigid bodies. The following theorem, stated without
a proof, extends property 2 to any system S .
Theorem 17.2. For any system S of rigid bodies subject to smooth constraints, we
have
P .r/ D 0 (17.29)

for all virtual velocity fields.

Collecting the results (17.24) and (17.29), the following condition is satisfied:

P .m/ C P .a/ D 0 (17.30)

for any virtual velocity field.

17.8 Lagrange’s Equations

In this section we prove that, for any system of constrained rigid bodies with
n degrees of freedom, (17.30) leads to a system of n second-order differential
equations in the unknown functions q 1 .t/; : : : ; q n .t/.
In view of (17.21) and (17.22), we write (17.30) in the following form:

N Z
X
.F.a/  Rr/  w dc D 0; (17.31)
i D1 Ci

which, resorting to (17.8) and (17.13), can also be written as

P t/  h .q; q;
.Qh .q; q; R t//wh D 0; 8.w1 ; : : : ; wn / 2 <n ; (17.32)
17.8 Lagrange’s Equations 299

where
N Z
X @r
P t/ D
Qh .q; q; F.a/  dc; (17.33)
i D1 Ci @q h
N Z
X @r
P q;
h .q; q; R t/ D Rr  dc: (17.34)
i D1 Ci @q h

For the arbitrariness of .w1 ; : : : ; wn / in (17.32), the functions q 1 .t/; : : : ; q n .t/

satisfy the following system of second-order differential equations:

P q;
h .q; q; R t/ D Qh .q; q;
P t/; h D 1; : : : ; n: (17.35)

To write (17.35) in a more expressive form, we introduce the kinetic energy of S

N Z
X
T D jPrj  jPrj dc;
i D1 Ci

which, recalling the Lagrangian expression (17.13) of the velocity field, becomes

1
T D ahk .q; t/qP h qP k C bh .q; t/qP h C T0 .q; t/; (17.36)
2
where
N Z
X @r @r
ahk .q; t/ D   dc D akh .q; t/; (17.37)
i D1 Ci @q h @q h
N Z
X @r @r
bh .q; t/ D   dc; (17.38)
i D1 Ci @q h @t
Z  2
1X
N
@r
T0 D  dc: (17.39)
2 i D1 Ci @t

Remark 17.1. When the constraints do not depend on time, we must use (17.5)
instead of (17.12). Consequently, @r=@t D 0, and the kinetic energy reduces to the
quadratic form
1
T D ahk qP h qP k  T2 : (17.40)
2
The kinetic energy is a positive quantity if the velocity field rP does not vanish, and
it is equal to zero if and only if rP D 0. It is also possible to prove that T D 0 if
and only if qP 1 D    qP n D 0. In other words, (17.40) is a positive-definite quadratic
form.
300 17 Lagrangian Dynamics

To give (17.35) a more convenient form, we start by noting that from (17.13)
there follows

@Pr @r
D h; (17.41)
@qP h @q
   
@Pr @ @r @ @r d @r
h
D h h
P
q h
C h
D ; (17.42)
@q @q @q @t @q dt @q h

so that also
 
@r d @r d @r
rR  D rP  h  rP 
@q h dt @q dt @q h
 
d @Pr @Pr
D rP  h  rP  :
dt @qP @q h

The preceding result implies that

N Z
X @r
h D Rr  dc
i D1 Ci @q h

X N Z X N Z
d @Pr @Pr
D Pr  h  Pr  h dc
i D1
dt Ci @qP i D1 Ci
@q

and we obtain the following expression for the quantities h :

d @T @T
h D  h: (17.43)
dt @qP h @q
From (17.35) and (17.43) we finally deduce the Lagrange equations:
d @T @T
P t/;
 h D Qh .q; q; h D 1; : : : ; n: (17.44)
dt @qP h @q
P t/ are called Lagrangian components of the active forces.
The quantities Qh .q; q;
Taking into account expression (17.36) of the kinetic energy, we can write (17.44)
as follows:
P t/  Ch .q; q;
ahk qR k D Qh .q; q; P t/  Kh .q; q;
P t/;
where the functions Ch .q; q; P t/ are given. On the other hand, the positive-definite
character of the quadratic form ahk qP h qP k implies that, at any configuration q, all the
principal minors of the matrix .ahk .q// have positive determinants; in particular,
det.ahk / ¤ 0. Consequently, the preceding system can be solved with respect to the
quantities qR h , that is, it can be written in the normal form
P t/:
qR h D g h .q; q; (17.45)
17.9 Conservative Forces 301

This result allows us to state that, under suitable regularity hypotheses on the
P t/ and Qh .q; q;
functions T .q; q; P t/, the system of Lagrange equations has one and
only one solution satisfying the initial data

q h .0/ D q0h ; qP h .0/ D qP0h (17.46)

that determine the initial position and velocity field of S .

17.9 Conservative Forces

We say that the positional forces acting on the system S of rigid bodies, with
constraints depending on time, are conservative if there exists a C 1 function U.q; t/,
called potential energy, such that

@U
Qh .q; t/ D  : (17.47)
@q h

Note that in (17.47) the time t appears as a parameter. If the active forces are
conservative, then @U=@qP h D 0. Consequently, introducing the Lagrange function

P t/ D T .q; q;
L.q; q; P t/  U.q; t/; (17.48)

the Lagrange equations (17.44) assume the following form:

d @L @L
 D 0; h D 1; : : : ; n: (17.49)
dt @qP h @q h

These equations show the fundamental importance of a Lagrangian description

of dynamical phenomena. In fact, they prove that the whole dynamical behavior
of a mechanical system subject to smooth constraints and conservative forces is
contained in a Lagrangian function.
The following theorem is very important in many applications of (17.49).
Theorem 17.3. Let S be a system of N rigid bodies subject to smooth holonomic
constraints. Further, denote by SL the system obtained by eliminating all the
constraints of S and by   .1 ; : : : ; L / the Lagrangian coordinates of SL . If
SL is acted upon by conservative forces with potential energy U.; t/, then S is
subjected to conservative forces with potential energy U.q; t/ D U..q/; t/, where
1 D 1 .q; t/; : : : ; L D L .q; t/ are the parametric equations of Vn  <.
Proof. For the sake of simplicity, we suppose that S is formed by a single
constrained rigid body. Formula (17.11) holds for S , and we denote by r D
rO .r0 ; 1 ; : : : ; L ; t/ the corresponding equation for SL . If SL is acted upon by
conservative forces, then there exists a potential energy U.; t/ such that the
302 17 Lagrangian Dynamics

Lagrangian components Qi , i D 1; : : : ; 6N , of the active forces are given by [see

(17.33)] Z
@Or @U
Qi D F.a/  dc D  :
Ci @i @i
On the other hand, for the system S , we have that
Z Z
@r @Or @i
Qh D F.a/  dc D F.a/  dc
Ci @q h Ci @i @q h
i
@U @ @U
D D ;
@i @q h @qh

and the theorem is proved. t

u
1 P t/ is the generalized potential energy of the
We say that the C function U.q; q;
forces acting on S if
d @U @U
P t/ D
Qh .q; q;  : (17.50)
dt @qP h @q h
For forces deriving from a generalized potential energy, the Lagrange equations still
assume the form (17.48), with the Lagrangian function given by

P t/ D T .q; q;
L.q; q; P t/  U.q; q;
P t/: (17.51)

P t/,
To recognize the severe restrictions that (17.50) imposes on the function U.q; q;
we start by noting that it can be put in the form

@2 U k
P t/ D
Qh .q; q; P t/;
qR C uh .q; q;
@qP h @qP k

where uh .q; q;P t/ are suitable functions independent of qP h . Since the left-hand sides
of the preceding equations do not depend on qR h , the function U.q; q; P t/ must depend
linearly on qP h , i.e.,
@2 U
D 0;
@qP h @qP k
and we can state that

P t/ D UO .q; t/ C …h .q; t/qP h :

U.q; q; (17.52)

Introducing (17.52) into (17.50) we are led to the equation

@UO d…h @…k k

P t/ D 
Qh .q; q; h
C  qP ;
@q dt @q h
17.9 Conservative Forces 303

which can also be written as follows:

 
@UO @…h @…k @…h
P t/ D 
Qh .q; q; C  qP k C : (17.53)
@q h @q k @q h @t

This formula shows that the forces having a generalized potential energy, apart from
the term @…h =@t, are formed by conservative forces with a usual potential energy
of UO .q; t/ and by gyroscopic forces whose Lagrangian components are
 
@…h @…k
Qh .q; q;
P t/ D  qP k  hk qP k : (17.54)
@q k @q h

These forces have the remarkable property that their power vanishes

Qh qP h D hk qP h qP k D 0 (17.55)

since hk is skew-symmetric and qP h qP k is symmetric.

We conclude this section with two important examples of forces having a
generalized potential.
(a) Charge moving under the action of an electromagnetic field.
Let S be a material point freely moving relative to the Cartesian frame
Ox 1 x 2 x 3 , and denote by r the position vector of P in this frame. Since
the Lagrangian coordinates coincide with the Cartesian coordinates and the
Lagrangian components of the active force F acting on P are the Cartesian
components of F relative to Ox 1 x 2 x 3 , we can write (17.52) and (17.53) in
vector form:

U.r; rP ; t/ D UO .r; rP ; t/ C …  rP ; (17.56)

@…
F.r; rP ; t/ D r UO C r  …  rP C : (17.57)
@t
Further, the electromagnetic force acting on a moving charge is given by
Lorentz’s force
F D q.E C rP  B/ D q.E  B  rP /; (17.58)
where E is the electric field and B the magnetic induction. Owing to Maxwell’s
equations, these fields can be obtained by a scalar potential ' and a vector
potential A by the formulae

@A
E D r'  ;
@t
B D r  A:
304 17 Lagrangian Dynamics

Taking into account these relations, Lorentz’s force (17.58) can be written as
follows:
@
F D r.q'/ C r  .qA/ C .qA/: (17.59)
@t
Comparing (17.59) with (17.57), we conclude that Lorentz’s force has the
following generalized potential energy:

U.r; rP ; t/ D q'  qA  rP : (17.60)

(b) Fictitious forces

Let Ox1 x2 x3 be a frame of reference whose origin O is at rest with respect
to an inertial frame O
. If m denotes the mass of a material point P moving
relative to Ox1 x2 x3 and r is the position vector of P in Ox1 x2 x3 , then the
fictitious force acting on P is [see (12.45)]

P  r C !  .!  r/ C 2!  rP 
F D mŒ!
P  r C .! ˝ !  ! 2 I/  r C 2!  rP :
D mŒ! (17.61)

Bearing in mind (17.57), it is a simple exercise to verify that the generalized

potential energy of F is given by

U.r; rP ; t/ D UO .r; t/ C ….r; t/  rP ; (17.62)

where
1
UO D mr  .! ˝ !  ! 2 I/  r; (17.63)
2
… D m!  r: (17.64)

17.10 First Integrals

Let S be a dynamical system with n degrees of freedom described by a Lagrangian

P t/, and denote by Vn  < the configuration space-time of S . A
function L.q; q;
function f W T .Vn /  < ! < is a first integral of the Lagrange equations if it is
constant along their solutions:

P
f .q.t/q.t/; t/ D f .q0 ; qP 0 ; 0/; (17.65)

where q.t/ is a solution satisfying the initial data q h .0/ D q0h ; qP h .0/ D qP0h , i D
1; : : : ; n.
In other words, a first integral supplies a conservation law that holds in any mo-
tion of S . From a mathematical point of view, it is a first-order differential equation
17.10 First Integrals 305

that may substitute one of the Lagrange equations, simplifying their integration.
Both these remarks show that searching for first integrals is a fundamental problem
of Lagrangian dynamics.
We say that a Lagrangian coordinate q h is cyclic or ignorable if it does not appear
in the Lagrangian function L. If we introduce the generalized momentum conjugate
to q h
@L
ph D h ; (17.66)
@qP
from Lagrange’s equation we obtain that ph is a first integral when L is independent
of q h :
P t/ D ph .q0 ; qP 0 ; 0/:
ph .q; q; (17.67)
This result shows how important is the choice of a suitable system of Lagrangian
coordinates in analyzing the motion of a dynamical system. Now we prove the
existence of a fundamental first integral.
Theorem 17.4. Let S be a system of rigid bodies with smooth and fixed holonomic
constraints. If the active forces derive from a generalized potential independent of
time [see (17.52)]
P D UO .q/ C …h .q/qP h ;
U.q; q/
then the total energy
E D T C UO (17.68)
is constant during any motion of S .
Proof. Along any motion of S we have that

dL @L h @L
D qP C h qR h
dt @q h @qP
   
@L h d @L h d @L
D qP C P
q  qP h :
@q h dt @qP h dt @qP h

Taking into account the Lagrange equations, the preceding relation reduces to the
condition  
dL d @L h
D qP ;
dt dt @qP h
which can be written equivalently as

@L h
qP  L D const: (17.69)
@qP h

Further, if the constraints do not depend on time, formula (17.40) of the kinetic
energy reduces to the quadratic form T2 , and we have that

L D T  U D T2  UO .q/  …h .q/qP h :
306 17 Lagrangian Dynamics

In conclusion, (17.69) becomes

@L h
qP  L D ahk qP h qP k  …h qP h  T2 C UO C …h qP h D T C UO ;
@qP h

and the theorem is proved. t

u
Remark 17.2. The term …h qP h does not appear in the total energy E since the power
of the gyroscopic forces vanishes.

17.11 Symmetries and First Integrals

In this section we prove a fundamental theorem that relates the first integrals of
Lagrange’s equations to the symmetries of a Lagrangian function.
As usual, we consider a dynamical system S with smooth and fixed holonomic
constraints and acted upon by conservative forces. We denote by Vn the config-
uration space and by T .Vn / its tangent fiber bundle, which is a 2n-dimensional
manifold called the velocity space of S . We recall that any point of T .Vn / is a pair
.q; v/ of a point q 2 Vn and a tangent vector v belonging to the tangent space Tq .Vn /
at q (Chap. 6). The dynamics of S is completely described by a Lagrangian function
L which can be regarded as a C 1 real-valued function on T .Vn /.
Let 's .q/ W <  Vn ! Vn be a one-parameter global transformation group of Vn ,
and let X be its infinitesimal generator (Chap. 7). The group 's .q/ is said to be a
symmetry group of L if

L.q; v/ D L.'s .q/; .'s / .q/v/; (17.70)

8.q; v/ 2 T .Vn /, 8s 2 <.

In (17.70), .'s / .q/ is the differential at the point q 2 Vn of the map 's .q/
and .'s / .q/v is the vector of T's .q/ corresponding to v 2 Tq .Vn / (Chap. 7 and
Fig. 17.5). If we set qs0h .q/ D .'s .q//h , then the linear map .'s / .q/ is represented
by the Jacobian matrix .@qs0h =@q k / and (17.70) becomes
 
0h @qs0h k
L.q ; v / D L qs .q/; k v :
h h
(17.71)
@q

Theorem 17.5 (E. Noether). Let S be a system of rigid bodies with smooth, fixed
holonomic constraints and subject to conservative forces. If 's .q/ W <  Vn ! Vn
is a symmetry group of the Lagrangian function L W T .Vn / ! <, then the function
f W T .Vn / ! < given by
@L
f .q; v/  h X h (17.72)
@v
17.11 Symmetries and First Integrals 307

orbits

v
q X

trajectories

Fig. 17.5 Noether’s theorem

is a first integral of the Lagrangian equations, that is,

P D f .q.0/; q.0//;
f .q.t/; q.t// P

along any solution of Lagrange’s equations.

Proof. First, we evaluate (17.71) along a solution q h .t/ of Lagrange’s equations:

@qs0h k
L.q h .t/; vh / D L.qs0h .q.t//; v /: (17.73)
@q k

Differentiating (17.73) with respect to s and evaluating the result at s D 0, we obtain

 
@L h @L @X h h
0D X C P
q
@q h @qP h @q h P /
q.t /;q.t
 h
@L h @L dX
D X C h
@q h @qP dt q.t /;q.tP /
   
@L d @L d @L h
D  X Ch
X :
@q h dt @qP h dt @qP h P /
q.t /;q.t

The theorem is proved since the functions q h .t/ satisfy Lagrange’s equations. t
u
Example 17.4. Consider a dynamical system whose Lagrangian function L does
not depend on the ignorable coordinate q 1 . Then, the one-parameter group of
transformations
q 01 D q 1 C s; q 02 D q 2 ; : : : ; q 0n D q n
308 17 Lagrangian Dynamics

is a symmetry group of L. The infinitesimal generator of this group is the vector

field X D .1; 0; : : : ; 0/ tangent to the configuration space Vn . Consequently, the first
integral associated with this symmetry group is

@L
f .q; v/ D D p1 ;
@qP 1

that is, the corresponding conjugate generalized momentum.

Example 17.5. Let P be a material point, constrained to move on a smooth plane
 in which Cartesian coordinates .x; y/ are introduced. Suppose that in these
coordinates the Lagrangian function is

1
LD m.xP 2 C yP 2 / C .˛x C ˇy/;
2

where ˛ and ˇ are two constants. The one-parameter group of transformations

x 0 D x C ˇs; y 0 D y  ˛s

is a symmetry group of L whose infinitesimal generator is X D .ˇ; ˛/.

Consequently, the corresponding first integral is

@L 1 @L 2
f D X C X D m.ˇ xP  ˛ y/:
P
@xP @yP

Example 17.6. Let S be a system of N material points r1 ; : : : ; rN with masses

m1 ; : : : ; mN . We suppose that, in cylindrical coordinates .r;  z/, the forces acting on
S are conservative with a potential energy U.r1 ; : : : ; rN ; z1 ; : : : ; zn /. Then, n D 3N
and Vn D <3N , and the Lagrangian function is given by

1X  2 
N
LD mi rPi C ri2 Pi2 C zP2i  U.r1 ; : : : ; rN ; z1 ; : : : ; zn /:
2 i D1

Since all the coordinates 1 ; : : : ; n are cyclic, the one-parameter group of transfor-
mations

ri0 D ri ;
i0 D i C s;
z0i D zi ; i D 1; : : : ; N;
17.11 Symmetries and First Integrals 309

is a symmetry group of L. The infinitesimal generator of this group is X D

.0; : : : ; 0; 1; : : : ; 1; 0; : : : ; 0/. Consequently, the corresponding first integral of
Lagrange’s equation is
XN
f D mi ri2 Pi :
i D1

It is easy to verify that this first integral coincides with components Kz of the total
angular momentum along the Oz-axis:

X
N
Kz D mi .xi yPi  yi xP i /:
i D1

Remark 17.3. We underline that the preceding theorem states that to any symmetry
group of L there corresponds a first integral of Lagrange’s equations. It is possible
to prove that also the inverse implication holds provided that we generalize the
definition of symmetry group. We face this problem in the Hamiltonian formalism.
It is also important to note that, although we can prove the equivalence between
symmetry groups and first integrals by extending the definition of symmetry group,
we are no more able, in general, to obtain the very simple expression (17.72) of the
first integral corresponding to this more general symmetry group.
The extension of Noether’s theorem to the case of a Lagrangian function
depending on time can be obtained by including the time among the Lagrangian
coordinates in the following way. Let L.q; v; t/ W T .Vn /< ! < be the Lagrangian
function of a dynamical system S over the .2n C 1/-manifold T .Vn /  <. Consider
the extension LO of L to the .2n C 2/ manifold T .Vn  </ such that
 v
O
L.q; t; v; u/ D L q; t; v; u: (17.74)
u
The following identities can be easily proved:

@LO @L @LO @L
h
D h
u; D h; (17.75)
@q @q @vh @v
@LO @L @LO @L vh
D u; D L h : (17.76)
@t @t @u @v u

We now verify that if the functions .q h .t// are a solution of the Lagrangian
equations relative to L, then .q h .t/; t/ are a solution of the Lagrangian equations
O In fact, in view of (17.75) and (17.76), regarding the first n equations
relative to L.
O
relative to L we can say that

d @LO @LO d @L @L
 D  D 0:
dt @qP h @q h dt @qP h @q h
310 17 Lagrangian Dynamics

Further, when we note that u D t ! tP D 1, the .n C 1/th equation is

!  
d @LO @LO d @L @L
 D L  h qP h 
dt @qP h @t dt @qP @t
uD1
     
dL d @L h @L @L h @L h d @L h
 qP  D qR C h qP  qP
dt dt @qP h @t @qP h @q dt @qP h
 
d @L @L
D  qP h D 0
dt @qP h @q h

since the first n Lagrange equations are verified by the functions .q h .t//.
We say that the one-parameter group of transformation 's W <  .Vn  </ !
.Vn  </ on the space-time of configurations .Vn  </ is a symmetry group of L if
O that is, if
it is a symmetry group of L,
 ˛ 
O O ˛ ˇ @'s ˇ
L.q ; v / D L 's .q /; ˇ v ;
˛ ˛
(17.77)
@v

where ˛; ˇ D 1; : : : ; n C 1, .q ˛ / D .q; t/, and .v˛ / D .v; u/. Applying Noether’s

theorem to L, O we determine the following first integral of the Lagrange equations
O
relative to L:
@LO
f D ˛ X ˛; (17.78)
@v
with X D .X ˛ / the infinitesimal generator of 's .q ˛ /. Taking into account (17.75)
and (17.76), we can give (17.78) the alternative form
 
@L h @L h
f D X C L  v X nC1 (17.79)
@vh @vh

along the curve .q h .t/; t/.

In particular, if L does not depend on time, then a symmetry group of L is given
by the following one-parameter group of transformations:

q 0h D q h ; t 0 D t C s:

Since the infinitesimal generator of this group is the vector field X D .0; : : : ; 0; 1/,
we obtain from (17.79) the conservation of the total mechanical energy

@L h
ED v  L:
@vh
17.12 Lagrange’s Equations for Nonholonomic Constraints 311

17.12 Lagrange’s Equations for Nonholonomic Constraints

Let S be a dynamical system with both holonomic and nonholonomic constraints,

and let Vn be the configuration space of the system S when we take into account
only the holonomic constraints. In Sect. 17.4, we assigned nonholonomic constraints
by r  n linear relations (17.9), which define, at any point q 2 Vn , a subspace
Vq of the tangent space Tq .Vn /. The fundamental consequence of the presence of
nonholonomic constraints is that in relation (17.32) the components .w1 ; : : : ; wn / of
the virtual velocity are not arbitrary since they must belong to Vq . In other words,
the condition
P t/  h .q; q;
.Qh .q; q; P q;
R t//wh D 0 (17.80)
must be verified for any .w1 ; : : : ; wn / 2 U.q/, that is, for all the vectors .w1 ; : : : ; wn /
confirming the r linear conditions

aih wh D 0; i D 1; : : : ; r  n; (17.81)

expressing the nonholonomic constraints. Consequently, the coefficients of wh in

(17.80) do not vanish but are linear combinations of (17.81):

P t/  h .q; q;
Qh .q; q; P q;
R t/ D 1 a1h C    C r arh ; h D 1; : : : ; n: (17.82)

The coefficients 1 ; : : : ; r , which can depend on q h , qP h , qR h , and t, are called

Lagrange’s multipliers. In conclusion, the Lagrange equations, in the presence of
nonholonomic constraints, are

d @T @T X r
 D Q h  i aih ; h D 1; : : : ; n: (17.83)
dt @qP h @q h i D1

These equations, together with the conditions due to the nonholonomic constraints,

aih qP h D 0; i D 1; : : : ; r; (17.84)

define a system of nCr differential equations in the nCr unknowns q 1 .t/; : : : ; q n .t/,
1 ; : : : ; r .
Example 17.7. Consider the homogeneous disk S rolling without slipping on a
plane already considered in Fig. 17.2. Since S is supposed to remain in a vertical
plane, the weight of S has no effect on the motion. Therefore, the Lagrange function
L of S reduces to the kinetic energy, which can be written as

1 2  1 
LD xP 1 C xP 22 C A.p 2 C q 2 / C C r 2 ;
2 2
312 17 Lagrangian Dynamics

where A and C are the momenta of inertia of S , and p, q, and r are the components
of the angular velocity ! in the body frame C x10 x20 x30 (Fig. 17.2). Since e3 D
sin 'e01 C cos 'e02 , the angular velocity can be written as

! D 'e P 3 D P sin 'e 0 C P cos 'e 0 C 'e

P 30 C e P 30 ;
1 2

and the Lagrange function becomes

1 2  1 2 
LD xP 1 C xP 22 C AP C C 'P 2 :
2 2
From (17.83) and the conditions determined at the end of Sect. 17.4, we obtain the
following system:

mxP 1 D 1 ; (17.85)

mxP 2 D 2 ; (17.86)
C 'R D a.1 cos   2 sin /; (17.87)
AR D 0; (17.88)
xP 1 D a'P cos ; (17.89)
xP 2 D a'P sin : (17.90)

Equation (17.88) implies that

.t/ D P0 C 0 ; (17.91)

where the meaning of the constants P0 and 0 is evident. Introducing (17.85) and
(17.86) into (17.87), we obtain the equation

C 'R D ma.xR 1 cos   xR 2 sin /

which, in view of (17.89) and (17.90), can also be written as

.C C ma2 /'R D 0;

so that
'.t/ D 'P0 t C '0 ; (17.92)
where again the meaning of 'P0 and '0 is evident. Finally, from (17.91), (17.92),
(17.89), and (17.90), we have that
a'P
x1 .t/ D  sin.P0 t C 0 /;
P0
a'P
x2 .t/ D cos.P0 t C 0 /
P0
17.13 Small Oscillations 313

if P0 ¤ 0 and

x1 .t/ D x10 C 'P0 t cos.P0 t C 0 /;

x2 .t/ D x20 C 'P0 t sin.P0 t C 0 /

if P0 D 0. In conclusion, the disk rotates uniformly about the vertical axis containing
its center C , rolls uniformly about the horizontal axis containing C , and the contact
point A moves uniformly along a straight line if P0 D 0 and moves uniformly on a
circumference of radius ja'= P P0 j if P0 ¤ 0.

17.13 Small Oscillations

Let S be a dynamical system with smooth and fixed constraints. We say that a
configuration C of S is an equilibrium configuration if, placing S at rest in the
configuration C at the instant t D 0, it remains in this configuration at any time.
Let Vn be the configuration space and denote by q the point of Vn representative of
the configuration C . Then, q is an equilibrium configuration if

q h .0/ D qh ; qP h .0/ D 0 ! q h .t/ D q ; 8t  0: (17.93)

Theorem 17.6. The configuration q is an equilibrium configuration of S if and

only if
Qh .q ; 0/ D 0; h D 1; : : : ; n: (17.94)
Proof. If q is an equilibrium configuration, i.e., if (17.93) holds, then the functions
q h .t/ D qh , h D 1; : : : ; n, must be a solution of Lagrange’s equations. Introducing
the functions q h .t/ into Lagrange’s equations, we obtain (17.94). Conversely, if
this last condition is satisfied, then the Lagrange equations admit q h .t/ D qh ,
h D 1; : : : ; n, as a solution. But this solution satisfies the initial data q h .0/ D
qh ; qP h .0/ D 0, h D 1; : : : ; n, and, consequently, it is the only solution of Lagrange’s
equations satisfying the initial data q h .0/ D qh ; qP h .0/ D 0, h D 1; : : : ; n. t
u
To apply the results of Chap. 10, we reduce Lagrange’s equations to an equivalent
system of 2n first-order differential equations. If we set qP h D vh , then (17.45) can
be written in the form

qP h D vh ; (17.95)
vP h D g h .q; v/: (17.96)

We say that the configuration C of S is a stable equilibrium configuration if

.q ; 0/ is a stable equilibrium configuration of system (17.95), (17.96). This means
that for any neighborhood I.q ; 0/ of .q ; 0/ 2 T .Vn /, there is a neighborhood
314 17 Lagrangian Dynamics

I0 .q ; 0/  I.q ; 0/ such that

8.q0 ; v0 / 2 I0 .q ; 0/ ) .q.t; q0 ; v0 /; v.t; q0 ; v0 // 2 I.q ; 0/: (17.97)

In Chap. 10, we remarked that checking the stability of an equilibrium configuration

requires a knowledge of the solutions. Since, in general, it is impossible to exhibit
the solutions of a given system of first-order differential equations, it is fundamental
to supply criteria of stability. In the following theorem, Dirichlet’s stability criterion
is proved.
Theorem 17.7. Let S be a mechanical system with smooth and fixed holonomic
constraints subject to active conservative forces with a potential energy U.q/. If
U.q / is an effective minimum of U.q/, then q is a stable equilibrium position of
S.
Proof. First, if U.q/ has a minimum at q , then
 
@U
D Qh .q / D 0; (17.98)
@q h q

and .q / is an equilibrium configuration (Theorem 17.6). Without loss of generality,

we can suppose that qh D 0, h D 1; : : : ; n. Now we prove that the total energy

V .q; v/ D T .q; v/ C U.q/ (17.99)

confirms the following conditions:

1. V .0; 0/ D 0;
2. V .q; v/ > 0 in a neighborhood of .0; 0/;
3. VP D 0 along any solution .q h .t/; vh .t// of (17.95), (17.96),
i.e., V .q; v/ is a Lyapunov function for system (17.95), (17.96). First, we have that
V .0; 0/ D U.0/ D 0 since the potential energy is defined up to an additive constant.
Further, the kinetic energy is a positive-definite quadratic form, and the hypothesis
that U.q/ has an effective minimum at 0 implies that U.q/ > 0 in a neighborhood of
0, and the second property is proved. Finally, the third property is confirmed since
the total energy is a first integral. t
u
Let qh D 0, h D 1; : : : ; n, be a configuration corresponding to a minimum
of the potential energy U.q/. For the preceding theorem it is a stable equilibrium
configuration. Consequently, for initial data .q0 ; v0 / in a suitable neighborhood I0 of
.0; 0/, the corresponding solutions remain in an assigned neighborhood I containing
I0 . We suppose that I is such that we can neglect in the following Taylor expansions
the terms of order higher than two in the variables q h , qP h :
17.13 Small Oscillations 315

1  h k
T D a qP qP ;
2 hk
   2 
@U 1 @ U
U D U.0/ C q C
h
qhqk ;
@q h 0 2 @@q h q k 0

where a D ahk .0/.

Since U.q/ is defined up to an arbitrary constant, we can suppose U.0/ D 0.
Further, U.q/ has a minimum at 0, and then .@U=@q h /0 D 0. Consequently, the
preceding expansions become

1  h k
T D a qP qP ; (17.100)
2 hk
1  h k
U D bhk q q ; (17.101)
2
where  
 @2 U
bhk D : (17.102)
@q h @q k 0
The quadratic form (17.100) is positive definite since all the principal minors of

the matrix .ahk / have positive determinants (Sect. 17.8). Also, the quadratic form
(17.101) is positive definite since U.q/ has an effective minimum at 0.
The motions corresponding to the approximate Lagrangian function

1  h k 1  h k
L .q; q/
P D a qP qP  bhk q q ; (17.103)
2 hk 2
which are called small oscillations about the stable equilibrium configuration 0, are
solutions of the following linear differential equations:
 k  k
ahk qP D bhk q : (17.104)

The analysis of small oscillations becomes trivial by the following considera-

tions. Let En be a Euclidean vector space equipped with the scalar product
 h k
u  v D ahk u v ; (17.105)

where u; v 2 En and uh ; vh denote their components in a basis .eh / of En . Let us

consider the eigenvalue equation
 k  k
bhk u D ahk u : (17.106)


Since the matrix .bhk / is symmetric and the quadratic form (17.10) is positive
definite, all its eigenvalues 1 ; : : : ; n are positive, and it is possible to find an

orthonormal basis .uh / of En , formed by eigenvectors of .bhk /. In this basis, the
316 17 Lagrangian Dynamics

 
matrices .ahk / and .bhk / have a diagonal form
0 1
1    0

.b hk / D @          A ;
0    n
0 1
1  0
.ahk / D @          A ;
0  1
the Lagrangian function (17.103) becomes

X  h 2 1 X  2
n n
P D 1
L .q; q/ qP  h q h ; (17.107)
2 i D1 2 i D1

and the corresponding Lagrange equations assume the simple form

p
qRh D  h q h : (17.108)

In conclusion, we have stated that small oscillations about a stable equilibrium

configuration, corresponding to an effective minimum of the potential energy, are

obtained by pcombining n harmonic motions along the eigenvectors of .bhk / with
frequencies h . These frequencies are called normal frequencies and harmonic
motions normal modes. The calculations presented in this section can be carried
out using the notebook SmallOscill.

17.14 Hamilton’s Principle

Newton’s laws, Euler’s equations, and Lagrange’s equations, which respectively

govern the dynamics of systems of material points, of a single rigid body, and
of systems of constrained rigid bodies, have a common aspect: they describe the
dynamical evolution by differential equations, that is, by relations between the
unknown functions describing the motion and their first and second time derivatives.
On the other hand, in physics there are many situations in which the problem we
confront does not spontaneously lead to a system of differential equations but to the
minimum value of an integral relation. In discussing some historical examples, we
discover a new formulation of mechanical laws.
• We start with the problem of the brachistochrone posed by Johann Bernoulli and
solved by Newton and Jakob Bernoulli. A brachistochrone curve, or a curve of
fastest descent, is the curve  between two points A and B that is covered in
the least time by a body that starts from the first point A with zero speed and
17.14 Hamilton’s Principle 317

A O
xb x

mg B

Fig. 17.6 Brachistochrone

is constrained to move along the curve to the second point B, with A no lower
than B, under the action of constant gravity and assuming no friction. In the
vertical plane  containing  , we introduce a frame of reference Oxy as shown in
Fig. 17.6. Let y.x/ be the equation of the curve  and denote by s the curvilinear
abscissa from the origin O. With our choice of the origin of coordinates and the
condition sP D 0 at the initial time, the conservation of energy reduces to the
relation
1 2
sP  mgy D 0:
2
Further, the length ds of the element of arc along  is given by
p
ds D 1 C y 02 dx:

Bearing in mind the preceding two relations, we obtain the time T to cover 
from A to B:
Z xb p
1 C y 02 .x/
TŒy.x/ D p dx: (17.109)
0 2gy.x/
By (17.109), a real number T corresponds to any C 1 .0; xb / function y.x/,
assuming the values y.0/ D 0 and y.xb / D yb , provided that the integral on the
right-hand side of (17.109) exists. Our problem will have a solution if we find a
function with the preceding properties that minimizes the value of T.
• As a second example we consider Fermat’s principle. Let S be an optical medium
with refractive index n.r/. This principle states that the path  taken between two
points A and B by a ray of light is the path that can be traversed in the least time.
318 17 Lagrangian Dynamics

Denoting by .x1 .t/; x2 .t/; x3 .t// the parametric equations of  , the time spent to
cover  with speed v D c=n is given by
Z tb q
T D n.x1 .t/; x2 .t/; x3 .t// xP1 2 C xP2 2 C xP3 2 dt: (17.110)
ta

It is well known that this principle includes in a very compact form all the laws
of geometric optics (e.g., reflection, refraction). Once again the path of a ray
minimizes the integral relation (17.110).
• Let  be a curve given by the function y.x/, a  x  b. Denote by S the surface
of revolution obtained rotating  about the Ox-axis. Determine the curve  for
which the area of S , given by
Z b p
S Œy.x/ D 2y.x/ 1 C y 02 dx; (17.111)
a

assumes the minimum value.

We define as functional any map

F W F ! <;

where F is a normed space of functions. Formulae (17.109)–(17.111) are examples

of functionals. Searching for a minimum of a functional is called a variational
problem. The elements of F at which F attains a minimum are said to be extremals
of the functional F .
Many situations can be described by a variational principle. For example, a soap
ball is spherical to minimize its surface for given internal and external pressures,
and a thin elastic membrane with a fixed boundary assumes a configuration
corresponding to a minimum of its area.
In view of the preceding examples, we are led to search for the extremals of a
functional of the form Z tb
I D P
L.q.t/; q.t// dt; (17.112)
ta
 
where q.t/ D q 1 .t/; : : : ; q n .t/ and the functions q h .t/ are of class C 1 Œta ; tb  and
satisfy the boundary conditions q.ta / D qa D .qa1 ; : : : ; qan / and q.tb / D qb D
.qb1 ; : : : ; qbn /. In what follows, we consider any function q.t/ as a curve of <n .
We do not introduce the wide mathematical apparatus that has been developed to
analyze the problem of the existence of extremals. We prefer to follow a very simple
approach, proposed by Euler and Lagrange, to obtain differential equations whose
solutions are the extremals we are searching for.
Let f W .s; t/ 2 Œ ;   Œta ; tb  be a family
 of curves, depending
 on the parameter
s, such that the parametric equations f 1 .s; t/; : : : ; f n .s; t/ belong to C 1 Œta ; tb .
17.14 Hamilton’s Principle 319

qn

qt qb

s
qa
t
q

Fig. 17.7 Family of curves f.s; t /

Further, we suppose that f.s; t/ contains the function q.t/ for s D 0. More explicitly,
we suppose that

f.s; ta / D qa ; f.s; tb / D qb ; 8s 2 Π; I (17.113)

f.0; t/ D q.t/: (17.114)

In other words, we are considering a family f.s; t/ of curves of <n , depending on

the parameter s, starting from the point qa and ending at the point qb . This family
reduces to the curve q.t/ for s D 0 (Fig. 17.7).
Now, introducing the family f.s; t/ into functional (17.112), we obtain a function
of the parameter s: Z tb  
I.s/ D L f.s; t/; Pf.s; t/ dt; (17.115)
ta

and we can state that q.t/ is an extremal of functional (17.112) if and only if
 
dI
D0 (17.116)
ds sD0

for any choice of the family f.s; t/ of curves satisfying properties (17.113) and
(17.114).
Since
@Pf @ @f @ @f
D D ;
@s @s @t @t @s
320 17 Lagrangian Dynamics

we can write
Z " #
dI tb
@L @f @L @Pf
D C dt
ds ta @f @s @Pf @s
 Z tb 
@L @f tb @L @ @L @f
D C  dt: (17.117)
@Pf @s ta ta @f @t @Pf @s

But, in view of (17.113), we have

 tb
@L @f
D 0;
@Pf @s ta

and (17.117), for s D 0, gives

  Z tb 
dI @L d @L @f
D  dt: (17.118)
ds sD0 ta @q dt @qP @s

To obtain the preceding relation, we substituted @=@t with d=dt since, for s D 0, all
the functions under the integral depend only on t; further, in view of (17.114), we
have that f.0; t/ D q.t/, Pf.0; t/ D q.t/.
P Comparing (17.116) and (17.118), we see
that a solution q.t/ of the Lagrange equations is an extremal of functional (17.112).
Conversely, if q.t/ is an extremal of functional (17.112), then (17.118) is satisfied
for every f.s; t/. Consequently, the right-hand side of (17.118)) vanishes for every
f.s; t/, and it is possible to prove that the function q.t/ is a solution of the Lagrange
equations.
In conclusion, the extremals of (17.112) are the solutions of the following
boundary value problem:

@L d @L
 D 0; q.ta / D qa ; q.tb / D qb : (17.119)
@q dt @qP

No general theorem of existence and uniqueness exists for the preceding boundary
value problem.
The preceding results prove the following theorem.
Theorem 17.8 (Hamilton’s principle). Let S be a dynamical system with La-
grangian function L. Then, an effective motion for which q.ta / D qa and q.tb / D qb
is an extremal of functional (17.112).
Functional (17.112) is called an action functional.
17.15 Geometric Formulations of Lagrangian Dynamics 321

17.15 Geometric Formulations of Lagrangian Dynamics

Up to now, our formulation of Lagrangian dynamics has used in a marginal way the
structures of differential geometry. Essentially, we have associated a differentiable
manifold Vn , the configuration space, with any dynamical system S . The dimension
n of Vn gives the degrees of freedom of S , i.e., the number of independent
parameters defining the configuration of S . Finally, the motion of S is represented
by a curve q.t/ of Vn , called the dynamical trajectory of S , satisfying the Lagrange
equations. This geometric formulation of Lagrangian dynamics is local since it uses
local coordinates .q h / on Vn ; in other words, there is no intrinsic formulation of this
theory. In this section, we present two interesting geometric versions of Lagrangian
dynamics in which any mechanical aspect assumes a geometric form.
Let S be a dynamical system with smooth and fixed holonomic constraints acted
upon by conservative active forces so that the dynamics of S is described by a
Lagrangian function L.q; q/. P We denote by Vn the configuration space of S . In
local coordinates .q 1 ; : : : ; q h / of Vn , the kinetic energy of S is

1
T D ahk .q/qP h qP k ; (17.120)
2
where the quadratic form on the right-hand side of (17.120) is positive definite and
the coefficients ahk are symmetric. Further, the functions ahk .q/ define a .0; 2/-
tensor over Vn since the kinetic energy is an invariant scalar with respect to the
coordinate transformations of Vn and qP h are the contravariant components of the
Lagrangian velocity relative to a coordinate basis .@=@q h /. These properties allow
us to equip Vn with the following Riemannian metric (Chap. 6):

ds 2 D ahk .q/dq h dq k : (17.121)

With the introduction of this metric, a scalar product is introduced in the tangent
space Tq .Vn / at any point q 2 Vn ; moreover, for any vector v 2 Tq .Vn / the following
relations between contravariant and covariant components hold:

vh D ahk vk ; vh D .a1 /hk vk ; (17.122)

 
where .a1 /hk is the inverse matrix of .ahk /. Consequently, the generalized
momenta
@L
ph D h D ahk qP k (17.123)
@qP
can be regarded as the covariant components of the Lagrangian velocity, that is, of
the vector v tangent to the dynamical trajectory of S . In formulae, we write

vh D qP h ; vh D ahk vk D ph : (17.124)
322 17 Lagrangian Dynamics

With these notations, the Lagrange equations (17.44) assume the following form:

@aj k j k
vP h  vP vP D Qh .q; v/: (17.125)
@x h

In Chap. 9, we proved that the Christhoffel symbols ijh and the metric coefficients
ahk of a Riemannian manifold are related by the equation

1 @aj k j k p p
vP vP D apj hk vj vk D hk vp vk ; (17.126)
2 @x h
in view of which we can write (17.125) in the final form

rvh dvh p
  hk vp vk D Qh .q; v/: (17.127)
dt dt
Taking into account that on the left-hand side of (17.127) are the covariant
components of acceleration in Lagrangian coordinates .q h /, we conclude that the
acceleration of the point q.t/, representing in the configuration space Vn the motion
of the material system S , is equal to the Lagrangian force .Qh /. In other words,
the motion of q.t/ in the configuration space Vn is governed by Newton’s law for a
material point with unit mass.
In particular, in the absence of forces, the dynamical trajectory q.t/ is a geodesic
of the metric (17.121).
In the preceding considerations, the constraints to which the dynamic system is
subject and the reactive forces exerted by them were geometrized by introducing a
configuration space equipped with the Riemannian metric (17.121). Now we want
to prove that it is possible to geometrize the active forces, provided that they derive
from a potential energy U.q/.
To prove this statement, we note that, for conservative forces, the total energy E
is constant during motion, i.e, T .q; q/P C U.q/ D E. Consequently, the dynamical
trajectory, corresponding to a fixed value of the total energy, is contained in the
region Wn  Vn defined by the inequality

T D E  U.q/  0 ) U.q/  E: (17.128)

Now we can prove the following theorem.

Theorem 17.9 (Maupertuis’ principle). Let S be a mechanical system with
smooth and fixed holonomic constraints, subject to conservative forces with poten-
tial energy U.q/. Then the dynamical trajectories of S corresponding to a given
total energy E are geodesics of the configuration space Vn , equipped with the
Riemannian metric

d 2 D 2.E  U /ahk dq h dq k  F ahk dq h dq k : (17.129)

17.15 Geometric Formulations of Lagrangian Dynamics 323

These geodesics are covered with the following velocity:

d
D 2.E  U /  F: (17.130)
dt

Proof. The geodesic of metric (17.129), connecting two points q0 and q1 of Vn , is

the extremal of the functional (Chap. 9)
s
Z q1
dq h dq k
I Œq.s/ D 2.E  U /ahk d; (17.131)
q0 d d
where the parameter along the curves is the curvilinear abscissa relative to metric
(17.129). We recall (Chap. 9) that, owing to the form of the preceding functional, the
extremals are independent on the parameterization. In other words, the geodesics are
defined as loci of points. The Lagrange equations corresponding to the functional
(17.131) are [see (6.93)]

d 1 @ dq 0j
.F ahk q 0k /  .F aj k /q 0j q 0k D 0; q 0j D :
d 2 @q h d

These equations can also be written as follows:

d F @aj k 0j 0k 1 @F
.F ahk q 0k /  q q D aj k q 0j q 0k : (17.132)
d 2 @q h 2 @q h

On the other hand, from the conservation of the total energy we have that

2T D aij qP j qP k D 2.E  U /:

Further,  D .t/, and the preceding condition assumes the form

 2
0j 0k d
aj k q q D 2.E  U /:
dt

In view of (17.129), F aj k q 0j q 0k D 1, and the previous equation becomes

 2
d
D 4.E  U /2 ;
dt

and we obtain (17.130) when we recall that T D .E  U /  0. Now we determine

the equations of geodesics, corresponding to metric (17.129), and we show that they
coincide with the Lagrange equations. In fact, these equations,

d @T @T @U
 D  h;
dt @qP h @q h @q
324 17 Lagrangian Dynamics

can also be written as follows:

d 1 @aj k j k @U
.ahk qP k /  qP qP D  h :
dt 2 @q h @q

Introducing the parameter .t/ and taking into account (17.130), we obtain the
equations
d 1 @aj k 2 0j 0k @U
F .F ahk q 0k /  F q q D  h;
d 2 @q h @q
which can be put in the final form
d F @aj k 0j 0k 1 @U 1 @F
.F ahk q 0k /  q q D D : (17.133)
d 2 @q h F @q h 2F @q h
The theorem is proved comparing (17.33) and (17.32). t
u
Remark 17.4. Given the mechanical system S and a value E of the total energy,
metric (17.129) is determined. Assigning the initial data q h .0/ D q0h and q 0h .0/ D
q00h , that is, the initial position and direction of the initial velocity, we determine
by (17.32) the geodesic starting from this initial position in the given direction. In
this way, we know the dynamical trajectory but we do not know how it is covered.
The condition that during the motion the energy has a given value implies that the
geodesic must be covered with the law (17.130).

17.16 Legendre’s Transformation

In this section, we analyze the path leading from the Lagrangian formulation of
mechanics to the Hamiltonian one. This path, in the case of constraints independent
of time, consists of the following fundamental steps.
1. First, we transform the Lagrange equations into an equivalent system of 2n
first-order differential equations by considering the Lagrangian velocities qP as
auxiliary unknowns. A solution xL .t/ of this system is a curve of the velocity
space, i.e., of the fiber bundle T Vn of the configuration space. In this space, the
Lagrange equations assume the form
xP L D XL .xL /; (17.134)
where XL .x/ is a vector field on T Vn .
2. Then we define the Legendre transformation, i.e., a diffeomorphism L W T Vn !
T  Vn between the tangent fiber bundle T Vn and the cotangent fiber bundle
T  .Vn /.
3. Finally, by the differential

L W T.q;v/ .T Vn / ! TL.q;v/ .T  Vn /; (17.135)

17.16 Legendre’s Transformation 325

TVn
XL
XH
q ,v

xL t xH t

Fig. 17.8 Legendre transformation

we obtain a new system of first-order differential equations,

xP H D XH .xH /; (17.136)

where XH D L .q; v/ 2 T  Vn . It is evident that the solutions of this new system

are curves of T  Vn (Fig. 17.8).
To extend the foregoing procedure to constraints depending on time, it is sufficient
to substitute Vn with Vn  <.
Let S be a dynamical system described by a Lagrangian function L.q; q; P t/. The
Lagrangian equations

d @L @L
 h D 0; h D 1; : : : ; n;
dt @qP h @q

are explicitly written as

LqP h q k qP k C LqP h qP k qR k C Lt q h  Lq h D 0; h D 1; : : : ; n; (17.137)

where we use the notation @f =@x D fx . From L.q; q; P t/ D T .q; q;

P t/  U.q; t/ and
(17.36) we have that
P t/;
LqP h qP k D ahk .q; q;
where det.ahk / ¤ 0; consequently, there exists the inverse matrix .ahk / D .a1 /hk .
Multiplying (17.137) by .ahl /, we obtain the Lagrangian equations in normal form:

qR h D ahl .Lq l  Lt qP l  LqP l q k qP k /; h D 1; : : : ; n: (17.138)

Remark 17.5. The kinetic energy is a scalar quantity, i.e., it is invariant with
respect to changes in the Lagrangian coordinates. Consequently, the quadratic form
1
P h qP k and the linear form bh qP h of (17.36) are invariant with respect to these
2 ahk q
transformations. Recalling the criteria that give the tensorial character of a quantity
and noting that .qP h / is a vector of Vn , we can state that .ahk / and .bh / are,
326 17 Lagrangian Dynamics

respectively, the components of a covariant 2-tensor and a covariant vector of Vn .

Then the generalized momenta
ph D LqP h D ahk qP k C bh

are the components of a covector of Vn .

Lagrangian function independent of time. Introducing the auxiliary unknowns
qP h D vh , Eqs. (17.138) are equivalent to the first-order system
qP h D vh ; (17.139)
vP h D ahl .Lq l  Lvl q k vk /; (17.140)

in the 2n unknowns q h .t/; vh .t/, h D 1; : : : ; n. These unknowns are the parametric

equations of a curve  belonging to T Vn , whereas the right-hand sides of (17.140)
define a vector field XL on T Vn whose components in a natural basis .@=@q h ; @=@vh /
of T Vn are
XL D .vh ; ahl .Lq l  Lvl q k vk /  .X h ; Y h /: (17.141)
The vector field XL has the property that, at any point xL  .q; v/ 2 T Vn , the first
set of the components relative to the natural basis of T Vn is equal to the second set
of the coordinates of the point of T Vn at which xL is tangent to T Vn . Such a field is
called a semispray vector field.
Independently of the adopted coordinates, system (17.139), (17.140) can be
written as
xP L D XL .x/: (17.142)
To simplify the notations, we set M D T Vn and M  D T  Vn and adopt on
both fiber bundles natural coordinates .q h ; vh / and .q h ; ph /, respectively. Then we
introduce the Legendre transformation

L W M ! M (17.143)

that, in natural coordinates, is defined as follows (Remark 17.5):

.q h ; ph / D L.q h ; vh / D .q h ; Lvh .q; v//: (17.144)

The differential of this mapping

L W .X h ; Y h / 2 T.q;v/ .M / ! .X h ; Y h / 2 TL.q;v/ .M  / (17.145)

is such that
    
X h ıkh 0 Xk
D
Yh Lvh q k Lvh vk Yk
D .X h ; Lvh q k X k C Lvh vk Y k /: (17.146)
17.16 Legendre’s Transformation 327

Taking into account (17.144) and (17.146), we conclude that the Legendre transfor-
mation L associates to a point xL D .q h ; vh / 2 M a point x D .q h ; ph / 2 M 
and to a tangent vector X D .X h ; Y h / 2 Tx M a tangent vector X 2 Tx M  . If
this correspondence is applied to the (17.142), we obtain the transformed equation
in M 
xP  D X .x /; (17.147)
where x D .q h ; ph /. Further, recalling that Lvh vk D ahk and taking into account
(17.141), we obtain
 
X D vh ; Lvh q k vk C ahk akl .Lq l  Lvl q m vm / D .vh ; Lq h /: (17.148)

Consequently, in natural coordinates, (17.147) leads to the system

qP h D vh ; (17.149)
pPh D Lq h : (17.150)

Introducing the Hamiltonian function


H.q; p/ D pk vk  L.q; v/ vDv.q;p/ ; (17.151)

where vh D vh .q; p/ is the inverse function of ph D Lvh D ahk vk , we have that

@vk @vk
Hq h D pk h
 Lq h  Lvk h D Lq h ;
@q @q
@vk @vk
Hph D vh C pk  Lv k D vh ;
@q h @q h

and system (17.149), (17.150) reduces to the equations

@H
qP h D ; (17.152)
@ph
@H
pPh D  ; (17.153)
@q h

which are called canonical Hamiltonian equations.

Lagrangian Function Dependent on Time In this case, the Lagrangian function
P t/ is defined on M  <. Consequently, we again introduce the auxiliary
L.q; q;
unknowns vh D qP h but write system (17.138) in the form
328 17 Lagrangian Dynamics

qP h D vh ; (17.154)
 
vP h D a hk
Lq k  Lt vk  Lvk q l v ;
l
(17.155)
tP D 1: (17.156)

The solutions x.t/ D .q.t/; v.t/; t/ of this system are curves of M  < that are the
integral curves of the vector field
   
XL D .X h ; Y h ; 1/ D vh ; ahk Lq k  Lt vk  Lvk q l q l ; 1 : (17.157)

Using these notations, system (17.154)–(17.156) can be written in the form

xP D XL .xx; t/: (17.158)

In natural coordinates of M and M  , the Legendre transformation

L W M  < ! M  < (17.159)

is defined as
 
L.q h ; vh ; t/ D q h ; Lvh .q; v; t/; t ; (17.160)

and its differential is

0 10 k 1
ıkh 0 0 X
L .X / D @ Lvh q k Lvh vk Lvh t A @ Y k A
0 0 1 X
 h 
D X ; Lvh q k X k C Lvh vk Y k C Lvh t X; X : (17.161)

It is a simple exercise to verify that the Legendre transformation reduces system

(17.154)–(17.156) to the form

@H
qP h D ; (17.162)
@ph
@H
pPh D  (17.163)
@q h
tP D 1; (17.164)

where now H D H.q; p; t/.

17.17 Exercises 329

17.17 Exercises

1. Let P be a material point subject to a Newtonian force having its center at

point O. Introducing spherical coordinates .r; '; / with the origin at O, prove
that the kinetic energy T and the potential energy are

1  2 
T D m rP C r 2 'P 2 C r 2 P 2 sin2  ;
2
h
U D ;
r

where h is a constant. Write the corresponding Lagrange equations and

recognize the meaning of the first integral corresponding to the cyclic variable
'. Which is the configuration space?
Hint: The spherical and Cartesian coordinates are related by the relations

x D r cos ' sin ;

y D r sin ' sin ;
z D r cos :

Further, the square of velocity of P is xP 2 C yP 2 C zP2 .

2. Prove that the kinetic energy and the potential energy of the spherical pendulum
P are
1 2 2 
T D mr 'P C P 2 sin2  ;
2
U D mgr cos ;

where g is the gravity acceleration and r the radius of the sphere S on which
P is constrained to move. Write the Lagrange equations and determine the
configuration space.
3. Using the notations of Sect. 16.2, prove that the kinetic energy and the potential
energy of a compound pendulum are

1
T D C 'P 2 ;
2
U D mgh cos ':

Determine the Lagrange equation and the configuration space.

4. Let S be a heavy gyroscope (Sect. 16.7). Denoting by , ', and  Euler’s
angles, prove that the kinetic energy and the potential energy are
330 17 Lagrangian Dynamics

y
B
B
G
l
A
p
j
A x
O
i

Fig. 17.9 Constrained heavy bar

1  P2 2 
T D A sin  C P 2 C C.'P C P cos /2 ;
2
U D mghG cos :

Write the Lagrange equations and the first integrals related to the cyclic
coordinates.
Hint: It is sufficient to recall (15.21)

1 2 
T D Ap C Bq 2 C C r 2
2
and the Euler relations of Sect. 12.4

p D P sin  sin ' C P cos ';

q D P sin  cos '  P sin ';
r D P cos  C ':
P

5. Let S be a homogeneous heavy bar with points A and B constrained to move

without friction on the Ox- and Oy-axes, respectively (Fig. 17.9). Let 2l be
the length of S . Determine the motion of S by the balance equations. Then,
noting that S 1 is the configuration space of S , determine the Lagrange equation
governing the motion of S .
Hint: The momenta of ˆ A and ˆ B relative to center of mass G are
!
GA  ˆ A j D l.sin 'i  cos 'j/  ˆ A j D lˆ A sin 'k;
!
GB  ˆ B i D l. sin 'i C cos 'j/  ˆ B i D lˆB cos 'k;

respectively, where k D i  j, and I is the momentum of inertia relative to

an axis orthogonal to Oxy and containing G. Then the components along the
axes Ox, Oy, and Oz (with Oz orthogonal to the plane Oxy) of the balance
equations are
17.17 Exercises 331

B
y G
l
j A

O i x Q

Fig. 17.10 Bar moving in a plane

mxR G D ˆB ;
myRG D p C ˆA ;
I 'R D l.ˆ A sin '  ˆB cos '/:

Introduce into the third equation the expressions of ˆA and ˆB deduced from
the first two equations. Finally, the Lagrange equation is obtained by proving
that
1 
T D I C ml 2 'P 2 ;
2
U D pl cos ':

6. A homogeneous bar AB is constrained to move on a smooth plane Oxy

(Fig. 17.10). At points A and B two elastic forces are applied with centers
O and Q, respectively. Further, 2l is the length of AB, M its mass, G the
center of mass, and C the momentum of inertia relative to G. Show that the
configuration space is <2  S 1 and the kinetic and potential energies are

1 1
T D M.xP 2 C yP 2 / C C 'P 2 ;
2 2
1

U D k1 .x  l cos '/2 C .y  l sin '/2
2
1

D k2 .x C l cos '  L/2 C .y C l sin '/2 ; (17.165)
2

where k1 and k2 are the elastic constants of the two springs.

Hint: The velocity of G is given by

rP G D xi
P C yj;
P
332 17 Lagrangian Dynamics

j B
l G G l
O x

i A C

Fig. 17.11 Two constrained bars

and applying Koenig’s theorem the expression of T is obtained. It is a simple

exercise to obtain the preceding expression of the potential energy.
7. Let S be the plane system in Fig. 17.10 formed by two homogeneous bars
of equal length (Fig. 17.11). Prove that the configuration space is <  S 1
(a cylinder), and verify that the kinetic energy and the potential energy are
given by

T D T1 C T2 ;
U D U1 C U2 ;

with
1 1
T1 D .C1 C ml 2 /'P 2 C mxP 2  ml sin ' xP ';
P
2 2
1  1
T2 D C C ml 2 .9 sin2 ' C cos2 '/ 'P 2 C mxP 2  3ml xP 'P sin ';
2 2
k
U1 D x 2 ;
2
k
U2 D .4l cos '  x/2 :
2
Hint: The center of mass G1 has the following coordinates:

xG1 D x C l cos ';

yG1 D l sin ';

whereas the coordinates of G2 are

xG2 D x C 3l cos ';

yG2 D l sin ':

Apply Koenig’s theorem to each bar.

8. Let AB be a homogeneous heavy bar rotating about an axis a orthogonal to
the plane Oxy and containing point A, which, in turn, is constrained to move
along the Ox-axis (Fig. 17.12). Prove that the configuration space is <  S 1
17.17 Exercises 333

j
O x A l
i
G

p B

Fig. 17.12 Pendulum with point A moving along Ox

(cylinder) and the kinetic energy and the potential energy are

1 1
T D .C C ml 2 /'P 2 C mxP 2 C ml xP 'P sin ';
2 2
k 2
U D x  pl cos ':
2
Write the Lagrange equations and determine the equilibrium configurations.
Hint: The coordinates of the center of mass are xG D x C l cos ', yG D
l cos '. Use Koenig’s theorem.
9. Let P1 and P2 be two heavy material points, with masses m1 and m2 ,
respectively, moving without friction on a circumference of radius r. Between
the two points acts a spring as shown in Fig. 17.13. Prove that the configuration
space is a torus S 1  S n and the kinetic energy T and the potential energy U
are
1 1
T D m1 r 2 'P12 C m2 r 2 'P22 ;
2 2
k 2
U D r .'2  '1 /2 C m1 gr sin '2 C m2 r sin '2 : (17.166)
2
Determine the Lagrange equations and the equilibrium positions.
10. A homogeneous rigid disk S rotates about an axis orthogonal to the plane of S
and containing its center O. Along a diameter of S a material point P moves
without friction (Fig. 17.14). Denoting by M and m the masses of S and P ,
respectively, prove that the kinetic energy and the potential energy are

1 2 1
T D mPs C .C C M s 2 /'P 2 ; (17.167)
2 2
k 2
U D r .'2  '1 /2 C m1 gr sin '2 C m2 r sin '2 : (17.168)
2
Determine Lagrange’s equations and find a first integral of them.
334 17 Lagrangian Dynamics

r P
j

O i x

Fig. 17.13 Two heavy mass points on a circumference

P
s
O ϕ

Fig. 17.14 Disk and a material point

11. Prove that the potential energy of a double pendulum (Fig. 17.15) is

U D p1 l1 cos '  p2 .2l1 cos ' C l2 cos /:

Determine the configuration space, the kinetic energy, the equilibrium configu-
rations, and the small oscillations around the stable equilibrium configurations.
Hint: Applying Koenig’s theorem to both bodies of the double pendulum with
masses m1 and m2 , we obtain

1 1 1 1
T D m1 rPG2 1 C C1 'P 2 C m2 rPG2 2 C C2 P 2 :
2 2 2 2
On the other hand, denoting k D i  j, we have that

rP G1 D 'kl
P 1 .sin 'i  cos 'j/ D l1 '.cos
P 'i C sin 'j/
17.17 Exercises 335

y
j
O i x

G
p

p l2
B

Fig. 17.15 Double pendulum

and

P 
rP G2 D rP A C k
!
AG 2
D 2'l P
P 1 .cos 'i C sin 'j/ C k.sin i  cos j/
D .2l1 'P cos ' C l2 P cos /i C .2l1 'P sin ' C l2 P sin /j:

12. Write the kinetic energy and the potential energy of a bi-atomic molecule
supposing that the interaction force between the two atoms is elastic.
13. Let O z be an inertial frame of reference and denote by Oxyz another frame
uniformly rotating about Oz with respect to O z. Determine the generalized
potential of the fictitious forces acting on a mass point P moving in the plane
Oxy.
Hint: Apply (17.63) and (17.64).
Chapter 18
Hamiltonian Dynamics

18.1 Symplectic Structure of the Phase Space

Let Vn be the configuration space of a dynamical system S subject to fixed and

smooth constraints and to conservative forces. In the last section of the preceding
chapter, we proved that the Legendre map transfers the dynamics of S from the

tangent fiber bundle M2n D T Vn to the cotangent fiber bundle M2n D T  Vn .

Henceforth we call M2n the phase space of the dynamical system S . In this space,
the dynamical trajectories of S are solutions of the Hamilton equations (17.152)
and (17.153).1
We recall some results regarding the tangent and cotangent fiber bundles
(Chap. 6). If .U; q/ is a chart on the configuration space Vn , then .@=@q h /q and
.dq h /q denote a basis of the tangent vector space Tq .Vn / and a basis of the cotangent

vector space Tq .Vn /, respectively. An element x 2 M2n is a pair .q; p/, with q 2 Vn
and p 2 Tq .Vn /, that is determined by the 2n numbers .q; p/ 2 U  <n . The charts


.U  <n ; .q; p// are called natural charts. It is evident that an atlas on Vn generates
 
an atlas of natural charts on M2n , called a natural atlas on M2n .
n
A natural basis relative to a natural chart .U < ; .q; p// is formed by the vectors
.@=@q h ; @=@ph / tangent to the coordinate curves.

In a natural chart .U  <n ; .q; p// of M2n , consider the differential form2

! D ph dq h : (18.1)

1
For the contents of Chapters 18–20 see also [1, 3, 19, 30].

2
The most general differential form on M2n is

! D Xh .q; p/dq h C X h .q; p/dph :

Then (18.1) is a particular nonexact differential form.

A. Romano, Classical Mechanics with Mathematica R

, Modeling and Simulation 337
in Science, Engineering and Technology, DOI 10.1007/978-0-8176-8352-8 18,
© Springer Science+Business Media New York 2012
338 18 Hamiltonian Dynamics

  T
Similarly, in another natural chart U  <n ; .qh ; p/ , with U U ¤ ;, we define
a new differential form
! D p h dq h : (18.2)
Recalling that a transformation from natural coordinates to natural coordinates is
expressed by the equations

q h D q h .q/; (18.3)
@q k
ph D .q/pk I (18.4)
@q h
S
in the intersection U U between the two charts, we have that

! D !; (18.5)

and we can state that the local differential forms (18.1) determine a unique

differential form ! on the whole manifold M2n . Together with this global differential

form, we consider on the whole M2n the 2-form

 D d!; (18.6)

which in any natural chart has the following coordinate representation:

dq h ^ dph : (18.7)

In view of (18.6), the 2-form  is closed, that is,

d D 0: (18.8)

Further, if  is intended as a skew-symmetric tensor, then in the natural basis (dq h ˝

dq k ; dq h ˝ dpk ; dph ˝ dqh ; dph ˝ dpk / of T.q;p/ M  , its components are given by
the 2n  2n skew-symmetric matrix
 
0 I
D ; (18.9)
I 0

where 0 and I denote, respectively, an n  n zero matrix and an n  n unit matrix.

Theorem 18.1. The matrix  has the following properties:

det  D 1; 2 D I; 1 D : (18.10)

Proof. Equation (18.10)1;2 follows at once from (18.9). Further, we also have that
./ D I, and (18.10)3 is proved. t
u
18.2 Canonical Coordinates 339


Definition 18.1. The pair .M2n ; /, where  is a global closed and nondegenerate
2-form, is called a symplectic manifold. It is the fundamental geometric structure of
Hamiltonian mechanics independent of time.

18.2 Canonical Coordinates


Definition 18.2. A local chart .U; .q; p// on M2n is a canonical or symplectic chart
if in the coordinates .q; p/ the components of  are given by matrix (18.9).
In the preceding section, it was proved that natural coordinates are canonical.
To determine other canonical coordinates, denote by .q; p/ natural coordinates and
consider the coordinate transformation

q D q.q; p/; (18.11)

p D p.q; p/: (18.12)

We prove the following theorem.

Theorem 18.2. The coordinate transformation (18.11), (18.12) is canonical if and
only if the Jacobian matrix
0 1
@q @q
B @q @p C
SDB @ @p @p A
C (18.13)
@q @p

satisfies one of the following conditions:

 D .S1 /T S1 ; (18.14)

T 1
S  D S ; (18.15)
 D SST : (18.16)

Proof. Introducing the notations .x ˛ / D .q; p/, .x ˛ / D .q; p/, ˛ D 1; : : : ; 2n, the
transformation (18.11), (18.12) can be written as follows:

x ˛ D x ˛ .x ˇ /: (18.17)

Since  is a .0; 2/-tensor, it is transformed according to the rule

@x  @x  @x  @x 
˛ˇ D ˛   D ˛   :
@x @x ˇ @x @x ˇ
340 18 Hamiltonian Dynamics

Then transformation (18.13) is canonical if and only if ˛ˇ D ˛ˇ , and (18.14)
is proved. Multiplying (18.14) on the left by ST , we obtain (18.15). Finally,
from (18.14) we have that
1 D S1 ST ;
and condition (18.16) is proved when we recall (18.10)3. t
u
1
Definition 18.3. Let fh .q; p/, gh .q; p/ be h D 1; : : : ; n, 2n C functions on an
open region of M  . We call the function

X n  
@fj @gj @fj @gj
Œq h ; pk    (18.18)
j D1
@q h @pk @pk @q h

a Lagrange bracket of 2n functions with respect to a pair of variables .q h ; pk /. In a

similar way, we define Lagrange’s bracket with respect to a pair .q h ; q k / or .ph ; pk /.
Definition 18.4. Let f .q; p/ and g.q; p/ be two C 1 functions on an open region of
M  . We call the function

X n  
@f @g @f @g
ff; gg   : (18.19)
j D1
@q j @pj @pj @q j

a Poisson bracket.
It is a simple exercise to prove the following theorem.
Theorem 18.3. Conditions (18.14)–(18.16) are equivalent, respectively, to

Œq h ; q k  D 0; Œq h ; pk  D ıkh ; Œph ; pk  D 0I (18.20)

@p k @ph @q k @ph
h
D  k; h
D ;
@q @q @q @p k
@p k @q h @q k @q h
D k; D I (18.21)
@ph @q @ph @p k
n o n o
q h ; q k D 0; q h ; p k D ıkh ; fp h ; p k g D 0: (18.22)

We conclude by noting that the composition of two symplectic transformations is

still symplectic; the identity is symplectic, and the inverse of a symplectic transfor-
mation is also symplectic. In other words, the set of symplectic transformations is a
group with respect to the composition. Such a group is called a symplectic group of
canonical transformations.
18.3 Generating Functions of Canonical Transformations 341

18.3 Generating Functions of Canonical Transformations

For now, we do not know if there are canonical charts besides the natural
ones. Further, although the preceding criteria are necessary and sufficient for the
canonicity of a transformation, it is not an easy task to obtain by them a canonical
chart. In this section, we show a different approach to overcoming this difficulty.

We have already stated that the charts .U; .q; p// and .U ; .q; p// of M2n are
symplectic if and only if [see (18.7) and (18.9)]

dq k ^ dph D dq k ^ dp h (18.23)

T
in U U . This condition is equivalent to

d.ph dq h / D d.p h dq h /;

which, in turn, can also be written as

p h dq h D ph dq h  df; (18.24)

T
where f is a C 1 function in U U . We show that this condition, which is
equivalent to (18.23), allows us to generate a symplectic transformation for a
convenient choice of the function f . For this reason, f is called a generating
function of the canonical transformation.
First, we note that system (18.11), (18.12) defines a canonical transformation if
it is invertible, i.e., if
0 1
@q @q
B @q @p C
det B C
@ @p @p A ¤ 0; (18.25)
@q @p

and it confirms condition (18.23) or, equivalently, (18.24). To understand how we

can generate a symplectic transformation starting from a function f , we must
consider system (18.11), (18.12) as a system of 2n equations in 4n unknowns. In
particular, we already know that this system, in view of (18.25), establishes a one-to-
one correspondence between the 2n variables .q; p/ and the variables .q; p/. Now,
we suppose that this defines a one-to-one correspondence between a suitable choice
of 2n variables among the 4n variables .q; p; q; p/ and the remaining ones.
For instance, suppose that (18.11) and (18.12) satisfy the further condition

 
@q h
det ¤ 0: (18.26)
@pk
342 18 Hamiltonian Dynamics

Then, locally, (18.11) and (18.12) can be written as

p D p.q; q/; (18.27)

p D p.q; q/; (18.28)

and the canonicity condition (18.24) becomes

p h dq h D ph dq h  df .q; p.q; q//  ph dq h  dfO.q; q/: (18.29)

Expanding the differential dfO, we obtain the following equality:

! !
@fO h @fO
ph C dq  ph  h dq h D 0;
@q h @q

in which dq h and dq h are arbitrary. Therefore, (18.29) is equivalent to the following

conditions:

@fO
ph D  .q; q/; (18.30)
@q h
@fO
ph D .q; q/: (18.31)
@q h

To put the canonical transformation (18.30), (18.31) into the form (18.11), (18.12),
we must suppose that fO.q; q/ satisfies the following condition:
!
@2 fO
det ¤ 0: (18.32)
@q h @q k

As a further example, we assume .q; p/ as fundamental variables and suppose

that (18.11), (18.12) can be written as

p D p.q; p/; (18.33)

q D q.q; p/: (18.34)

Then, (18.24) gives the condition

p h dq h D ph dq h  df .q; p.q; p//  ph dq h  dfO.q; p/;

which is equivalent to

ph dq h C q h dp h D dfQ.q; p/; (18.35)

18.4 Isomorphism Between Vector Fields and Differential Forms 343

where
fQ.q; p/ D fO.q; p/ C q h .q; p/p h :
Expanding the differential of fQ.q; p/ and recalling the arbitrariness of dq h and dp h ,
we obtain that (18.29) is equivalent to the system

@fQ
ph D .q; p/; (18.36)
@q h
@fQ
qh D .q; p/: (18.37)
@p h

To put the canonical transformation (18.36), (18.37) into the form (18.11), (18.12),
we now suppose that fQ.q; p/ satisfies the condition
!
@2 fQ
det ¤ 0: (18.38)
@q h @p k

In conclusion, there are infinite canonical charts on M  besides the natural ones.

18.4 Isomorphism Between Vector Fields

and Differential Forms

Up to now, the skew-symmetric 2-tensor or the 2-form  has played a fundamental

 
role in selecting the canonical charts on M2n . Further, at any point x 2 M2n , it

defines an antiscalar product x .X; Y/, 8X; Y 2 Tx M2n , in any vector tangent

space Tx M2n that becomes a symplectic space. In this section, we present another
fundamental role played by .
We already know (Chap. 6) that on a Riemannian manifold Vn the metric tensor
g defines at any point x 2 Vn an isomorphism between the tangent space Tx .Vn / and
its dual space Tx .Vn /. Similarly, on the symplectic manifold .M2n 
; / we consider
the linear map

X 2 Tx M2n ! ! 2 Tx M2n


such that

!.Y/ D x .X; Y/; 8 Y 2 Tx M2n : (18.39)
Let .U; x ˛ /, ˛ D 1; : : : ; 2n, be any local chart on M  in a neighborhood of x. If we
denote by .@=@x h / the corresponding natural basis at x and by dx ˛ the dual basis,
then (18.39) has the following coordinate representation:

!˛ D .x /ˇ˛ X ˇ : (18.40)

344 18 Hamiltonian Dynamics

Since, in view of (18.10), det..x /˛ˇ / D 1, the linear mapping (18.40) is an

isomorphism.
In a symplectic chart .U; x ˛ / D .U; q; p/, we have that
   
@ @ @
D ; ;
@x ˛ h
@q @ph
.dx ˛ / D .dq h ; dph /;
@ @
X D Xh C Xh ;
@q h @ph
! D !h dq h C ! h dph ;

and (18.40) becomes

 
0 I
.!1 ; : : : ; !n ; ! 1 ; : : : ; ! n / D .X 1 ; : : : ; X n ; X1 ; : : : ; Xn / ;
I 0

so that

!h D Xh ; (18.41)
h h
! DX : (18.42)

Henceforth we denote the 1-form corresponding to X by isomorphism (18.39) and

the vector X corresponding to ! by the inverse isomorphism, respectively with the
notations (X flat, ! sharp)
! D X[ ; X D !] : (18.43)

We conclude this section by noting that, since  is a closed 2-form, from (8.28)
we obtain
LX  D d.iX /: (18.44)

But in any system of coordinates, we have that

.iX /˛ D X ˇ ˇ˛ ;

so that
i X  D X[ : (18.45)

In conclusion, from (18.44) and (18.45) we obtain the useful result

LX  D d.X[ /; 
8 X 2 M2n : (18.46)
18.5 Global and Local Hamiltonian Vector Fields 345

18.5 Global and Local Hamiltonian Vector Fields

 
Definition 18.5. Let M2n be the phase state. A vector field X 2 M2n is globally

Hamiltonian if there exists a differentiable function H.x/ 2 F M2n such that

XH D .dH /] : (18.47)

The function H.x/ is called a Hamiltonian function of X.

The following theorem is evident.
 
Theorem 18.4. The set H M2n of Hamiltonian fields on M2n coincides with the

image of the exact 1-forms on M2n by the linear isomorphism ]. In particular,
 
H M2n is a subspace of M2n .
In symplectic coordinates .q h ; ph /, from (18.41) and (18.42) we obtain
0 1 0 1
@H @H
B @q C ] B @p C
dH D B C B
@ @H A ! XH D @ @H
C:
A (18.48)

@p @q

The integral curves .t/ of a globally Hamiltonian vector field satisfy the
equation
P .t/ D XH ..t// D .dH /] ..t//; (18.49)

which in symplectic coordinates assumes the canonical form

@H
qP h D ;
@ph
@H
pP h D  :
@q h

Definition 18.6. A dynamic Hamiltonian system is a triad .M; ; H /.


Definition 18.7. A vector field X 2 M2n is said to be locally Hamiltonian if

LX  D 0; (18.50)

i.e., if  is invariant along the integral curves of the vector field X or, equivalently,
if it is invariant with respect to the one-parameter group of diffeomorphisms
generated by X.
346 18 Hamiltonian Dynamics

Theorem 18.5. If X is globally Hamiltonian, then it is locally Hamiltonian. If X


is locally Hamiltonian, then 8x 2 M2n there exist a neighborhood U of x0 and a
1
function H of class C .U / such that

X D .dH /] ; 8 x 2 U: (18.51)

Proof. The hypothesis that X is globally Hamiltonian implies X D .dH /] , and then
X[ D dH . From (18.46) it follows that LX  D 0. Conversely, in view of (18.46),
the condition LX  D 0 is equivalent to d.X[ / D 0, and this condition, in turn,
implies the existence of a function H for which X[ D dH only locally. t
u

Definition 18.8. A one-parameter group of diffeomorphisms 't of M2n is said to be
symplectic if  is invariant with respect to any diffeomorphism 't , that is, if (18.50)
holds.

18.6 Poisson Brackets


Definition 18.9. 8f , g 2 F M2n , the function

ff; gg D .df ] ; dg ] / D .Xf ; Xg / (18.52)

is the Poisson brackets of f and g.

Theorem 18.6. In a symplectic system of coordinates, (18.52) gives
n 
X 
@f @g @f @g
ff; gg D  : (18.53)
@q h @ph @ph @q h
hD1

Proof. In fact,

@f @f
df D dq h C h dph ;
@q h @q
@g h @g
dg D h
dq C h dph :
@q @q

Owing to (18.41) and (18.42), in symplectic coordinates we have that

@f @ @f @
df ] D  ;
@ph @q h @q h @ph
@g @ @g @
dg ] D h
 h ;
@ph @q @q @ph
18.6 Poisson Brackets 347

and in matrix form (18.52) gives

0 1
   @g

@f @f 0 I B h C
B @q C ;
ff; gg D  h @ @g A (18.54)
@ph @q I 0 
@ph

and (18.53) is proved. t

u

Theorem 18.7. 8f , g 2 F M2n , we have that

ff; gg D LXg f D LXf g; (18.55)

 
Xf ; Xg D Xff;gg : (18.56)

Proof. In view of (18.39) and (18.43), we have that

ff; gg D .df ] ; dg ] / D df .Xg / D Xg f D LXg f:

From the skew symmetry of  follows (18.55). We omit the proof of (18.56). t
u

Theorem 18.8. 8f; g; h 2 F M2n and 8a; b 2 <, the following identities hold:

ff; f g D 0; (18.57)
ff; gg D  fg; f g ; (18.58)
ff; ag C bhg D a ff; gg C b ff; hg ; (18.59)
ff; fg; hgg C fg; fh; f gg C fh; ff; ggg D 0: (18.60)

Proof. Equations (18.57)–(18.59) follow at once from (18.52). Further, in view

of (18.55) and (18.56), we have

ff; fg; hgg D LXf fg; hg D LXf LXg h;

fg; fh; f gg D  fg; ff; hgg D LXg LXf h;

fh; ff; ggg D LXff;gg h D Xff;gg h

D ŒXf ; Xg  D .LXf LXg  LXg Lxf /h;

and (18.60) is proved. t

u
Remark 18.1. We know (Theorem 6.4) that the binary operation
  
.X; Y/ 2 M2n  M2n ! ŒX; Y 2 M2n (18.61)
348 18 Hamiltonian Dynamics


gives the < vector space M2n a Lie algebra structure. Further, in view of (18.57),
this operation associates a globally Hamiltonian field to a pair of globally Hamil-
tonian fields. In other words, these fields form a Lie subalgebra with respect to the
bracket operation. On the other hand, Theorem 18.8 states that the < vector space

F M2n is also a Lie algebra with respect to the Poisson bracket
  
.f; g/ 2 F M2n  F M2n ! ff; gg 2 F M2n :

Finally, from (18.57) it follows that the mapping


 W f 2 F M2n ! X D .df /] 2 M2n

(18.62)

is a morphism between the preceding two Lie algebras. It is not an isomorphism

because the kernel of  is given by locally constant functions since

.f / D .df /] D 0

implies df D 0.

18.7 First Integrals and Symmetries


Definition 18.10. Let f 2 F M2n be a differentiable function such that df ¤ 0,

8x 2 M2n . Then, f is said to be a first integral of the dynamical system

.M2n ; ; H / if
f ..t// D cos t (18.63)
along the integral curves of the Hamiltonian field XH .
In other words, f is a first integral if any integral curve .t/ of the Hamiltonian
vector field XH that crosses the level surface f .x/ D cos t is fully contained in this
surface S . Equivalently, the restriction to S of the vector field XH is tangent to S .
From a physical point of view, a first integral gives a conservation law.
Condition (18.63) can be equivalently expressed by the conditions

fP D XH f D LXH f D 0: (18.64)

Recalling (18.56), we can state the following theorem.


Theorem 18.9. The function f 2 F M2n , such that df ¤ 0, is a first integral if
and only if
ff; gg D 0: (18.65)
Theorem 18.10. We have that

1. H is a first integral of the dynamical system .M2n ; ; H /;
18.7 First Integrals and Symmetries 349

 
2. If f is a first integral of .M2n ; ; H /, then H is a first integral of .M2n ; ; f /;

3. If f and g are first integrals of .M2n ; ; H /, then ff; gg is also a first integral

of .M2n ; ; H /.
Proof. Since fH; H g D  fH; H g, we have that fH; H g D 0, and the first
statement is proved. Again, property 2 follows from the skew symmetry of the
Poisson brackets. Finally, the third statement follows at once from (18.60) and the
conditions ff; H g D fg; H g D 0. t
u
Remark 18.2. We underline that the first integral ff; gg could be dependent on the
first integrals f and g.
Remark 18.3. The first integral H.x/ D const states the conservation of the me-

chanical energy of the mechanical system .M2n ; ; H / with forces and constraints
independent of time. In fact, for such a system we have that [see (17.151)]
 
H.q h ; ph / D ph vh  L.q; v/ vDv.qp/

in any system of natural coordinates. But in our hypotheses,

1
LDT U D ahk vh vk  U.q/;
2

where U.q/ is the potential energy of the acting forces. Further, we have that ph D
@L=@vh D ahk vk . If we denote by .ahk / the inverse matrix of .ahk /, then vh D
ahk pk , and the Hamiltonian function assumes the form

1 hk
H DT CU D a ph pk C U.q/:
2

Remark 18.4. If in a system of symplectic coordinates H.q; p/ does not depend on

one of the coordinates q h , for instance on q 1 , then the coordinate q h is said to be
cyclic. In this case, the Hamiltonian equations show that the conjugate momentum
is constant.

Definition 18.11. Let f 2 F M2n be a function such that df ¤ 0. Suppose that
the globally Hamiltonian field Xf D .df /] is complete on M2n 
and denote by
 
's W <  M2n ! M2n the symplectic group generated by Xf (Definition 18.8). We
say that 't is a symmetry group for the Hamiltonian function H of the dynamical

system .M2n ; ; H / if

H.x/ D H.'s .x//; 8 s 2 <; 8 x 2 M2n (18.66)

or, equivalently, if
LXf H D 0: (18.67)
The following theorem extends Noether’s theorem to the Hamiltonian formalism.
350 18 Hamiltonian Dynamics


Theorem 18.11. The function f is a first integral of the dynamical system .M2n ; ;
H / if and only if H admits a symmetry group generated by the complete globally
Hamiltonian field Xf .
Proof. It is sufficient to note that (18.64), (18.65), and (18.67) imply the following
chain of identities:

ff; H g D 0 , LXf H D 0 , H.x/ D H.'s .x//: (18.68)

t
u

Remark 18.5. The last identity holds if Xf is complete. If M2n is compact, then any
Hamiltonian vector field is complete. In general, all Hamiltonian fields that are not
complete can be only locally associated with a symmetry of H.
Remark 18.6. In Theorem 18.11, it is fundamental to suppose that the infinitesimal
generator X of a symplectic group is Hamiltonian. In fact, if X generates a
symplectic group of symmetry of H but it is not Hamiltonian, then we have that

LX H D 0; LX  D 0:

The second condition, in view of (18.50), allows us to state that X is locally

Hamiltonian, so that the existence of the first integral is only locally proved.
Remark 18.7. Noether’s theorem states the equivalence between first integrals and
symplectic symmetries generated by complete and globally Hamiltonian vector
fields. This theorem is different from Theorem 17.5, in which it is proved that
the symmetry of a Lagrangian function generated by a vector field X on the
configuration space Vn only implies the existence of a first integral given by the
formula
@L h
f D X :
@qP
This is due to a different definition of symmetry, which in Theorem 17.5 involves
only transformations of the Lagrangian coordinates on Vn instead of transformations
of the coordinates .q; v/ on the fiber bundle T Vn . However, when this reduced class
of transformations is taken into account, we have the advantage that the preceding
formula explicitly gives the first integral. In contrast, in Theorem 18.11, the first
integral is obtained by integrating the system of partial differential equations

.df /] D Xf ;

which in symplectic coordinates is written

@f @f
Xh D ; Xh D  :
@ph @q h
18.8 Poincaré’s Absolute and Relative Integral Invariants 351

s F U
s F

uk
s
M*n
U F
u

Fig. 18.1 Absolute integral invariant ˛

18.8 Poincaré’s Absolute and Relative Integral Invariants


Definition 18.12. A k-form ˛ 2 ƒk M2n is an absolute integral invariant of the

complete vector field X on M2n if
LX ˛ D 0: (18.69)

In other words, denoting by 's .x/ the one-parameter group of diffeomorphisms

generated by X, we can say that ˛ is an absolute integral invariant if

's .˛.'s .x/// D ˛.x/; 8 x 2 M2n


; 8 s 2 <; (18.70)
where 's is the codifferential of 's (Fig. 18.1).
Theorem 18.12. Let † be an arbitrary differentiable k-dimensional submanifold

of M2n . Then ˛ is an absolute integral invariant of the complete vector field X if
and only if
Z Z
˛D ˛; 8 s 2 <: (18.71)
† 's .†/

Proof. To simplify the proof, we suppose that † is a k-cube .U; ; F / (Sect. 8.4),
so that † D F .U /. From the definition of the integral of a k-form ˛ we obtain
Z Z Z
˛D .'s ı F / ˛ D F  '  ˛:
's ı† U U

Consequently, if ˛ is an integral invariant,then (18.70) implies that

Z Z Z
˛D F ˛ D ˛;
's ı† U †
352 18 Hamiltonian Dynamics

and the first part of the theorem is proved. Conversely, if (18.71) holds for any U ,
then Z Z
F ˛ D F  's ˛:
U U
From the arbitrariness of U and the continuity of the functions under the integral
we obtain F  ı 's ˛ D F  ˛, that is, 's ˛ D ˛. t
u
Definition 18.13. The .k  1/-form ˛ is a relative integral invariant of the

complete vector field X on M2n if

d.LX ˛/ D 0: (18.72)

Recalling (8.30), the preceding condition becomes

LX .d˛/ D 0; (18.73)

and, in view of (18.69), we can say that ˛ is a relative integral invariant if and only
if d˛ is an absolute integral invariant.

Theorem 18.13. The .k  1/-form ˛ on M2n is a relative integral invariant of the
complete vector field X if and only if
Z Z
˛D ˛; (18.74)
@† 's .@†/


where † is an arbitrary k-dimensional submanifold of M2n and @† is its boundary.
Proof. Again, we suppose that † is a k-cube .U; ; F /. Then, from (18.73) and
Theorem 18.12, we have the condition
Z Z
d˛ D d˛;
† 's .†/

which, applying Stokes’ theorem (Chap. 8), becomes

Z Z
˛D ˛;
@† 's .@†/

and the theorem is proved. t

u

Theorem 18.14. Let X be a complete vector field on M2n . Then the following
properties are equivalent:
1. XH is locally Hamiltonian.
2. The 2-form  is an absolute integral invariant of X.
3. The 1-form !, such that  D d!, is a relative integral invariant.
18.9 Two Fundamental Theorems 353

Proof. If X is locally Hamiltonian, then (18.50) holds, so that  is an absolute

integral invariant [see (18.69)] and ! is a relative integral invariant [see (18.73)].
The other implications are evident. t
u
The preceding theorem allows us to define the complete locally Hamiltonian

vector fields X on M2n as vector fields characterized by one of the following
conditions:
Z Z
D ; (18.75)
† 's .@†/
Z Z
!D !; (18.76)
† 's .@†/


where † is an arbitrary two-dimensional submanifold of M2n and 's is the
(symplectic) one-parameter group of diffeomorphisms generated by X.
In many books, the Hamiltonian fields are introduced starting with one of the
foregoing properties. This is possible only if we refer to a coordinate neighborhood,

not to the whole manifold M2n .

18.9 Two Fundamental Theorems


Theorem 18.15 (Liouville’s Theorem). Let .M2n ; ; H / be a Hamiltonian dy-
namical system, and suppose that the Hamiltonian vector field XH is complete. Then
the 2n-form
n.n1/

O D .1/
2
 n (18.77)
nŠ

is an absolute integral invariant of XH , that is, it confirms one of the following

conditions:
Z Z
LX O D 0; O D
 O
; (18.78)
W 's .W /


where W is an arbitrary 2n-dimensional submanifold of M2n and 's a one-
parameter symplectic group of diffeomorphisms generated by XH . Further, in
symplectic coordinates,

O D dq 1 ^    ^ dq n ^ dp1 ^    ^ dpn :
 (18.79)

Remark 18.8. Let .q h ; ph / be a symplectic system of coordinates of M2n 

, and
1 n
consider the parallelepiped having edges .@=@q ; : : : ; @=@q ; @=@p1 ; : : : ; @=@pn /.
Then (18.79) shows that  O is a volume n-form for which the volume of the
354 18 Hamiltonian Dynamics

preceding parallelepiped is equal to one. Consequently, if W is a 2n-cube .U; ; F /,

where U  <2n , then the volume of the 2n-dimensional submanifold W is given by
Z
dq 1    dq n dp1    dpn : (18.80)
F .U /

Finally, regard W as a set of initial data .q0h ; p0h / of the Hamilton equations of

the system .M2n ; ; H /. Since 's is generated by XH , its orbits coincide with
the dynamical trajectories of the Hamilton equations. Therefore, 's ı W is the
submanifold formed by the configurations at the instant s of all the motions starting
from the points of W . In conclusion, the preceding theorem states that the volume
of this region does not vary over time.
Proof. Since LXH is a derivation (Chap. 8) and (18.50) holds, we have that
n.n1/

O D .1/
2
LXH  LXH n
nŠ
n.n1/
.1/ 2
D .LXH  ^     ^  C    C  ^    ^ LXH / D 0;
nŠ

O is an absolute integral invariant. Further, in symplectic coordinates,

and 

n D .dq 1 ^ dp1 C    C dq n ^ dpn / ^

   ^ .dq 1 ^ dp1 C    C dq n ^ dpn /: (18.81)

The only nonvanishing terms of the preceding product have the form dq i1 ^ dpi1 ^
   ^ dq in ^ dpin , where each of the indices i1 ; : : : ; in is different from the others.
Consequently, the right-hand side of (18.81) contains nŠ terms. Consider the term
dq 1 ^ dp1 ^    ^ dq n ^ dpn . It assumes the form dq 1 ^ dq n ^    ^ dp1 ^ dpn
by 1 C 2 C    C .n  1/ D n.n  1/=2 inversions in each of which the term
changes its sign. On the other hand, any other term can be reduced to the form
dq 1 ^ dp1 ^    ^ dq n ^ dpn by suitable exchanges of the factors dq i ^ dpi that do
not modify the sign of the considered term. In conclusion, (18.81) can be put in the
form
n.n1/
n D nŠ.1/ 2 dq 1 ^    ^ dq n ^ dp1 ^    ^ dpn ;
and (18.80) is proved. t
u
Theorem 18.16. The equilibrium positions of a Hamiltonian system cannot be
asymptotically stable.
Proof. In fact, if x  2 M2n 
is an asymptotically stable position of a Hamiltonian
system, then there exists a neighborhood I of x  such that the dynamical trajectories
starting from the points x0 2 I tend to x  when t ! 1. In other words, the volume
of 's .I / goes to zero when t ! 1, against Liouville’s theorem. t
u
18.9 Two Fundamental Theorems 355

s W

s W

Fig. 18.2 Phase portrait of a simple pendulum

Example 18.1. The phase portrait of a simple pendulum is represented in Fig. 18.2
(Sect. 14.9). Since the lower the pendulum period the higher the amplitude of
its oscillations, the dynamical trajectories starting from initial data contained in
region W and closer to the origin describe a greater fraction of the oscillation
than trajectories starting from points of W farther from the origin. Therefore, the
evolution of the set of initial data W is that sketched in Fig. 18.2 for the time instants
s1 < s2 .

Definition 18.14. Let .M2n ; ; H / be a Hamiltonian dynamical system, and de-
note by 't a symplectic group generated by the complete Hamiltonian vector field

XH . We say that the submanifold W of M2n is invariant with respect to 't if

't .W / D W; 8t 2 <: (18.82)

For instance, the level manifolds f D const, where f is a first integral of

 
.M2n ; ; H /, are examples of invariant submanifolds of M2n .

Theorem 18.17 (Poincaré). Let W  M2n be a compact and invariant submani-
fold with respect to 't . Then, for
T any submanifold S0  W and 8
> 0, there exists
an instant t1 >
such that S0 't1 .S0 / ¤ ;.
Proof. Owing to Liouville’s theorem, for an arbitrary value
> 0 of t, the regions
Sn D 'n
.S0 /, n D 0; 1; : : : ; n; : : :, have the same volume (Fig. 18.3) and are
contained in W T . First, we show that there are two positive integers n0 and n0Ck
such that Sn0 Sn0 Ck ¤ ;. In fact, in the contrary case, all the domains Sn should
be distinct. Since they have the same volume vol.S0 /, the total volume of all the sets
S1 ; : : : ; Sn ; : : : should be equal to infinity,
T against the hypothesis that W is invariant
and compact. If in the condition Sn0 Sn0 Ck ¤ ; we have n0 D 0, then the theorem
is proved. If n0  1, since '
is a diffeomorphism and '
Cs D 't ı 's , we obtain
the results
356 18 Hamiltonian Dynamics

S
W

t W
n W
M*n

Fig. 18.3 Poincaré’s theorem

Sn0 \ Sn0 Ck ¤ ; ) '

ı '.n0 1/
.S0 / \ '
ı '.n0 Ck1/
.S0 / ¤ ;
) '.n01/
.S0 / \ '.n0 Ck1/
.S0 / ¤ ; )
   ) S0 [ 'n0
.S0 / ¤ ;;

and the theorem is proved. t

u

Theorem 18.18. All trajectories starting from the points of the region S0  W
(except for the points of a subset with measure equal to zero), where W is an
invariant region of the Hamiltonian vector field, go back to S0 an infinite number of
times.
Proof. In fact, if a set U0  S0 exists for which this property is not verified, we
obtain an absurd result in view of the previous theorem. u
t
The preceding theorems show that, when orbits are not periodic, the phase por-
trait is very complex, especially if we take into account that, applying the uniqueness
theorem to Hamiltonian equations, trajectories corresponding to different initial data
cannot intersect each other.

18.10 Volume Form on Invariant Submanifolds


Let .M2n ; ; H / be a Hamiltonian system, and denote by †E the level .2n  1/

submanifold of M2n defined by the equation

H.x/ D E: (18.83)

We have already said that all dynamical trajectories with initial data belonging to †E
lie on †E . Equivalently, the restriction X†E of XH to the points of †E is tangent
to †E .
18.10 Volume Form on Invariant Submanifolds 357

t E+ E

X E E

Fig. 18.4 Invariant volume form on submanifold


Theorem 18.19. If †E is a regular submanifold of M2n , then there exists a .2n 
1/-form on †E that is an invariant volume form of the vector field X†E .
Proof. Introduce into a neighborhood U of a point x 2 †E the system of coordi-
nates .x 1 ; : : : ; x 2n1 ; H / such that .x 1 ; : : : ; x 2n1 / are coordinates in U \ †E and
the value H D E corresponds to U ˚\†E . Denoting by 0 a region contained in U\
†E , we consider the domain 0 D xj.x 1 ; : : : ; x 2n1 / 2 0 ; E  H  E C E 

M2n (Fig. 18.4). Liouville’s theorem states that the volume V0 of 0 is equal to the
volume Vt of t D 't .0 /, where 't is the one-parameter group generated by the
Hamiltonian field XH . If we denote by .q; p/ canonical coordinates in U , we can
write
Z Z ECE Z
Vt D dq 1    dq n dp1    dpn D dH J dx 1    dx 2n1 ;
t E 't . 0 /

where J is the determinant of the Jacobian matrix of the coordinate transformation

.q; p/ ! .x 1 ; : : : ; x 2n1 ; H /. Applying the mean value theorem to the integral, we
obtain Z
Vt D E J dx 1    dx 2n1 ;
't . 0 /

where J denotes the value of J at a suitable point of the interval .E; E C E/.
Finally, the invariant volume of 't . 0 / is
Z
Vt
!.'t . 0 // D lim D JH DE dx 1    dx 2n1 : (18.84)
E!0 E 't . 0 /

In conclusion, the invariant volume .2n  1/-form of the vector field X†E is

E D JH DE dx 1 ^    ^ dx 2n1 : (18.85)

t
u
358 18 Hamiltonian Dynamics

Now, we determine a more useful form of (18.85). Let .q; p/ be a symplectic


system of coordinates in M2n and suppose that the equation of the level submanifold
H.q; p/ D E can be written in the following parametric form

q h D q h ; h D 1; : : : ; n;
p˛ D p˛ ; ˛ D 1; : : : ; n  1; (18.86)
pn D pn .q h ; p˛ ; E/:

Consider the coordinate change .q; p/ ! .q h ; p˛ ; H / such that

q h D q h ; h D 1; : : : ; n;
p˛ D p˛ ; ˛ D 1; : : : ; n  1; (18.87)
pn D pn .q h ; p˛ ; H /:

It is evident that the determinant of the Jacobian matrix of (18.87) is

@pn
J D :
@H
In conclusion, in these coordinates E assumes the form
 
@pn
E D dq 1 ^    ^ dq n ^ dp1 ^    dpn1 : (18.88)
@H H DE

Example 18.2. The Hamiltonian function of the simple pendulum with mass and
length equal to 1, is
1
H D p 2  g cos q:
2
Then, (18.86) assume the form
p
q D q; p D 2.E C g cos q/

and E becomes
1
E D p dq:
2.E C g cos q/
Remark 18.9. In many books (e.g., [31]), instead of (18.84), the following alterna-
tive invariant volume form E on the submanifold H.q; p/ D E is proposed:
Z
d
!. / D ; (18.89)
jrH j

where is any measurable region on the submanifold †E and

18.11 Exercises 359

v
u n " 2  2 #
uX @H @H
jrH j D t C :
@q h @ph
hD1

Formula (18.89) is proved under the following assumptions.


• The symplectic coordinates .q; p/ are Cartesian coordinates in M2n .
• The level submanifold H.q; p/ D E is a Riemannian manifold equipped with

the Riemannian metric .gij / induced by M2n .
Regarding the first assumption, we note that if the symplectic coordinates .q; p/
are supposed to be Cartesian, the components of the metric tensor in the Euclidean

space M2n are gij D ıij . It is very easy to prove that if we introduce new symplectic
coordinates .q0 ; p0 /, then the components of the metric tensor cannot be equal to ıij .
The second assumption implies that, if pn D pn .q h ; p1 ; : : : ; pn1 ; E/ is the local
equation of †E , then the induced Riemannian metric .gij / on †E leads us to the
following results:
n 
X  n1 
X 
@pn 2 @pn 2
g D det.gij / D 1 C C ; (18.90)
@q h ˛D1
@p˛
hD1

p
d D gdq 1    dq n dp1    dpn1 : (18.91)
On the other hand, the relations between the partial derivatives appearing
in (18.90) and the implicit form H.q h ; ph / D E of †E are

@pn @H=@q h @pn @H=@p˛

D  ; D : (18.92)
@q h @H=@pn @p˛ @H=@pn
Introducing (18.92) into (18.90) and (18.91), we can write (18.89) as
Z
@pn 1
!. / D dq    dq n dp1    dpn1 ;
@H

and we find again the invariant volume form (18.88). In conclusion, although (18.89)
is based on wrong and useless hypotheses, it leads to the right definition of volume
form.

18.11 Exercises

1. Verify that in any system of coordinates .x ˛ /, ˛ D 1; : : : ; 2n on M2n


, Hamilton’s
equations assume the form
@H
xP ˛ D ˛ˇ ˇ : (18.93)
@x
Hint: See (18.49) and (18.47).
360 18 Hamiltonian Dynamics

2. Verify that in any system of coordinates .x ˛ /, ˛ D 1; : : : ; 2n on M2n


, Poisson’s
brackets assume the form

@f @g
ff; gg D ˛ˇ ˇ : (18.94)
@x ˛ @x

3. Verify that if the forces acting on a mechanical system H.q h ; ph / derive from
a generalized potential, then H D const reduces to T C UO .q h ; ph / D const
[see (17.68)].
4. Show Noether’s theorem starting from (18.66) and adopting arbitrary coordinates
.x ˛ / on M2n

.
Hint: Differentiating (18.66) with respect to s we obtain (see preceding exercises)

@f @H
fP D ˛ ˛ˇ ˛ D ff; H g :
@x @x

5. Suppose that the Hamiltonian function H.q h ; ph / does not depend on the
symplectic coordinate q 1 . Verify, by Noether’s theorem, that the corresponding
conjugate momentum is constant.
Hint: The one-parameter group of diffeomorphisms

q 0h D q h C sı 1h ; ph0 D ph ; h D 1; : : : ; n;

is a symmetry group of H whose infinitesimal generator is the vector field

X D .1; 0; : : : ; 0/:

This field is globally Hamiltonian since it is defined by the system

@f @f
ı 1h D ; 0D
@ph @q h

whose solution f D p1 is a first integral.

6. Prove that for the Hamiltonian system for which

1 2
1 1 2

H D p C p22 C .q / C .q 2 /2
2 1 2
the volume 3-form on the invariant submanifold H D E is
1
E D q dq 1 ^ dq 2 ^ dp1 :
2E  .p12 C .q 1 /2 C .q 2 /2 /
Chapter 19
The Hamilton-Jacobi Theory

19.1 The Hamilton–Jacobi Equation

We have repeatedly emphasized that determining the explicit solutions of Hamil-

tonian equations is a task beyond the capabilities of mathematics. This situation
justifies all attempts to obtain information about these solutions. For instance, a
knowledge of the first integrals allows us to localize the dynamical trajectories in

the phase space M2n . In turn, Nöther’s theorem proves that the presence of first
integrals is strictly related to the existence of symmetries. In this chapter, we analyze
the Hamilton–Jacobi theory whose purpose it is to determine a set of canonical
coordinates in which the Hamiltonian equations have such a simple form that we
can obtain solutions without effort. However, determining this set of coordinates
requires solving a nonlinear partial differential equation. Although, in general, this
problem could be more difficult than the direct integration of the Hamiltonian
equations, there are some interesting cases in which we are able to exhibit its
solution by adopting the method, proposed by Jacobi, of separated variables.
Let us consider the Hamiltonian equations

@H @H
qP h D ; pPh D  h ; h D 1; : : : ; n; (19.1)
@ph @q

in the symplectic coordinates .q; p/ of M2n .
We wish to determine a symplectic transformation of coordinates

q D q.q; p/; p D p.q; p/ (19.2)

such that, in the new coordinates, the Hamiltonian function depends only on the
coordinates p. Recalling that, under a transformation of symplectic coordinates
(19.2), the new Hamiltonian function is obtained by applying transformation (19.2)

A. Romano, Classical Mechanics with Mathematica R

, Modeling and Simulation 361
in Science, Engineering and Technology, DOI 10.1007/978-0-8176-8352-8 19,
© Springer Science+Business Media New York 2012
362 19 The Hamilton-Jacobi Theory

to the variables q; p in the old Hamiltonian function H.q; p/, we are searching for
a transformation of symplectic coordinates for which

H .q; p/ D H.q.q; p/; p.q; p// D H .p/: (19.3)

In these new coordinates, the Hamiltonian equations become

@H
qP D  .p/; pP D 0; (19.4)
@p

and their solutions are

q D t C ˛; p D ˇ; (19.5)
where ˛ and ˇ are constant vectors depending on the initial data in the coordinates
.q; p/. We obtain the solutions of the Hamiltonian equations in the coordinates .q; p/
by introducing (19.5) into (19.2)

q D q.t C ˛; ˇ/; p D p.t C ˛; ˇ/: (19.6)

Recognizing the advantage of such a transformation of coordinates, we must

suggest a way to determine it. To this end, we recall that we can obtain a symplectic
transformation of coordinates by the relations (Sect. 18.3)

@W .q; p/ @W .q; p/
pD ; qD ; (19.7)
@q @p

where W .q; p/ is the generating function of (19.2) satisfying the condition

 
@2 W
det ¤ 0: (19.8)
@q@p

Introducing (19.7)1 into (19.3), we can state that the generating function W .q; p/
must be a solution of the Hamilton–Jacobi partial differential equation
 
@W .q; p/
H q; D H .p/: (19.9)
@q

Remark 19.1. The first-order partial differential equation (19.9) is nonlinear since it
depends quadratically on the first derivatives @W=@q h . Consequently, determining
its solution is more difficult than solving Hamiltonian equations. Further, before
Jacobi, the method proposed by Cauchy, Monge, and Ampére to determine the
solution of a first-order partial equation required the integration of a system of
ordinary differential equations, the characteristic system, which can be associated
with the starting partial differential equation (Appendix A). Now, in the case of
(19.9), this characteristic system coincides with the Hamiltonian equations we wish
19.1 The Hamilton–Jacobi Equation 363

to solve. In other words, we are faced with a vicious circle. However, in the next
section, we present an alternative method proposed by Jacobi that allows us to
overcome this gap.
Remark 19.2. On the right-hand side of (19.9) is a Hamiltonian function in the
new symplectic coordinates we are searching for. It is unknown since its form
depends on the transformation itself. In other words, there are infinite symplectic
transformations of coordinates satisfying (19.9), so we can arbitrarily choose the
form of H .p/, and our choice will influence the transformation of coordinates. In
particular, Jacobi adopted the following choice of this function:

H .p/ D p n  E; (19.10)

where E is the total mechanical energy.

Remark 19.3. The solution of a partial differential equation depends on an arbitrary
function, not on n arbitrary constants, as is required by (19.9). In fact, the simple
first-order partial differential equation

@u
D xy (19.11)
@x

in the unknown u.x; y/ has the general solution

1 2
u.x; y/ D x y C F .y/;
2

where F .y/ is an arbitrary function of the variable y. A particular solution of (19.9)

can be obtained, for instance, using the function u0 .y/ D u.0; y/. In fact, taking
into account this condition, we obtain

1 2
u.x; y/ D x y C u0 .y/:
2
More generally, we consider the equation

@u @u
C D 0;
@x @y

which, introducing the constant vector field X D .1; 1/, can also be written in the
form
X  ru D 0: (19.12)
Consider the system of ordinary differential equations (Appendix A)

xP D 1; yP D 1
364 19 The Hamilton-Jacobi Theory

whose solution is given by the family  of straight lines x D x0 C t; y D y0 C t,

where x0 and y0 are arbitrary constants. In view of (19.12), we can say that the
function u is constant along the family  of straight lines:

du
D 0:
dt

Let .x0 .s/; y0 .s// be a curve that intersects at any point one and only one
straight line of . Then, if the values u0 .s/ of u.x; y/ are given along the curve
.x0 .s/; y0 .s//, the function u.x; y/ is completely determined. Once again, the
solution of (19.12) depends on an arbitrary function u0 .s/.
The preceding considerations show that we are interested not in finding the
general integral of the Hamilton–Jacobi equation but in determining a family of
its solutions, depending on n arbitrary constants. In the next section, we show that
such a family of solutions, called a complete integral, exists for a wide class of
Hamiltonian functions. We do not prove that, in general, the complete integral of a
partial differential equation exists and that the general integral is the envelope of the
family of solutions contained in the complete integral.

19.2 Separation Method

In the preceding section, we showed that knowledge of a complete integral of

the Hamilton–Jacobi equation (19.9) leads us to a solution of the Hamiltonian
equations. In this section, we present the separation method or the method of
separation of variables to obtain a complete integral for a wide class of Hamiltonian
functions. The mechanical systems whose Hamiltonian functions belong to this
class are said to be separable systems.
We say that the separation method is applicable to Eq. (19.9) if its complete
integral has the form

W .q; p/ D W1 .q 1 ; p/ C    C Wn .q n ; p/: (19.13)

It is evident that, in general, the solution of (19.9) cannot be given in the form
(19.13) for any Hamiltonian function. However, we can prove that for a wide class of
Hamiltonian functions it is possible to obtain the complete integral of (19.9) starting
from (19.13).
Theorem 19.1. If the Hamiltonian function has the form

H.q; p/ D gn .: : : g3 .g2 .g1 .q 1 ; p1 /; q 2 ; p2 /; q 3 ; p3 / : : : ; q n ; pn /; (19.14)

then the complete integral of (19.9) can be obtained by (19.13).

19.2 Separation Method 365

Proof. If (19.14) holds, then the Hamilton–Jacobi equation (19.9) can be written as
follows:
       
1 @W 2 @W 3 @W n @W
gn : : : g3 g2 g1 q ; 1 ; q ; 2 ; q ; 3 : : : ; q ; n D p n :
@q @q @q @q
(19.15)
This equation is satisfied if
 
@W
g1 q 1 ; 1 D p 1 ;
@q
 
@W
g2 q 2 ; 2 ; p 1 ; p 2 D p 2 ;
@q
: : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : ::
 
n @W
gn q ; n ; p1 ; : : : ; p n1 D p n :
@q

Supposing that the conditions

@gi
¤ 0; i D 1; : : : ; n; (19.16)
@pi

hold, the preceding system assumes the equivalent form

@W
D G1 .q 1 ; p 1 /;
@q 1
@W
D G2 .q 2 ; p 1 ; p 2 /;
@q 2
: : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : ::
@W
D Gn .q n ; p 1 ; : : : ; p n /: (19.17)
@q n

From the first equation we obtain

W D W1 .q 1 ; p 1 / C WO 1 .q 2 ; : : : ; q n ; p 1 ; : : : ; p n /; (19.18)

where Z
W1 .q ; p 1 / D
1
G1 .q 1 ; p 1 /dq 1 ;

and WO 1 .q 2 ; : : : ; q n ; p 1 ; : : : ; p n / is an arbitrary function. Substituting (19.18) into

(19.17)2 we have that

W D W1 .q 1 ; p 1 / C W2 .q 2 ; p 1 ; p 2 / C WO 2 .q 3 ; : : : ; q n ; p 1 ; : : : ; p n /; (19.19)
366 19 The Hamilton-Jacobi Theory

where Z
W2 .q 2 ; p 1 ; p 2 / D G2 .q 2 ; p 1 ; p 2 /dq 2 : (19.20)

Proceeding in the same way, we finally obtain

W D W1 .q 1 ; p 1 / C w2 .q 2 ; p 1 ; p 2 / C    C Wn .q n ; p 1 ; : : : ; p h /; (19.21)

where
Z
Wh .q ; p 1 ; : : : ; p h / D
h
Gh .q h ; p 1 ; : : : ; p h /dq h ; h D 1; : : : ; n; (19.22)

and the theorem is proved. t

u
The implicit form of the equations of motion is obtained by substituting (19.5)
into the symplectic transformation of coordinates

@Wh h @Wh h
ph D .q ; p 1 ; : : : ; p h /; qh D .q ; p 1 ; : : : ; p h /; h D 1; : : : ; n:
@q h @p h
(19.23)

19.3 Examples of Separable Systems

1. The Hamiltonian function of a harmonic oscillator S is

p2 kq 2
H.q; p/ D C ; (19.24)
2m 2m

where m is the mass of S and k is the elastic constant of the force acting on S .
Then, the Hamilton–Jacobi equation is
 2
1 @W kq 2
C D p  E: (19.25)
2m @q 2

The complete integral of (19.25)

Z p
W .q; p/ D 2mE  mkq 2 dq

generates the canonical transformation .q; p/ ! .q; p/ given by

@W p
pD D 2mE  mkq 2 ;
@q
Z
@W mdq
qD D p : (19.26)
@p 2mE  mkq 2
19.3 Examples of Separable Systems 367

On the other hand, in the coordinates .q; p/ the equations of motion are [see
(19.5)]
q D t C ˛; p D E: (19.27)
Introducing these functions into (19.26), we obtain
Z r Z
m m dq
t C˛ D p dq D q ;
2mE  mkq 2 k 2E
 q2 k

so that r r
2E k
qD sin .t C ˛/:
k m
2. Let P be a material point of mass m moving in a plane  in which we adopt
polar coordinates .r; '/ under the action of a Newtonian force with potential
energy U.r/. Then the Lagrangian function and the Hamiltonian function of P
are given by

1
LD m.Pr 2 C r 2 'P 2 /  U.r/;
2
!
1 p'2
H D pr C 2 C U.r/:
2
(19.28)
2m r

Consequently, the Hamilton–Jacobi equation becomes

" 2  2 #
1 @W 1 @W
C 2 C U.r/ D p 2  E: (19.29)
2m @r r @'

Comparing (19.15) and (19.29), we obtain that

!
1 p2
g1 D p' D p 1 ; g2 D pr2 C 21 C U.r/;
2m r

and (19.13) becomes

s
Z
p 21
W .r; '; p 1 ; E/ D p 1 ' C 2m.E  U.r//  dr:
r2
The canonical transformation generated by this function is
s
@W p 21
pr D D 2m.E  U.r//  ; (19.30)
@r r2
@W
p' D D p1; (19.31)
@'
368 19 The Hamilton-Jacobi Theory

Z
@W p1
q D
1
D s dr C '; (19.32)
@p 1
p2
r2 2m.E  U.r//  21
r
Z
@W m
q2 D D s dr: (19.33)
@p 2
p2
2m.E  U.r//  21
r

On the other hand, in the coordinates .q; p/ the equations of motion have the
form
q 1 D ˛ 1 ; q 2 D t C ˛;
(19.34)
p 1 D ˇ1 ; p 2 D E:

Introducing (19.34) into (19.32) and (19.33), we obtain the implicit form of the
orbit r D r.'/ and of the function r D r.t/.
3. Consider a material point P with mass m that is constrained to move on a surface
of revolution S :
x D r cos ';
y D r sin ';
z D f .r/;
where x, y, and z are the Cartesian coordinates of P with respect to a frame of
reference Oxyz, ' is the angle between the plane , containing P and the axis
Oz, and the plane Oxz, and, finally r is the distance of P from Oz. If no active
force is acting on P , then the square of velocity is v2 D xP 2 C yP 2 C zP2 , and the
Lagrangian function of P becomes

1  
LD m .1 C f 02 .r//Pr 2 C r 2 'P 2 : (19.35)
2
Since
@L
pr D D m.1 C f 02 .r//2 ;
@Pr
@L
p' D D mr 2 ';
P
@'P

the Hamiltonian function can be written as

!
1 pr2 p'2
H.r; '; pr ; p' / D C :
2m 1 C f 02 r2

Verify that the solution of the corresponding Hamilton–Jacobi equation is

Z s 
p p1
W D p1' C 2mE  2 .1 C f 02 /dr:
r
19.4 Hamilton’s Principal Function 369

19.4 Hamilton’s Principal Function

In Chap. 17, we considered the derivative at s D 0 of the integral (17.115). In this

integral f.s; t/, .s; t/ 2 Œ;   Œt1 ; t2  ! Vn , denotes a one-parameter family of
curves, starting from a point q1 at the instant t1 and arriving at the point q2 at the
instant t2 . In other words, the initial and final points, as well as the time interval
Œt1 ; t2 , are the same for all the curves f.s; t/. In this section, we consider the wider
family of curves such that, 8s 2 Œ; , the initial and final points and the time
interval in which each curve is defined

f.s; t/ W t 2 Œt1 .s/; t2 .s/ ! Vn (19.36)

depend on s. Then, instead of (17.115), we take into account the other integral:
Z t2 .s/
If .s/ D L.f.s; t/; Pf.s; t// dt: (19.37)
t1 .s/

Repeating the calculations from Sect. 17.14, we obtain

   2
dIf  2 @L @f
P i /; t i / 1 C
D L.q.t i /; q.t P i /; t i / .0; t i /
.q.t i /; q.t
ds 0 @qP @s 1
Z t2  
@L d @L @f
C  .0; t/ dt; (19.38)
t1 @q dt @qP @s

where
q.t/ D f.0; t/; t i D ti .0/; i D 1; 2: (19.39)
This expression, introducing the Hamiltonian function [see (17.151)]

H.f.s; t/; Pf.s; t/; t/ D p  Pf  L; (19.40)

assumes the equivalent form

   2
dIf  2 dt i
P i /; t i / 1  H.q.t i /; q.t
D p.q.t i /; q.t P i /; t i / .0; t i /
ds 0 ds 1
Z t2  
@L d @L @f
C  .0; t/ dt; (19.41)
t1 @q dt @qP @s

where we used the notation

@f @f dt i
D .0; t i / C .0; t i / .0; t i /: (19.42)
@s @t ds
370 19 The Hamilton-Jacobi Theory

In Sect. 17.14, we remarked that the boundary value problem of finding a solution
q.t/ of the Lagrange equations, satisfying the boundary conditions q.t1 / D q1 and
q.t2 / D q2 , is a problem profoundly different from the initial value problem relative
to the same equations. In fact, for the former a general theorem insures the existence
and uniqueness of the solution, whereas for the latter no general theorem holds.
Consequently, the boundary value problem could have no solution, one solution,
or infinite solutions (focal points). Regarding this problem, we assume that for any
choice of q1 2 D1  Vn , q2 2 D2  Vn , there is one and only one solution of
Lagrange’s equations starting from q1 at the instant t 1 and arriving at q2 at the
instant t 2 . If this hypothesis is verified, then for any choice of q1 2 D1 , q2 2 D2 ,
we can evaluate the integral (19.37) along the unique solution q.t/ of Lagrange’s
equations such that q.t 1 / D q1 and q.t 2 / D q2 . In this way, the integral (19.37)
defines a function
Z t2
S W .t 1 ; t 2 ; q1 ; q2 / 2 <  D1  D2 !
2
P
L.q.t/; q.t/; t/ dt; (19.43)
t1

which is called Hamilton’s principal function. It is evident that the differential dS

of S at the point .t 1 ; t 2 ; q1 ; q2 / 2 <2  D1  D2 coincides with the differential of
the function If .s/, provided that all the curves f.s; t/ are solutions of Lagrange’s
equations (effective motions). Then, in view of (19.41), we obtain

dS D Œpdq21  ŒH dt21 ; (19.44)

where [see (19.42)]

@f @f
dq D ds D .0; t i /ds C .0; t i /dt i ; (19.45)
@s @t
dt i
dt i D .0; t i /ds: (19.46)
ds

The initial and final points of the motions f.s; t/ are given by qi .s/ D f.s; ti .s//.
Consequently, the differential (19.45) gives the variations of these points in going
from a given motion to a close one. Finally, changing our notations as follows

t 1 D t0 ; t 2 D t; q1 D q0 ; q2 D q; p1 D p0 ; p2 D p; (19.47)

from (19.44) we obtain

@S
.t0 ; t; q0 ; q/ D p; (19.48)
@q
@S
.t0 ; t; q0 ; q/ D p0 ; (19.49)
@q0
19.4 Hamilton’s Principal Function 371

@S
.t0 ; t; q0 ; q/ D H.q; p; t/; (19.50)
@t
@S
.t0 ; t; q0 ; q/ D H.q0 ; p0 ; t0 /: (19.51)
@t0

If Hamilton’s principal function satisfies the condition

 
@S
det ¤ 0; (19.52)
@q h @q0k

then we can give (19.49) the form

q D q.t0 ; t; q0 ; p0 /: (19.53)

In conclusion, if Hamilton’s principal function S.t0 ; t; q0 ; q/ is known and

satisfies (19.52), then
• The equation of motion (19.53) is given in implicit form by (19.48) and (19.49);
• The correspondence between the initial data .q0 ; p0 / 2 D0 and .q; p/ 2 D is
symplectic, that is, the motion in the phase state defines a sequence of symplectic
transformations between the initial data and the corresponding positions at the
instant t;
• The function S.t; q; q0 / is a complete integral of the equation
 
@S @S
C H q; D 0: (19.54)
@t @q

Now we suppose that the Hamiltonian function H does not depend on time. If we
denote by E D H.q0 ; p0 / the value of the total energy corresponding to the initial
data and we recall that the total energy is constant during motion, then (19.50) and
(19.51) yield
S.t0 ; t; q0 ; q/ D E.t  t0 / C W .q0 ; q/ (19.55)
where, in view of (19.54), the function W is a complete integral of the Hamilton–
Jacobi equation  
@W
H q; .q0 ; q/ D E: (19.56)
@q
We conclude this section with a geometric description of the Hamilton–Jacobi
equation. Let Vn be the configuration space of the dynamical system S described by
the Lagrangian function

1
P D
L.q; q/ ahk qP h qP k  U.q/; (19.57)
2
372 19 The Hamilton-Jacobi Theory

N
t
q

Vn

Fig. 19.1 Moving surface †

where U.q/ is the potential energy of the forces acting on S , and let Vn be equipped
with the Riemannian metric (Sect. 17.15)

ds 2 D ahk .q/dq h dq k : (19.58)

Before proceeding, we need to make clear the meaning of the normal speed cn
of a moving surface †.t/ in the Riemannian space Vn . Denoting by f .q; t/ D 0
the implicit equation of †.t/, the unit normal vector N to †.t/ has covariant
components given by
f;h f;h
Ni D D ; (19.59)
hk
a f;h f;k jrf j
where f;h D @f =@q h and .ahk / is the inverse matrix of .ahk /. Now, at the arbitrary
point q 2 †.t/ we consider an arbitrary curve ./ such that (Fig. 19.1)

.t/ D q; P
.t/ D N.q/: (19.60)

The intersection points of the curve ./ with †./,  > t, satisfy the equation
f ../; / D 0. Differentiating this equation with respect to , setting  D t, and
recalling (19.60), we obtain the condition

f;h P h C f;t D 0;

which can also be written as

ft
cn  N   D  : (19.61)
jrf j

It is evident that cn is the speed of †.t/ along the normal N at q, i.e., it is the normal
velocity of †.t/. Since the surface is moving, we necessarily have @f =@t ¤ 0.
Consequently, we can locally put the equation of †.t/ in the form

tD .q/; (19.62)
19.4 Hamilton’s Principal Function 373

qt
t
q

Vn

Fig. 19.2 Moving surface t D .q/

where, in view of (19.61), the function .q/ is a solution of the eikonal equation

1
jr j2 D ahk ;h ;k D : (19.63)
cn2 .q/

We set t0 D 0 in (19.55) and suppose that q0 is a fixed point of Vn (Fig. 19.2).

Then we consider the moving surface † .t/ of the equation

1 
tD W .q0 ; q/  .q0 ; q/: (19.64)
E

It is evident that † .t/ reduces to the point q0 for t D 0. Further, omitting the
dependency on q0 , from (16.63) it follows that its normal speed cn is given by
p p
E E
cn D D p :
jr  .q/j a hk  ;h ;
k

In view of (19.48) and (19.55),

@ 
pD ; (19.65)
@q

and we have that

 
ahk ;h ;k D ahk ph pk D 2T D 2.E  U.q//: (19.66)

Finally, the normal speed can also be written as

p
E
cn D p : (19.67)
2.E  U.q//

In conclusion, Eq. (19.64) defines a moving surface in Vn advancing with normal

speed (19.67).
374 19 The Hamilton-Jacobi Theory

qt

q t

Vn

Fig. 19.3 Huygens’ principle

Let q.t/ be a dynamical trajectory starting from the point q0 . The contravariant
components of the vector tangent to this trajectory are .qP h /, whereas in metric
(19.58), the covariant components of this vector are ph D ahk qP h . Comparing this
result with (19.65), we can state that the surfaces † .t/ are orthogonal to the
trajectories sorting from q0 .
Remark 19.4. We have proved that the normal speed cn of a point q 2 † .t/ is
parallel to a dynamical trajectory that, starting from a point q0 , intersects † .t/ at
that point q. However, cn is not equal to the velocity of the dynamical system along
the dynamical trajectory. In fact, the velocity v of a point moving along a dynamical
trajectory has contravariant components .qP h /, so that

v2 D ahk qP h qP k D 2T D 2.E  U.q//:

Consequently, in view of (19.67), we obtain the relation

p
E
cn D :
v

We now prove Huygens’ principle, which makes the following statement

(Fig. 19.3).
Let †0 be an .n  1/-dimensional surface of the initial data q0 .u/, u D
.u1 ; : : : un1 /, of the configuration space Vn . Then the moving surface †.t/ of
equation t D .q0 ; q/, q0 2 †0 , is the envelope of all the surfaces † .t/
originating from the different points of †0 .
First, we denote the collection of all the surfaces originating from an arbitrary
but fixed point of †0 by the equation

O .u; q/ D 
.q0 .u/; q/:
19.4 Hamilton’s Principal Function 375

To prove the preceding statement, it is sufficient to recall that the envelope of the
surfaces t  O .u; q/ upon varying the parameters .u1 ; : : : ; un1 / can be obtained by
eliminating these parameters in the system

t D O .u; q/; (19.68)

@O
.u; q/ D 0: (19.69)
@u

Supposing that (16.69) can be solved with respect to the parameters u1 ; : : : ; un1

u˛ D u˛ .q/; ˛ D 1; : : : ; n  1;

substitution of these equations into (19.68) yields

t  O .u; q/ D t  .q/ D 0: (19.70)

To prove that the function .q/ is a solution of the eikonal equation, we note that
from (19.69) we obtain

@ @ O @u˛ @O @O
h
D ˛ h
C h D h:
@q @u @q @q @q

This result states that, at the point q, the surface t D .q/ has the same tangent
plane as the surface t  O .u; q/, which originates from a point q0 2 †0 . Further,
this surface satisfies the eikonal equation, and this is true for .q/.
Chapter 20
Completely Integrable Systems

20.1 Arnold-Liouville’s Theorem

This chapter contains some of the most advanced topics in analytical mechanics.
Due to the introductory character of the book, the theorems requiring a deep
knowledge of geometry and algebra are partially proved or only stated.

Definition 20.1. Let .M2n ; ; H / be a Hamiltonian system that admits p  n
first integrals f1 .x/; : : : ; fp .x/ of class C 1 . We say that these first integrals are
independent if the differentials df1 ; : : : ; dfp are linearly independent at any point

x 2 M2n .

Theorem 20.1. Let M2n be of class C k . If the independent first integrals f1 ; : : : ; fp ,
k 
p  n, are of class C , k  1, then the subset of M2n

Mf .c/ D fx 2 M2n ; fh .x/ D ch ; h D 1; : : : ; pg; (20.1)

where c1 ; : : : ; cp are given constants, is a p-dimensional C k manifold.

Proof. Let x0 be a point of Mf .c/ , and denote by .x ˛ /, ˛ D 1; : : : ; 2n, a system

of coordinates of M2n in a neighborhood of x0 . Then the coordinate representation
fO1 .x /; : : : ; fOp .x / of the functions f1 .x/; : : : ; fp .x/ is of class C k and the set
˛ ˛

Mf .c/ is locally defined by the system fOh D ch , h D 1; : : : ; p. Further, owing

to the linear independence of the differentials df1 ; : : : ; dfp , the Jacobian matrix
.@fOh =@x ˛ / has a rank equal to p. From what we proved in Chap. 6, it follows that

M2n is a p-dimensional C k manifold. t
u
Definition 20.2. The manifold Mf .c/ is called a level manifold and is determined
by the p independent first integrals f1 .x/; : : : ; fp .x/.
Definition 20.3. The C k first integrals f1 .x/; : : : ; fp .x/ are said to be in involu-
tion if

A. Romano, Classical Mechanics with Mathematica R

, Modeling and Simulation 377
in Science, Engineering and Technology, DOI 10.1007/978-0-8176-8352-8 20,
© Springer Science+Business Media New York 2012
378 20 Completely Integrable Systems

ffh ; fk g D 0; h; k D 1; : : : ; p; h ¤ k: (20.2)

Definition 20.4. A Hamiltonian system .M2n ; ; H /
is a completely integrable
system if it admits n independent first integrals in involution.
We have already discussed (Sect. 18.7) the importance of first integrals in

localizing the dynamical trajectories of a Hamiltonian system .M2n ; ; H /. In fact,

if f .x/ is a first integral of .M2n ; ; H /, then a dynamical trajectory starting from
a point of the manifold f .x/ D c is fully contained in this manifold. This result,
which still holds when the Hamiltonian system possesses p first integrals, pushes

us to determine the structure of Mf .c/ in order to localize the region of M2n in

which the dynamical trajectories of .M2n ; ; H / are contained. In this regard, the
following theorem is fundamental.

Theorem 20.2 (Arnold–Liouville). Let .M2n ; ; H / be a completely integrable
]
Hamiltonian system, and denote by Xfh the Hamiltonian vector fields .df /h , h D
1; : : : ; n. If the level manifold
˚ 

Mf .c0 / D x 2 M2n ; fh .x/ D ch0 ; h D 1; : : : ; p ; (20.3)

corresponding to the constants c0 D .c10 ; : : : ; ch0 /, is connected and compact, then

the following statements hold.
1. Mf .c0 / is invariant with respect to the one-parameter group of diffeomorphisms
ht .x/ generated by the vector fields Xfh .
2. Mf .c0 / is an n-dimensional torus
˚ 
T n .c0 / D .' 1 ; : : : ; ' n / mod 2 ; (20.4)

where ' 1 ; : : : ; ' n are angular coordinates on T n .c0 /.

3. There exist an open 2n-dimensional neighborhood U of Mf .c0 / and a diffeo-
morphism F W U ! T n .c0 /  I.c0 /, where I.c0 / D .c10  ı1 ; c10 C ı1 / 
    .cn0  ın ; cn0 C ın /, ıh > 0; in other words, the .' 1 ; : : : ' n ; c1 ; : : : ; cn /,
.c1 ; : : : ; cn / 2 I.c0 /, are local coordinates on M2n 
in a neighborhood of Mf .c0 /
0
and the level manifolds Mf .c/ , c 2 I.c / are tori.

4. The dynamical trajectories of .M2n ; ; H / that lie on Mf .c/ in the local
1 n
coordinates .' ; : : : ' ; c1 ; : : : ; cn / have the following parametric equations:

' h D wh .c/t C '0h ; ch D const; h D 1; : : : ; n; (20.5)

where the quantities '0h are constant.

Proof. We limit ourselves to sketching the proof of this very difficult theorem.
First, the Hamiltonian fields Xfh , h D 1; : : : ; n, are independent since the map
] is an isomorphism (Sect. 18.4), and the differentials dfh , h D 1; : : : ; n, are
linearly independent. Further, the map  defined by (18.63) is the Liea algebra

of vector fields differentiable functions on M2n , equipped with Poisson’s bracket,
20.1 Arnold-Liouville’s Theorem 379

into the Lie algebra of vector fields equipped with bracket of the
 corresponding
vector fields. Therefore, from ffh ; fk g D 0 there follows Xfh ; Xfk D 0. Let ht .x/
be a one-parameter group of diffeomorphisms generated by Xfh . Since Mf .c0 / is
compact, the vector fields Xfh are complete and the corresponding groups ht .x/ are
global. Along any orbit of kt .x/ we have that

dfh t
. .x// D LXfk fh D ffh ; fk g D 0;
dt k
so that fh .ht .x// D const. Consequently, if for t D 0 we have fh .k0 .x// D
fh .x/ D ch , then, 8t > 0, we also have fh .kt .x// D ch and the orbit starting
from x 2 Mf .c0 / belongs to Mf .c0 / . In other words, Mf .c0 / is invariant with respect
to all the global groups ht .x/, h D 1; : : : ; n, and the infinitesimal generators Xfh ,
h D 1; : : : ; n, of these groups are tangent to Mf .c0 / .
At this point, we can state that on Mf .c0 / there exist n independent tangent vector
 
fields Xfh satisfying the conditions Xfh ; Xfk D 0. It is possible to prove that
the only compact n-dimensional manifold on which there are n vector fields with
the foregoing properties is an n-dimensional torus T .c0 / D S 1      S 1 . In the
angular coordinates ' 1 ; : : : ; ' n on T n .c0 / the vector fields Xfh have the coordinate
representation
@
Xfh D Xfkh .c0 / ; (20.6)
@' k
where the components Xfhh are constant. Further, the linear independence of the dif-
ferentials df1 ; : : : ; dfn , after a suitable renumbering of the symplectic coordinates,
leads to put the system

fh .q; p/ D ch ; h D 1; : : : ; n; (20.7)

in the form
q D q;
p D p.q; c/;
at least for c belonging to a suitable interval I.c0 /. Then it is possible to prove that,
in this interval, the preceding system defines compact manifolds T n .c/ that inherit
all the properties of T n .c0 /. In particular, on each torus T n .c0 /, the vector fields Xfh
have the coordinate representation
@
Xfh D Xfkh .c0 / ; (20.8)
@' k
and the dynamical trajectories have the form (20.5). t
u
Remark 20.1. Any Hamiltonian system with one degree of freedom (n D 1) is
completely integrable since the energy integral H.q; p/ D E holds. If n D 2, then
a Hamiltonian system is completely integrable provided that it admits a first integral
f .q; p/ D c independent of the energy integral H.q; p/ D E. In fact, since f is
380 20 Completely Integrable Systems

O
O

c’

Fig. 20.1 Phase portrait of a simple pendulum

a first integral, ff; H g D 0. Finally, it is evident that a Hamiltonian system with

n D 3 is completely integrable, provided that it admits two first integrals f1 , f2
such that ff1 ; f2 g D 0.
Example 20.1. The Hamiltonian function of a simple pendulum of unit mass and
unit length is
1
H D p 2  g cos ;
2
where the symbols have the usual meaning and the phase space M2 is a cylinder
S  <.
Its phase portrait is shown in Fig. 20.1, where the level curve  , given by
H D E  , separates the oscillatory motions from the progressive ones. The curve
 is formed by three connected components, that is, the point  D , which is
identical to  D , and the two curves c and c 0 , which are both diffeomorphic to
<. The curves corresponding to values of energy less than E  are tori.
Example 20.2. Let P be a material point with unit mass moving on the straight line
r under the action of a conservative force. The phase portrait of P , corresponding
to a potential energy U.q/ represented on the left-hand side of Fig. 20.2, is shown
on the right-hand side of the same figure. For values E < 0 of the total energy E
we obtain one-dimensional tori.

Example 20.3. Let S be a mechanical system with a Hamiltonian function given by

n
1 X 2 
H.q; p/ D ph C 2h qh2 :
2
hD1
20.1 Arnold-Liouville’s Theorem 381

Uq
p E
E
E
O E q

q* E
q
E q*

Fig. 20.2 Phase portrait of a one-dimensional motion

It is easy to verify that the functions fh .q; p/ D ph2 C 2h .q h /2 , h D 1; : : : ; n, are

n independent first integrals such that ffh ; fk g D 0, h ¤ k. In other words, the
system S is completely integrable. We notice that the Hamiltonian function is itself
a first integral, but it depends on the preceding first integrals since it coincides with
their sum. Each equation fh .q; p/ D ch defines an ellipsis in the plane .q h ; ph /.
Consequently, the level manifold is the product of these ellipses, that is, it is an
n-dimensional torus.
Example 20.4. Let S be a heavy gyroscope. Then, if ', , and  are the Euler
angles and we take into account (15.21) and the kinematic relations of Sect. 12.4,
then the Hamiltonian function of S can be written in the form
" #
1 1 2 A sin2  C C cos2  2 2
H D p C p' C p  2 cos p p'
2 A sin2  C

CP z0G cos ;

where P is the weight of S , z0G the abscissa of the center of mass of S along
the gyroscopic axis, and A and C are the moments of inertia of S . From the
independence of the Hamiltonian function of and ' follow the first integrals

p Dc ; p' D c' ;

where c and c' are constant. Further, since in the ˚ Hamiltonian

 formalism the
variables q and p are independent, we have also p ; p' D 0. In conclusion,
the first three integrals H , p , and p' are independent and in involution. By some
calculations, it is possible to verify that the level manifolds are connected and
compact so that they are tori.
382 20 Completely Integrable Systems

20.2 Angle-Action Variables

Generally, the coordinates .'; c/, defined by the Arnold–Liouville theorem, are not
symplectic; in other words, functions (20.5) are not solutions of a Hamiltonian
system. It must also be remarked that the aforementioned theorem proves the
existence of these variables, but it does not specify their relation with the coordinates
.q; p/ in which the dynamical problem is initially formulated. In this section, we

prove the existence in a neighborhood V  M2n of the level manifold Mf .c0 / of a
new set of coordinates .; I/, called angle-action variables, such that the following
statements hold.
• Are symplectic coordinates.
• The action variables define the tori.
• The variables  are angular variables on the tori.
• The relation between the initial coordinates .q; p/ and the new angle-action
coordinates is known.
To prove the existence of these coordinates, we consider the immersion map

i W T n .c/ ! M2n

: (20.9)


If .q; p/ are local symplectic coordinates in M2n , then the parametric equations of
n
T .c/, i.e., the coordinate formulation of the mapping (20.9), has the form

q D q.'; c/; (20.10)

p D p.'; c/: (20.11)

Theorem 20.3. Let ! be the Liouville 1-form. Then, the restriction i  ! to T n .c/
of the differential 1-form ! is closed on T n .c/, i.e.,

di  !.u; v/ D 0; (20.12)

for any pair of vectors .u; v/ tangent to T n .c/.

Proof. In fact, in view of Exercise 8.5, we have that d.i  !/ D i  .d!/, so that,
denoting by  any 2-chain on T n .c/ (Chap. 8), Stokes’ theorem gives
Z Z Z
i ! D d.i  !/ D i  .d!/:
@  

On the other hand, d! D , and the preceding equation yields

Z Z

i !D i  : (20.13)
@ 
20.2 Angle-Action Variables 383

Since the 2-form i   on T n .c/ is the restriction to T n .c/ of , we have

i  .u; v/ D .u; v/ (20.14)

for any pair of vectors u, v, tangent to T n .c/. In proving the Arnold–Liouville

theorem, we showed that the vector fields Xfh are independent and tangent to
the tori. Consequently, at any point of T n .c/, they determine a basis of the
corresponding space that is tangent to T n .c/, and we can write that

u D uh Xfh ; v D vh Xfh :

Introducing these representations of u and v into (20.14), we obtain

   
i  .u; v/ D uh vk  Xfh ; Xfk D uh vk  .dfh /] ; .dfk /] :

In view of Definition (18.52) of Poisson’s bracket, we can state that

i  .u; v/ D uh vk ffh ; fk g D 0

since the first integrals are in involution, and the theorem is proved. t
u
The results (20.12) and (20.13) imply that
Z
i ! D 0 (20.15)
@

along any closed curve   @, which is the boundary of an oriented 2-chain on
a torus.
In the set of curves on T n .c/ we introduce the following equivalence relation:
two curves  and 1 are equivalent if they are homotopic.1 We denote by Π the
equivalence class of all the curves homotopic to  .
Let h be a closed curve with parametric equations ' h D s, ' k D ak , if h ¤ k,
where 0  s  2 and 0  ak  2 are constant, and denote by Œh  the
equivalence class of all the curves homotopic to h . It is evident that none of the
curves of Œh  can be reduced to a point by a continuous transformation.
Theorem 20.4. Let  and 1 be two arbitrary curves of Œh . Then
Z Z Z
!D ! !: (20.16)
 1 Œh 

1
A curve 1 D x1 .s/, s 2 Œa; b, on a manifold Vn is homotopic to a curve  D x.s/, s 2 Œa; b,
of Vn if there exists a continuous mapping F W Œa; b  Œ0; 1 ! Vn such that F .0; s/ D x.s/ and
F .s; 1/ D x1 .s/, in other words, if 1 reduces to  by a continuous transformation.
384 20 Completely Integrable Systems

n n

Fig. 20.3 Homotopic curves

Proof. Let  be the oriented 2-chain on T n .c/ whose boundary @ is  [ 1 . Then,

by Stokes’ theorem and (20.12), we have that (Fig. 20.3)
Z Z Z
i  D i !  i  ! D 0;
  1

and the theorem is proved. t

u
In view of this result, the following definition is consistent.
Definition 20.5. We define as action variables the n quantities
Z
Ih D !; h D 1; : : : ; n: (20.17)
Œh 

Taking into account the meaning of the variables .'; c/, the parametric equations of
a torus T n .c/ can be written as

q D q.'; c/; (20.18)

p D p.'; c/; (20.19)

where the variables c determine the torus T n .c/ and the angles ' locate a point of
it. Consequently, a curve  2 Œh  will have parametric equations

q.s/ D q.'.s/; c/;

p.s/ D p.'.s/; c/;

! D ph dq h , and the action variables will only depend on the quantities c, that is,
on the torus. Formally, we can write that

I D J.c/: (20.20)
20.2 Angle-Action Variables 385

Since it is possible to prove that the Jacobian determinant of (20.20) is different

from zero, there exists the inverse mapping

c D c.I/; (20.21)

which allows us to use the action variables to define the torus, instead of the
quantities c, as well as to write the parametric equations of the torus in the equivalent
form

O
q D q.'; I/; (20.22)
O
p D p.'; I/: (20.23)

Now we search for a function W .q; I/ generating a new set of symplectic

coordinates .; I/ by the usual formulae
@W
pD .q; I/; (20.24)
@q
@W
D .q; I/; (20.25)
@I
where
 
@W
det ¤ 0: (20.26)
@q h @Ik
The coordinates defined by (20.25) are called angle variables.
As generating function W .q; I/ we take
Z Zx

W .q; I/ D i !D ph .q; I/dq h ; (20.27)
 x0

where x0 ; x 2 T n .c/ and  is an arbitrary curve on the torus starting from x0 and
ending at x. Due to Theorem 20.3, integral (20.7), that is, the value of the function
W , does not depend on the curve  , provided that  can be reduced to a point
by a continuous transformation. In contrast, if  contains a curve belonging to the
equivalence class Œh , then we obtain a variation W of the value of W given by

W D Ih : (20.28)

In other words, the generating function W assumes infinite values at any point
of the torus. As a consequence of this property of W , the values of the variables
 corresponding to a point of the torus T n .c/ are infinite. The difference k  h
between two of these values is
@W @ @Ik
k  h D k D k W D D ıkh (20.29)
@Ih @Ih @Ih
386 20 Completely Integrable Systems

when the curve  contains a curve of the equivalence class Œk .

We conclude by remarking that, since the coordinates .; I/ are symplectic, the
Hamiltonian function in these new variables is given by

H .; I/ D H.q.; I/; p.; I/: (20.30)

On the other hand, a torus is determined by giving the action variables I and the
total energy is constant on a torus. Consequently,

H D H .I/; (20.31)

and the Hamilton equations in the symplectic coordinates .; I/ assume the form

@H
P D ; (20.32)
@I
IP D 0: (20.33)

Finally, the parametric equations of the dynamical trajectories are

 D .I0 /t C  0 ; (20.34)
I D I0 ; (20.35)

where  0 and I0 are constant.

The quantities
@H
.I/ D .I/
@I
are the fundamental frequencies of a completely integrable Hamiltonian system.
Remark 20.2. From (20.30), (20.31), and (20.24) it follows that the generating
function W is a complete integral of the Hamilton–Jacobi equation
 
@W
H q; .q; I/ D H .I/: (20.36)
@q

20.3 Fundamental Frequencies and Orbits


Definition 20.6. Let S D .M2n ; ; H / be a completely integrable Hamiltonian
n
system, and denote by T .I/ a torus corresponding to given values of the action
variables. The resonance module of the vector .I/ 2 <n of the fundamental
frequencies of S is the subset M of Zn such that

M D fm 2 Zn ; m  .I/ D 0g: (20.37)

20.3 Fundamental Frequencies and Orbits 387

The dimension of M, called resonance multiplicity, gives the number of indepen-

dent relations m.I/ D 0, with integer coefficients, existing among the fundamental
frequencies.
We omit the proof of the following theorem.

Theorem 20.5. Let .M2n ; ; H / be a completely integrable Hamiltonian system,
n
and denote by T .I/ the torus corresponding to given values of the action variables.
Then all the orbits belonging to T n .I/ are
• Periodic if and only if dimM D n  1,
• Everywhere dense T n .I/ if and only if dimM D 0,
• Everywhere dense on a torus with dimension n  dimM contained in T n .I/ if
0 < dimM < n  1.
Definition 20.7. Let .t/ D .t/ be an orbit on T n .I/ corresponding to the initial
datum .0/ D  0 . The temporal mean of the function F .; I/ along .t/ is defined
by the limit
Z T
FO .t0 ;  0 / D lim F ..t/; I/ dt: (20.38)
T !1 t
0

Theorem 20.6. If the function F .; I/ is measurable with respect to the Lebesgue
measure, then, for almost all the initial data  0 2 T n .I/,2 the temporal mean exists
and does not depend on t0 .
Definition 20.8. Let F .; I/ be a function on T n .I/ that is measurable with respect
to the Lebesgue measure. Then the spatial mean of F .; I/ is defined as follows:
Z
1
F D F .; I/ d; (20.39)
mis.T n .I// T n .I/

where mis.T n .I// is the measure of T n .I/.

Theorem 20.7. If dimM D 0, that is, if the orbits are everywhere dense on the
torus T n .I/, then the temporal mean does not depend on the orbits and is equal to
the spatial mean.
A mechanical system for which the temporal and spatial means are equal is
said to be ergodic. We can say that a completely integrable Hamiltonian system
is ergodic when the level manifolds defined by its n first integrals are tori and
dimM D 0.
Theorem 20.8. Let S be a completely integrable Hamiltonian system, and denote
by .; I/ its angle-action variables and by  the vector of its fundamental

2
That is, except for the points of T n .I/ belonging to a subset with a Lebesgue measure equal to
zero.
388 20 Completely Integrable Systems

frequencies. Suppose that dimM D l < n, i.e., that the fundamental frequencies
confirm l independent linear relations

m˛h
h D 0; ˛ D 1; : : : ; l; h D 1; : : : ; n; (20.40)

where the coefficients m˛h are integer numbers. Then there exists a canonical
transformation .; I/ ! . 0 ; I0 /, to new angle-action variables, in which the
Hamiltonian function is
 0 
h0 D H 0 IlC1 ; : : : ; In0 : (20.41)

Proof. Introduce the symplectic transformation .; I/ ! . 0 ; I0 / defined as follows:

@W 0
 0˛ D ; (20.42)
@I˛0
@W 0
 0lCˇ D 0 ; (20.43)
@IlCˇ
@W 0
Ih D ; (20.44)
@ h

where ˛ D 1; : : : ; l, ˇ D 1; : : : ; n  l, h D 1; : : : ; n, and the generating function

W 0 is
W 0 D m˛h  h I˛0 C IlCˇ
0
 lCˇ : (20.45)

In view of (20.45), system (20.42)–(20.44) becomes

lCˇ
 0˛ D m˛h  h ;  0lCˇ D  lCˇ ; Ih D m˛h I˛0 C IlCˇ
0
ıh : (20.46)

Taking into account (20.34), (20.40), and (20.46), we obtain that

P 0˛ D m˛h P ˛ D m˛h
h D 0; (20.47)
P 0lCˇ D P lCˇ D
lCˇ : (20.48)

On the other hand, recalling that the new coordinates are symplectic and taking into
account (20.46), we have in these new coordinates the Hamiltonian function

H 0 . 0 ; I0 / D H.I. 0 ; I0 // D H 0 .I0 /: (20.49)

In view of (20.47) and (20.48), the Hamilton equations relative to this Hamiltonian
function become
20.4 Angle-Action Variables of the Harmonic Oscillator 389

@H 0
P 0˛ D ;
@I˛0 D 0
@H 0
P 0lCˇ D 0 D
0lCˇ ;
@IlCˇ
0
@H
IPh0 D  0h D 0;
@
and the theorem is proved. t
u

20.4 Angle-Action Variables of the Harmonic Oscillator

The harmonic oscillator is a completely integrable Hamiltonian system

(Remark 20.1). Its Hamiltonian function is

1 2
H.q; p/ D .p C q 2 /; (20.50)
2

and the first integral H D E defines a one-dimensional torus  on the cylinder

S  < with the parametric equation
p p p
p D 2E  q 2 ;  E  q  E: (20.51)

In view of (20.17), the action variable is

Z Z p Z  p
E p E p
I.E/ D pdq D p 2E  q 2  p 2E  q 2 D 2E: (20.52)
  E E

Since E D I =2, from (20.27) we obtain the generating function

Z Z r
q p q
I
W .q; I / D 2E  u2 du D  u2 du: (20.53)
0 0 

The corresponding symplectic transformation .q; p/ ! .; I / is given by the

equations
r r 
@W I @W 
pD D  u2 ;  D D arcsin q : (20.54)
@q  @I I
390 20 Completely Integrable Systems

Finally, H.I / D I =2, and the Hamilton equations are

I
IP D 0; P D : (20.55)
2
More generally, the Hamiltonian function of an n-dimensional harmonic oscilla-
tor (Example 20.3) is
n
1 X 2 
H.q; p/ D ph C .h q h /2 : (20.56)
2
hD1

We have already shown that this Hamiltonian system is completely integrable and
that the first integrals fh D ph2 C .h q h /2 D ch define a torus with parametric
equations given by
q p p
c c
ph D ch  .h q h /2 ;  h  q h  h ; h D 1; : : : ; n: (20.57)
 
Consequently, the action variables are defined by the relations
p h
p
Zc= q  Z c=h
q
Ih .ch / D ch  .h q h /2  ch  .h q h /2 ; (20.58)
p p
 c=h c=h

which yield
ch
Ih .ch / D  ; h D 1; : : : ; n: (20.59)
h
Finally, we have that
s
Zq X
n
 h Ih
W .q; I/ D  .h uh /2 duh : (20.60)

q0 hD1

It is an easy exercise to determine the angle variables  and the Hamiltonian

equations in the variables .; I/.

20.5 Angle-Action Variables for a Material Point

in Newtonian Fields

Let P be a material point with mass m moving in the space under the action of a
Newtonian force. If in the space we adopt spherical coordinates .r; '; / with its
origin at the center of force, then the Lagrange function of P is

1 2
k
LD rP 3 C r 2 P 2 C r 2 sin2  'P 2 C :
2 r
20.5 Angle-Action Variables for a Material Point in Newtonian Fields 391

Consequently, the Hamiltonian function can be written as

!
1 2 p2 p'2 k
H D pr C 2 C 2
 ; (20.61)
2m r 2
r sin  r

where the momenta ph D @L=@qP h are

pr D mrP ; p' D mr 2 sin2  '; P

P p D mr 2 : (20.62)

Since the coordinate ' is cyclic, the conjugate momentum p' is constant. To prove
that P is a completely integrable Hamiltonian system, we must verify that there is a
third integral that is in involution with p' (Remark 20.1). We can easily obtain this
first integral recalling that in a central force field the total angular momentum KO
with respect to the center of force O vanishes. But KO D r  mPr, where r is the
position vector; further, in the spherical coordinates .r; '; /, we have r D rer , rP D
rP er C r sin  'e P  , where er , e' , and e are unit vectors along the coordinate
P ' C r e
curves, and we can write the angular momentum as
P '  sin e /:
KO D mr 2 .e (20.63)

If KO is constant, in particular, its length is constant. Then, in view of (20.62), we

at once verify that
p' 2
jKO j2 D p2 C : (20.64)
sin2 
In conclusion, we have the following three first integrals of our Hamiltonian
system:

f1  p' D ˛' ; (20.65)

p'2
f2  p2 C D ˛2 ; (20.66)
sin2 
!
1 p2 p'2 k
f3  H D pr2 C 2 C  D ˛r : (20.67)
2m r r 2 sin2  r

It is not difficult to verify that the preceding first integrals are independent;
further, ff1 ; f3 g D ff2 ; f3 g D 0 since f1 and f2 are first integrals and
( )
2
p'2
ff1 ; f2 g D p' ; p C
sin2 
3
" !
X @p' @ p'2
2
D p C
hD1
@q h @ph sin2 
392 20 Completely Integrable Systems

! #
@ p'2 @p'
 h p2 C
@q sin2  @ph
!
@ p'2 @p'
D p2 C D 0:
@ sin2  @p

since p' and p are independent variables. In conclusion, f1 , f2 , and f3 are in

involution.
System (20.65)–(20.67) can also be written in the form

p' D ˛' ; (20.68)

1 q
p D ˛2 sin2   ˛'2 ; (20.69)
j sin j
q
pr D 2m˛r r 2  2mkr  ˛2 ; (20.70)

provided that this yields

˛2 sin2   ˛'2  0; 2m˛r r 2  2mkr  ˛2  0: (20.71)

The first inequality in (20.71) is verified when

ˇ ˇ ˇ ˇ
ˇ ˛' ˇ ˇ ˛' ˇ
 ˇ ˇ  sin   ˇˇ ˇˇ ;
ˇ ˇ
˛ ˛
i.e., when
˛'  ˛ ; 0    0 ; (20.72)
where 0 D arcsin j˛' =˛ j. Further, if

mk 2
˛r   ; (20.73)
2˛2

then the roots of the trinomial 2m˛r r 2  2mkr  ˛2 are real and are equal to
2 s 3
k 4 4˛r ˛2
r1;2 D  1 1C 5: (20.74)
2˛r mk 2

Applying the Cartesian rule to the trinomial, we recognize that, if ˛r  0, then one
of the roots (20.74) is positive and the other one is negative. Since r is a positive
quantity, only the root r2 is acceptable. Moreover, the trinomial is positive for r > r2
and the orbit is unbounded. In contrast, if ˛r < 0, then both roots are positive, and
the trinomial is not negative when

0 < r1  r  r2 : (20.75)
20.5 Angle-Action Variables for a Material Point in Newtonian Fields 393

Henceforth we refer to this case in which [see (20.72)]

mk 2
  ˛r < 0: (20.76)
2˛2

All the preceding considerations allow us to put the level manifold in the parametric
form
l' D ';
 D ;
r D r;
p' D ˛' ;
q
1
p D ˛2 sin ˛'2 ;
j sin j
q
1
pr D 2m˛r r 2  2mkr  ˛2 ; (20.77)
r
where
0  '  2; 0    0 ; 0 < r1  r  r2 : (20.78)
3
The preceding formulae show that the level manifold T .˛r ; ˛' ; ˛ / is compact so
that it is a torus. The coordinate curves obtained by varying one of the variables r,
', and  are closed curves defining three fundamental cycles Œr , Œ' , and Œ .
The action variables are
Z Z 2
I' D p' d' D ˛' d' D 2 ˛' ;
Œ'  0
s
Z
˛'2
I D p0d D2 R ˛2  d;
Π Π0 sin2 
s  
Z Z r2
k ˛2
Ir D pr dr D 2 2m ˛r C  2 : (20.79)
Œr  r1 r r

It is possible to verify that the preceding relations yield the following expressions
of the action variables:
s
2m
I' D 2 ˛' ; I D 2.˛  ˛' /; Ir D .I C I' / C k ; (20.80)
˛r

from which we obtain

I' I C I' 2 2 mk 2
˛' D ; ˛ D ; ˛r D : (20.81)
2 2 .I C I' C Ir /2
394 20 Completely Integrable Systems

Comparing (20.81) and (20.67), we deduce the Hamiltonian function in the

angle-action variables
2 2 mk 2
H .I' ; I ; Ir / D  : (20.82)
.I C I' C Ir /2

If we denote by ' ,  , and r the angle variables corresponding to I' , I , and Ir ,

the Hamilton equations are

4 2 mk 2
P' D P D Pr D ;
.I C I' C Ir /2
IP' D IP D IPr D 0: (20.83)

Since all the fundamental frequencies are equal, from Theorem 20.5 we obtain that
the orbits are closed on the torus and the motions are periodic.

20.6 Bohr–Sommerfeld Quantization Rules

It can be verified experimentally by a spectroscope that the radiation emitted by

hot hydrogen gas is compounded by waves having discrete frequencies. Rydberg
and Ritz empirically determined the following formula, which includes all the
frequencies of the light emitted by hydrogen gas:
 
e
1 1
  D RH  ; (20.84)
c p2 q2

where e
 is the wave number of the radiation,
its frequency, c the light velocity in
a vacuum, and
RH D 109;677:576 cm1
is the Rydberg constant. If in (20.84) we set p D 1, we obtain the frequencies of the
Lyman series
 

1
D RH 1  2 ; q D 2; 3; : : : :
c q
For p D 2 we obtain the Balmer series
 

1 1
D RH  2 ; q D 3; 4; : : : :
c 2 q

Analogously, to p D 3; 4; 5 correspond the emission spectra of Paschen, Bracket,

and Pfund, respectively.
This discrete character of the frequencies of the emitted light was not compatible
with the physics of the beginning of 1900. In the attempt to give a theoretical
20.6 Bohr–Sommerfeld Quantization Rules 395

justification of the experimental results, Bohr and Sommerfeld made the following
assumptions:
• An electron rotating about the nucleus of a hydrogen atom cannot move along
an arbitrary orbit compatible with classical mechanics. Further, the electron does
not radiate electromagnetic waves during its accelerated motion on an allowed
orbit, in disagreement with classical electrodynamics.
• Electromagnetic radiation is emitted or adsorbed by an electron only when it
goes from an allowed orbit to another one. More precisely, let E be the energy
difference between the values of energy corresponding to two allowed orbits.
Then, in going from one to the other, the electron radiates electromagnetic waves
with frequency
j Ej

D ; (20.85)
h
where h is Planck’s constant.
At this point it is fundamental to assign a selection rule or a quantization rule
to determine the allowed orbits and then to evaluate the possible values of E.
The rule proposed by Bohr and Sommerfeld refers to completely integrable systems.
Therefore, it can be applied to a hydrogen atom.
Denote by I1 ; : : : ; In the action variables of a completely integrable Hamiltonian
system S . The Bohr–Sommerfeld rule states that the allowed orbits satisfy the
following conditions:

Ih D nh „; nh D 1; 2; : : : ; „ D h=2; (20.86)

which, for a hydrogen atom (see preceding section), become

I' D n' „; I D n „; Ir D nr „: (20.87)

Introducing these relations into (20.82), we obtain the possible values of the energy

2 2 mk 2
En D  ; n D n' C n C nr : (20.88)
n2 „2
On the other hand, if we denote by e the charge of the electron, we have

e2
kD ;
40
where 0 is the dielectric constant of the vacuum, and (20.88) becomes

2 2 me 4 1
En D  : (20.89)
02 h2 n2
396 20 Completely Integrable Systems

In conclusion, we can evaluate the energy variation j Ej D jEq  Ep j in going

from one allowed orbit to another, and introducing this value into (20.85) we finally
obtain the formula
 

2 2 me 4 1 1
D  ; (20.90)
c c0 h2 p2 q2
which agrees with (20.84) and allows one to evaluate the Rydberg constant.
In spite of the excellent results derived from this theory, it appears to be
unsatisfactory for the following reasons:
• It introduces selection rules that are extraneous to the spirit of classical mechan-
ics.
• It is applicable solely to completely integrable Hamiltonian systems.

20.7 A Sketch of the Hamiltonian Perturbation Theory

In the last section of this chapter, we present a brief introduction to Hamiltonian per-
turbation theory. Our approach does not analyze the correctness of the mathematical
procedures that are used. In this regard, we recall, without proof, the fundamental
KAM theorem (Kolmogoroff–Arnold–Moser), which makes clear the hypotheses
under which the results here presented hold.
The Hamiltonian theory of perturbation has as its aim the analysis of a Hamiltonian
system S that is close to a completely integrable Hamiltonian system S0 . It is based
on the assumption that the motions of S are close to the motions of S0 . In other
words, it is based on the preconception that small causes produce small effects.
Let S0 be a completely integrable Hamiltonian system, and denote by H0 .I0 / its
Hamiltonian function depending only on the action variables I0 of S0 . Henceforth
S0 will be called an unperturbed system. Then we consider another Hamiltonian
system S whose Hamiltonian function has the form

H. 0 ; I0 / D H0 .I0 / C H1 . 0 ; I0 /; (20.91)

where  0 are the angle variables of S0 and  is a small nondimensional parameter.

The system S is called a perturbed system or quasi-integrable system.
Example 20.5 (Fermi–Pasta–Ulam, 1955). Let S be a chain of nC2 material points
Pi , i D 0; : : : ; n C 1, constrained to move on the Ox-axis. Suppose that any point
Pi interacts with Pi 1 and Pi C1 by nonlinear elastic springs. The points P0 and
PnC1 are supposed to be fixed. Finally, all the particles have the same mass m.
The Hamiltonian function of this system is
n
X n 
X 
ph2 k
H.q; p/ D C .qhC1  qh /2 C .qhC1  qh /˛ ; (20.92)
2m 2
hD1 hD0

where ˛ is an integer greater than 2.

20.7 A Sketch of the Hamiltonian Perturbation Theory 397

Example 20.6. Let P be a planet (e.g., the Earth) moving under the action both of
the Sun S and a massive planet Pm (e.g., Jupiter). If we suppose that the motion of
Pm is not influenced by P , then the position vector rJ of Pm is a known function of
time. Let q, p be the position vector of P and its momentum, and denote by m D 1,
mJ , and mS the masses of P , Pm , and the Sun, respectively. Then the Hamiltonian
function of P is
1 2 K 1
H.q; p; t/ D jpj   K ; (20.93)
2 jqj jq  rJ j

where K is a constant and  D mJ =mS .

It is quite natural to wonder if it is possible to determine a coordinate transfor-
mation . 0 ; I0 / ! .; I/, where the variables  are still periodic with a period 2
such that in the new variables the Hamiltonian function of the perturbed system S
becomes a function of the only variables I, up to second-order terms in , i.e.,

H.; I/ D H0 .I/ C H1 .I/ C  2 H2 .; I/: (20.94)

We can consider applying this procedure again searching for a new coordinate
transformation .; I/ ! . 0 ; I0 / in which the Hamiltonian function becomes

H. 0 ; I0 / D H0 .I0 / C H1 .I0 / C  2 H2 .I0 / C  3 H3 . 0 ; I0 /: (20.95)

In other words, the idea underlying the Hamiltonian perturbation theory consists
in presupposing the existence of a sequence of completely integrable systems
whose motions increasingly approximate the motions of the perturbed nonintegrable
system S . After making clear the aim of the theory, we show how to determine
the first transformation . 0 ; I0 / ! .; I/ without analyzing all the mathematical
difficulties encountered. First, we search for a generating function S. 0 ; I/ of
such a coordinate transformation. Since this function must generate an identity
transformation when  D 0, we set
n
X
S. 0 ; I/ D 0h Ih C S1 . 0 ; I/ C    : (20.96)
hD0

In fact, this function generates the symplectic transformation

@S1
I0h D Ih C  C ; (20.97)
@0h
@S1
h D 0h C C ; (20.98)
@Ih
398 20 Completely Integrable Systems

which reduces to an identity where  D 0. Taking into account the preceding

relations, we have the following Taylor expansions:
n 
X 
@H0 @S1
H0 .I0 / D H0 .I/ C  C O. 2 /; (20.99)
@I0h I0 DI @0h
hD1

H1 . 0 ; I0 / D H1 . 0 ; I/ C O. 2 /: (20.100)

Introducing these relations into (20.94) we obtain the Hamiltonian function in terms
of the new variables
n
!
X @S1
H0 .I/ C 
h .I/ C H1 . 0 ; I/ C O. 2 /; (20.101)
@0h
hD1

where we have introduced the fundamental frequencies [see (20.35)] of the unper-
turbed system S0
 
@H0

h .I/ D : (20.102)
@I0h I0 DI
In view of this condition, we see that (20.94) is confirmed if S1 . 0 ; I/ is a solution
of the fundamental equation of the first-order perturbation theory:
n
X @S1

h .I/ . 0 ; I/ D H1 .I/  H1 . 0 ; I/: (20.103)
@0h
hD1

This equation contains two unknowns: the generating function S1 . 0 ; I/ and the
function H1 .I/. To determine the former function, we start by noting that to the

values . 0 ; I/ and . 0 C 2; I/ corresponds the same point of the phase state M2n .

Consequently, all the real functions on M2n are periodic functions of the variables
 0 with the same period 2. In turn, this property implies that any function on

M2n admits a Fourier expansion in the variables  0 (Appendix B). In particular, the
Fourier expansion of S1 starts with the constant mean value. Since the derivatives of
this value with respect to the variables  0 are zero, the mean value of the left-hand
side of (20.103) vanishes and we have that

H1 .I/ D < H1 . 0 ; I/ >; (20.104)

where < H1 . 0 ; I/ > is the mean value of the perturbation. Introducing the new
function
HO . 0 ; I/ D H1 . 0 ; I/ < H1 . 0 ; I/ >; (20.105)
Eq. (20.103) becomes
n
X @S1

h .I/ . 0 ; I/ D HO 1 . 0 ; I/: (20.106)
@0h
hD1
20.7 A Sketch of the Hamiltonian Perturbation Theory 399

To determine the solution S1 . 0 ; I/ of this equation, we introduce the following

Fourier expansions (Appendix B):
1
X
HO 1 . 0 ; I/ D hk .I/ei k 0 ; (20.107)
kD1
1
X
S1 . 0 ; I/ D sk .I/ei k 0 ; (20.108)
kD1

where the components of the n-dimensional vector k D .k1 ; : : : ; kn / are integers.

Introducing these expansions into (20.106) we have that
n
X n
X X1
@S1
 h .I/ . 0 ; I/ D i
h .I/ sk .I/ei k 0
@0h
hD1 hD1 kD1
1 X
X n
Di
h kh sk .I/ei k 0
kD1 hD1
1
X
Di .  k/sk .I/ei k 0 :
kD1

From these relations we derive the following expressions of the coefficients sk .I/ of
expansion (20.108):
ihk .I/
sk .I/ D : (20.109)
.I/  k
This expression generates some perplexities regarding the convergence of
series (20.108). In fact, if the fundamental frequencies are linearly dependent, then
there exists at least one vector k such that

  k D
1 k1 C    C
n kn D 0;

and some denominators of (20.109) may vanish. Even if the fundamental frequencies
are linearly independent, some denominators of (20.109) may assume values arbi-
trarily close to zero. For instance, suppose that we are concerned with a Hamiltonian
system with two degrees of freedom
p such that the fundamental frequencies assume
the values
1 D 1 and
2 D 2. Setting
p
ak1 D 2k1 C k2

we obtain
p k2
2 D  C a:
k1
400 20 Completely Integrable Systems

Consequently, with a convenient choice of the integers k1 and k2 , we can make a

as small as we wish. This means that some denominators of (20.109) could become
very small.
Example 20.7. Consider a nonlinear oscillator with a Hamiltonian function given by

1 2 1
H.q; p/ D .p C q 2 / C  q 4 ; (20.110)
2 4

where q is the abscissa of the oscillator. In view of (20.55),

D 1=2,  D 20 ,
and (20.110) becomes

H.0 ; I0 / D
I0 C 
2 I02 sin4 0  H0 .I0 / C H1 .0 ; I0 /; (20.111)

where   0  .
The mean value of the perturbation term is [see (20.104)]
Z 
1 2 2 3 2 2
< H.0 ; I / >D
I sin4 0 d0 D
I ; (20.112)
2  8

and then (see the notebook Fourier.nb)

 
1 1 1
HO D 
2 I 2 ei 20 C ei 20  ei 40  ei 40 : (20.113)
4 4 4

20.8 Overview of KAM Theorem

Let us return to action-angle coordinates. Theorem 20.5 shows that the flow on each
torus crucially depends on the arithmetical properties of the fundamental frequen-
cies . There are essentially two cases: (1) If the frequencies  are nonresonant,
or rationally independent on a torus, then, on this torus, each orbit is dense and
the flow is ergodic. (2) If the frequencies  are resonant, or rationally dependent,
then the torus decomposes into an l-parameter family of invariant tori. Each orbit
is dense on such a lower dimensional torus, but not in T n .I/. A special case arises
when there exist l D n  1 independent resonant relations. Then each frequency

1 ; : : : ;
n is an integer multiple of a fixed nonzero frequency, and the whole torus is
filled by periodic orbits with the same period. Can we state that a small perturbation
of an integrable system does not substantially modify the preceding description of
an integrable system? The remarks at the end of the preceding section engender
suspicions about the possibility of approximating the behavior of a perturbed system
by the behavior of suitable completely integrable systems. These suspicions are
strengthened by the following considerations.
20.8 Overview of KAM Theorem 401

Definition 20.9. Consider a perturbed Hamiltonian system with Hamiltonian

function (20.91). The Hamiltonian function H0 .I/ of the unperturbed system is said
to be nondegenerate if
   2 
@
h @ H0 .I/
det D det ¤0 (20.114)
@Ik @Ih @Ik
in a domain D  <n .
In view of condition (20.114), the frequency map

hWD! (20.115)

is a local diffeomorphism between D and some open frequency domain   <n . In

other words, the tori of the unperturbed system corresponding to I 2 D are in a one-
to-one correspondence with the frequencies  2 . From the preceding hypothesis
it follows that nonresonant tori and resonant tori of all types all form dense subsets in
phase space. Indeed, the resonant ones sit among the nonresonant ones like rational
numbers among irrational numbers.
Theorem 20.9 (Poincarè). Let H.I; ; / be periodic with respect to the variables
 and analytic with respect to . Then, if H0 .I/ is nondegenerate, there is no analytic
first integral independent of H .
The preceding theorem states that a perturbed Hamiltonian system can not be
integrable, at least if we consider analytic first integrals. Consequently, our request
to approximate the behavior of a perturbed Hamiltonian system with the behavior
of a sequence of completely integrable systems seems more and more vague.
Other negative results due to Poincarè prove that, also for small values of the
parameter , invariant tori do not persist. More precisely, Poincarè proved that in
the presence of resonance, invariant tori are in general destroyed by an arbitrarily
small perturbation in the sense that, except for a torus with an n  1 family of
periodic orbits, usually only a finite number of periodic orbits survive a perturbation
and the others disintegrate, originating a chaotic behavior. In other words, in a
nondegenerate system, a dense set of tori is usually destroyed.
Since dense sets of tori are destroyed, there seems to be no hope for other tori to
survive. However, in 1954 Kolmogorov proved that the majority of tori survives. In
fact, those tori survive whose frequencies  are not only nonresonant but strongly
nonresonant in the sense of the following definition.
Definition 20.10. Let n > 1. A vector  2 <n satisfies a diophantine condition
(with constant  > 0 and exponent  n  1) or a small divisor condition if,
8m 2 Z n , m ¤ 0,

jm  j  ; (20.116)
jmj
where jmj D jm1 j C    C jmn j.
402 20 Completely Integrable Systems

Denote by C;   a set, which can be proved to be nonempty, of all frequencies

satisfying condition (20.116), for fixed  and . Since the Hamiltonian function
H0 .I/ is nondegenerate, there exists a map

h1 W C; ! A; ; (20.117)

where
˚ 
A; D h1 .C; / D I0 W .I0 / 2 C; : (20.118)

It can be proved that the set A; has a very complex structure: it is closed,
perfect, nowhere dense, and a Cantor set with positive measure.3
Now we can state the KAM theorem.
Theorem 20.10 (KAM). Let S be a perturbed Hamiltonian system with
Hamiltonian function H.I; ; / analytic and nondegenerate. Denote by > n  1
e  given constants. Then a positive constant c . / exists such that, 8 I 2 A; , all
the tori T n .I/ are invariant, though slightly
p deformed. Moreover, they fill the region
D  T n .I/ up to a set of measure O. /.
Remark 20.3. Since A; has no interior points, it is impossible to state whether an
initial position falls onto an invariant torus or into a gap between such tori. In other
words, theorem KAM states that p the probability that a randomly chosen orbit lies
on an invariant torus is 1  O. /.

3
An example of Cantor’s set can be obtained as follows. Start with the unit interval Œ0; 1 and choose
0 < k < 1. Remove the middle interval of length k=2. The length of the two remaining intervals is
1k=2. From each of them remove the middle interval of length k=8. The four remaining intervals
will have a total length of 1  k=2  k=4. From each of them remove the middle interval of length
k=32. The remaining length will be 1  k=2  k=4  k=8, and so on. After infinitely many steps,
the remaining set will have the length

1  k=2  k=4  k=8 C    D 1  k > 0:

 Part II
Mechanics
Chapter 21
Elements of Statistical Mechanics
of Equilibrium

21.1 Introduction

Statistical mechanics is an important part of mechanics. Its origin can be found

in the old Greek dream of describing the macroscopic behavior of real bodies
starting from the properties of the elementary constituents of matter: the atoms
(Democritus). If this approach were practicable, it would be possible to reduce
the variety of macroscopic properties of real bodies to the few properties of the
elementary components of matter. However, this point of view exhibits some
apparently insuperable difficulties related to the huge number of atoms composing
macroscopic bodies. These difficulties justify the opposite approach undertaken by
continuum mechanics, which will be partially described in Chap. 23, devoted to
fluid mechanics.1
Let S be a perfect gas at equilibrium in a container with adiabatic walls.
Denote, respectively, by V , p, and  the volume, pressure, and temperature of
S . Experimentally, we verify that, at equilibrium, these quantities are uniform and
constant and satisfy a state equation

p D p.; V /: (21.1)

In a more complex system, for instance a mixture of two different gases S1 and S2 ,
(21.1) is substituted by the more general state equation

p D p.; V; c/; (21.2)

where c is the concentration of one of the gases S1 constituting the mixture, i.e.,
the ratio between the mass m1 of S1 and the total mass m of the mixture. We recall
that the concentrations c1 and c2 of the two gases confirm the condition c1 C c2 D
1. In any case, we can state that, from a macroscopic point of view, the uniform

1
For the contents of this chapter see [10, 31, 50, 53].

A. Romano, Classical Mechanics with Mathematica R

, Modeling and Simulation 403
in Science, Engineering and Technology, DOI 10.1007/978-0-8176-8352-8 21,
© Springer Science+Business Media New York 2012
404 21 Elements of Statistical Mechanics of Equilibrium

equilibrium of a gas S is described by a finite number of variables satisfying one or

more state equations.
On the other hand, from a microscopic point of view, S is formed by a very large
collection of (monoatomic, biatomic, . . . ) interacting molecules. Their number is
so large that any mechanical description of their behavior in time appears rather
utopian. The statistical mechanics of equilibrium sets itself the aim of deriving
the macroscopic behavior of gases from their molecular structure simply by taking
advantage of the large number of molecules whose evolution we are interested in
describing.
Whatever the procedure we adopt to describe the microscopic behavior of S at
equilibrium, we must
• Define the mechanical system S describing the body S microscopically,
• Obtain the macroscopic quantities as mean values of microscopic variables,
• Prove that these mean values satisfy the state equations.

21.2 Kinetic Theory of Gases

In the kinetic theory of the equilibrium of gases, the microscopic system S is formed
by N molecules incessantly moving inside a container having a volume V . Each
molecule, with f degrees of freedom, freely moves through the volume V until
it encounters another molecule or hits the walls of the container. Any collision is
supposed to be elastic, and the mean path between two subsequent collisions is
called a free mean path. Further, a collision among more than two molecules
is considered to be highly improbable, and the number of collisions per unit of time
is supposed to be very large. The preceding considerations lead us to formulate the
hypothesis of molecular chaos:
• At macroscopic equilibrium, all molecules of S have the same probability of
moving with a given velocity and of occupying an elementary volume dV  V .
• All possible velocities are equiprobable for a given molecule.
For the sake of simplicity we consider the case f D 3, i.e., we suppose that the
molecules are material points. From the hypothesis that, at equilibrium, the density
of the gas S is uniform there follows that the number dN.x/ of molecules contained
in an elementary volume having a vertex at x 2 V and edges dx, dy, and dz is given
by the formula
N
dN.x/ D dx dy dz: (21.3)
V
Besides the physical space V , we consider a three-dimensional Euclidean space
<3 in which we introduce a Cartesian frame of reference Ovx vy vz . A possible
velocity of a molecule is given by the position vector v of a point P of <3 relative
to O. It is evident that the state of a single particle is given by a point V  <3 .
!
Let dN.v/ be the number of molecules with velocity OQ D v, where Q belongs
21.2 Kinetic Theory of Gases 405

to the parallelepiped having a vertex at P and edges dvx , dvy , and dvz . Then, we call
the function g.v/ a velocity distribution function such that

dN.v/ D Ng.v/dvx dvy dvz : (21.4)

In other words, g.v/ is a probability density in the velocity space <3 .

Since at equilibrium there is no preferred direction of velocity, the function g.v/
depends on the length of velocity, i.e., on v2 . To determine the function g.v2 /, we
consider a collision between two molecules and denote by u, v and u0 , u0 their speeds
before and after the collision, respectively. Applying the conservation of momentum
and energy, and recalling that the two molecules have the same mass, we obtain

u C v D u0 C v0 ; (21.5)
2 2 02 02
u Cv D u Cv : (21.6)

The total number of collision per unit time is proportional to the product g.u2 /g.v2 /.
Similarly, the inverse collision process, in which two molecules, after colliding with
speeds u0 and v0 , have speeds u and v, has a probability proportional to the product
g.u02 /g.v02 /. But at equilibrium the number of the first kind of collision must be
equal to the number of the second type of collision to obtain a complete mixing as
required by the hypothesis of molecular chaos. Therefore, we have that

g.u2 /g.v2 / D g.u02 /g.v02 /: (21.7)

In the preceding equation we set v0 D 0. Then we determine u02 in terms of u2 and

v2 by (21.6), and we write (21.7) as

g.u2 /g.v2 / D g.u2 C v2 /g.0/:

After differentiating this relation with respect to u2 , we put u2 D 0 into the result
and obtain the condition
g 0 .v2 / g 0 .0/
D  ˛: (21.8)
g.v2 / g.0/
An elementary integration of this equation leads to the formula
2
g.v2 / D Ae˛v ; (21.9)

where A and ˛ are integration constants. To determine these constants, we note that
(21.4), in view of (21.9), assumes the form
2
dN.v/ D NAe˛v dvx dvy dvz :

Integrating on the whole velocity space, we obtain the formula

Z
2
1D e˛v dvx dvy dvz ;
<3
406 21 Elements of Statistical Mechanics of Equilibrium

which, when we adopt spherical coordinates in <3 , becomes

Z 1
2
1 D 4A v2 e˛v dv: (21.10)
0

We note that the integral on the right-hand side of (21.10) is convergent when ˛ > 0.
Further, from the well-known formulae
Z   1=2 Z
1
˛x 2
1
2 1   1=2
e dx D ; x 2 e˛x dx D (21.11)
1 ˛ 1 2˛ ˛

we obtain
 ˛ 3=2
AD ;

so that we can write
 ˛ 3=2 2
dN.v/ D N e˛v dvx dvy dvz ; (21.12)

where the constant ˛ is still to be determined.
Owing to (21.12), we can state that the mean value f .v/ of any function f .v/ in
the velocity space is given by the formula
Z  ˛ 3=2 Z
1 2
f .v/ D f .v/dN.v/ D f .v/e˛v dvx dvy dvz : (21.13)
N <3  <3

In particular, the mean value  of the kinetic energy  D mv2 =2 of a molecule of

mass m is given by the relation
Z
1  ˛ 3=2 2
D mv2 e˛v dvx dvy dvz ;
2  <3

which in spherical coordinates becomes

Z
m  ˛ 3=2 1
2
D 4 v4 e˛v dv:
2  0

Since it can be proved that

Z 1
2 3  ˛ 1=2
v4 e˛v dv D ;
0 8˛ 2 

we finally obtain the result

3m
D : (21.14)
4˛
21.2 Kinetic Theory of Gases 407

z
δ P
x

vx

p
y

Fig. 21.1 Collisions on the wall p

On the other hand, from the macroscopic theory of perfect gases, we know that
the total energy E is given by the formula
3
ED N k; (21.15)
2
where k is the Boltzmann constant and  the absolute temperature of the gas.
Since  D E=N , by comparing (21.15) and (21.14), we determine the value of
the constant ˛:
m
˛D ; (21.16)
2k
and (21.12) assumes the final form
 m 3=2 m 2
dN.v/ D N e 2k v dvx dvy dvz : (21.17)
2k
Now we prove that (21.17) leads to the state equation of perfect gases. To prove
this statement, it is sufficient to evaluate the pressure of the gas on the walls of
the container. Microscopically, this pressure is due to the momentum adsorbed
by the walls during the many collisions of molecules with the walls themselves.
We suppose that these collisions are elastic and consider a plane wall s orthogonal
to the Ox-axis. Let dZ.v/ be the number of collisions per unit time and unit surface
!
of s of the molecules with speed v D OP , with vx > 0 and P belonging to the
parallelepiped ı, which has a unit face on s and height vx (Fig. 21.1). It is evident
that all the molecules contained in ı at instant t reach the wall s in unit time so that
dZ.v/ is equal to this number. Since the volume of ı is vx , we obtain
vx
dZ.v/ D dN.v/: (21.18)
V
408 21 Elements of Statistical Mechanics of Equilibrium

On the other hand, in each collision a particle gives the wall s the momentum 2mvx ,
and then the total momentum transferred to the wall by all the molecules contained
in ı with vx > 0 is
dN.v/
2mvx dZ.v/ D 2mv2x : (21.19)
V
Consequently, the total momentum transferred to the wall s by all the particles
contained in ı with any possible velocity v for which vx > 0 is obtained by
integrating (21.19):
 ˛ 3=2 2mN Z 1 2
Z 1
2
Z 1
2
pD e˛vy dvy e˛vz dvz v2x e˛vx dvx : (21.20)
 V 1 1 0

In view of (21.11) and recalling (21.16), from (21.20) we obtain that

kN
pD : (21.21)
V
But Boltzmann’s constant k is related to Avogadro’s number N0 and to the constant
R of perfect gases by the relation k D R=N0 , and (21.21) assumes the form of the
state equation of a perfect gas
N
pD R  nR: (21.22)
N0

21.3 Boltzmann–Gibbs Distribution

In this section we present an approach to statistical mechanics due to Gibbs. This

approach is still based on statistical hypotheses but it resorts to the phase space
of Hamiltonian mechanics. Let S be a gas at macroscopic equilibrium in a given
volume V with adiabatic walls. As usual, we suppose that S contains a very large
number N of identical molecules with n degrees of freedom. Finally, denote by S

the microscopic system of these N molecules, M2nM the phase space of S, and
by 2n the phase space of any molecule. In particular, if the particles of S are

monoatomic, then M2nN D <6N and 2n D <6 ; if the particles of S are biatomic,

then M2nN D .<6  T 2 /N and 2n D <6  T 2 , where T 2 is a two-dimensional

torus. We can always order the symplectic coordinates .q; p/ of M2nN in such a
way that the first 2n of them refer to the first particle, the coordinates from 2n to
4n refer to the second particle, and so on. We also suppose that the potential energy
describing the interaction between the particles goes to infinity when the distance
goes to zero and differs from zero only in a very small region about any particle. In
this hypothesis there is no collision among the particles, the Hamiltonian formalism

can be applied, and the dynamical orbit is a regular curve of the phase state M2nN .
21.3 Boltzmann–Gibbs Distribution 409

Since S is contained in a reservoir with adiabatic walls, its energy H.q; p/ has
a constant value E. Then, the admissible states of S are represented by the points
of the level manifold †E , defined by the equation H.q; p/ D E, which is supposed
to be compact. Owing to the large value of N , we have no possibility either of
determining the initial data of the particles of S or of integrating the Hamiltonian
equations of S. Once again, we resort to statistical assumptions to have information
about the evolution of S.

In view of our choice of the symplectic coordinates of M2nN , to any point

.q; p/ 2 M2nN we can associate N points .qh ; : : : ; qhCn ; ph ; : : : ; phCn /, h D
1; : : : ; N , belonging to the phase space 2n . These points represent the states of
each particle when the whole system is in the state .q; p/. When this state varies on
†E , the corresponding points of 2n vary in a set   2n , which is bounded since
†E is compact. Let B be a cube containing the set  and consider a partition P of
B into elementary cells Ci , i D 1; : : : ; L, having edges ! and volume ! 2n . We
suppose that
1. Any molecule of S can be singled out with respect to the others;
2. The length ! of the edge of any cell Ci is small with respect to the edge of
the cube B, so that we can suppose that all the particles inside Ci have the same
energy, but it so large that any cell contains many points of .
The first assumption states that it is possible to attribute a target to any
molecule so that we can count them, distinguishing each particle from the others.
The following considerations show the relevance of the second assumption.
Suppose that TNi particles of S are in the states represented by Ni points
belonging
T to C i , i D 1; : : : ; L. Remembering the meaning of the points of
Ci , the numbers Ni must satisfy the following relations:

N1 C    C NL D N; (21.23)
N1 1 C    C NL L D E: (21.24)

It is evident that a single point T on the surface †E corresponds to such a distribution

of particles in the sets Ci , i D 1; : : : ; L. Since we can distinguish T one
particle from all others, we can change the particles belonging to the set Ci 
without changing their number. After this T operation, we have the same distribution
N1 ; : : : ; NL of different particles in Ci , i D 1; : : : ; L. It is evident that a differ-
ent point of †E corresponds to this new distribution. Denoting by D.N1 ; : : : ; NL /
the number of possible ways to obtain a given distribution N1 ; : : : ; NL , we say that
the distribution N1 ; : : : ; NL is the most probable if it corresponds to the maximum
of D.N1 ; : : : ; NL /.
The total number D.N1 ; : : : ; NL / of possible ways to obtain a given distribution
N1 ;T: : : ; NL , is equal to the number
T of combinations obtained setting N1 particle into
C1 , N2 particles into C2 , and so on. This number is given by the relation
    
N N  N1 N  N1     NL1
D.N1 ; : : : ; NL / D 
N1 N2 NL
410 21 Elements of Statistical Mechanics of Equilibrium

which yields the following result

NŠ
D.N1 ; : : : ; NL / D : (21.25)
N1 ŠN2 Š    NL Š
Before going on, we recall Stirling’s formula which states that, for the positive
integer n ! 1, it is
p 1 p
nŠ  2nnC 2 en  2nn en : (21.26)

Consequently, we can state that

p
ln nŠ D n ln n  n C ln 2  n ln n  n: (21.27)

Taking into account (21.27) and (21.25), we obtain

L
X
ln D.N1 ; : : : ; NL / D N ln N  Ni ln Ni : (21.28)
i D1

We are interested in evaluating the distribution N1 ; : : : ; NL corresponding to the

maximum value of D.N1 ; : : : ; NL /, under the constraints (21.23) and (21.24). It is
evident that this distribution can also be obtained by evaluating the maximum value
of ln D.N1 ; : : : ; NL /, that is, of (21.28), under the same constraints. Introducing the
Lagrangian multipliers ˛ and ˇ, we have to find the maximum of the function
L
X
F .N1 ; : : : ; NL / D N ln N  .Ni ln Ni C ˛Ni C ˇNi i / C ˛N C ˇE: (21.29)
i D1

Equating to zero the partial derivatives of (21.29) with respect to the variables Ni ,
we obtain the system

ln Ni C 1 C ˛ C ˇi D 0; i D 1; : : : ; L; (21.30)

so that the values of N1 ; : : : ; NL at which F .N1 ; : : : ; Nl / reaches an extremum are

Ni D e.1C˛/ eˇi ; (21.31)

where the constants ˛ and ˇ can be obtained from (21.23) and (21.24). We omit the
proof that the extremum (21.29) is a maximum.
Following Boltzmann and Gibbs, we state that S evolves in such a way
that its states are always very close to the state of S determined by the most
probable distribution N1 ; : : : ; NL . This assumption is supported by the following
considerations.
Suppose that a distribution N1 ; : : : ; NL of particles into the cells C1 ; : : : ; CL is
given, satisfying conditions (21.23) and (21.24). Then, since the coordinates of all
the particles are assigned, a point of †E is determined that belongs to the e cell

of M2nN intersecting †E and having a volume ! 2nN . It is important to remark
21.3 Boltzmann–Gibbs Distribution 411

that by changing the particles into the cells C1 ; : : : ; CL without changing their
number, we change their coordinates obtaining a different cell Ci of volume ! 2nN .
In conclusion, to all distributions corresponding to N1 particles into C1 , N2 particles

in C2 , and so on, we can associate a family of cells of M2nN with a volume

.N1 ; : : : ; NL / D D.N1 ; : : : ; NL /! 2nN : (21.32)


In particular, the volume of the cells of M2nN corresponding to distribution (21.31)
is given by
.N1 ; : : : ; NL / D D.N1 ; : : : ; NL /! 2nN : (21.33)
Now we evaluate the ratio

.N1 ; : : : ; NL / D.N1 ; : : : ; NL /
  D : (21.34)
.N1 ; : : : ; NL / D.N1 ; : : : ; NL /

Considering the logarithm of both sides and the Taylor expansion of the right-hand
side, we obtain

X L  
 @
ln D D ln D C ln D .Ni  Ni /
i D1
@H i 

L  
1X @2
C ln D .Ni  Ni /.Nh  Nh /: (21.35)
2 i D1 @Ni @Nh 

On the other hand,

@
ln D D  ln Ni  1;
@Ni
@2 1
ln D D  ıhi ;
@Ni @Nh Nh
and (21.35) becomes
L
X
ln D D ln D  C .˛ C ˇi / .Ni  Ni /
i D1
L
X
1 1
 .Ni  Ni /2 :
2 i D1
Ni

Finally, taking into account conditions (21.23) and (21.24), we obtain the condition
 2
PL PL
  12 1  2
i D1 N  .Ni Ni /  12 i D1 Ni
 Ni
Ni
1
D e i De ; (21.36)

412 21 Elements of Statistical Mechanics of Equilibrium

which shows that the ratio = is very close to zero for any distribution
N1 ; : : : ; NL different from N1 ; : : : ; NL . The following remark shows the interest
of the preceding result. We know that the orbit of the system S lies on the manifold

†E  M2nN . In view of (21.36), we can state that this orbit is almost contained in
those cells Ci corresponding to distribution (21.33).
Multiplying (21.33) by ! 2n , in the limit N ! 1, we obtain

dN.q; p/ D aeˇ.q;p/ dqdp; (21.37)

where a D A! 2n . Integrating on the space 2n , we determine the value of a:

N
aD R ; (21.38)
2n eˇ.q;p/ dqdp

and (21.37) gives the Boltzmann–Gibbs distribution:

dN 1
DR ˇ.q;p/ dqdp
eˇ.q;p/ dqdp; (21.39)
N 2n e

which allows us to evaluate the mean value f of any function f .q; p/:
Z
1
f DR ˇ.q;p/ dqdp
f .q; p/eˇ.q;p/ dqdp: (21.40)
2n e 2n

As in Sect. 22.2, we can prove that ˇ D 1=kT .

Remark 21.1. Suppose that the degrees of freedom of any particle are equal to 3,
the momentum p D mPr, and the energy  reduces to the kinetic energy. Then,
integrating (21.40) on the space V with respect to all the momenta except the
momentum of a particle, we obtain again the Maxwell distribution (21.17).
Remark 21.2. We notice that the Boltzmann–Gibbs distribution does not depend on
the choice of canonical coordinates since the determinant of the Jacobian matrix of
a canonical transformation of coordinates .q; p/ ! .q; p/ is equal to one. Further,
from Liouville’s theorem, there follows that (21.40) does not change during motion.

21.4 Equipartition of Energy

Let S be a system of identical molecules Pi , i D 1; : : : ; n, at macroscopic

equilibrium at the absolute temperature . Assume that any molecule has n degrees
of freedom, and denote by 2n the phase state of each of them and by .q; p/
canonical coordinates on 2n . For instance, if Pi is monoatomic, then it has three
translational degrees of freedom; if Pi is biatomic, then it has three translational
degrees of freedom, two rotational degrees of freedom, and one vibrational degree
of freedom. Further, we assume that the interaction among the different parts of Pi
21.4 Equipartition of Energy 413

is described by a potential energy U.q/ that has a minimum at a position 0. Then

we can suppose that the small oscillations of the molecule Pi are described by a
Lagrangian function
n
1 X 
P D
L.q; q/ ai .qP i /2  i .q i /2 (21.41)
2 i D1

[see (17.107)], where the quantities q i are the coefficients of the kinetic energy
evaluated in the equilibrium configuration and i > 0 denotes the principal
frequencies. Consequently, the Hamiltonian function is
n  
1X 1 i 2 i 2
P D
H.q; p/ .pP / C i .q / : (21.42)
2 i D1 ai

Now we prove the fundamental theorem of the equipartition of energy.

Theorem 21.1. Let S be a system of N particles whose Hamiltonian function has
the form (21.42) in a suitable system of Lagrangian coordinates. Then, if S is
at macroscopic equilibrium at the absolute temperature , the mean energy  of
the particle Pi is obtained by supposing that to each quadratic term of (21.42)
corresponds the mean energy k=2, that is,  is given by

 D nk: (21.43)

Proof. The mean energy  is obtained by (21.40) with f given by (21.42). To

evaluate the right-hand side of (21.40), we first note that
 
Z 1 Pn pi2 i 2
 2k i D1 ai C i .q /
e dqdp
2n
n Z
Y p2
n Z
Y i .q i /2
 2a ik
D e i dpi e 2k dq i : (21.44)
i D1 < i D1 <

Further, we have that

 2 
Z n   1 Pn
1 X ph2
pi i /2
 i D1 C i .q
C h .q h /2 e
2k ai
dqdp
2n 2 ah
hD1
 
Z n Pn pi2
1 X ph2  2k
1
i D1 ai C i .q i /2
D e dqdp
2n 2 ah
hD1
 
Z n Pn pi2
1X 1
 2k i D1
i 2
ai C i .q /
C h .q h /2 e dqdp: (21.45)
2n 2
hD1
414 21 Elements of Statistical Mechanics of Equilibrium

The first integral on the right-hand side of the preceding equation can be transformed
as follows:
 2 
Z n 1 Pn
1 X ph2  2k
pi i 2
i D1 ai C i .q /
A e dqdp
2n 2 ah
hD1
n Z
X 2
Y Z n Z
Y
1 ph2  2ka
ph
ip2
 2ka i .qi /2
D e h dph e i dpi e 2k dqi : (21.46)
2 < ah < i D1 <
hD1 i ¤h

Taking into account (21.44) and (21.46) we obtain that

n Z
1X
2
A 1 ph2  2ka
ph
  D e h dph :
Pn pi2 2 R p2 ah
R 1
 2k i D1
i 2
ai C i .q / hD1  ah <
<e dph
h
2n e dqdp

Finally, by (21.11), we can give the preceding equation the final form

A n
  D k:
Pn pi2 2
R 1
 2k i D1 ai C i .q i /2
2n e dqdp

The theorem is proved repeating the same calculations for the second integral on the
right-hand side of (21.45). t
u
The preceding theorem allows us to calculate the specific heat c of gases. In fact,
in view of (21.43), the total energy E of a volume of gas containing N molecules
with n degrees of freedom is given by

E D N nk; (21.47)

and the specific heat is

dE
D nk: cD (21.48)
d
In this chapter we have merely sketched the ideas underlying the statistical
mechanics of equilibrium. More precisely, we have shown how it is possible to
describe the equilibrium of macroscopic bodies adopting a simple mechanical
microscopic model, satisfying classical mechanics, and some convenient statistical
axioms that are necessary to manage the huge number of elementary particles
(molecules) associated with the bodies. In doing this, we have omitted consideration
of many important topics forming the wide subject of statistical mechanics because
of the introductory character of this book. The reader interested in the many aspects
of equilibrium and nonequilibrium statistical mechanics may refer to the extensive
bibliography on the topic (see, for instance, [10, 31, 50, 53]).
Chapter 22
Impulsive Dynamics

22.1 Balance Equations of Impulsive Dynamics

If f .t/ is a function of time, then we denote by

f D f .t2 /  f .t1 / (22.1)

the variation of f .t/ in the time interval .t1 ; t2 /.

Let B be a rigid body acted upon by the forces .Pi ; Fi /, i D 1; : : : ; s, where
the force Fi is applied at the point Pi belonging to the region C occupied by B.
Integrating the balance equations (15.48) and (15.49) in the time interval .t1 ; t2 /, we
obtain
Xs Z t2
mPrG D Fi dt; (22.2)
i D1 t1

s Z t2
X
KO D .Pi  O/  Fi dt; (22.3)
i D1 t1

where O is a fixed point or the center of mass G of B. The integral

Z t2
Ii D Fi dt (22.4)
t1

is called the impulse of the force Fi in the time interval .t1 ; t2 /.

Definition 22.1. We say that the motion of the rigid body B is impulsive in the
short time interval .t1 ; t2 / if in this interval there is a finite variation in the velocity
field of B that is not accompanied by an appreciable variation in its position. When
this happens, the short time interval .t1 ; t2 / is called an exceptional interval.

A. Romano, Classical Mechanics with Mathematica R

, Modeling and Simulation 415
in Science, Engineering and Technology, DOI 10.1007/978-0-8176-8352-8 22,
© Springer Science+Business Media New York 2012
416 22 Impulsive Dynamics

A typical example of an impulsive motion is given by the collision between two

billiard balls. In this case, the exceptional interval coincides with the time interval
during which the two balls are in contact with each other (a fraction of a second).
Definition 22.2. The velocity fields of the body B at the instants t1 and t2 are called
the anterior velocity field and the posterior velocity field, respectively.
Impulsive dynamics is based on the following assumptions:
1. An exceptional time interval .t1 ; t2 / can be identified with the initial instant t1 .
Then, such an instant is called an exceptional instant.
2. During an exceptional interval, the body B remains at the position it occupies at
the initial instant t1 .
In the preceding assumptions, any quantity depending on the velocity field of B
generally exhibits a finite discontinuity at an exceptional instant. This implies that
in analyzing an impulsive motion we are forced to drop the hypothesis of regularity
of the velocity field. On the other hand, assumption 1 requires particular behavior
of the forces acting during the exceptional time interval .t1 ; t2 /. In fact, in view of
(22.2) and (22.4), a variation in the velocity field is compatible with a finite impulse
of the acting forces only if these forces assume very large values in the time interval
.t1 ; t2 /. Forces having this characteristic are called impulsive forces.
Before proceeding, we make clear the preceding considerations using a simple
example. A material point P of mass m that is constrained to move along the axis
Ox is at rest at the initial time t D 0. In the time interval .0; T / point P is acted
upon by the sinusoidal force

F D A sin t; (22.5)
T
so that in .0; T / the velocity and position of P are given by the functions
 
AT t
x.t/
P D 1  cos ; (22.6)
m T
AT AT 2 t
x.t/ D t  2 sin : (22.7)
m  m T
Further, since the impulse of the force F is
Z T
 2AT
I D A sin t D ; (22.8)
0 T 
we obtain a constant value 2˛= of I for any T , provided that we choose A D ˛=T ,
that is, if we increase the applied force when the time interval .0; T / reduces to zero.
In the limit T ! 0, formulae (22.6) and (22.7) give

2˛ I
xP D D ; x D 0: (22.9)
m m
22.2 Fundamental Problem of Impulsive Dynamics 417

Impulsive forces, as ordinary forces, can be active and reactive. In the first case,
they are real impulsive forces applied to B at an exceptional instant. In the second
case, they are produced by the introduction of new constraints at an exceptional
instant. An example of a reactive impulsive force is a ball bouncing on the floor.
From (22.2) and (22.3) and assumptions 1 and 2 we can write the balance
equations of impulsive dynamics of a rigid body B in the following form:
s
X
mPrG D Ii ; (22.10)
i D1
s
X
KO D .Pi  O/  Ii ; (22.11)
i D1

where Ii , i D 1; : : : ; s, are the active and reactive impulses acting on the body B at
an exceptional instant.
For a system S of N rigid bodies Bi , i D 1; : : : ; N , we must apply the preceding
equations to any solid Bi , recalling that the active and reactive impulses acting on
Bi are external to Bi . The reactive impulses can be external to S or determined by
other solids of S. In this last case, they satisfy the action and reaction principle.
In conclusion, if S is a constrained system of N rigid bodies B1 ; : : : ; BN , then
for each body we can write
si
X
mPrGi D Iij  Ii ; (22.12)
j D1
si
X .I/
KOi D .Pj  Oi /  Iij  MOi ; (22.13)
j D1

.I/
i D 1; : : : ; N , where Ii and MOi denote, respectively, the total impulse and the total
impulse torque of the active and reactive impulses acting on Bi .

22.2 Fundamental Problem of Impulsive Dynamics

Let S D fB1 ; : : : ; BN g be a system of constrained rigid bodies and suppose that, at

the exceptional instant t, S is subject to active and reactive impulses. We make the
following assumptions:
• Active and reactive impulses act on points of the bodies Bi of S.
• The constraints are smooth.
• The reactive impulses are due to bilateral or unilateral constraints introduced at
the exceptional instant.
418 22 Impulsive Dynamics

Fig. 22.1 Collision between two solids

The first assumption simplifies the analysis of the problem we are faced with.
The second assumption implies that the reactive impulse acting at the point Ph 2 Bi
is orthogonal to the boundary of the region Ci occupied by Bi at the exceptional
instant. Regarding the third assumption we note that an example of a unilateral
constraint is given by the collision of a body of S with an external obstacle or with
another body of S. A bilateral constraint can be introduced by fixing one or two
points of a body of S.
With the preceding assumptions we can formulate the fundamental problem of
impulsive dynamics of constrained rigid bodies as follows:
Given the position of S , its anterior velocity field of S, the active impulses acting
on S, and the smooth constraints at an exceptional instant, determine the posterior
velocity field of S.
To solve this problem we have at our disposal:
1. Equations (22.12) and (22.13) applied to each body Bi of S;
2. The restrictions on the reactive impulses deriving from the hypothesis that the
constraints are smooth.
Simple examples (see Exercises 1–3 at the end of the chapter) show that the
preceding data are not sufficient to determine the posterior velocity field when
unilateral constraints, i.e., when there are collisions between bodies of S or between
a body of S and an external obstacle, are present at an exceptional instant. To solve
the fundamental problem of the impulsive dynamics of rigid bodies, we need to
add Newton’s laws on collisions between solids. To formulate these laws, consider
two solids Bi and Bj of S and denote by Pi and Pj the contact points of the
bodies Bi and Bj colliding at an exceptional instant. Further, we denote by ni the
internal normal vector to the boundary of Bi at the point Pi and by nj the internal
normal vector to the boundary of Bj at the point Pj (Fig. 22.1). It is evident that
ni D nj . Finally, we denote by rP  i and r PC
i the anterior and posterior speeds of
Pi , respectively, that is, the velocities of Pi before and after the exceptional instant.
Similarly, we denote by rP  PC
j and r j the speeds of Pj before and after the collision.
The two quantities

w r
ij D ni  .P P
j r i /; wC rC
ij D ni  .P PC
i r j / (22.14)
22.3 Lagrangian Formulation of Impulsive Dynamics 419

are called the speed of approach and the speed of separation of the two solids Bi
and Bj , respectively. Then, Newton’s law states that:
There exists a positive number eij 2 Œ0; 1, called the coefficient of restitution
and depending on the physical constitution of Bi and Bj , such that

wC
ij D eij wij : (22.15)

The collision is said to be perfectly elastic if eij D 1 and perfectly inelastic if

eij D 0. It is evident that if the body Bi collides with an external obstacle, then
Newton’s laws can still be applied. However, in this case, the speed rP j in (22.14)
coincides with the velocity of the obstacle at the collision point with Bi .
When we consider condition (22.15), for any contact between solids of S or of
a solid of S with an external obstacle, (22.12) and (22.13), and the restrictions on
the reactive impulses deriving from the hypothesis that the constraints are smooth,
then we obtain a system of equations that allow us to solve the fundamental problem
of impulsive dynamics. We verify this statement by solving some exercises at the
end of the chapter. The theoretical proof of the preceding statement requires the
Lagrangian formalism and will be given in the next sections.

22.3 Lagrangian Formulation of Impulsive Dynamics

Let u.r/ be an arbitrary vector field depending on the position vector r of a point
belonging to region C . Denote by

ır D ırO C ı'  .r  rO / (22.16)

an infinitesimal rigid displacement of region C , where ırO is the infinitesimal

displacement of an arbitrary point O 2 C and ı' the infinitesimal rotation of C .
The following formula holds
Z
u.r/  ır dC D R  ırO C MO  ı'; (22.17)
C

where R is the resultant vector of the field u.r/ and MO its momentum with respect
to O. It is evident that if u.r/ is only defined at a finite set of points P1 ; : : : ; Ps of C ,
then the preceding formula is valid provided that the integral is substituted with the
summation over index i of Pi . To prove (22.15), it is sufficient to introduce (22.16)
into the left-hand side of (22.17) to obtain
Z  Z 
ırO  u.r/ dC C ı'  .r  rO /  u.r/ dC;
C C

and (22.17) is proved.

420 22 Impulsive Dynamics

Now let S D fB1 ; : : : ; BN g be a system of N rigid bodies acted upon by forces

satisfying the hypotheses of impulsive dynamics. Denote by Ci the region occupied
by the body Bi at the exceptional instant t and i its mass density. The total force and
the torque of the vector field i PrdC , which denotes the variation of momentum of
the region dC  C1 \    \ CN at the instant t, are
N
X N
X
mi PrGi ; KGi ; (22.18)
i D1 i D1

where mi is the mass of Bi . By (22.17), for any infinitesimal rigid displacement of

all the bodies of S,1 we obtain the result
N Z
X N
X
i Pr  ır dC D .ırOi  mi PrGi C ı'i  KGi / : (22.19)
i D1 Ci i D1

This result allows us to conclude that (22.12) and (22.13) are equivalent to the
condition
XN Z N
X
i Pr  ır dC  Ii  ıri D 0; (22.20)
i D1 Ci i D1

in which ır is an arbitrary infinitesimal rigid displacement of any body of S.

Definition 22.3. An infinitesimal rigid displacement of Bi is virtual if it is
compatible with the constraints of S at an exceptional instant.
Although (22.20), for this class of displacements, is no longer equivalent to (22.12)
and (22.13), it leads to a condition that supplies the Lagrangian formulation of the
impulsive dynamics.
In fact, let Vn be the configuration space of the constrained system S at the
exceptional instant t, and denote by q 1 ; : : : ; q n arbitrary Lagrangian coordinates
of Vn . Then the position vector r of any point of S in its configuration at the instant
t is a function of the Lagrangian coordinates (Sect. 17.3), and any infinitesimal
displacement
n
X @r h
ır D ıq (22.21)
@q h
hD1

is compatible with all the constraints of S, i.e., it is a virtual displacement.

Introducing (22.21) into (22.20), we obtain the condition
n N Z
! n
X X @r X
i Pr  h dC ıq h D Ih ıq h ; (22.22)
i D1 Ci
@q
hD1 hD1

1
We explicitly note that these displacements could not satisfy the constraints of S.
22.4 Analysis of the Lagrange Equations 421

where
N
X @r
Ih D Ii  (22.23)
i D1
@q h
are the Lagrangian components of the impulses.
Since with (17.41) we proved that
@r @Pr
D h;
@q h @qP
we can write
N Z
X @Pr @T
i Pr  dC D  h D ph ; (22.24)
i D1 Ci @qP h @qP

where T is the kinetic energy of S and ph , h D 1; : : : ; n, are the momenta of S

(Chap. 17).
Introducing (22.24) into (22.22) and using the arbitrariness of the variations ıq h ,
we finally obtain the Lagrangian equations of impulsive dynamics:
ph D Ih ; h D 1; : : : ; n: (22.25)

22.4 Analysis of the Lagrange Equations

The impulses acting on the system S at the exceptional instant t are active and
reactive. Further, the reactive impulses can be determined by bilateral or unilateral
constraints. In other words, we can decompose Ih into the summation
.rb/ .ru/
Ih D Iha C Ih C Ih ; (22.26)
.a/
where Ih , h D 1; : : : ; n, are the Lagrangian components of the active impulses,
.rb/
Ih are the Lagrangian components of the reactive impulses produced by bilateral
.ru/
constraints, and Ih are the Lagrangian components of the reactive impulses
generated by unilateral constraints. On the other hand, if the constraints are smooth,
.rb/
Ii is orthogonal to ıri , i D 1; : : : ; N , and all the corresponding Lagrangian
components vanish. Consequently, system (22.25) reduces to
.a/ .ru/
ph D Ih C Ih ; h D 1; : : : ; n: (22.27)

In the absence of unilateral constraints, the preceding system allows us to

determine the quantities phC when the active impulses and the values ph are known.
Recalling that the Lagrangian velocities qP h and the kinetic momenta ph are related
by the relations (Chap. 17)
qP h D ahk pk ; (22.28)
422 22 Impulsive Dynamics

we can conclude that (22.27) solve the fundamental problem of impulsive dynamics,
at least in the absence of unilateral constraints, i.e., in the absence of collisions.
We want to show that we can solve this problem also in the presence of unilateral
smooth constraints, provided that we add Newton’s laws of collision to the Eqs.
(22.27). For the sake of simplicity, we suppose that only the bodies Bi and Bj of the
system S collide at the instant t, when they occupy regions Ci and Cj , respectively.
Denote by Pi 2 Ci and Pj 2 Cj the collision points. Then, since the constraints are
smooth, the reactive impulses at the contact points are parallel to the internal unit
normal vectors ni and nj to the surfaces of Ci , respectively. Further, these impulses
verify the action and reaction principle and ni D nj . In view of these remarks, we
can write that
 
.ru/ .ru/ @ri .ru/ @rj .ru/ @ri @rj
Ih D Ii ni  h C Ij nj  h D Ii ni   : (22.29)
@q @q @q h @q h
Introducing (22.29) into (22.27), we obtain the system
 
.a/ .ru/ @ri @rj
phC D ph C Ih C Ii ni   ; h D 1; : : : ; n: (22.30)
@q h @q h
System (22.30) shows that the collision between Bi and Bj introduces the only
.rv/
unknown Ii . Therefore, we need an equation to equate the equations and the
unknowns. In view of the Lagrangian formula of the Lagrangian velocity,
@r h
rP D qP ; (22.31)
@q h
(22.14) and (22.28), Newton’s law of collision (22.15) leads to the following
equation:
   
@ri @rj hk @rj @ri
ni   a pkC D cij ni   ahk pk ; (22.32)
@q h @q h @q h @q h

which, together with (22.30), gives a system to determine the n C 1 unknowns

.ru/
phC , h D 1; : : : ; n, and Ii . It is evident that the preceding considerations can
be extended to the case of more collisions.

22.5 Exercises

1. Let S be a homogeneous circular disk at rest on a horizontal plane . Suppose

that at the exceptional instant t an active impulse I, parallel to , is applied at a
point P on the boundary of S . Determine the posterior speed field assuming that
 is smooth.
22.5 Exercises 423

G O P x

Fig. 22.2 Homogeneous disk on a plane

We choose the Cartesian axes as in Fig. 22.2. First, we solve the problem by
cardinal equations that, with our notations, give

mxP G D Ix ;
myPG D Iy ;
C'P D rIy ;

where m is the mass of S , C the moment of inertia relative to the Oz-axis, and
r the radius of S . Since xP G D yPG D 'P  D 0, from the preceding equations we
obtain
Ix
xP GC D ;
m
Iy
yPGC D ;
m
rIy
'P C D :
C
2. Solve the preceding problem using the Lagrangian formalism.
We choose as Lagrangian coordinates xG , yG , and ', where xG and yG are
Cartesian coordinates with the origin at a point O ¤ G. Then the position vector
of any point of S is

r D xG i C yG j C .cos 'i C sin 'j/; (22.33)

where i and j are the unit vectors along the Cartesian axes and 0    r.
Further, the kinetic energy of S can be written as
1  2  1
T D m xP G C yPG2 C C 'P 2 : (22.34)
2 2
First, verify that

pxG D mxP G ; pyG D myPG ; p' D C ';

P
424 22 Impulsive Dynamics

y
l
S
.
rP-
r
n x
G O P
ru
I

Fig. 22.3 Collision of a homogeneous disk

and the Lagrangian components of the active impulse are

 
@r
Ix G D I  D I  i D Ix ;
@xG P
 
@r
IyG D I  D I  j D Iy ;
@yG P
 
@r
I' D I  D I  . sin 'i C  cos 'j/P D rIy :
@' P

Applying (22.27), we obtain the same equations we derived from the balance
equations.
3. Let S be a homogeneous circular disk of radius r and mass m moving on a
horizontal smooth plane . Suppose that at instant t, disk S collides with a
vertical plane whose intersection with  is the straight line l. Determine the
posterior velocity field knowing the velocity field before the collision.
We refer to Fig. 22.3. The balance equations of impulsive dynamics give

mxP G D I .ru/ ; myPG D 0; C 'P D 0;

where C is the moment of inertia relative to the axis orthogonal to  at point G.

Equivalently, we have that

I .ru/
xP GC D xP G  ; yPGC D yPG ; 'P C D 'P  : (22.35)
m
On the other hand, Newton’s collision law (22.15) gives
C
 n  r
P D cn  rP ; (22.36)

where c is the coefficient of restitution relative to the collision between S and the
straight line l. Denoting by i and j the unit vectors along the Ox- and Oy-axes,
respectively, and recalling that
22.5 Exercises 425

rP P D xP G i C yPG j C 'k
P  .xG i C yG j/;

where k D i  j, relation (22.36), in view of (22.35), can also be written as

xP GC D c xPG : (22.37)

By this equation and (22.35), we solve the problem of determining the posterior
speed of S .
4. Solve Problem 3 adopting the Lagrangian formalism.
Introduce a new Cartesian frame Oxy, O ¤ G, with axes parallel to the axes
adopted in Fig. 22.4. Then, xG , yG , and ' are Lagrangian coordinates for S and
the position vector r of the points of S and the kinetic energy T are given by
(22.33) and (22.34), respectively. Consequently, we have that

pxG D mxP G ; pyG D myPG ; p' D C ':

P

On the other hand, since n D i and the straight line l is at rest, the Lagrangian
components of the reactive impulse I.ru/ D I .ru/ n, due to the unilateral
constraint, are
 
.ru/ .ru/ @r
Ix G D  I i  D Ix ;
@xG P
 
.ru/ .ru/ @r
IyG D I i  D 0;
@yG P
 
.ru/ .ru/ @r
I' D I i
@' P
D I .ru/ i  . sin 'i C  cos 'j/P D 0:

In conclusion, (22.27) reduce to

mxP G D I .ru/ ; yPG D 0; 'P D 0:

It remains to apply the Lagrangian formulation of Newton’s law (22.32), which

is now written as
   
@r h C @r
i .qP / D ci  .qP h / : (22.38)
@q h P @q h P

Recalling that .qP h / D .xP G ; yPG ; '/,

P it is a simple exercise to verify that this
equation coincides with (22.37).
5. Determine the posterior speed field after the collision of two homogeneous
circular disks S1 and S2 moving on a smooth horizontal plane .
426 22 Impulsive Dynamics

y
.
S rP-

r .
rP-
n x

G O
I ru G

P P

Fig. 22.4 Collision between two disks

Denote by m1 and m2 the masses of S1 and S2 and by r1 and r2 their radii.

Finally, denote by C1 and C2 their moments of inertia relative to the centers
of mass G1 and G2 (Fig. 22.4). The balance equations of impulsive dynamics
applied to each disk give
.ru/
mi xP Gi D Ii ;
mi yPGi D 0;
Ci 'i D 0;
.ru/ .ru/
i D 1; 2, where I1 D I2 , owing to the action and reaction principle. Show
that Newton’s law of collision allows one to obtain the values of xP GCi after the
collision.
6. Solve the preceding problem by the Lagrange equations.
Chapter 23
Introduction to Fluid Mechanics

23.1 Kinematics of a Continuous System

In previous chapters, we only considered systems of point particles or rigid bodies.

In this chapter, we analyze the fundamental principles underlying mechanical
models in which, together with the extension of bodies, their deformability is taken
into account.1
Let S be a deformable continuous system moving relative to a Cartesian frame
of reference Ox1 x2 x2 . Denote by C.0/ and C.t/ the regions occupied by S at the
initial time and at the arbitrary instant t, respectively. If x 2 C.t/ is the position
vector at instant t of the point X 2 C.0/, then the equation of motion of S has
the form
x D x.X; t/: (23.1)
We suppose that function (23.1) is of class C 2 with respect to its variables and
confirms the condition
 
@xi
J D det > 0; (23.2)
@Xj
so that (23.1) is invertible for any instant t.
Any quantity of the continuous system S can be represented in Lagrangian
or Eulerian form, depending on whether it is expressed as a function of X, t or
(x, t), i.e., if it is a field assigned on the initial configuration C.0/ or on the current
configuration C.t/:

D .x; t/ D .x.X; t/; t/ D e.X; t/: (23.3)

For instance, the velocity and the acceleration of the particle X 2 C.0/ are given
by the following partial derivatives:

1
For the contents of this chapter see [43, 44, 51].

A. Romano, Classical Mechanics with Mathematica R

, Modeling and Simulation 427
in Science, Engineering and Technology, DOI 10.1007/978-0-8176-8352-8 23,
© Springer Science+Business Media New York 2012
428 23 Introduction to Fluid Mechanics

@x @2 x
v De
v.X; t/ D ; a De
a.X; t/ D : (23.4)
@t @t 2
Because of the invertibility assumption of (23.1), the preceding fields can be
represented in the Eulerian form

v D v.x; t/ D e
v.x.X; t/; t/; a D a.x; t/ D e
a.x.X; t/; t/: (23.5)

Consequently, the Eulerian form of acceleration is given by

@v
aD C v  rv; (23.6)
@t
where
@v
v  rv D vj : (23.7)
@xj
More generally, the time derivative of the Eulerian field .x; t/ is

P D @ Cvr ; (23.8)
@t
and we can say that the material derivative in the spatial representation contains two
contributions: the first one is a local change expressed by the partial time derivative
and the second one is the convective derivative.
Relevant features of motion are often highlighted by referring to particle paths
or to streamlines. For this reason we introduce the following definitions.
The vector field v.x; t/, at a fixed time instant t, is called a kinetic field.
A particle path is the trajectory of an individual particle of S . In Lagrangian
terms, particle paths can be obtained by integration of (23.4):
Z t
x D x0 C vQ .X; t/dt: (23.9)
0

If the velocity field is expressed in the Eulerian form v D v.x; t/, the
determination of particle paths requires the integration of a nonautonomous system
of first-order differential equations:
dx
D v.x; t/: (23.10)
dt
A streamline is defined as the continuous line, at a fixed instant t, whose tangent
at any point is in the direction of velocity at that point.
Based on this definition, streamlines represent the integral curves of the kinetic
field, i.e., the solution of the autonomous system
dx
D v.x; t/; t D const; (23.11)
ds
23.2 Velocity Gradient 429

where s is a parameter along the curve. We note that two streamlines cannot
intersect; otherwise, one would face the paradoxical situation of a velocity with
two directions at the intersection point.
The motion is defined as stationary if
@
v.x; t/ D 0; (23.12)
@t
which is equivalent to saying that all particles of S that during the time evolution
cross the position x 2 C have the same velocity.
Since in the stationary motion the right-hand sides of both (23.11) and (23.12)
are independent of t, the two systems of differential equations are equivalent to each
other, so that the particle paths and streamlines coincide.

23.2 Velocity Gradient

Two relevant kinematic tensors will be used extensively. The first one is a symmetric
tensor, defined as the rate of deformation or stretching
 
1 @vi @vj
Dij D C D Dj i ; (23.13)
2 @xj @xi

and the second one is a skew-symmetric tensor, defined as a spin or vorticity tensor
 
1 @vi @vj
Wij D  D Wj i : (23.14)
2 @xj @xi

By means of these two tensors, the spatial gradient of velocity can be conveniently
additively decomposed into
rv D D C W: (23.15)
By recalling the definition of differential

v.x C dx; t/ D v.x; t/ C dx  rv (23.16)

and using the decomposition (23.15), we obtain

v.x C dx; t/ D v.x; t/ C !  dx C Ddx; (23.17)

where ! is a vector such that !  dx D Wdx. It can be verified that

1
!D r  v; (23.18)
2
430 23 Introduction to Fluid Mechanics

which shows that, in a neighborhood of x 2 C , the local kinetic field is composed

by a rigid motion, with angular velocity given by (23.17), so that it is a function of
time and x, as well as of the contribution Ddx.
In the sequel, it will be useful to refer to the following expression for Eulerian
acceleration, derived from (23.6)

@vi @vi @vi @vi @vj @vj

ai D C vj D C vj C vj  vj
@t @xj @t @xj @xi @xi
@vi 1 @v2
D C 2Wij vj C :
@t 2 @xj

Since Wij vj D ihj !h vj , we obtain

@vi 1 @v2
ai D C 2ihj !h vj C ;
@t 2 @xj
and, referring to (23.17), we obtain the relation
@v 1
aD C .r  v/  vC rv2 : (23.19)
@t 2

23.3 Rigid, Irrotational, and Isochoric Motions

For the analysis we are going to present, it is of interest to introduce Liouville’s

formula
JP D J r  v: (23.20)
To prove (23.19), first observe that, by definition,
ˇ ˇ ˇ ˇ
ˇ @v1 @v1 @v1 ˇ ˇ @x1 @x1 @x1 ˇ
ˇ ˇ ˇ ˇ
ˇ @X1 @X2 @X3 ˇ ˇ @X1 @X2 @X3 ˇ
  ˇ ˇ ˇ ˇ ˇ
@x @x @x @x ˇ ˇ @x @x2 @x2 ˇ
d i ˇ 2 2 2 ˇ ˇ 2 ˇ
JP D det Dˇ ˇ CCˇ ˇ:
dt @XL ˇ @X1 @X2 @X3 ˇ ˇ @X1 @X2 @X3 ˇ
ˇ ˇ ˇ ˇ
ˇ @x3 @x3 @x3 ˇ ˇ @v3 @v3 @v3 ˇ
ˇ ˇ ˇ ˇ
ˇ @X @X @X ˇ ˇ @X @X2 @X3 ˇ
1 2 3 1

Moreover,
@vh @vh @xj
D
@XL @xj @XL
and, since each determinant can be written as J @vi =@xi (no summation on i ), (23.19)
is proved.
23.3 Rigid, Irrotational, and Isochoric Motions 431

Theorem 23.1. The motion of S is (globally) rigid if and only if

D D 0: (23.21)

Proof. To prove that (23.20) is a necessary condition, we start from the velocity
field of a rigid motion vi .x; t/ D vi .x0 ; t/ C ij l !j .t/.xl  x0l /. Then the velocity
gradient
@vi
D ij k !j
@xk
is skew-symmetric, and we obtain that
@vi @vk  
2Dik D C D ij k C kj i !j D 0:
@xk @xi
To prove that D D 0 is a sufficient condition, by (23.16) and (23.17), we have that
@vi
D Wij ; Wij D Wj i : (23.22)
@xj

System (23.21) of nine differential equations with the three unknown functions vi .x/
can be written in the equivalent form

dvi D Wij dxj ; (23.23)

so that (23.21) have a solution if and only if we can integrate the differential
forms (23.22). If region C of the kinetic field is simply linearly connected, a
necessary and sufficient condition for (23.22) to be integrable is
@Wij @Wih
D :
@xh @xj
By cyclic permutation of indices, two additional conditions follow:
@Whi @Whj @Wj h @Wj i
D ; D :
@xj @xi @xi @xh
Summing up the first two, subtracting the third one, and taking into account (23.21)2,
we obtain the condition
@Wij
D 0;
@xh
which shows that the skew-symmetric tensor Wij does not depend on spatial
variables and eventually depends on time. Then, integration of (23.22) gives
vi .x; t/ D v0i .t/ C Wij .t/.xj  x0j /;

and the motion is rigid. t

u
432 23 Introduction to Fluid Mechanics

The motion of S is irrotational if

1
!D r  v D 0: (23.24)
2
Again supposing that region C is simply linearly connected, it follows that the
motion is irrotational if and only if

v D r'; (23.25)

where ' is a velocity potential, also defined as the kinetic potential.

Consider any region c.t/  C.t/, which is the mapping of c.0/  C.0/,
throughout the equations of motion. This region is said to be a material volume
because it is always occupied by the same particles. If during the motion of S the
volume of any arbitrary material region does not change, then the motion is said to
be isochoric or isovolumic, i.e.,
Z
d
dc D 0:
dt c.t /

By a change of variables .xi / ! .Xi /, the previous requirement is also written as

Z
d
J dc D 0;
dt c.0/

and, since the volume c.0/ is fixed, differentiation and integration can be exchanged
and, because of (23.19), it holds that
Z Z
J r  v dc D r  v dc D 0; 8 c.t/  C.t/:
c.0/ c.t /

Thus, the conclusion is reached that a motion is isochoric if and only if

r  v D 0: (23.26)

Finally, an irrotational motion is isochoric if and only if the velocity potential

verifies Laplace’s equation
' D r  r' D 0; (23.27)

whose solutions are known as harmonic functions.

23.4 Mass Conservation

To derive the mass conservation law for a continuous system S , the basic assumption
is introduced that the mass of S is continuously distributed over the region C.t/ it
23.4 Mass Conservation 433

occupies at instant t. In other words, we assume the existence of a function  .x; t/,
called the mass density, which is supposed to be of class C 1 on C.t/.
According to this assumption, if any arbitrary region c.0/ 2 C.0/ of S is mapped
onto c.t/ during motion x.X; t/, then the mass of c.0/ at instant t is given by
Z
m.c.0// D .x; t/ dc; (23.28)
c.t /

and the mass conservation principle postulates that, during motion, the mass of any
material region does not change over time
Z
d
.x; t/ dc D 0: (23.29)
dt c.t /

The mass conservation law for an arbitrary fixed volume v assumes the form
Z Z
d
.x; t/ dc D  .x; t/ v  Ndc: (23.30)
dt v @v

Starting both from (23.29) and (23.30), when the density function is continuous
in the region C.t/, we obtain that mass conservation is expressed by the following
local equation:
@
C r  .v/ D P C r  v D 0: (23.31)
@t
Before formulating the dynamical laws of a perfect fluid according to Euler, we
prove the fundamental transport theorem:
Theorem 23.2. If .x; t/ is a C 1 Eulerian field defined on the material region C.t/
occupied by the continuous system S , then the following result holds:
Z Z
d
 dc D  P dc; (23.32)
dt c.t / c.t /

for any material region c.t/  C.t/.

Proof. In fact, we have that
Z Z
d d
 dc D J dc:
dt c.t / dt c.0/

Taking into account (23.19) and (23.31), the preceding equation gives
Z Z Z Z
d
 dc D .J r  v C J P C J P /dc D J P dc D  P dc;
dt c.t / c.0/ c.0/ c.t /

and the theorem is proved. t

u
434 23 Introduction to Fluid Mechanics

23.5 Equations of Balance of a Perfect Fluid

In this section we introduce a model, due to Euler, that gives a good description
of the behavior of gases and liquids, at least under suitable kinematic conditions
clarified in the sequel. We do not present this model starting from the modern
general principles of continuum mechanics since we prefer to follow a historical
approach.
We start with the first fundamental assumption on which Euler’s model of a
perfect fluid is based. Let @c.t/ be the boundary of an arbitrary region c.t/  C.t/,
where C.t/ is the region occupied by the fluid S at instant t, and let n be the external
unit normal to @c.t/. Then the total force F exerted by the fluid external to c.t/ on
the fluid contained in c.t/ is given by
Z
F@c.t / D  p.x/nd; (23.33)
@c.t /

where the pressure p depends only on the point x 2 @c.t/. In other words, (23.33)
states that Pascal’s law also holds in dynamical conditions. Besides the surface
force (23.33), it is assumed that there exist forces called body forces that are
distributed over the volume C.t/. Denoting by b the force per unit mass, the total
volume force acting on c.t/ is expressed by the integral
Z
Fc.t / D b dc: (23.34)
c.t /

In particular, if the only volume force is the weight, then b D g, where g is the
gravity acceleration.
The second fundamental assumption of fluid dynamics states that the balance
equations of mechanics hold for any material volume c of the fluid.
Taking into account (23.33), we can write the balance equations of a fluid in the
form
Z Z Z
d
v dc D  pn d C b dc; (23.35)
dt c.t / @c.t / c.t /
Z Z
d
.x  x0 /  v dc D  .x  x0 /  pn d
dt c.t / @c.t /
Z
C .x  x0 /  b dc; (23.36)
c.t /

where x0 is the position vector of an arbitrary pole.

From the law of balance (23.35), valid for any material volume c.t/  C.t/,
when we assume that the functions under the integrals are continuous, take into
account (20.32), and apply Gauss’s theorem to the surface integral, we obtain the
following local equation at any point x 2 C.t/:
23.5 Equations of Balance of a Perfect Fluid 435

a D rp C b: (23.37)

Similarly, assuming that the pole x0 is fixed in the frame of reference R relative
to which we evaluate the motion of the fluid S , the i th component of (23.36) gives
the local condition

@p
ij h .xj  x0j /ah D ij h ıj h p  ij h .xj  x0j /
@xh
Cij h .xj  x0j /bh ; (23.38)

which is identically satisfied in view of (23.37).

In conclusion, collecting the results of this and the previous section, we have the
following equations to evaluate the dynamical evolution of a perfect fluid:

P D r  v; (23.39)
vP D rp C b: (23.40)

This is a system of nonlinear partial differential equations [see (23.12)] in the

unknowns .x; t/, v.x; t/, and p.x; t/. In other words, this system is not closed.
This circumstance is quite understandable since the preceding equations hold for
any perfect fluid S . We need to introduce in some way the nature of S to obtain the
balancing between equations and unknowns.
If S is incompressible, i.e., if S is a liquid, then the mass density  is constant
and system (23.39), (23.40) becomes

r  v D 0; (23.41)
1
vP D  rp C b: (23.42)


In this case, we obtain a system of two nonlinear partial differential equations in the
unknowns v.x; t/, and p.x; t/. In other words, one liquid differs from another only
in its density.
When the fluid S is compressible, i.e., if S is a gas, then the balancing between
the equations and unknowns is obtained giving a constitutive equation relating the
pressure to the density of the gas:

p D p./: (23.43)

This equation can only be determined by experimentation. For instance, for an

ideal gas
p D R; (23.44)
436 23 Introduction to Fluid Mechanics

where R is the universal gas constant and  the absolute temperature. For a van der
Waals gas S
R  a
pD  ; (23.45)
M v  b v2
where M is the molar mass of S , v D 1= is the specific volume, and a and b are
two constants depending on the gas.

23.6 Statics of Fluids

In this section, we consider conservative body forces, i.e., forces for which a
potential energy U.x/ exists such that

b D rU.x/: (23.46)

The statics of a perfect fluid subject to conservative body forces is governed by

the following equation [see (23.39) and (23.40)]:

rp D rU.x/; (23.47)

where  is constant for a liquid and a given function p D p./ for a gas. Since
experimental evidence leads us to assume that dp=d > 0, the function p D p./
can be inverted.
If we introduce the notation
Z
dp
h./ D ; (23.48)

then
dh 1
rh.p/ D rp D rp;
d 
and (23.47) becomes
r.h.p/ C U.x// D 0: (23.49)
This equation shows that in any connected region occupied by fluids, the following
condition holds:
h.p/ C U.x/ D const: (23.50)
In particular, for a liquid, (23.50) becomes [see (23.48)]
p
C U.x/ D const: (23.51)


We now propose to analyze some consequences of (23.50) and (23.51).

23.6 Statics of Fluids 437

• In a fluid S , the level surfaces of the potential energy coincide with the isobars.
• Let S be a liquid, subject to its weight, on whose boundary acts a uniform
pressure p0 (for instance, atmospheric pressure). If the vertical axis Oz is
downward oriented, then the relation U.x/ D gz C const implies that the free
surface of S is a horizontal plane.
• Let S be a liquid in a cylinder subject to its weight and to a uniform pressure on
its boundary. If the cylinder containing S is rotating about a vertical axis a with
uniform angular velocity !, then its free surface is a paraboloid of rotation about
!
a. In fact, the force acting on the arbitrary particle P 2 S is given by g C ! 2 QP ,
where g is the gravity acceleration and Q the projection of P on a. It follows that

!2 2
U D gz  .x C y 2 /;
2
and (23.51) proves that the free surface is a paraboloid.
• Let S be a liquid in a container subject to its weight and to atmospheric pressure
p0 on its free surface. Assume that the vertical axis Oz is downward oriented,
and choose the origin in such a way that the arbitrary constant of the potential
energy is equal to zero when z D 0. Then from (23.51) it follows that
p.z/ p0
 gz D ;
 
i.e.,
p.z/ D p0 C gz; z > 0; (23.52)
and Stevin’s law is obtained: Pressure increases linearly with depth by an amount
equal to the weight of the liquid column acting on the unit surface.
• Let S be an ideal gas at equilibrium at constant and uniform temperature when
subject to its weight. Assuming that the vertical axis Oz is upward oriented, we
get U.z/ D gz C const, and, taking into account (23.44), (23.50) becomes
Z p
dp
R D gz;
p0 p

so that
 gz 
p.z/ D p0 exp  : (23.53)
R
Finally, we prove an important consequence of (23.51).
Theorem 23.3 (Archimedes’ principle). The buoyant force on a body submerged
in a liquid is equal to the weight of the liquid displaced by the body.

Proof. To prove this statement, consider a body S submerged in a liquid, as shown

in Fig. 23.1. If p0 is the atmospheric pressure, then the force acting on S is given by
Z Z
FD p0 N d  .p0 C gz/N d; (23.54)
e i
438 23 Introduction to Fluid Mechanics

Ce

Ci

Fig. 23.1 Archimedes’ principle

where N is the outward unit vector normal to the body surface , i is the submerged
portion of , and e is the portion above the waterline. By adding and subtracting
on the right-hand side of (23.54) the integral of p0 N over  (Fig. 23.1), (23.54)
becomes
Z Z
FD p0 N d  .p0 C gz/N d; (23.55)
@Ce @Ci

so that knowledge of F requires the computation of the integrals in (23.55). To do

this, we define a virtual pressure field (continuous on ) in the interior of the body:
p D p0 on Ce ;
p D p0 C gz on Ci :
Applying Gauss’s theorem to the integrals in (23.55), we obtain
Z Z Z Z
p0 N d D rp0 dV D 0; p0 N d D rp0 dV D 0;
@Ce Ce @Ci Ci
Z Z
gzN d D g rz dV:
@Ci Ci

Finally,
F D gVi k; (23.56)
where k is the unit vector associated with Oz. t
u
Equation (23.56) gives the resultant force acting on the body S . To complete
the equilibrium analysis, the momentum MO of pressure forces with respect to an
!
arbitrary pole O must be explored. If r D OP , then
23.7 Fundamental Theorems of Fluid Dynamics 439

Z Z
MO D p0 r  N d  r  .p0 C gz/N d: (23.57)
e i

Again, by adding and subtracting on the right-hand side of (23.57) the integral of
p0 r  N over , (23.57) becomes
Z Z
MO D p0 r  N d  r  .p0 C gz/N d:
@Ce @Ci

Applying Gauss’s theorem, we obtain

Z Z
@xj
 p0 ij l dV D p0 ij l ıj l dV D 0;
Ce @xl Ce
Z
@ 
ij l xj .p0 C gz/ dV
Ci @xl
Z

D ij l ıj l .p0 C gz/ C ij l xj gı3l dV
Ci
Z
D g ij 3 xj dV:
Ci

Finally,
M0 D g Œx2C i  x1C j Vi ; (23.58)
where i and j are orthonormal base vectors on the horizontal plane and
Z Z
x1C Vi D x1 dV; x2C Vi D x2 dV: (23.59)
Ci Ci

Expression (23.58) shows that the momentum of the pressure forces vanishes
if the line of action of the buoyant force passes through the centroid of the body.
The centroid of the displaced liquid volume is called the center of buoyancy.
In summary: A body floating in a liquid is at equilibrium if the buoyant force is
equal to its weight and the line of action of the buoyant force passes through the
centroid of the body. It can be proved that the equilibrium is stable if the center of
buoyancy is above the centroid and is unstable if the center of buoyancy is below
the centroid.

23.7 Fundamental Theorems of Fluid Dynamics

The momentum balance equation (23.40), when applied to a perfect fluid subjected
to conservative body forces, is written as

vP D rp  rU I (23.60)

440 23 Introduction to Fluid Mechanics

with the additional introduction of (23.48), it holds that

vP D r.h.p/ C U /: (23.61)

Recalling the preceding definitions, we can prove the following theorems.

Theorem 23.4 (W. Thomson, Lord Kelvin). In a barotropic flow under conserva-
tive body forces, the circulation around any closed material curve is preserved,
i.e., it is independent of time:
Z
d
v  ds D 0: (23.62)
dt

Proof. If is a material closed curve, then there exists a closed curve  in C  such
that is the image of  under the motion equation D x.  ; t/. It follows that
Z Z Z  
d d @xi d @xi
vi dxi D vi dXj D vi dXj
dt dt  @Xj  dt @Xj
Z   Z  
@xi @xP i @vi
D vP i C vi dXj D vP i C vi dxj ;
 @Xj @Xj @xj

and, taking into account (23.61), it is proved that

Z Z   Z  
d 1 2 1 2
v  ds D vP C rv  ds D  r h.p/ C U  v  ds D 0: (23.63)
dt 2 2

t
u
Theorem 23.5 (Lagrange). If at a given instant t0 a motion is irrotational, then it
continues to be irrotational at any t > t0 , or, equivalently, vortices cannot form.
Proof. This can be regarded as a special case of Thomson’s theorem. Suppose that
in the region C0 occupied by fluid at instant t0 the condition ! D 0 holds. Stokes’s
theorem requires that
Z

0 D v  ds D 0
0

for any material closed curve 0 . But Thomson’s theorem states that
D 0 for all
t > t0 , so that also !.t/ D 0 holds at any instant. t
u
Theorem 23.6 (Bernoulli). In a steady flow, along any particle path, i.e., along the
trajectory of an individual element of fluid, the quantity

1 2
H D v C h.p/ C U (23.64)
2
23.7 Fundamental Theorems of Fluid Dynamics 441

is constant. In general, the constant H changes from one streamline to another, but
if the motion is irrotational, then H is constant in time and over the whole space of
the flow field.
Proof. Recalling (23.19) and the time independence of the flow, we can write
(23.61) as
1
vP D .r  v/  v C rv2 D r.h.p/ C U /: (23.65)
2
A scalar multiplication by v gives the relation
 
1 2
vP  r v C h.p/ C U D 0;
2
which proves that (23.64) is constant along any particle path. If the steady flow is
irrotational, then (23.65) implies H D const through the flow field at any time. ut
In particular, if the fluid is incompressible, then Bernoulli’s theorem states that
in a steady flow along any particle path (or through the flow field if the flow is
irrotational) the quantity H is preserved, i.e.,

1 2 p
H D v C C U D const: (23.66)
2 

The Bernoulli equation is often used in another form, obtained by dividing (23.66)
by the gravitational acceleration

hz C hp C hv D const;

where hz D U=g is the gravity head or potential head, hp D p=g is the pressure
head, and hv D v2 =2g is the velocity head.
In a steady flow, a stream tube is a tubular region † within a fluid bounded by
streamlines. We note that streamlines cannot intersect each other. Because @=@t D
0, the mass conservation (23.31) gives r  .v/ D 0, and by integrating over a
volume V defined by the sections 1 and 2 of a stream tube (Fig. 23.2), we obtain
Z Z
QD v  n d D v  n d: (23.67)
1 2

This relation proves that the flux is constant across any section of the stream tube.
If the fluid is incompressible, then (23.67) reduces to
Z Z
v  n d D v  n d: (23.68)
1 2

The local angular speed ! is also called the vortex vector, and the related integral
curves are vortex lines; furthermore, a vortex tube is a surface represented by all
442 23 Introduction to Fluid Mechanics

N
v
1

Fig. 23.2 Stream tube

vortex lines passing through the points of a (nonvortex) closed curve. Recalling
that a vector field w satisfying the condition r  w D 0 is termed solenoidal and that
2r  ! D r  r v D 0, we conclude that the field ! is solenoidal. Therefore, vortex
lines are closed if they are limited, and they are open if unconfined. We observe that
Fig. 23.3 can also be used to represent a vortex tube if the vector v is replaced by !.
The following examples illustrate some relevant applications of Bernoulli’s
equation.
1. Consider an open vessel with an orifice at depth h from the free surface of the
fluid. Suppose that fluid is added on the top to keep the height h constant. Under
these circumstances, it can be proved that the velocity of the fluid leaving the
vessel through the orifice is equal to that of a body falling from the elevation h
with initial velocity equal to zero (this result is known as Torricelli’s theorem
because it was found long before Bernoulli’s work). Assuming that at the
free surface we have v D 0 and z D 0, it follows that H D p0 =, and
applying (23.66) we derive the relation

p0 v2 p0
H D D  gh C ;
 2 
p
so that v D 2gh.
2. In a horizontal pipe of variable cross section, the pressure of an incompressible
fluid in steady motion decreases in a converging section.
First, the mass conservation (23.68) requires that

v1 1 D v2 2 ; (23.69)

so that the fluid velocity increases in a converging section and decreases in the
diverging section.
23.7 Fundamental Theorems of Fluid Dynamics 443

Furthermore, since U D z D const along the stream tube, (23.66)

implies that
v21 p1 v2 p2
C D 2C ;
2  2 
and this proves that the pressure decreases in a converging section. This result
is applied in Venturi’s tube, where a converging section acts as a nozzle by
increasing the fluid velocity and decreasing its pressure.
Theorem 23.7 (Helmoltz’s first theorem). The flux of a vortex vector across any
section of a vortex tube is constant.
Proof. Let 1 and 2 be two sections of a vortex tube T , and consider the closed
surface † defined by 1 , 2 , and the lateral surface of T . Applying Gauss’s theorem
we have
Z Z Z
1
!  N d D r  ! dV D r  .r  v/ dV D 0;
† V 2 V

where N is the unit outward vector normal to †. The definition of a vortex tube
implies that ! is tangent to † at any point, so that the theorem is proved since
Z Z Z
!  N d D !  N1 d D !  N2 d D 0;
† 1 2

where N1 is the unit vector normal to 1 , pointing toward the interior of the tube,
and N2 is the outward unit vector normal to 2 . t
u
From this theorem it also follows that the particle vorticity increases if the vortex
curves are converging.
Theorem 23.8 (Helmoltz’s second theorem). Vortex lines are material lines.
Proof. At the instant t0 D 0, the vector ! is supposed to be tangent to the surface
0 . Denote by .t/ the material surface defined by the particles lying upon 0 at the
instant t0 . We must prove that .t/ is a vortex surface at any arbitrary instant. First,
we verify that the circulation
along any closed line 0 on 0 vanishes. In fact, if
A is the portion of 0 contained in 0 , it holds that
Z Z Z

D v  ds D r  v  N d D 2 !  N d D 0
0 A A

since ! is tangent to A. According to Thomson’s theorem, the circulation is

preserved along any material curve, so that, if .t/ is the image of 0 , it follows
that Z Z Z
v  ds D r  v  Nd D 2 !  Nd D 0:
.t / A.t / A.t /
Since A.t/ is arbitrary, !  N D 0, and the theorem is proved. t
u
444 23 Introduction to Fluid Mechanics

The theorem can also be stated by saying that vortex lines are constituted by the
same fluid particles and are transported during motion. Examples include smoke
rings, whirlwinds, and so on.

23.8 Boundary Value Problems for a Perfect Fluid

The motion of a perfect compressible fluid S subjected to body forces b is governed

by the momentum equation [see (23.40)]
1
vP D  rp./ C b (23.70)

and the mass conservation
P C r  v D 0: (23.71)
Equations (23.70) and (23.71) are a first-order system for the unknowns v.x; t/ and
.x; t/, and to find a unique solution, both initial and boundary conditions must be
specified.
If we consider the motion in a fixed and compact region C of a space (e.g., a
liquid in a container with rigid walls), then the initial conditions are

v.x; 0/ D v0 .x/; .x; 0/ D 0 .x/ 8 x 2 C; (23.72)

and the boundary condition is

vND0 8 x 2 @C; t > 0: (23.73)

This boundary condition states that the fluid can perform any tangential motion
on a fixed surface whose unit normal is N.
The problem is then to find in C  Œ0; t the fields v.x; t/ and .x; t/ that satisfy
the balance equations (23.70) and (23.71), the initial conditions (23.72), and the
boundary condition (23.73).
If the fluid is incompressible . D const/, then Eq. (23.70) becomes
1
vP D  rp C b; (23.74)

while the mass conservation (23.71) leads us to the condition

r  v D 0: (23.75)

The unknowns of system (23.74) and (23.75) are given by the fields v.x; t/ and
p.x; t/, and the appropriate initial and boundary conditions are

v.x; 0/ D v0 .x/ 8 x 2 C;
vN D 0 8 x 2 @C; t > 0: (23.76)
23.9 Two-Dimensional Steady Flow of a Perfect Fluid 445

A more complex problem arises when a part of the boundary is represented by a

moving or free surface f .x; t/ D 0. In this case, finding the function f is a part of
the boundary value problem. The moving boundary @C 0 , represented by f .x; t/ D
0, is a material surface since a material particle located on it must remain on this
surface during the motion. This means that its velocity cN along the unit normal
N to the free surface is equal to v  N ; that is, f .x; t/ must satisfy the condition
[see (4.30)]
@
f .x; t/ C v.x; t/  rf .x; t/ D 0:
@t
In addition, on the free surface it is possible to prescribe the value of the pressure,
so that the dynamic boundary conditions are
@
f .x; t/ C v.x; t/  rf .x; t/ D 0;
@t
p D pe 8 x 2 @C 0 ; t > 0; (23.77)

where pe is the prescribed external pressure.

On the fixed boundary part, the previous impenetrability condition (23.74)
applies, and if boundary C extends to infinity, then conditions related to the
asymptotic behavior of the solution at infinity must be added.

23.9 Two-Dimensional Steady Flow of a Perfect Fluid

The following two conditions define an irrotational steady motion of an incompress-

ible fluid S :
r  v D 0; r  v D 0; (23.78)
where v D v.x/. The first condition allows us to deduce the existence of a velocity
or kinetic potential '.x/ such that

v D r'; (23.79)

where ' is a single- or a multiple-valued function, depending on whether or not the

motion region C is connected.2
In addition, taking into account (23.78)2, it holds that

' D r  r' D 0: (23.80)

R
2
If C is not a simply connected region, then the condition r v D 0 does not imply that vds D
0 on any closed curve since this curve could not be the boundary of a surface contained in C .
In this case, Stokes’s theorem cannot be applied.
446 23 Introduction to Fluid Mechanics

Equation (23.80) is known as Laplace’s equation, and its solution is a harmonic

function.
Finally, in dealing with a two-dimensional (2D) flow, the velocity vector v at any
point is parallel to a plane  and is independent of the coordinate normal to this
plane. In this case, if a system Oxyz is introduced, where the x- and y-axes are
parallel to  and the z-axis is normal to this plane, then we have

v D u.x; y/i C v.x; y/j;

where u and v are the components of v on x and y, and i and j are the unit vectors
of these axes.
If now C is a simply connected region of the plane Oxy, then conditions (23.78)
become
@u @v @u @v
 C D 0; C D 0: (23.81)
@y @x @x @y
These conditions allow us to state that the two differential forms !1 D udx C vdy
and !2 D vdx C udy are integrable, i.e., there is a function ', called the velocity
potential or the kinetic potential, and a function , called the stream potential or
the Stokes potential, such that

d' D udx C vdy; d D vdx C udy: (23.82)

From (23.78)2 it follows that the curves ' D const are at any point normal to the
velocity field. Furthermore, since r'  r D 0, the curves D const are flow
lines.
It is relevant to observe that (23.82) suggest that the functions ' and satisfy
the Cauchy–Riemann conditions
@' @ @' @
D ; D ; (23.83)
@x @y @y @x
so that the complex function

F .z/ D '.x; y/ C i .x; y/ (23.84)

is holomorphic and represents a complex potential. Then the complex potential can
be defined as a holomorphic function whose real and imaginary parts are the velocity
potential ' and the stream potential , respectively. The two functions ' and are
harmonic and the derivative of F .z/,

@' @
V  F 0 .z/ D Ci D u  i v D jV j ei ; (23.85)
@x @x

represents the complex velocity, with jV j being the modulus of the velocity vector
and  the angle that this vector makes with the x-axis.
23.9 Two-Dimensional Steady Flow of a Perfect Fluid 447

level lines of
level lines of

Fig. 23.3 Uniform motion

Within the context of considerations developed in the following discussion, it

is relevant to remember that the line integral of a holomorphic function vanishes
around any arbitrary closed path in a simply connected region since the Cauchy–
Riemann equations are necessary and sufficient conditions for the integral to be
independent of the path (and therefore it vanishes for a closed path).
The preceding remarks lead to the conclusion that a 2D irrotational flow of
an incompressible fluid is completely defined if a harmonic function '.x; y/ or a
complex potential F .z/ is prescribed, as is shown in the following examples.
Example 23.1 (Uniform motion). Given the complex potential

F .z/ D U0 .x C iy/ D U0 z; (23.86)

it follows that V D U0 , and the 2D motion

v D U0 i (23.87)

is defined, where i is the unit vector of the Ox-axis. The kinetic and Stokes
potentials are ' D U0 x and D U0 y, and the curves ' D const and D const
are parallel to Oy and Ox (Fig. 23.3), respectively. This example shows that the
complex potential (23.86) can be introduced to describe a 2D uniform flow, parallel
to the wall y D 0.
Example 23.2 (Vortex potential). Let a 2D flow be defined by the complex potential

F .z/ D i ln z D i ln rei D  i ln r; (23.88)

2 2 2 2
448 23 Introduction to Fluid Mechanics

level lines of
level lines of

Fig. 23.4 Vortex

where r and  are polar coordinates. It follows that

y
'D D arctan ;
2 2 x

D i ln r D  ln.x 2 C y 2 /;
2 2
and the curves ' D const are straight lines through the origin, while the curves
D const are circles whose center is the origin (Fig. 23.4).
Accordingly, the velocity components become
@'
y
sin 
uD D D ;
@x 2 x 2 C y 2 2 r
@'
x
cos 
vD D D :
@y 2 x 2 C y 2 2 r
It is relevant to observe that the circulation around a path bordering the origin is
given by
Z
v  ds D
;

so that it does not vanish if

¤ 0. This does not contradict the condition r 
v D 0 since a plane without an origin is no longer a simply connected region. In
Sect. 9.7, further arguments will be addressed to explain why the circulation around
an obstacle does not vanish in a 2D flow.
23.9 Two-Dimensional Steady Flow of a Perfect Fluid 449

y
Q >0

level curves of
level curves of

Fig. 23.5 Source

The complex potential (23.88) can then be used with advantage to describe the
uniform 2D flow of particles rotating around an axis through the origin and normal
to the plane Oxy.
Example 23.3 (Sources and sinks). For pure radial flow in the horizontal plane, the
complex potential is taken to be
Q Q Q
F .z/ D ln z D ln rei D .ln r C i/; Q > 0;
2 2 2
so that
Q Q Q Q y
'D ln r D log.x 2 C y 2 /; D D arctan :
2 2 2 2 x
The curves ' D const are circles, whereas the curves D const are straight lines
through the origin. The velocity field is given by
Q
vD er ;
2 r
where er is the unit radial vector (Fig. 23.5). Given an arbitrary closed path around
the origin, the radial flow pattern associated with the preceding field is said to be
either a source, if
Z
v  n d D Q > 0; (23.89)

or a sink, if Q < 0. The quantity Q=2 is called the strength of a source or sink.
450 23 Introduction to Fluid Mechanics

Q >0

Q <0 v

level curves of
level curves of

Fig. 23.6 Doublet

Example 23.4 (Doublet). A doublet is the singularity obtained by taking to zero

the distance between a source and a sink having the same strength. More precisely,
consider a source and a sink of equal strength, the source placed at a point A, with
z1 D aei ˛ , and the sink placed at B, with z2 D aei ˛ . The complex potential for
the combined flow is then
Q Q
F .z/ D ln.z C aei ˛ /  ln.z  aei ˛ /  f .z; a/: (23.90)
2 2
The reader is invited to find the kinetic and Stokes potential (Fig. 23.6).
If points A and B are very near to each other, i.e., a ' 0, it follows that
Q 1
F .z/ D f .z; 0/ C f 0 .z; 0/a D 2aei ˛ :
2 z

In particular, if a ! 0, so that Qa= ! m, then we obtain the potential of a

doublet (source–sink):
m
F .z/ D ei ˛ : (23.91)
z
Furthermore, if ˛ D 0, e.g., if the source and the sink are on the x-axis, then
m my mx
F .z/ D D 2 i 2 'Ci ; (23.92)
z x C y2 x C y2
and the flow lines D const are circles through the origin.
23.9 Two-Dimensional Steady Flow of a Perfect Fluid 451

Fig. 23.7 Flow about a cylinder

Example 23.5. Consider a 2D flow with complex potential

U0 a 2
F .z/ D U0 z C ; (23.93)
z
i.e., the flow is obtained as the superposition of a uniform flow parallel to the x-axis
(in the positive direction) and the flow due to a dipole (source–sink) system of
strength m D U0 a2 . Assuming z D x C iy, then from (23.93) it follows that
   
a2 a2
' D U0 x 1 C 2 ; D U 0 y 1  :
x C y2 x2 C y 2
The flow lines D const are represented in Fig. 23.7 (for U0 D a D 1/.
Note that the condition x 2 Cy 2 D a2 corresponds to the line D 0. Furthermore,
if the region internal to this circle is substituted by the cross section of a cylinder,
then (23.93) can be assumed to be the complex potential of the velocity field around
a cylinder. The components of the velocity field are given by
 2 2  
@' 2x  y a2
uD D U0 1  a D U0 1  2 cos 2 ;
@x r4 r
2
@' a
vD D U0 2 sin 2:
@y r
It should be noted that, although singular points associated with a doublet do not
actually occur in real fluids, they are interesting because the flow pattern associated
with a doublet is a useful approximation far from singular points, and it can be
combined to advantage with other nonsingular complex potentials.
Example 23.6. All the previous examples, with the exception of Example 23.2,
refer to 2D irrotational flows whose circulation along an arbitrary closed path is
vanishing. We now consider the complex potential

U0 a 2
z
F .z/ D U0 z C Ci ln : (23.94)
z 2 a
452 23 Introduction to Fluid Mechanics

y
1.5
1
0.5
0
x
−0.5
−1
−1.5
−2 −1 0 1 2

Fig. 23.8 Flow about a cylinder with vorticity

This represents a flow around a cylinder of radius a obtained by superposing a

uniform flow, a flow generated by a dipole, and the flow due to a vortex. The dipole
and the vortex are supposed to be located on the center of the section. In this case,
according to (23.89), the vorticity is
, and it can be supposed that the vortex is
produced by a rotation of the cylinder around its axis. Flow lines are given by the
complex velocity
dF a2

V D D U0 .1  2 / C i :
dz z 2z
To find stagnation points (V D 0), we must solve the equation dF=dz D 0. Its roots
are given by
 q 
1 
zvD0 D i
˙ 16U02 2 a2 
2 : (23.95)
4U0 
Three cases can be distinguished:
2 < 16 2 a2 U 2 ,
2 D 16 2 a2 U 2 , and
2 >
16 2 a2 U 2 , whose corresponding patterns are shown in Figs. 23.8–23.10 (assuming
U0 D a D 1/.
At this stage the reader should be aware that these examples, although of practical
relevance, are based on equations of steady-state hydrodynamics, which in some
cases give rise to inaccurate or paradoxical results. As an example, in the next
section we will analyze D’Alembert’s paradox, according to which the drag of a
fluid on an obstacle is zero. To remove this paradox, it will be necessary to improve
the model of a fluid taking into account the effects of viscosity.
23.10 D’Alembert’s Paradox and the Kutta–Joukowky Theorem 453

y
1.5
1
0.5
0
x
−0.5
−1
−1.5
−2 −1 0 1 2

Fig. 23.9 Flow about a cylinder with vorticity

y
1.5
1
0.5
0
x
−0.5
−1
−1.5
−2 −1 0 1 2

Fig. 23.10 Flow about a cylinder with vorticity

23.10 D’Alembert’s Paradox and the Kutta–Joukowky

Theorem

In this section, we prove that the model of a perfect liquid leads to the paradox
that a body experiences no resistance while moving through a liquid (D’Alembert’s
paradox). This paradox shows that, although the model of a perfect liquid leads
to results that are in agreement with many experimental results, in some cases it
fails. This is due to the fact that this model does not take into account the viscosity
of liquids. In other words, there are some cases where it is possible to neglect the
effects of viscosity on motion and the model of a perfect liquid can be used. In other
cases, it is fundamental to take into account these effects to obtain again agreement
between theory and experimental results.
454 23 Introduction to Fluid Mechanics

N
Σ

V
C

Fig. 23.11 Flow about a cylinder with vorticity

Let S be a solid of volume C placed in an irrotational and steady flow of a liquid

in the absence of body forces (Fig. 23.11). Then (23.74) becomes
@vi @p
vj D :
@vj @xi
Further, taking into account the incompressibility condition (23.75), the preceding
equation gives
@
.p C vi vj / D 0: (23.96)
@xj
Let V be an arbitrary fixed volume V containing the body S . Then, integrat-
ing (23.96) on the region V  C , applying Gauss’s theorem, and recalling that
v  N D 0 on the whole boundary of C , we obtain the integral condition
Z Z
pN d C .pI C v ˝ v/  N d D 0: (23.97)
@C @V

But the first integral is the opposite of the force F acting on S , so that the preceding
condition gives
Z
FD .pI C v ˝ v/  N d: (23.98)
@V
On the other hand, Bernoulli’s theorem (23.66), in the absence of body forces,
states that
1
p D p0 C .V 2  v2 /; (23.99)
2
where p0 and V denote, respectively, the pressure and the uniform velocity in
a liquid at infinity. We do not prove that in our hypothesis, liquid flow has the
following asymptotic behavior:

v D V C O.r 3 /; (23.100)
23.10 D’Alembert’s Paradox and the Kutta–Joukowky Theorem 455

where r is the distance of a point in the flow relative to an arbitrary origin.

Introducing (23.100) into (23.98), taking into account the asymptotic behavior
(23.99), and assuming that V is a sphere, we obtain that
Z
F D .p0 I  V ˝ V/ N d C O.r 1 /: (23.101)
@V

Finally, applying Gauss’s theorem, the preceding condition gives

F D 0; (23.102)

and D’Alembert’s paradox is proved.

We conclude this section stating without proof (see [43]) that in a planar flow of
a liquid, (23.100) and (23.99) become

v D ViC .yi C xj/ C O.r 2 /; (23.103)

2 r 2
1

p D p0 C  yV C O.r 2 /; (23.104)
2 2 r 2
where i and j are unit vectors along the Ox- and Oy-axes in the plane of the flow,
the speed V at infinity of the flow is parallel to i, and
I

D v  ds

is the circulation along any curve surrounding body C . Then it is possible to prove
that, instead of (23.103), we obtain the Blausius formula

F D 
V j: (23.105)

This formula shows that although a steady flow of an inviscid fluid predicts no
drag on an obstacle in the direction of relative velocity in an unperturbed region, it
can predict a force normal to this direction. This is a result obtained independently
by W.M. Kutta in 1902 and N.E. Joukowski (sometimes referred as Zoukowskii) in
1906, known as the Kutta–Joukowski theorem. Such a force is called lift, and it is
important for understanding why an airplane can fly.
Before closing this section, we observe that our inability to predict the drag of an
inviscid fluid in the direction of relative velocity does not mean we should abandon
the perfect fluid model. Viscosity plays an important role around an obstacle, but,
far from the obstacle, the motion can still be conveniently described according to
the assumption of an inviscid fluid.
456 23 Introduction to Fluid Mechanics

23.11 Waves in Perfect Gases

This section is devoted to wave propagation in perfect fluids. Waves are considered
to be small perturbations of an undisturbed state, corresponding to a homogeneous
fluid at rest. This assumption allows us to linearize the motion equations and to
apply elementary methods.3
Consider a compressible perfect fluid at rest, with uniform mass density 0 , in the
absence of body forces. Assume that motion is produced by a small perturbation, so
that it is characterized by a velocity v and a density  that are only slightly different
from 0 and 0 , respectively. More precisely, such an assumption states that v and
 D   0 are first-order quantities together with their first-order derivatives.
In this case, the motion equations

ŒclvP D rp./ D p 0 ./r;

P C r  v D 0 (23.106)

can be linearized. In fact, neglecting the first-order terms of jvj and , we obtain

p 0 ./ D p 0 .0 / C p 00 .0 / C    ;

@ @
P D P D C vr D C ;
@t @t
@v @v
vP D C v  rv D C ;
@t @t
so that (23.106) become
@v
0 D p 0 .0 /r;
@t
@
C 0 r  v D 0: (23.107)
@t
Consider now a sinusoidal 2D wave propagating in the direction of the vector n
with velocity U and wavelength :

2 2
v D a sin .n  x  U t/;  D b sin .n  x  U t/:
 

3
For an analysis of wave propagation in nonlinear media, see [44].
23.11 Waves in Perfect Gases 457

We want a solution to the previous system with this form. To find this solution, we
note that

@v 2 2 @ 2 2
D  U a cos .n  x  U t/; D  U b cos .n  x  U t/;
@t   @t  
2 2 2 2
rv D n  a cos .n  x  U t/; r D nb cos .n  x  U t/;
   
so that system (23.107) becomes

0 U a D p 0 .0 /bn;
U b C 0 a  n D 0;

i.e.,
.p 0 .0 /n ˝ n  U 2 /Ia D 0;
.p 0 .0 /  U 2 b D 0:
Assuming a reference frame whose axis Ox is oriented along n, we obtain

U D 0; a ? n;
p
U D ˙ p 0 .0 /; a D an; (23.108)

where the eigenvalue 0 has a multiplicity of 2. Thus we prove the existence of

dilational waves propagating at a speed
p
U D p 0 .0 /:

The velocity U is called the sound velocity, and the ratio m D v=U is known
as the Mach number. In particular, the motion is called subsonic or supersonic,
depending on whether m < 1 or m > 1.
Chapter 24
An Introduction to Special Relativity

24.1 Galilean Relativity

A relativistic theory has as its aim to show that it is possible to give the general
laws of physics a form independent of any observer, that is, of the adopted frame
of reference. Up to now, this objective has not been completely realized, even from
the general theory of relativity.1 In this chapter, a relativistic theory is intended as a
theory for which we have defined:
1. The class R of the frames of reference in which the physical laws have the same
form,
2. The transformations of space and time coordinates in going from a frame of
reference belonging to R to another one in the same class,
3. The physical laws relative to any frame of reference of R,
4. The transformation formulae of the physical quantities appearing in physical
laws upon changing the frame of reference in the class R.
For instance, in Chap. 13, we showed that Galilean relativity assumes that R is
the class of inertial frames of reference, the space-time coordinate transformations
are the Galilean ones, i.e., formulae (12.5) with a constant matrix Qij , the physical
laws are Newtonian laws, and the mass and forces are objective.
Following discovery of the Maxwell equations governing all optical and electro-
magnetic phenomena, it was quite natural to try to verify if they could be accepted
within the framework of Galilean relativity. We now verify that the Maxwell
equations do not satisfy the principle of Galilean relativity.2

1
See, for instance, [37].
2
For the topics of this chapter see [15, 22, 27, 36, 38, 39, 41, 42, 47, 48, 56].

A. Romano, Classical Mechanics with Mathematica R

, Modeling and Simulation 459
in Science, Engineering and Technology, DOI 10.1007/978-0-8176-8352-8 24,
© Springer Science+Business Media New York 2012
460 24 An Introduction to Special Relativity

Let S be a charged continuous medium moving relative to an inertial observer I .

If we denote by  the charge density and by J the current density, then the charge
conservation is expressed by the following continuity equation:
@
C r  J D 0: (24.1)
@t
Further, in a vacuum, the Maxwell equations have the form
@E
r  H D 0 ; (24.2)
@t
@H
r  E D 0 ; (24.3)
@t
where H is the magnetic field, E the electric field, 0 the dielectric constant, and 0
the magnetic permeability of a vacuum. Further, applying the divergence operator
r to both sides of (24.2) and (24.3), we obtain

r  E D 0; r  H D 0; (24.4)

provided that we suppose that the electromagnetic fields vanish before a given
instant in the past.
It is a very simple exercise to verify that, under a Galilean transformation (12.5)
(with a constant matrix Qij ), leading from inertial frame I to another inertial frame
I 0 , we have that

r D r0; (24.5)
@ @
D 0  u  r 0; (24.6)
@t @t
where u is the uniform velocity of I 0 relative to I . Then, in view of (24.5) and (24.6),
under a Galilean transformation I ! I 0 , (24.1) becomes
@
C r 0 .J  u/ D 0:
@t 0
0
This equation again assumes the form (24.1) in the inertial frame I if and only if 
and J are transformed according to the formulae

0 D ; J0 D J  u: (24.7)

Consequently, if these transformation formulae were experimentally confirmed,

we could conclude that the continuity equation satisfied the Galilean principle of
relativity.
On the other hand, if the foregoing procedure is applied to the Maxwell
equations (24.2) and (24.3), we obtain
24.1 Galilean Relativity 461

@E
r 0  H D 0  0 u  rE;
@t 0
@H
r 0  E D 0 0 C 0 u  rH:
@t
Taking into account the vector identity

r  .a  b/ D b  ra  a  rb C ar  b  br  a;

where a and b are arbitrary regular fields, and recalling (24.7) and that the relative
velocity u is constant, the preceding equations become
@E
r 0  .H  0 u  E/ D 0 ;
@t 0
@H
r 0  .E C 0 u  H/ D 0 0 :
@t
It is evident that there is no transformation formula of the fields E and H for
which these equations assume the same form in the inertial frame I 0 . In conclusion,
Maxwell’s equations do not agree with Galilean relativity.
We now derive a further disconcerting consequence of Maxwell’s equations.
Applying the operator r to both sides of (24.2) and (24.3), recalling the vector
identity
r  .r  a/ D r.r  a/  a;
where  is the Laplace operator, and taking into account (24.4), we obtain the
following equations:

@2 H
H D 0 0 ;
@t 2
@2 E
E D 0 0 ;
@t 2
showing that the electric and magnetic fields propagate in a vacuum with a speed

1
cD p ;
0 0

independently of the motion of electromagnetic field’s sources. In particular, since

we have proved that Maxwell’s equations are not compatible with Galilean relativ-
ity, we must conclude that they are valid at most in one inertial frame. Finally, we
note that all the preceding results apply to light waves since Maxwell proved that
optical waves are special electromagnetic waves.
462 24 An Introduction to Special Relativity

24.2 Optical Isotropy Principle

The wave character of light propagation was established during the eighteenth
century, when scientists became convinced that all physical phenomena could
be described by mechanical models. Consequently, it appeared natural to the
researchers of that time to suppose that empty space was filled by an isotropic
and transparent medium, the ether, which supported light waves. This hypothesis
seemed to be confirmed by the circumstance that the forces acting on charges and
currents could be evaluated by supposing that electromagnetic fields generated a
deformation state of the ether, which was described by the Maxwell stress tensor.
However, this description was unacceptable for the following reasons:
• The high value of light speed required a very high density of ether.
• Such a high density implied the existence of both longitudinal and transverse
waves, whereas Maxwell’s equations showed the transverse character of electro-
magnetic waves.
Also damaging to the hypothesis of the ether as a material medium was the fact
that Maxwell’s equations were not covariant, i.e., invariant in form, under Galilean
transformations, as we saw in the previous section. Consequently, electromagnetic
phenomena did not confirm a relativity principle, so they were assumed to hold
only in one frame of reference, the optically isotropic frame. It is well known
that any attempt to localize this frame of reference failed (e.g., Michelson, Morley,
Kennedy, Fitzgerald), so that researchers were faced with a profound physical
inconsistency.
Einstein gave a brilliant and revolutionary solution to the preceding problem
by accepting the existence of an optically isotropic system and exploiting the
consequences of this assumption. More precisely, he postulated that:
There exists at least an optically isotropic frame of reference.
Owing to the preceding postulate, it is possible to find a frame of reference I in
which light propagates in empty space with constant speed along straight lines in
any direction. In particular, in this frame I it is possible to define a global time t
by choosing an arbitrary value c of the speed of light in empty space. In fact, the
time measured by a clock located at a point O 2 I will be accepted if a light signal
that is sent from O at t D t0 toward an arbitrary point P , where it is again reflected
by a mirror toward O, reaches O at the instant t D t0 C 2OP =c, for any t0 . Let us
suppose that a set of identical clocks is distributed among different points in space.
A clock at any point P can be synchronized with a clock located at the fixed point
O by sending a light signal at the instant t0 from O and requiring that it arrive at P
at the instant t D t0 C OP =c. We do not prove that the time variable defined by the
foregoing procedure is independent of the initial point O and the initial instant t0 .
In view of what was stated in Sect. 24.1, if we accept the Galilean transformations
in passing from a frame of reference I to another frame of reference I 0 , which
moves with respect to I with constant velocity u, then there exists at most one
24.2 Optical Isotropy Principle 463

isotropic optical frame. Conversely, if we drop the assumptions on which the

Galilean transformations are based, then it is possible to prove (see [22], p. 403)
that there are infinite optically isotropic frames of reference. More precisely,
if .x 1 ; x 2 ; x 3 ; t/ and .x 01 ; x 02 ; x 03 ; t 0 / are the spatial and temporal coordinates
associated with the same event by two observers I and I 0 , then the finite relations
between these coordinates are expressed by the Lorentz transformations
 
j uj
x D xO C Qj ıh C .  1/ 2 uh x h  Qji  uj t;
0i 0i i
(24.8)
u
uh x h
t 0 D t0   C  t; (24.9)
c2
where xOi , t0 , and ui are constant, Qji are the coefficients of a constant orthogonal
matrix, u D juj, and
1 1
Dr  u 2  p1  ˇ 2 > 1: (24.10)
1
c

Remark 24.1. It is not an easy task to prove (24.8) and (24.9) starting from the
preceding postulate. However, if we add the hypothesis that the required transfor-
mation is linear, then it becomes very simple to derive the Lorentz transformations,
as is proved in most books on special relativity.3 This assumption is equivalent to
requiring that optically isotropic frames also be inertial frames.
Remark 24.2. Let us consider any event that has constant space-time coordinates in
the frame I 0 . Then, differentiating (10.1), we obtain
  j
ui dx
ıji C .  1/ 2 uj D  ui : (24.11)
u dt
It is an easy exercise to verify that the preceding system admits the following
solution:
dx i
D ui ; (24.12)
dt
so that we can say that any point of I 0 moves with respect to I with constant velocity
.ui /. Therefore, the relative motion of I 0 with respect to I is a uniform translatory
motion with velocity .ui /. Similarly, we can prove that any point of I moves with
respect to I 0 with constant velocity u. In particular, xO0i denotes the coordinates of
the origin of I with respect to I 0 , when t D 0.
Remark 24.3. In the limit c ! 1, the Lorentz transformations reduce to the
Galilean ones.

3
A simple proof is shown at the end of this section.
464 24 An Introduction to Special Relativity

We now introduce some particular Lorentz transformations. We define as a

Lorentz transformation without rotation any Lorentz transformation obtained
from (10.1) by supposing that the 3  3 orthogonal matrix .Qji / reduces to the
identity matrix
 
ui
x 0i D xO0i C ıhi C .  1/ 2 uh x h   ui t; (24.13)
u
uh x h
t 0 D t0   C  t: (24.14)
c2
To understand the meaning of the preceding definition, we denote by .Sji / the
constant 3  3 orthogonal matrix defining the rotation x i D Sji x j of the spatial
axes of I , in which the uniform velocity u of I 0 with respect to I becomes parallel
to the axis Ox 1 :
0 11
u   
@ 0 A D Sji uj : (24.15)
0
If we note that u2 D uh uh and uh x h are invariant with respect to a change in the
spatial axes, then, applying the rotation .Sji / to both sides of (24.13), that is, by
performing the same rotation on the spatial axes of I 0 and I , we obtain
 
ui
x 0i D x 0iO C ıhi C .  1/ 2 uh x h   ui t; (24.16)
u
uh x h
t 0 D t0   C  t: (24.17)
c2
These formulae, when we recall (24.10), presuppose that xO0i D t0 D 0, and,
for the sake of simplicity, we omit the overline, leading us to the special Lorentz
transformations
 
x 01 D  x 1  ut ; (24.18)
x 02 D x 2 ; (24.19)
03 3
x Dx ; (24.20)
 u 
t 0 D  t  2 x1 ; (24.21)
c
provided that the two inertial frames I and I 0 have the same origin when t D t 0 D 0,
the axes O 0 x 01 are laid upon one another during the relative motion with uniform
velocity u parallel to Ox 1 , O 0 x 02 is parallel to Ox 1 , and O 0 x 02 is parallel to Ox 2 .
These transformations include all the significant relativistic aspects since the most
general Lorentz transformations can be obtained by arbitrary rotations of the spatial
axes of the frames I and I 0 and arbitrary translations of the origins O and O 0 .
24.2 Optical Isotropy Principle 465

We conclude this section with an elementary proof of (24.18)–(24.21) supposing

that the transformation between the two inertial frames I and I 0 is linear. First of
all, we note that, since the motion of I 0 relative to I is along the Ox 1 -axis, the
coordinates x 2 and x 3 are transformed according to (24.19) and (24.20). Owing to
the supposed linearity of the transformations, for the remaining variables x 1 and t
we have that
x 01 D Ax 1 C Bt; (24.22)
0 1
t D C x C Dt; (24.23)
where the constants A, B, C , and D must be determined. First, the origin O 0 of I 0
moves relative to I with uniform velocity u parallel to Ox 1 . Consequently, when
x 01 D 0, (24.22) gives
x1 B
D  D u: (24.24)
t A
On the other hand, the origin of I moves relative to I 0 with velocity u. Therefore,
from (24.22), in which x 1 D 0, we obtain that

x 01 B
0
D D u: (24.25)
t D
Comparing (24.24) and (24.25), we have that

B D uA; D D A;

and (24.22) and (24.23) become

x 01 D A.x 1 C ut/; (24.26)

t 0 D C x 1 C At: (24.27)

To find the remaining constants A and C , we consider a planar light wave

propagating along the Ox 1 -axis with speed c. According to the optical isotropy
principle, the propagation of this wave is described by the equations

.x 1 /2  c 2 t 2 D 0; .x 01 /2  c 2 .t 0 /2 D 0 (24.28)

in the frames I and I 0 , respectively. Inserting (24.26) and (24.27) into (24.28)2, we
obtain

.A2  c 2 C 2 /.x 1 /2  2A.Au C c 2 C /x 1 t C A2 .u2  c 2 /t 2 D 0: (24.29)

This equation reduces to (24.28)1 if

A2  c 2 C 2 D 1; Au C c 2 C D 0; A2 .u2  c 2 / D c 2 ;
466 24 An Introduction to Special Relativity

i.e., if
1 u
AD r  ; C D ;
u2 c2
1
c2
and (24.19) and (24.20) are proved.

24.3 Simple Consequences of Lorentz’s Transformations

In this section, we show that the Lorentz transformations lead to the conclusion that
the concepts of space and time intervals are relative to the observer. Let xA01 and
xB01 > xA01 be the abscissas of two points A0 and B 0 belonging to the axis O 0 x 01 of
the inertial frame I 0 , and denote by x 01 D xB01  xA01 the distance between these
points, measured by rulers at rest relative to I 0 . The length x 01 will be called the
rest or proper length of the segment A0 B 0 . Now we define the length x 1 of the
segment A0 B 0 with respect to an inertial frame I , relative to which this segment
moves with uniform velocity u along the axis Ox 1  O 0 x 01 . The quantity x 1 is
identified by the distance between the two points A and B of the Ox 1 -axis that are
occupied by A0 and B 0 , respectively, at the same instant t in the inertial frame I .
For t constant, (24.18) gives
x 01 D x 1 > x 1 ; (24.30)
0 0
so that the length of A B relative to I is 1= times smaller than the length relative
to I . Since the dimensions of the lengths in the direction perpendicular to the
velocity do not change, a volume V is connected with the proper volume V 0 by
the relation
V 0 D V: (24.31)
Consider now a rest or proper time interval t 0 D t20  t10 evaluated by a clock C
at rest at the point xA01 of the inertial frame I 0 . To obtain the corresponding instants t1
and t2 evaluated in the inertial frame I , it is sufficient to apply the inverse of (24.21)
and remember that I moves relative to I 0 with uniform velocity u. In this way we
obtain
 
u2
t1 D  t10  2 xA01 ;
c
 
u2
t2 D  t20  2 xA01 :
c
Consequently, we obtain
1
t 0 D t < t; (24.32)

where t D t2  t1 . In other words, a moving clock goes slower than a clock at rest.
24.3 Simple Consequences of Lorentz’s Transformations 467

Now we present two simple but interesting applications of (24.31) and (24.32).
• Let P be an unstable particle produced in the atmosphere at a distance l from
the terrestrial surface. Denote by I 0 and I the rest inertial frame of P and a
terrestrial frame, respectively. Suppose that the mean life t 0 (evaluated in I 0 )
and the speed v of P (evaluated in I ) are such that l > vt 0 . If we assume
classical kinematics, then it is impossible for P to reach the surface of the Earth.
Conversely, resorting to (24.31) and (24.32), we can justify the arrival of P at the
terrestrial surface. In fact, for the observer at rest relative to P , the particle must
cover a shorter distance l 0 D l= < l in the time t 0 . In contrast, for an observer
at rest on the Earth’s surface, the particle P must cover the distance l but lives
longer, i.e., t= .
• Suppose that at the points xA1 < xB1 of the Ox 1 -axis of the inertial frame I
two events happen at the instants tA and tB , respectively. For an observer in the
inertial frame I 0 moving relative to I with velocity u, the same events happen at
the instants tA0 and tB0 , given by
 u   u 
tA0 D  tA  2 xA ; tB0 D  tB  2 xB ;
c c
so that
 xB  xA 
tB0  tA0 D  tB  tA  u : (24.33)
c2
This condition shows that two events that are simultaneous for I are not
simultaneous for I 0 if xB ¤ xA . The following question arises: supposing that
tB > tA , is it possible to find an observer I 0 for which the order of the events is
inverted? This circumstance is confirmed if and only if the following condition
is satisfied:
u
tB  tA < 2 .xB  xA /: (24.34)
c
Because u=c  1, then also u=c 2  1=c, and the preceding condition gives
xB  xA
tB  tA < : (24.35)
c
This necessary condition for an inversion of the time order of two events allows
us to state that if in an inertial frame I 0 we have an inversion of the time
order, then the time interval between the two events is necessarily less than
the time taken by light to cover the distance xB  xA . Conversely, from (24.35)
follows (24.34), at least for a certain value of u. We remark that (24.35) cannot
be satisfied for two events that are related to each other by a cause-and-effect
relationship. In other words, in an inertial frame, no physical perturbation
propagates with a speed greater than the light speed c.
468 24 An Introduction to Special Relativity

24.4 Relativistic Composition of Velocities and Accelerations

From (24.21) we obtain that

  
dt u  rP 1
D  1  : (24.36)
dt 0 c2
Owing to the preceding relation, by simple but tedious calculations, it is possible
to derive from (24.18)–(24.21) the transformation formulae for the velocity and the
acceleration under a Lorentz transformation. For instance, starting from the special
Lorentz transformations, we obtain the following special formulae of the velocities:

xP 1  u
xP 01 D ; (24.37)
uxP 1
1 2
c
1 xP 2
xP 02 D ; (24.38)
 uxP 1
1 2
c
1 xP 3
xP 03 D : (24.39)
 uxP 1
1 2
c
By a further derivation with respect to time, we can derive the following formulae
for acceleration in going from inertial frame I to inertial frame I 0 :
 3=2
u2
1 2
c
xR 01 D  1
3 xR ; (24.40)
xP 1 u
1 2
c
0 1
u2 xP 2 u
1 2 B 2
xR 02 D  c BxR C c 2 xR 1 CC; (24.41)
2 @
xP 1 u A
xP 1 u 1 2
1 2
c c
0 1
u2 xP 3 u
1 2 B 3
xR 03 D  c BxR C c 2 xR 1 CC: (24.42)
2 @ xP 1 u A
xP 1 u 1 2
1 2
c c

Remark 24.4. It is evident that the preceding formulae reduce to Galilean ones
when c ! 1.
Now we present some simple consequences of the preceding formulae.
24.4 Relativistic Composition of Velocities and Accelerations 469

x x’

c
a

I I’ x x’

x x’

O O’ u

Fig. 24.1 Star aberration

• It is a simple exercise to verify that a spherical light wave in I,

.x 1 /2 C .x 2 /2 C .x 3 /2  c 2 t 2 D 0;

is described in the frame I 0 by the equation

.x 01 /2 C .x 02 /2 C .x 03 /2  c 2 t 02 D 0;

in agreement with the principle of optical isotropy.

• Let a be a light ray emitted by a star, and let us suppose that it reaches the
terrestrial surface in a direction orthogonal to the translational velocity u of the
Earth (Fig. 24.1). I 0 is the rest inertial frame of the Earth, and I is the rest inertial
frame of the star. Then, in (24.37)–(24.39) we have that

xP 1 D xP 2 D 0; xP 3 D c:

Introducing into (24.37)–(24.39) the preceding components of speed relative to

the star frame I , we obtain the components of the ray speed in the terrestrial
frame
c
xP 01 D u; xP 02 D 0; xP 03 D  :

Consequently, the angle  between the velocity rP 0 in the frame I 0 and O 0 x 03 is
given by the relation
 
xP 01 u u u2 u
tan  D 03 D  ' 1C 2 ' :
xP c c 2c c
This result explains the aberration of fixed stars.
470 24 An Introduction to Special Relativity

• Let S be a homogeneous and isotropic optical medium, and let n be its refractive
index. Suppose that S moves uniformly with a speed u along the Ox 1 -axis of the
inertial frame I and denote by I 0 the rest frame of S , with O 0 x 01 parallel to Ox 1 .
Consider a light ray a propagating along O 0 x 01 with speed c=n. Then, applying
the inverse formulae of (24.37)–(24.39), the speed of a along Ox 1 is given by
c
Cu c  u 
1
xP D n
u ' n C u 1  cn ;
1C
cn
that is,
 
c 1
xP 1 D Cu 1 2 :
n n
This result represents a simple explanation of Fizeau’s experiment, which, before
special relativity, required very difficult theoretical interpretations that had not
been experimentally confirmed. One of these attempts at interpretation required
the partial drag of the ether by a moving medium (Sect. 24.2).

24.5 Principle of Relativity

In classical mechanics a relativity principle holds in the set R of inertial frames.

These frames are related to each other by the Galilean transformations. Einstein
extends this principle to any field of physics. More precisely, he supposes that
The fundamental equations of physics have the same form in the whole class
of the inertial frames. Analytically, the fundamental equations of physics must be
covariant under Lorentz transformations.
To make clear the deep meaning of this principle, let us consider a physical law
that in the inertial frame I is expressed by the differential relation
 
@A @B @A @B
F A; B; : : : ; i ; i ; : : : ; ; ; : : : D 0; (24.43)
@x @x @t @t
where A, B; : : : are physical fields depending on the spatial coordinates and time. If
we denote by A0 , B 0 ; : : : the corresponding fields evaluated by the inertial observer
I 0 , then this law satisfies the relativity principle if in the new frame I 0 it assumes
the form
 
@A0 @B 0 @A0 @B 0
F A0 ; B 0 ; : : : ; 0i ; 0i ; : : : ; 0 ; 0 ; : : : D 0; (24.44)
@x @x @t @t
where A0 , B 0 ; : : : are the quantities A, B; : : : evaluated by the inertial observer I 0 .
We underline that the principle of relativity does not state that the evolution of
a physical phenomenon is the same in all inertial frames. It only states that the
fundamental physical laws, which are expressed by differential equations, have the
24.5 Principle of Relativity 471

same form in any inertial frame. As a consequence, two inertial observers who
repeat an experiment in the same initial and boundary conditions arrive at the same
results.
We note that the possibility to verify the covariance of the physical law (24.43) is
related to knowledge of the transformation law of the quantities A, B; : : : in going
from inertial frame I to inertial frame I 0 . In other words, from a mathematical
viewpoint, a physical law could be covariant with respect to more transformation
groups, provided that the transformed quantities A0 , B 0 ; : : : are suitably defined.
Therefore, we can establish the covariance of a physical law only by verifying
that the assumed transformation laws for the quantities A, B; : : : are experimentally
confirmed.
To make clear this aspect of the theory, we recall that in Sect. 24.1 we proved
that the continuity equation is covariant with respect to Galilean transformations
provided that the the charge density  and the current density J are transformed
according to (24.7). In contrast, if we start from the special Lorentz transforma-
tions (24.18)–(24.21), instead of (24.5) and (24.6), then we derive the formulae
 
@ @ u @
D   ; (24.45)
@x 1 @x 01 c 2 @t 0
@ @
2
D ; (24.46)
@x @x 02
@ @
3
D ; (24.47)
@x @x 03
 
@ @ @
D  u : (24.48)
@t @t 01 @x 01
Then, under a Lorentz transformation, (24.1) becomes
@ h  u 1 i @
 
0
   2
J C 01  J 1  u
@t c @x
@ 2 @ 3
C 02 J C 03 J D 0: (24.49)
@x @x
We see that the continuity equation is covariant under Lorentz transformations if
and only if
 u 
0 D    2 J 1 ; J 01 D .J 1  u/; J 02 D J 2 ; J 03 D J 3 : (24.50)
c
Denoting by ak the component of vector a parallel to the relative velocity u and by
a? the component orthogonal to u, the preceding relations can also be written as
follows:
 
Ju
0 D    2 ; J0k D .J0k  u/; J0? D J? : (24.51)
c
472 24 An Introduction to Special Relativity

We conclude that, a priori, i.e., from a mathematical point of view, the continuity
equation of charge could be covariant under both Galilean and Lorentz transforma-
tions, provided that we assume different transformation properties for the charge
density and the current vector. However, the experiment will compel us to decide
which of (24.7) and (24.50) we must choose. We emphasize that it could happen
that, still from a mathematical point of view, the considered equations would be
covariant under only one of the two transformation groups. In fact, in Sect. 24.1, we
showed that Maxwell’s equations in a vacuum, (24.2) and (24.3), are not covariant
under Galilean transformations. However, in the next section, we verify that they are
covariant under Lorentz transformations, provided that the electromagnetic fields
are transformed in a suitable way in going from an inertial frame to another inertial
frame.

24.6 Covariance of Maxwell’s Equations

In this section, we verify the covariance of Maxwell’s equations in a vacuum

under Lorentz transformations. In an inertial frame I , (24.3)2 and (24.4)2 have the
following coordinate form:

@E 3 @E 2 @H 1
2
 3
D 0 ;
@x @x @t
@E 1 @E 3 @H 2
 D 0 ;
@x 3 @x 1 @t
@E 2 @E 1 @H 3
 D 0 ;
@x 1 @x 2 @t
@H 1 @H 2 @H 3
C C D 0:
@x 1 @x 2 @x 3
Inserting (24.45)–(24.48) into the preceding system, we obtain

@E 3 @E 2 @H 1 @H 1
c  D 0  C 0  u ;
@x 02 @x 03 @t 0 @x 01
  
@E 1 @
3 3
 @ 2 uE 3
 .E C 0 uH / D 0  H C ;
@x 03 @x 01 @t 0 mu0 c 2
  
@
 2 3
 @E 1 @ 3 uE 2
 E  0 uH  02 D 0 0  H  ;
@x 01 @x @t mu0 c 2

@H 1 u @H 2 @H 3
 01
  2 C 02 C 03 D 0;
@x c @x @x
24.6 Covariance of Maxwell’s Equations 473

p
where c D 1= 0 0 is the light velocity in a vacuum. Solving the fourth of
the preceding equations with respect to  @H 1 =@x 01 and inserting the obtained
expression into the first equation, we have the following system:

@ h  3 i @ h  i @H 1
c 02
 E C 0 uH 2  03  E 2  0 uH 3 D 0  0 ;
@x @x @t
" !#
@E 1 @ h i @ uE 3
3 3 2
 .E C  0 uH / D  0 0  H C ;
@x 03 @x 01 @t mu0 c 2
" !#
@ h  2 3
i @E 1 @ 3 uE 2
 E  0 uH  02 D 0 0  H  ;
@x 01 @x @t mu0 c 2

@H 1 u @H 2 @H 3
 01
  2 C 02 C 03 D 0:
@x c @x @x

By adopting the same notations of (24.51), we can state that Maxwell’s equations
in a vacuum are covariant under Lorentz transformations if the electromagnetic field
is transformed according to the following formulae:

E0k D Ek ; E0? D  .E? C 0 u  H? / ; (24.52)

H0k D Hk ; H0? D  .H?  0 u  E? / : (24.53)

We conclude our considerations about relativistic electromagnetism in a vacuum4

by proving charge conservation. Consider a volume V occupied by a charge q,
distributed inside V with a charge density .r; t/. Let V move relative to an inertial
frame I with velocity rP . Denote by I0 .x; t/ the proper inertial frame of the point
x 2 V at an arbitrary instant t. Because the current density J becomes J D Pr and
the velocity u of I0 .x; t/ relative to I at the instant t is rP .x; t/, condition (24.50)1
gives

0 D :

Taking into account (24.31), we have that
Z Z
qD  dV D 0 dV0 D q0 ; (24.54)
V V0

and we can state that the total charge is invariant with respect to Lorentz transfor-
mations.

4
For the difficult topic of electromagnetism in matter, see, for instance, [27, 39, 44].
474 24 An Introduction to Special Relativity

24.7 Relativistic Dynamics

In this section we derive a relativistic equation describing the dynamics of a material

point in inertial frames following the approach proposed by Einstein.
Let us suppose that the material point P with mass m is a charge q moving
relative to the inertial frame I with velocity rP in the presence of an electromagnetic
field .E; H/. Then P is acted upon by the Lorentz force

F D q.E C 0 rP  H/: (24.55)

Denote by I0 the proper frame of P at an arbitrary instant t, that is, the inertial
frame that moves relative to I with a uniform velocity equal to the velocity rP at
instant t. In the proper frame I0 , the Lorentz force reduces to the electrostatic force

F0 D qE0 (24.56)

since the charge q carried by P is invariant [see (24.54)]. It is a simple exercise

to verify that from (24.52) and (24.53) we obtain the following transformation
formulae for electric force under a special Lorentz transformation I ! I0

F01 D F 1 ; (24.57)
F02 D F 2 ; (24.58)
F03 D F ;3
(24.59)

which, adopting the notation of (24.51), can be summarized as follows:

F0k D Fk ; F0? D  F? : (24.60)

Einstein makes the following assumptions:

1. All the forces, regardless of their nature, transform according to (24.60) under a
Lorentz transformation.
Also, for the change of inertial frame I ! I0 , formulae (24.40)–(24.42) give

xR 01 D  3 xR 1 ; (24.61)
xR 02 2 2
D  xR ; (24.62)
xR 01 2 3
D  xR ; (24.63)

which can equivalently be written as

rR 0k D  3 rR k ; rR 0? D  2 rR ? : (24.64)
24.7 Relativistic Dynamics 475

The second assumption on which Einstein bases his relativistic dynamics is:
2. The Newtonian equation of motion is valid in the proper frame I0 , that is, we
have that
m0 rR 0 D F0 : (24.65)
Applying (24.60) and (24.64)–(24.65), we obtain the following equations in
any inertial frame I :
m0  3 rR k D Fk ; m0  rR ? D F? : (24.66)

Now we prove that the preceding equations are the projections parallel and
orthogonal to rP , respectively, of the equation
d
.mPr/ D F; (24.67)
dt
where
1
m D m0  D m0 q (24.68)
jPrj2
1 c2

is the relativistic mass of the material point P .

To prove the preceding statement, we first note that
 1=2
d d jPrj2 rP  rR k
D 1 2 D 3 2 ; (24.69)
dt dt c c
jPrj2
2 D 1 C 2 : (24.70)
c2
Then also
 
d 2
. rP / D  rP  rR k rP C rR k C rR ? ; (24.71)
dt c2
so that we have
   2 
d 
. rP / D  rP  rR k rP k C rR k ; (24.72)
dt k c2
 
d
. rP / D  rR ? : (24.73)
dt ?

Introducing the unit vector u along rP we have that rP D jPrju, rR k D jRrk ju, and
recalling (24.70), we can write (24.72) as follows:
   2 
d  2
. rP / D  2
jP
r j C 1 rR k D  3 rR k : (24.74)
dt k c

This equation, together with (24.73), proves the equivalence between (24.66)
and (24.67).
476 24 An Introduction to Special Relativity

Now we derive an important consequence of the relativistic equation (24.67). By

a scalar product of (24.67) and rP , we obtain the equation

d
rP  .mPr/ D rP  F:
dt
Taking into account (24.74), the preceding equation becomes
 
d d
rP  .mPr/ D rP  .mPr/ D m0  3 rP  rR k D rP  F: (24.75)
dt dt k

In view of (24.69), we finally obtain the equation

dE
D rP  F; (24.76)
dt
where the scalar quantity
E D mc 2 D m0 c 2 (24.77)
is called the relativistic energy of the material point P . Also, we call the scalar
quantities
T D E  m0 c 2 ; E0 D m 0 c 2 (24.78)
the relativistic kinetic energy and the rest or proper relativistic energy. From the
approximate equality
1 jPrj2
 1 '
2 c2
we derive the condition
 
1 jPrj2
T D m0 jPrj2 C O :
2 c2
This relation shows that for a velocity much smaller than the light speed c, the
relativistic kinetic energy reduces to the classical one. In other words, (24.75) is the
relativistic formulation of the classical theorem on the variation of kinetic energy.

24.8 Minkowski’s Space-Time

In the preceding section, we described Einstein’s formulation of special relativity,

which starts from both physical considerations and Maxwell’s equations. Hence-
forth we discuss the elegant and useful geometric formulation of the special theory
of relativity supplied by Minkowski. This formulation represented the bridge that
Einstein crossed to arrive at a geometric formulation of gravitation.
24.8 Minkowski’s Space-Time 477

We start by analyzing the geometric structure of the four-dimensional space

underlying this model. We denote by V4 a generalized Euclidean four-dimensional
space in which an orthonormal frame of reference .O; e˛ /, ˛ D 1; : : : ; 4, can be
found such that the coefficients ˛ˇ of the scalar product, which is defined by a
symmetric covariant 2-tensor g, are given by the following matrix:
0 1
1 00 0
  B0 10 0 C
˛ˇ D B
@0
C: (24.79)
01 0 A
0 00 1

The space V4 is called Minkowski’s space-time and any point P 2 V4 is an event.

Moreover, any vector v of the vector space associated with V4 is said to be a four-
vector or 4-vector.
We call a Lorentz frame any frame of reference of V4 in which the coefficients
˛ˇ of the scalar product assume the form (24.79). Finally, a Lorentz transformation
is any (linear) coordinate transformation

x 0˛ D x00˛ C A˛ˇ x ˇ (24.80)

relating the coordinates .x ˛ / of any event in the Lorentz frame .O; e˛ / to the
coordinates .O 0 ; e0˛ / of the same event in the Lorentz frame .O 0 ; e0˛ /. The definition
of Lorentz frame requires that the matrix A˛ˇ must satisfy the condition

˛ˇ D A ˛ Aˇ   : (24.81)
 
˛
Since in any Lorentz frame the square of the distance s between two events x.1/
 
˛
and x.2/ assumes the form
  
ˇ ˇ
s 2 D ˛ˇ x.2/
˛ ˛
 x.1/ x.2/  x.1/
3 
X 2  2
i i 4 4
D x.2/  x.1/  x.2/  x.1/ ; (24.82)
i D1

we can say that the Lorentz transformations are orthogonal transformations of

V4 , provided that the orthogonality is evaluated by the scalar product ˛ˇ . Owing
to (24.81) or (24.82), we can conclude that the set of all these transformations is a
group.
The square v  v D g.v; v/ of the norm of a 4-vector v of V4 can be positive,
negative, or zero. In the first case, we say that v is a spacelike 4-vector. In the
second case, v is said to be a timelike 4-vector. Finally, if v  v D 0, then v is a null
478 24 An Introduction to Special Relativity

4-vector. At any point O 2 V4 , the set of the events P 2 V4 such that the 4-vectors
!
OP D v are null, i.e., the set of the 4-vectors for which
! !
OP  OP D 0; (24.83)

is a cone CO , with its vertex at O, which is called a light cone at O. Condi-

! !
tion (24.83) defines a cone since, if OP 2 CO , then also OP 2 CO . In a Lorentz
frame .O; e˛ /, the cone (24.83) is represented by the equation
3
X
.x i /2  .x 4 /2 D 0: (24.84)
i D1

!
The 4-vectors v D OP , corresponding to events P , which are internal to CO , are
timelike 4-vectors, whereas those corresponding to events P , which are external to
CO , are spacelike 4-vectors.
Let us fix an arbitrary timelike 4-vector eO , and let .O; eO / be a uniform vector field
of V4 . We say that, at any O 2 V4 , the 4-vector eO defines the direction of the future
at O. Moreover, the internal region COC of CO , to which eO belongs, is said to be
the future of O, whereas the remaining internal region CO of CO is the past of O.
Finally, the set of events external to CO is defined as the present of O.
The first three axes of a Lorentz frame .O; e/ are spacelike 4-vectors, whereas
the fourth axis e4 is a timelike 4-vector. Moreover, let ¢O;e4 be the three-dimensional
space formed by all the 4-vectors generated by linear combinations of the three
4-vectors e1 ; e2 ; e3 . It is easy to verify that ¢O;e4 is a properly Euclidean space, with
respect to the scalar product g, and that its elements are spacelike 4-vectors.
Before making clear the physical meaning of all the preceding definitions, we
must prove the following propositions.
Proposition 24.1. Let u be any timelike 4-vector, and let us denote by ¢O;u the
three-dimensional space of all 4-vectors that are orthogonal to u. Then, any v 2 ¢O;u
is a spacelike 4-vector.
Proof. Let .O; e˛ / be a Lorentz frame, and let u D ui ei C u4 e4  u? C 4
Pu3 e4i be the
decomposition of the 4-vector u in the basis .e˛ /. Since u?  u? D 1 .u /2 , the
4-vector u? is spacelike. With the same notation, we set v D v? C v4 e4 , so that the
following relations hold:

u  u D u?  u?  .u4 /2 < 0;
u  v D u?  v?  u4 v4 D 0;
v  v D v?  v?  .v4 /2 :

Consequently,
.u?  v? /2
v  v D v?  v?  ;
.u4 /2
24.8 Minkowski’s Space-Time 479

so that, by the application of Schwartz’s inequality to u?  v? , we obtain

.u?  u? /.v?  v? /
v  v  v?  v?  :
.u4 /2

Finally, this inequality implies that

v?  v?
vv Œu?  u?  .u4 /2
> 0;
.u4 /2

and the proposition is proved. t

u
Remark 24.5. Owing to the preceding proposition, we can say that any timelike 4-
vector u defines infinite Lorentz frames at event O. In fact, it is sufficient to consider
the frame .O; e1 ; e2 ; e3 ; u=juj/, where the mutually orthogonal unit vectors ei , i D
1; 2; 3, belong to ¢O;u .
Proposition 24.2. If v is a spacelike 4-vector at O 2 V4 , then it is possible to find
at least a timelike 4-vector such that u  v D 0.
Proof. Let .O; e˛ / be a Lorentz frame at O. Then, by adopting the same notations
used in the preceding proposition, we have that

v?  v?  .v4 /2 > 0: (24.85)

We must now prove that there exists at least a 4-vector u such that

u  u D u?  u?  .u4 /2 < 0; (24.86)

4 4
u  v D u?  v?  u v D 0: (24.87)

From (24.86) we derive that the 4-vector u is orthogonal to v if we choose

arbitrarily the components u1 ; u2 ; u3 , i.e., the spacelike 4-vector u? , provided that
the component u4 is given by
u?  v?
u4 D :
v4
For this choice of v4 , condition (24.86) becomes

.u?  v? /2
u  u D u?  u?  :
.v4 /2
Again by applying Schwartz’s inequality, we can write the preceding equation as
 
v?  v?
u  u D u?  u? 1  : (24.88)
.v4 /2

Finally, since u?  u? > 0, from (24.86) and (24.88) we have that u  u < 0. t
u
480 24 An Introduction to Special Relativity

Remark 24.6. For any spacelike vector v at event O, we can find infinite Lorentz
frames .O; e˛ / for which e1 D v=jvj. In fact, it is sufficient to take one of the
Ptimelike 4-vectors e4 orthogonal to v and choose in the three-dimensional
existing
space ¢O;e4 to which v belongs two other unit vectors that are orthogonal to each
other and orthogonal to v.
Let us suppose that we have introduced a direction of the future by the uniform
O so that it is possible to define the cones CO and COC at any event O.
4-vector u,
Proposition 24.3. Let u 2 COC . Then, v 2 COC if and only if

u  v < 0: (24.89)

Proof. In fact, uu < 0 since u 2 COC . But uw is a continuous function of w 2 COC ,
which is a connected set. Consequently, if, for a 4-vector v 2 COC , we had u  v > 0,
then there would exist a timelike 4-vector u? 2 COC such that u  u? D 0. But this is
impossible for Proposition 24.1.
Conversely, owing to the remark about Proposition 24.1, we can find a Lorentz
frame at O having e4 D u=juj as a timelike axis. In this frame, we have that u  u D
u4 v4 . On the other hand, since u; u 2 COC , we have u4 > 0 and v4 > 0, and the
proposition is proved. t
u
We denote by LC O the set of Lorentz frames .O; e˛ / at O whose axes e4
belong to the positive cone COC . The Lorentz transformation between two Lorentz
frames in LCO is said to be orthochronous. It is evident that the totality of these
transformations is a group.
Proposition 24.4. Let .O; e˛ / 2 LC O be a Lorentz frame. Then the Lorentz frame
.O; e0˛ / belongs to LC
O if and only if

A44 > 0: (24.90)

Proof. Since e04 D A˛4 e˛ , we have that e04  e4 D A44 . The proposition is proved when
we take into account Proposition 24.3. t
u

24.9 Physical Meaning of Minkowski’s Space-Time

To attribute a physical meaning to some geometrical objects associated with the

Minkowski space-time, we begin with the following remark. Let I and I 0 be two
inertial frames of reference. We have already said that the relation between the
coordinates .xAi ; tA / and .xA0i ; tA0 /, which observers I and I 0 associate with the same
event A, is a Lorentz transformation. It is a very simple exercise to verify that any
Lorentz transformation leaves invariant the following quadratic form:
24.9 Physical Meaning of Minkowski’s Space-Time 481

3
X
s2  .xAi  xBi /2  c 2 .tA  tB /2
i D1
3
X
D .xA0i  xB0i /2  c 2 .tA0  tB0 /2 ; (24.91)
i D1

relating the space-time coordinates .xAi ; tA / and .xBi ; tB / of event A in the Lorentz
frame I and the corresponding coordinates .xAi 0 ; tA0 /, .xBi 0 ; tB0 / of the same events
in I 0
Moreover, from (24.21) we obtain that

dt 0
D  > 0: (24.92)
dt

Furthermore, if we introduce the notation x 4 D ct, then the quadratic form

(24.91) becomes identical to (24.82).
Let us introduce a direction of the future at any point of V4 by the uniform
timelike field e4 . Owing to Proposition 24.2 and the associated remark, we can
define at any point of V4 a Lorentz frame .O; e˛ / 2 LC O . Let us introduce a one-
to-one correspondence among the inertial frames I in the physical space and the
orthogonal or Lorentz frames in LC O in the following way. First, we associate a fixed
inertial frame I to the Lorentz frame .O; e˛ / in LC 0
O . Let I be any inertial frame
whose relation with I is expressed by (24.8) and (24.9). Then to I 0 we associate
the Lorentz frame .O 0 ; e0˛ /, whose transformation formulae (24.80) with respect to
.O; e˛ / are given by

.xO˛ 0 / D .xOi 0 ; ct0 /; (24.93)

 j
 j
!
j
Qji ıh C .  1/ uu2uh Qji uc
.A˛ˇ / D : (24.94)
 uci 

Since
A44 D  > 0; (24.95)
the Lorentz frame .O 0
; e0˛ /
belongs to LC
O.
Conversely, let us assign the Lorentz frame .O 0 ; e0˛ / 2 LCO by (24.80), where
A44 > 0. To determine the corresponding inertial frame I 0 , we must determine
the right-hand sides of (24.93) and (24.94), i.e., the quantities t0 ; ui ; Qji , and xOi 0 .
Owing to the orthogonality of the matrix .Qji /, we must evaluate ten quantities
starting from the four coordinates .xO˛ 0 / and the six independent coefficients A˛ˇ
[see (24.81)].
482 24 An Introduction to Special Relativity

In particular, if I and I 0 are related by a special Lorentz transformation,

then (24.93) and (24.94) reduce to the following formulae:

.xO˛ 0 / D .0; 0/; (24.96)

0 1
 0 0  uc
B 0 10 0 C
.A˛ˇ / D B
@ 0 0 1 0 A:
C (24.97)
 uc 0 0 

The preceding considerations allow us to make clear the physical meaning of the
definitions we gave in the preceding section. In fact, to any point P 2 V4 , with
coordinates .x ˛ / in a Lorentz frame .O; e˛ / 2 LC O , there corresponds an event
that has coordinates .x i ; x 4 =c/ in the inertial frame I . In particular, all the events
belonging to the light cone CO have coordinates verifying (24.84). To these points
of V4 correspond all the events whose coordinates in the inertial frame I satisfy the
equation
3
X
.x i /2  c 2 t 2  r 2  c 2 t 2 D 0 (24.98)
i D1

or, equivalently, the two equations r  ct D 0 and r C ct D 0, which represent,

respectively, a spherical light wave expanding from O and a spherical light wave
contracting toward O.
Moreover, owing to Proposition 24.1–24.3 and the related remark, for any point
P 2 COC it is possible to find a Lorentz frame in LC O and then an inertial frame I
such that event P has in I the coordinates .0; 0; 0; t/, t > 0, so that it appears to
happen after the event at the origin. In contrast, if P 2 CO , then the corresponding
event appears to happen before the event at O for observer I . In other words, any
event belonging to COC happens after the event at O, for some inertial frames, and
any event in CO happens before the event at O, for some inertial frames.
In contrast, an event that belongs to the present of O, owing to Proposi-
tion 24.2 and the related remark, appears to happen at the same time to some
inertial observer I .
We conclude this section with a very important remark regarding a consequence
of the correspondence existing among the inertial frames and the Lorentz frames in
V4 . If we succeed in formulating the physical laws by tensor relations in V4 , then
they will be covariant with respect to Lorentz’s transformations; in other words,
they will satisfy the principle of relativity.

24.10 Four-Dimensional Equation of Motion

In the preceding sections, we presented Einstein’s approach to the relativistic

dynamics of a material point in any inertial frame I . We can summarize the results
24.10 Four-Dimensional Equation of Motion 483

of this approach stating that the Newtonian equations must be substituted by the
following ones:
d
.mv/ D F; (24.99)
dt
d
.mc 2 / D F  rP ; (24.100)
dt
where
m0
mD q (24.101)
jPrj2
1 c2

is the relativistic mass of P , m0 its rest mass, and rP the velocity of P . In this section,
we recall the four-dimensional formulation of (24.99), (24.100) in the space-time V4 .
Let
x i D x i .t/ (24.102)
be the equation of motion of a particle with respect to an inertial frame I . To the
trajectory (24.102) we can associate a curve  of V4 , which in the Lorentz frame
T  .O; e˛ / corresponding to I has equations

x i D x i .t/; x 4 D ct: (24.103)

The curve  is called a world trajectory or world line of P . It defines a curve of V4

that does not depend on the adopted Lorentz frame. The square norm of the tangent
vector .xP i ; c/ to  is
3
X
.xP i /2  c 2 < 0; (24.104)
i D1

and it is negative since the velocity of P in any inertial frame is less than the velocity
c of light in a vacuum. In other words, the curve  is timelike and its tangent vector
at any point P is contained in the map COC of the light cone at P .
Let x i .t / be the position of P at the instant t in the inertial frame I . We recall
that the rest frame or proper frame I of P at the instant t is an inertial frame that has
its origin at x i .t/ and moves with velocity v with respect to I . The corresponding
Lorentz frame ..x i .t /; ct /; e0 /  T has a time axis e4 tangent to  at the point x i .t /
since in T the vector e4 must have components .0; 0; 0; 1/. In going from T to T ,
the infinitesimal distance between two events on  is invariant, so that we have

3
!
X
2
ds D .xP /  c dt 2 D c 2 d 2 ;
i 2 2
(24.105)
i D1
484 24 An Introduction to Special Relativity

where  is the proper time, i.e., the time evaluated by observer I.t/. From (24.105)
we derive the relation
dt
D : (24.106)
d
If we adopt this time along , then the parametric equations (24.103) become

x ˛ D x ˛ ./: (24.107)

We define as a world velocity or 4-velocity the 4-vector

dx ˛
U˛ D ; (24.108)
d
which, in view of (24.106), has the following components in the Lorentz frame I :

U ˛ D . rP ; c/: (24.109)

Moreover,
U ˛ U˛ D  2 .jPrj2  c 2 / D c 2 < 0: (24.110)
It is a simple exercise to verify that we can write (24.99) and (24.100) in the
covariant form
dU ˛
m0 D ¥˛ ; (24.111)
d
where the 4-force is given by
 
˛ F  rP
.¥ / D  F; : (24.112)
c

24.11 Tensor Formulation of Electromagnetism in Vacuum

Let S be a continuous system of moving charges, and denote by  and rP the charge
density and the velocity field of S , respectively. If S moves in a vacuum, the
continuity equation (24.1) becomes
@
C r  .Pr/ D 0: (24.113)
@t
Consider a transformation (24.80) from a Lorentz frame I to another one I 0 , and
suppose that the quantities
 
˛ Pr
J D ; (24.114)
c
24.11 Tensor Formulation of Electromagnetism in Vacuum 485

are the components of a 4-vector, called a 4-current. In other words, we are

assuming that, in going from I to I 0 , we obtain the following results:
J 0˛ D A˛ˇ J ˇ : (24.115)

Then (24.113) can be written as the divergence in V4 of the 4-current

@J ˛
D 0: (24.116)
@x ˛
The form of this equation is independent of the Lorentz frame and, consequently,
verifies a principle of relativity.
Exercise 24.1. Verify that from (24.115) and (24.97) we obtain again (24.50).
Similarly, the Maxwell equations (24.1) and (24.4)2 can be written in the form

@F ˛ˇ
D 0; (24.117)
@x ˇ
where
0 1
0 H 3 H 2 cE 1
B H 3 0 H 1 cE 2 C
F ˛ˇ DB C
@ H 2 H 1 0 cE 3 A : (24.118)
cE 1 cE 2 cE 3 0
Consequently, if F ˛ˇ are supposed to be the components of a (skew-symmetric)
2-tensor, then (24.117) has the same form in any Lorentz frame and the relativity
principle is satisfied by (24.1) and (24.4)2. F ˛ˇ is called an electromagnetic tensor.
If we introduce the adjoint tensor of F ˛ˇ
1
F ?˛ˇ D  ˛ˇ  F  ; (24.119)
2
which is obtained from F ˛ˇ with the substitution cE ! H, then Maxwell’s
equations (24.2) and (24.4)1 assume the covariant form
@F ?˛ˇ
D 0: (24.120)
@x ˇ

Exercise 24.2. Verify that from

F 0˛ˇ D A˛ Aˇ F 
and (24.97) we obtain (24.52) and (24.53).
Exercise 24.3. Verify that the components of the Lorentz force (24.55) are equal to
the first three components of the 4-vector
q ˛ˇ
F˛ D F Uˇ :
c
Appendix A
First-Order PDE

A.1 Monge’s Cone

In this appendix, we give a sketch of the method proposed by Monge, Ampere, and
Cauchy to reduce the integration of a first-order PDE to the integration of a system
of ordinary equations.
Let F .x; u; p/ be a function of class C 2 .<2nC1 / verifying the following condi-
tions:
˚ 
Pn set F2 D .x; u; p/ 2 < ; F .x; u; p/ D 0 is not empty.
2nC1
1. The
2. i D1 Fpi ¤ 0; 8.x; u; p/ 2 F:
Under these hypotheses, the following general first-order PDE:

F .x; u; ru/ D 0 (A.1)

in the unknown u D u.x/ 2 C 2 .D/, with D  <n , will be considered.

First, any solution u D u.x/ of (A.1) defines a surface † that is called an
integral surface of (A.1). Moreover, the vector N  .ru; 1/  .p; 1/ of <nC1 is
normal to the integral surface †. Consequently, (A.1) expresses the relation existing
between the normal vectors to all the integral surfaces at any point .x; u/.
To understand the geometrical meaning of this relation, we start by noting that
the quantities .x0 ; u0 ; p0 / completely define a plane containing the point .x0 ; u0 / and
having a normal vector with components .p0 ; 1/. Then, for a fixed point .x0 ; u0 /,
the equation
F .x0 ; u0 ; p0 / D 0 (A.2)

defines a set …0 of planes containing .x0 ; u0 / and tangent to the integral surfaces of
(A.1), to which .x0 ; u0 / belongs. Owing to conditions 1 and 2, …0 is not empty, and

A. Romano, Classical Mechanics with Mathematica R

, Modeling and Simulation 487
in Science, Engineering and Technology, DOI 10.1007/978-0-8176-8352-8,
© Springer Science+Business Media New York 2012
488 A First-Order PDE

we can suppose that Fpn .x0 ; u0 ; p0 / ¤ 0. In turn, this result implies that (A.2), at
least locally, can be written in the form

pn D pn .x0 ; u0 ; p1 ; : : : ; pn1 /:

Consequently, the set …0 is formed by a family of planes depending on the

parameters .p1 ; : : : ; pn1 /. Any plane  2 …0 contains the point .x0 ; u0 / and has
.p1 ; : : : ; pn1 ; pn .p1 ; : : : ; pn1 /; 1/ as a normal vector, i.e., it is represented by
the equation

X
n1
f .X; U; p1 ; : : : ; pn1 /  p˛ .X˛  x0˛ / C pn .p˛ /.U  u0 / D 0; (A.3)
˛D1

where .X; U / are the coordinates of any point of .

The envelope of all these planes is defined by the system formed by (A.3) and
the equation
@pn
.X˛  x0˛ / C .Xn  x0n / D 0; ˛ D 1; : : : ; n  1; (A.4)
@p˛
which is obtained by differentiating (A.3) with respect to p˛ ; ˛ D 1; : : : ; n  1. By
Dini’s theorem, @pn =@p˛ D Fp˛ =Fpn , and (A.4) becomes

Fpn .X˛  x0˛ /  Fp˛ .Xn  x0n / D 0; ˛ D 1; : : : ; n  1: (A.5)

Equations (A.3) and (A.5) constitute a linear system of n equations in the

unknowns ..X˛  x0˛ /; .Xn  x0n //. The determinant of the matrix of this system’s
coefficients is
0 1
p1 p2 : : : pn1 pn
BF C
B pn 0 : : : 0 Fp1 C
B C
 D B 0 Fpn : : : 0 Fp2 C D .1/n1 Fpn2 .p  Fp / ¤ 0:
B C n
@ ::: ::: ::: ::: ::: A
0 0 : : : Fpn Fpn1

Consequently, the parametric equations of the envelope of …0 become

Fpi
Xi  x0i D .U  u0 /; i D 1; : : : ; n: (A.6)
p  Fp

These equations define a cone C0 since, if the vector V D .X  x0 ; U  u0 / is a

solution of (A.6), then also the vector V, where  2 <, is a solution. This cone C0
is called Monge’s cone at .x0 ; u0 /.
A.2 Characteristic Strips 489

charact. direction

xn

Fig. A.1 Monge’s cone

u
charact. strip

xn

charact. curve

Fig. A.2 Characteristic curve and strip

A.2 Characteristic Strips

The preceding considerations allow us to state that, if u D u.x/ is an integral surface

† of (A.1), then a plane tangent to † at the point .x0 ; u0 / belongs to …0 , i.e., it
is tangent to Monge’s cone at that point (Fig. A.1). Moreover, the characteristic
directions of Monge’s cone, which are tangent to †, define a field of tangent vectors
on † whose integral curves are called characteristic curves.
We can associate with any point of this curve .x.s/; u.s// a plane tangent to †
having director cosines .p; 1/  .ru.x.s//; 1/. The 1-parameter family of these
planes along a characteristic curve is said to be a characteristic strip (Fig. A.2).
490 A First-Order PDE

We now write the system of ordinary differential equations in the unknowns

.x.s/; u.s/; p.s// that define a characteristic strip on the integral surface †. Since
the tangent to a characteristic curve at a point belongs to Monge’s cone at that point,
from (A.6) we have that
dx
D Fp ; (A.7)
ds
du
D p  Fp : (A.8)
ds

Moreover, along a characteristic curve .x.s/; u.s// we have p.s/ D ru.x.s//,

and then
dpi
D uxi xj Fpj :
ds
On the other hand, the differentiation of (A.1) with respect to xi gives

Fxi C uxi Fu C uxi xj Fpj D 0:

Comparing the last two relations, we obtain the following system of 2n C 1

equations:

dx
D Fp ; (A.9)
ds
dp
D .Fx C Fu p/; (A.10)
ds
du
D p  Fp ; (A.11)
ds

in the 2n C 1 unknowns .x.s/; u.s/; p.s//, which is called a characteristic system

of (A.1).
We have proved that, if u.x/ is an integral surface of (A.1), then its characteristic
strips satisfy system (A.9)–(A.11). Moreover, any solution of (A.9)–(A.11) is a
characteristic strip of an integral surface of (A.1), as is proved by the following
theorem.
Theorem A.1. Let .x0 ; u0 ; p0 D ru.x0 // be a plane tangent to an integral surface
† at the point x0 ; u0 /. Then the solution of (A.9)–(A.11) corresponding to the initial
datum .x0 ; u0 ; p0 / is a characteristic strip of †.
Proof. It is sufficient to recall what we have already proved and to remark that
the characteristic strip determined by the initial datum .x0 ; u0 ; p0 / is unique for the
uniqueness theorem. t
u
Taking into account the function F .x.s/; u.s/; p.s// and recalling (A.9)–(A.11),
we verify that
A.2 Characteristic Strips 491

Theorem A.2.
F .x.s/; u.s/; p.s// D const: (A.12)
Remark A.1. If Eq. (A.1) is quasilinear

F .x; u; ru/ D a.x; u/  ru  b.x; u/ D 0; (A.13)

then system (A.9)–(A.11) becomes

dx
D a.x; u/; (A.14)
ds
du
D b.x; u/: (A.15)
ds
This is a system of n C 1 equations in the unknowns .x.s/; u.s// that can be solved
without the help of (A.10). In particular, if (A.13) is linear, then a and b depend
only on x. Therefore, (A.14) supplies the projection of the characteristic curves in
<n and (A.15) gives the remaining unknown u.s/.
Remark A.2. Let us suppose that (A.1) has the form

F .x; ru/ D 0: (A.16)

Denoting by t the variable xn , by p the derivative Fut , by x the vector .x1 ; : : : ; xn1 /,
and by p the vector .p1 ; : : : ; pn1 /, the preceding equation becomes

F .x; t; p; p/ D 0: (A.17)

If Fp ¤ 0, then (A.17) can be written as a Hamilton–Jacobi equation

p C H.x; t; p/ D 0; (A.18)

whose characteristic system is

dx
D Hp ; (A.19)
ds
dp
D Hx ; (A.20)
ds
du
D p  Hp C p; (A.21)
ds
dt
D 1; (A.22)
ds
dp
D Ht : (A.23)
ds
492 A First-Order PDE

Owing to (A.22), we can identify s with t and the preceding system becomes

dx
D Hp ; (A.24)
dt
dp
D Hx ; (A.25)
dt
du
D p  Hp  H; (A.26)
dt
dp
D Ht : (A.27)
ds
We note that (A.24) and (A.25) are Hamiltonian equations that can be solved
independently of the remaining ones.

A.3 Cauchy’s Problem

The Cauchy problem relative to (A.1) can be formulated as follows.

Let  be an .n  1/-dimensional manifold, contained in a region D of <n , and
let u0 .x/ 2 C 2 . / be an assigned function on  . Determine a solution u.x/ of (A.1)
whose restriction to  coincides with u0 .x/.
In other words, if
x0 D x0 .v˛ /; .v˛ / 2 V  <n1 ; (A.28)

is a parametric representation of  , then Cauchy’s problem consists in finding a

solution u.x/ of (A.1) such that
u.x0 .v˛ // D u0 .x0 .v˛ //: (A.29)
Geometrically, we can say that we want to determine an n-dimensional manifold
u.x/ satisfying (A.1) and containing the initial .n  1/-dimensional manifold  .
We prove that the requested solution can be obtained as the envelope of a suitable
family of characteristic strips. In this regard, we first note that the points of  , given
by the initial datum (A.29), supply the initial data for the unknowns x.s/; u.s/ of
(A.9), (A.11). However, up to now, we have had no initial data for the unknown
p.s/. We show that these data can also be obtained by (A.29). In fact, we recall that
the variable p, together with 1, defines the normal vector to the plane tangent to
the integral surface † at the point .x; u.x// 2 †. Moreover, † must be tangent to
the Monge cone along a characteristic direction at any point and, in particular, at
any point of  (Fig. A.3).
Consequently, we take p0 in such a way that, at any point of  , .p0 ; 1/ is
orthogonal to the vector .@x=@v˛ ; @u=@v˛ /, which is tangent to  . Expressed as
A.3 Cauchy’s Problem 493

u
charact. strip

xn

Fig. A.3 Cauchy data for (A.9)–(A.11)

formulae, we have that

@x0 @u0
 p0  D 0; (A.30)
@v˛ @v˛
F .x0 .v˛ /; u0 .v˛ /; p0 .v˛ // D 0; (A.31)

˛ D 1; : : : ; n  1.
If .x0 .v˛ /; u.v˛ /; p0 .v˛ // 2  verifies the preceding system and the condition
 
@x0
J D det Fp ¤ 0; (A.32)
@v˛

is satisfied at this point, then system (A.30), (A.31) can be solved with respect to
p0 .v˛ / in a neighborhood of the point .x0 .v˛ /; u.v˛ /; p0 .v˛ //. In this way we obtain
the initial data for (A.9)–(A.11).
The proof of the following theorem can be found in any book on PDEs:
Theorem A.3. If it is possible to complete the initial datum (A.29) by solving, with
respect to p0 .v˛ /, system (A.30), (A.31); then one and only one solution exists of the
Cauchy problem.
Appendix B
Fourier Series

A function f .t/ is said to be periodic of period 2T if

f .t C 2T / D f .t/; 8 t 2 <: (B.1)

The Fourier series of the periodic function f .t/ is the series
1  
a0 X k t k t
C ak cos C bk sin ; (B.2)
2 T T
kD1

where
Z T
1
a0 D f .f / dt; (B.3)
T T
Z T
1 k t
ak D f .t/ cos dt; (B.4)
T T T
Z T
1 k t
bk D f .t/ sin dt (B.5)
T T T

are the Fourier coefficients of the function f .t/. In particular, a0 =2 is equal to

the mean value of the function f .t/ in the period 2T . The following fundamental
theorem can be proved.
Theorem B.1. Let f W < ! < be a periodic function of period 2T , which is almost
continuous together with its first derivative. If the discontinuities of f .t/ and f 0 .t/
are finite, then the series (B.2) converges to f .t/ at any point t 2 < at which f .t/
is continuous, whereas it converges to

f .c C /  f .c  /
(B.6)
2
at any discontinuity point c.

A. Romano, Classical Mechanics with Mathematica R

, Modeling and Simulation 495
in Science, Engineering and Technology, DOI 10.1007/978-0-8176-8352-8,
© Springer Science+Business Media New York 2012
496 B Fourier Series

F t

ft

t
T T

Fig. B.1 First periodic extension of f .t /

F t
ft

T T

Fig. B.2 Second periodic extension of f .t /

Definition B.1. The function f .t/ is even if

f .t/ D f .t/; (B.7)

and it is odd if
f .t/ D f .t/; (B.8)

8t 2 <.
Remark B.1. If the function f .t/ is even, then the Fourier coefficients bk vanish for
any integer k. In fact, it is sufficient to decompose integral (B.5) into the sum of
the two integrals over the intervals .T; 0/ and .0; T / and recall that sin t is an odd
function. Similarly, if f .t/ is odd, then all the Fourier coefficients ak vanish.
Remark B.2. A Fourier series can be extended to any function f .t/ that in a
bounded interval Œ0; T  is continuous with its first derivative almost everywhere in
Œ0; T , provided that the discontinuities of f .t/ and f 0 .t/ are finite. In fact, it is
sufficient to apply series (B.2) to one of the following periodic functions that extend
the function f .t/ in the interval ŒT; T  as follows (Figs. B.1 and B.2):
B Fourier Series 497


f .t/; t 2 Œ0; T ;
F1 .t/ D
f .t/; t 2 ŒT; 0;

f .t/; t 2 Œ0; T ;
F2 .t/ D
f .t/; t 2 ŒT; 0:

It is more convenient in applications to resort to the complex form of a Fourier

series. This form is obtained as follows. From Euler’s identity
ei t D cos t i sin t (B.9)

there follows
1
X 1 
X 
k t k t k t
Ak ei T D A0 C Ak ei T C Ak ei T

kD1 kD1
1
X  
k t k t
D A0 C Ak cos C i sin
T T
kD1
 

k t k t
C Ak cos  i sin
T T
1
X k t
D A0 C .Ak C Ak / cos
T
kD1

k t
C i .Ak  Ak / sin :
T
This result is identical to (B.2) if we set
Ak C Ak D ak ; (B.10)
Ak  iAk D i bk ; (B.11)

that is, if and only if

a0
A0 D
; (B.12)
2
1
Ak D .ak  i bk /; (B.13)
2
1
Ak D .ak C i bk /: (B.14)
2
Equations (B.9)–(B.11) allow us to write (B.2) as
1
X k t
Ak ei T ; (B.15)
1
498 B Fourier Series

where
Z T
1 k t
Ak D f .t/ei T dt; (B.16)
2T T
Z T
1 k t
Ak D f .t/ei T dt D Ak : (B.17)
2T T

The complex form (B.15) can be at once extended to the case of a function
f .t1 ; : : : ; tn / of n real variables .t1 ; : : : ; tn /, which is periodic with the same
period T with respect to each variable. In fact, since f .t1 ; : : : ; tn / is periodic with
respect to t1 , we have that
1
X k1 t1
.1/
f .t1 ; : : : ; tn / D Ak1 .t2 ; : : : ; tn /ei T ;
k1 D1

.1/
where, in view of (B.16) and (B.17), the functions Ak1 .t2 ; : : : ; tn / are periodic in
the variables t2 ; : : : ; tn . Applying (B.15) to these functions, we obtain
1
X 1
X k1 t1 k2 t2
.2/
f .t1 ; : : : ; tn / D Ak1 k2 .t3 ; : : : ; tn /ei T ei T :
k2 D1 k1 D1

Finally, iterating the foregoing procedure, we can write that

1
X kt
f .t/ D Ak ei T ; (B.18)
k1 ;:::;kn D1

where t D .t1 ; : : : ; tn / and k D .k1 ; : : : ; kn / is a vector whose components are

integer numbers.
The notebook Fourier gives both the real and complex Fourier series of a func-
tion of one or more variables; in addition, it contains many worked-out examples.
References

1. Abraham, R., Marsden, J.E.: Foundation of Mechanics. Benjamin Cummings, Reading, Mass.
(1978)
2. Arnold, V.L.: Equazioni Differenziali Ordinarie. Mir, Moscow (1979)
3. Arnold, V.L.: Metodi Matematici della Meccanica Classica. Editori Riuniti, Rome (1979)
4. Barger, V., Olsson, M.: Classical Mechanics: A Modern Perspective. McGraw-Hill, New York
(1995)
5. Bellomo, N., Preziosi, L., Romano, A.: Mechanics and Dynamical Systems with Mathematica.
Birkhäuser, Basel (2000)
6. Birkhoff, G., Rota, G.C.: Ordinary Differential Equations. Wiley, New York (1989)
7. Bishop, R.L., Goldberg, S.I.: Tensor Analysis on Manifolds. MacMillan, New York (1968)
8. Boothby, W.M.: An Introduction to Differential Manifolds and Riemannian Geometry.
Academic, San Diego (1975)
9. Borrelli, R.L., Coleman, C.S.: Differential Equations: A Modeling Approach. Prentice-Hall,
Englewood Cliffs, NJ (1987)
10. Cercignani, C., Illner, R., Pulvirenti, M.: The Mathematical Theory of Diluite gases. Springer,
Berlin Heidelberg New York (1994)
11. Choquet-Bruhat, Y.: Gèometrie Diffèrentielle et Systèmes Extèrieurs. Dunod, Paris (1968)
12. Chow, T.L.: Classical Mechanics. Wiley, New York (1995)
13. Crampin, M., Pirani, F.: Applicable Differential Geometry. Cambridge University Press,
Cambridge (1968)
14. de Carmo, Manfredo P.: Differential Geometry of Curves and Surfaces. Prentice-Hall, Engle-
wood Cliffs, NJ (1976)
15. Dixon, W.G.: Special Relativity: The Foundations of Macroscopic Physics. Cambridge
University Press, Cambridge (1978)
16. Dubrovin, B.A., Novikov, S.P., Fomenko, A.T.: Geometria delle Superfici, dei Gruppi di
Trasformazioni e dei Campi. Editori Riuniti, Rome (1987)
17. Dubrovin, B.A., Novikov, S.P., Fomenko, A.T.: Geometria e Topologia delle Varietà. Editori
Riuniti, Rome (1988)
18. Dubrovin, B.A., Novikov, S.P., Fomenko, A.T.: Geometria Contemporanea. Editori Riuniti,
Rome (1989)
19. Fasano, A., Marmi, S.: Meccanica Analitica. Bollati Boringhieri, Turin (1994)
20. Fawles, G., Cassidy, G.: Analytical Dynamics. Saunders, Ft. Worth, TX (1999)
21. Flanders, H.: Differential Forms with Applications to the Physical Sciences. Academic,
New York (1963)
22. Fock, V.: The Theory of Space, Time and Gravitation, 2nd edn. Pergamon, Oxford (1964)
23. Goldstein, H.: Classical Mechanics, 2nd edn. Addison-Wesley, Reading, MA (1981)

A. Romano, Classical Mechanics with Mathematica R

, Modeling and Simulation 499
in Science, Engineering and Technology, DOI 10.1007/978-0-8176-8352-8,
© Springer Science+Business Media New York 2012
500 References

24. Goldstein, H., Poole, C., Safko, J.: Classical Mechanics, 3rd edn. Addison Wesley, Reading,
MA (2002)
25. Hale, J., Koçak, H.: Dynamics and Bifurcations. Springer, Berlin Heidelberg New York (1991)
26. Hill, R.: Principles of Dynamics. Pergamon, Oxford (1964)
27. Hutter, K., van der Ven, A.A., Ursescu, A.: Electromagnetic Field Matter Interactions in
Thermoelastic Solids and Viscous Fluids. Springer, Berlin Heidelberg New York (2006)
28. Jackson, E.A.: Perspectives of Nonlinear Dynamic, vol. I. Cambrige University Press,
Cambridge (1990)
29. Jackson, E.A:, Perspectives of Nonlinear Dynamic, vol. II. Cambrige University Press,
Cambridge (1990)
30. Josè, J.V., Saletan, E.: Classical Dynamics: A Contemporary Approach. Cambridge University
Press, Cambridge (1998)
31. Khinchin, A.I.: Mathematical Foundations of Statistical Mechanics. Dover, New York (1949)
32. Landau, L., Lifshitz, E.: Mechanics, 3rd edn. Pergamon, Oxford (1976)
33. MacMillan, W.D.: Dynamics of Rigid Bodies. Dover, New York (1936)
34. Marasco, A.: Lindstedt-Poincarè method and mathematica applied to the motion of a solid with
a fixed point. Int. J. Comput. Math. Appl. 40, 333 (2000)
35. Marasco, A., Romano, A.: Scientific Computing with Mathematica. Birkhauser, Basel (2001)
36. Möller, C.: The Theory of Relativity, 2nd edn. Clarendon, Gloucestershire, UK (1972)
37. Norton, J.D.: General covariance and the foundations of general relativity: eight decades of
dispute. Rep. Prog. Phys. 56, 791–858 (1993)
38. Panowsky, W., Phyllips, M.: Classical Electricity and Magnetism. Addison-Wesley, Reading,
MA (1962)
39. Penfield, P., Haus, H.: Electrodynamics of Moving Media. MIT Press, Cambridge, MA (1967)
40. Petrovski, I.G.: Ordinary Differential Equations. Prentice-Hall, Englewood Cliffs, NJ (1966)
41. Resnick, R.: Introduction to Special Relativity. Wiley, New York (1971)
42. Rindler, W.: Introduction to Special Relativity. Clarendon Press, Gloucestershire, UK (1991)
43. Romano, A., Lancellotta, R., Marasco, A.: Continuum Mechanics using Mathematica,
Fundamentals, Applications, and Scientific Computing. Birkhauser, Basel (2006)
44. Romano, A., Marasco, A.: Continuum Mechanics, Advanced Topics and Research Trends.
Birkhauser, Basel (2010)
45. Saletan, E.J., Cromer, A.H.: Theoretical Mechanics. Wiley, New York (1971)
46. Scheck, F.: Mechanics: From Newton’s Laws to Deterministic Chaos. Springer, Berlin
Heidelberg New York (1990)
47. Stratton, J.A.: Electromagnetic Theory. McGraw-Hill, New York (1952)
48. Synge, J.: Relativity: The Special Theory, 2nd edn. North-Holland, Amsterdam (1964) (Reprint
1972)
49. Synge, J.L., Griffith, B.A.: Principles of Mechanics. McGraw-Hill, New York (1959)
50. Thomson, C.J.: Mathematical Statistical Mechanics. Princeton University Press, Princeton,
NJ (1972)
51. Truesdell, C., Noll, W.: The Nonlinear Field Theories of Mechanics. Handbuch der Physik,
vol. III/3. Springer, Berlin Heidelberg New York (1965)
52. von Westenholz, C.: Differential Forms in Mathematical Physics. North-Holland, New York
(1981)
53. Wang, K.: Statistical Mechanics, 2nd edn. Wiley, New York (1987)
54. Whittaker, E.T.: A Treatise on the Analytical Dynamics of Particles and Rigid Bodies.
Cambridge University Press, Cambridge (1989)
55. Zhang, W-B.: Differential Equations, Bifurcations, and Chaos in Economy. World Scientific,
Singapore (2005)
56. Zhong Zhang, Y.: Special Relativity and Its Experimental Foundations. Advanced Series on
Theoretical Physical Science, vol. 4. World Scientific, Singapore (1997)
Index

Symbols anterior velocity field, 416

4-vector, 477 antiderivation, 105
4 current, 485 antiscalar product, 56
4force, 484 aphelion, 220
4velocity, 484 areal velocity, 167
C h curve, 71 asymptotically stable, 142
C k function, 67, 71 atlas, 68
p-form, 76 attractive, 142
r-chain, 112 automorphism, 12
r-cube, 111 autonomous differential equation, 138
(0,2)-tensor, 20 autoparallel curve, 122
(1,1)-tensor, 23
(2,0)-tensor, 22
B
balance equation of angular momentum, 207
A balance equation of linear momentum, 206
absolute derivative, 191 balance equations, 258
absolute integral invariant, 351 balance equations of impulsive dynamics, 417
absolute motion, 190 Barbashin-Krasovskii, 146
accelerated motion, 165 basis, 5
acceleration, 164 Bernoulli’s theorem, 440
action functional, 320 Blausius formula, 455
action variables, 384 body forces, 434
action-reaction principle, 200 body frame, 179
active forces, 256 body of reference, 161
addition, 3 Boltzmann-Gibbs distribution, 412
adjoint, 65 brachistochrone, 316
affine connection, 120 bracket, 77
algebra, 8
algebraic multiplicity, 53
alternating tensor, 29, 33 C
amplitude, 169 canonical basis, 43, 44
angle, 49 canonical chart, 339
angle variables, 385 canonical form, 43, 44
angle-action variables, 382 canonical parameters, 123
angular momentum, 207, 248 Cauchy problem, 138
angular velocity, 184 center of buoyancy, 439

A. Romano, Classical Mechanics with Mathematica R

, Modeling and Simulation 501
in Science, Engineering and Technology, DOI 10.1007/978-0-8176-8352-8,
© Springer Science+Business Media New York 2012
502 Index

center of mass, 206, 247 covector, 17

central ellipsoid of inertia, 255 curl tensor, 130
central force, 215 curvature tensor, 125
central inversion, 54 cyclic coordinate, 305, 349
central momentum, 251
characteristic equation, 53
characteristic polynomial, 52 D
charge conservation, 473 decelerated motion, 165
chart, 68 degenerate .0; 2/-tensor, 45
chracteristic curves, 489 degrees of freedom, 289
clocks, 161 diffeomorphism, 67
closed form, 108 differential, 75, 78
cochain, 112 dilational waves, 457
codifferential, 79 diophantine condition, 401
coefficient of restitution, 419 direct sum, 8
cohomology class, 108 direction of the future, 478
cohomology relation, 108 directrix of a conic, 218
cohomology space, 108 Dirichlet’s stability criterion, 314
commutative algebra, 8 divergence, 107, 130
compatible chart, 69 domain of attraction, 142
complete integral, 364 doublet, 450
complete vector field, 97 drag force, 211
completely integrable system, 378 dual form, 65
complex potential, 446 duality, 61
complex velocity, 446 dummy index, 5
components, 21, 22 dynamic friction coefficients, 258
compound motion, 170 dynamic Hamiltonian system, 345
compound pendulum, 263 dynamical time, 198
concentrated force, 256 dynamical trajectory, 289
configuration space, 289
configuration space-time, 293
connection coefficients, 120 E
conservative, 301 effective potential, 217
conservative force, 210 eigenvalue, 51
constitutive equation, 435 eigenvalue equation, 51
constraint, 289 eigenvalue problem, 52
constraints, 211 eigenvector, 51
constraints depending on time, 293 eikonal equation, 373
continuity equation, 460 elastic force, 215
contracted multiplication, 26 electromagnetic tensor, 485
contraction, 25 ellipsoid of inertia, 255
contravariant 2-tensor, 22 endomorphism, 12
contravariant components, 7 epimorphism, 12
contravarince law, 19 equation of motion, 162
convective derivative, 428 equilibrium configuration, 313
coordinate transformation, 68 equilibrium solution, 142
coordinates, 68 equipartition of energy, 413
Coriolis force, 211 ergodic, 387
Coriolis’s acceleration, 193 escape velocity, 222
cotangent vector space, 75 ether, 462
Coulomb’s friction laws, 257 Euclidean tensor, 62
covariance law, 19 Euclidean vector, 62
covariant 2-tensor, 20 Euclidean vector space, 48
covariant components, 49 Euler equations, 264
Index 503

Euler’s angles, 185 gradient, 130

Euler-Lagrange equations, 86 gradient of velocity, 429
Eulerian form, 427 gravitational mass, 235
Eulerian model, 197 gyroscope, 269
even function, 496 gyroscopic axis, 269
even permutation, 33 gyroscopic compass, 282
event, 477 gyroscopic forces, 303
exact differential form, 107
exceptional instant, 416
exceptional interval, 415 H
exterior algebra, 37 Hamilton’s principal function, 370
exterior derivative, 105 Hamilton’s principle, 320
exterior product, 29, 34, 36 Hamilton–Jacobi partial differential equation,
external total force, 205 362
Hamiltonian equations, 327
Hamiltonian function, 327, 345
F harmonic function, 131, 446
faces of r-cube, 111 harmonic functions, 432
Fermat’s principle, 317 harmonic motion, 169
fiber, 81 Hausdorff space, 68
fictitious forces, 211 herpolhode, 267
finite dimension, 5 Hodge’s star operator, 65
first cardinal equation of dynamics, 206 holonomic basis, 74
first Helmoltz theorem, 443 holonomic constraints, 289
first integral, 140, 304, 348 Huygens’ principle, 374
focal points, 86
focus of a conic, 218
force, 200 I
force law, 200 ideal constraint, 257
Foucault gyroscope, 282 ignorable coordinate, 305
four-vector, 477 image, 12
Fourier’s coefficients, 495 impulse, 415
Fourier’s series, 495 impulsive forces, 416
free mean path, 404 impulsive motion, 415
free rotations, 265 index, 46
free system of order n, 5 inertial forces, 211
frequency, 169 inertial mass, 235
frequency map, 401 infinite dimension, 5
fundamental frequencies, 386 infinitesimal generator, 96
future, 478 initial condition, 138
initial conditions, 204
initial datum, 138
G initial-value problem, 138
Galilean relativity principle, 205 instantaneous rotation center, 188
gas, 435 integral of a form, 113
general integral, 140 integral of an r-form, 110
generalized momentum, 305 interior product, 101
generalized orthonormal bases, 48 invariant, 100
generalized potential energy, 302 invariant region, 355
generating function, 341 inversion, 33
geodesic, 85 involution, 377
geometric multiplicity, 52 irrotational, 432
globally asymptotically stable, 142 isochoric, 432
globally Hamiltonian vector field, 345 isolated point, 198
504 Index

isomorphism, 12 Lyapunov function, 141, 144

isovolumic, 432 Lyapunov’s direct method, 144

K M
König’s theorem, 208 Mach number, 457
Kepler’s laws, 225 manifold, 68, 69
kernel, 12 mass, 203
kinetic energy, 208, 249 mass conservation principle, 433
kinetic field, 428 mass density, 247,
kinetic potential, 432 433
kinetic theory, 404 mass forces, 256
Kutta–Joukowsky theorem, 455 material point, 197
material volume, 432
matrix of the linear map, 11
L maximal atlas, 69
lab frame, 179 maximal solution, 139
Lagrange bracket, 340 mechanical determinism, 204
Lagrange equations, 300 metric tensor, 83
Lagrange function, 301 Minkowski’s inequality, 47
Lagrange’s multipliers, 311 mixed 2-tensor, 23
Lagrange’s theorem, 440 modulus, 49
Lagrangian components, 300 molecular chaos, 404
Lagrangian components of the impulses, 421 moment of inertia, 254
Lagrangian coordinates, 289 Monge’s cone, 488
Lagrangian equations of impulsive dynamics, monomorphism, 12
421 morphism, 11
Lagrangian form, 427 Mozzi’s axis, 181
Lagrangian velocities, 291 Mozzi’s theorem, 181
Laplace operator, 131
Laplace’s equation, 432, 446
Legendre transformation, 324, 326 N
length, 49, 84 natural atlas, 337
level manifold, 377 natural basis, 74
Levi–Civita symbol, 64 natural charts, 337
Lie derivative, 98 natural coordinates, 81
Lie’s algebra, 9 Newtonian force, 215
lift, 455 Newtonian model, 197
light cone, 478 nondegenerate .0; 2/-tensor, 45
line of nodes, 185 nondegenerate Hamiltonian, 401
linear form, 17 nonholonomic constraints, 292
linear map, 11 normal frequencies, 316
linear momentum, 205, 248 normal modes, 316
linearly dependent, 5 normal speed, 372
linearly independent, 5 null four-vector, 478
Lipschitz condition, 139 nutation, 275
liquid, 435
local time, 161
locally Hamiltonian vector field, 345 O
Lorentz frame, 477 objective function, 201
Lorentz transformation, 477 objective tensor, 200
Lorentz transformation without rotation, odd function, 496
464 odd permutation, 33
Lorentz’s force, 474 one-parameter group of diffeomorphisms, 95
Index 505

one-parameter local group of diffeomorphisms, Q

97 quadratic form, 46
opposite vector, 4 quantization rule, 395
optically isotropic frame, 462 quasi-integrable system, 396
orbit, 95, 138
oriented r-cube, 111
oriented vector space, 38 R
orthochronous, 480 radial velocity, 166
orthogonal, 48 rate of deformation, 429
orthogonal group, 54 reactive forces, 211, 256
orthogonal matrix, 54 real vector space, 4
orthogonal transformation, 53 reduced length, 263
reduced mass, 224
relative derivative, 191
P relative integral invariant, 352
paracompact space, 68 relative motion, 190
parallel along a curve, 121 relativistic energy, 476
parametric equations, 71 relativistic kinetic energy, 476
particle path, 428 relativistic mass, 475
past, 478 resonance module, 386
perihelion, 220 resonance multiplicity, 387
periodic function, 495 rest relativistic energy, 476
permutation, 33 Riemannian connection, 129
perturbed system, 396 Riemannian manifold, 83
phase portrait, 140 rigid body, 162
phase space, 337 rolling friction, 257
planar motion, 188 rolling spin, 257
Poincare, notebook, 150 rotation axis, 182
Poinsot’s motions, 265 rotational motion, 182
point mass, 197 rotations, 54
Poisson bracket, 340 rough constraint, 257
Poisson brackets, 346
polhode, 267
positional force, 210 S
positive definite, 46 scalar product, 47
positive semidefinite, 46 scalar velocity, 163
posterior velocity field, 416 scalars, 4
potential energy, 210, 301 scattering angle, 226
potential, stream or Stokes, 446 Schmidt’s orthonormalization procedure, 50
potential,velocity or kinetic, 446 Schwarz’s inequality, 46
principal frame of inertia, 251 second cardinal equation of dynamics, 207
principal moments of inertia, 251 second Helmholtz theorem, 443
principal steady axes, 269 secular terms, 150
principle of inertia, 198 selection rule, 395
principle of the gyroscopic effect, 279 semispray vector field, 326
principle of velocity composition, 191 separable systems, 364
product, 4, 8 separation method, 364
products of inertia, 250 shock parameter, 226
projection map, 81 signature, 84
proper frame, 474 simple eigenvalue, 52
proper rotations, 54 simple pendulum, 237
proper time, 484 sink, 449
pseudo-Euclidean vector space, 47 skew-symmetric tensor, 29, 33
pseudotensor density, 63 small divisor condition, 401
506 Index

small oscillations, 315 tensor product, 20, 22–24

smooth constraint, 257 terrestrial dynamics, 234
smooth contractible set, 109 theorem of kinetic energy, 259
solenoidal vector field, 442 Thomson–Kelvin theorem, 440
sound velocity, 457 timelike four-vector, 477
source, 449 Torricelli’s theorem, 442
space frame, 179 torsion tensor, 125
space-time, 477 total energy, 305
space-time frame of reference, 161 total torque, 207
spacelike 4-vector, 477 translational, 182
spanned, 8 transport theorem, 433
spatial mean, 387 transverse velocity, 166
special Lorentz transformations, 464
spectrum, 52
speed of approach, 419 U
speed of separation, 419 uniform vector field, 122
spherical motion, 185 unit vector, 8, 48
spin, 429 universal time, 162
spin friction, 257 unperturbed system, 396
squared distance, 84 unstable, 142
stable, 142
stable equilibrium configuration, 313 V
stagnation points, 452 vector field, 76
star-shaped set, 108 vector subspace, 7
state space, 138 vector velocity, 163
static friction coefficients, 258 vectors, 4
statistical mechanics, 404 velocity distribution function,
Stevin’s law, 437 405
stream line, 428 velocity or kinetic potential,
stream tube, 441 445
strength of a source or sink, 449 velocity space, 306, 324
stretching, 429 Venturi’s tube, 443
strict components, 31 virtual displacement, 420
subsonic motion, 457 virtual velocity, 294
subspace generated, 8 vortex line, 441
sum, 4 vortex potential, 447
supersonic motion, 457 vortex tube, 441
symmetric tensor, 41 vortex vector, 441
symmetry group, 306, 349 vorticity tensor, 429
symplectic basis, 57
symplectic chart, 339
symplectic group, 340, 346 W
symplectic manifold, 339 weight, 235
symplectic transformation, 57 work, 209
symplectic transformation group, 58 world line, 483
symplectic vector space, 56 world trajectory, 483
world velocity, 484
T
tangent fiber bundle, 81 Y
tangent vector space, 73 y vector of V , is said to be, 51
temporal mean, 387
tensor algebra, 25
tensor field, 76 Z
tensor of inertia, 250 zero vector, 4