Introduction To Differential Geometry With Tensor Applications

Introduction to Differential
Geometry with Tensor

Applications
Scrivener Publishing
100 Cummings Center, Suite 541J
Beverly, MA 01915-6106
Publishers at Scrivener
Martin Scrivener (martin@scrivenerpublishing.com)
Phillip Carmical (pcarmical@scrivenerpublishing.com)
Introduction to Differential
Geometry with Tensor
Applications
Dipankar De
This edition first published 2022 by John Wiley & Sons, Inc., 111 River Street, Hoboken, NJ 07030, USA
and Scrivener Publishing LLC, 100 Cummings Center, Suite 541J, Beverly, MA 01915, USA
© 2022 Scrivener Publishing LLC
For more information about Scrivener publications please visit www.scrivenerpublishing.com.
All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or
transmitted, in any form or by any means, electronic, mechanical, photocopying, recording, or other-
wise, except as permitted by law. Advice on how to obtain permission to reuse material from this title
is available at http://www.wiley.com/go/permissions.
Wiley Global Headquarters

111 River Street, Hoboken, NJ 07030, USA
For details of our global editorial offices, customer services, and more information about Wiley prod-
ucts visit us at www.wiley.com.
Limit of Liability/Disclaimer of Warranty

While the publisher and authors have used their best efforts in preparing this work, they make no rep
resentations or warranties with respect to the accuracy or completeness of the contents of this work and
specifically disclaim all warranties, including without limitation any implied warranties of merchant-
ability or fitness for a particular purpose. No warranty may be created or extended by sales representa
tives, written sales materials, or promotional statements for this work. The fact that an organization,
website, or product is referred to in this work as a citation and/or potential source of further informa
tion does not mean that the publisher and authors endorse the information or services the organiza
tion, website, or product may provide or recommendations it may make. This work is sold with the
understanding that the publisher is not engaged in rendering professional services. The advice and
strategies contained herein may not be suitable for your situation. You should consult with a specialist
where appropriate. Neither the publisher nor authors shall be liable for any loss of profit or any other
commercial damages, including but not limited to special, incidental, consequential, or other damages.
Further, readers should be aware that websites listed in this work may have changed or disappeared
between when this work was written and when it is read.
Library of Congress Cataloging-in-Publication Data
ISBN 9781119795629
Cover image: Geometry images provided by De

Background image: Abstract Spiral, Fabian Schmidt | Dreamstime.com
Cover design by Kris Hackerott
Set in size of 11pt and Minion Pro by Manila Typesetting Company, Makati, Philippines
Printed in the USA
10 9 8 7 6 5 4 3 2 1
This book is dedicated to the late Dr. G. Suseendran whose inspiration made
this book possible.
v
Contents
Preface xv
About the Book xvii
Introduction 1
Part I: Tensor Theory 7

1 Preliminaries 9
1.1 Introduction 9
1.2 Systems of Different Orders 9
1.3 Summation Convention Certain Index 10
1.3.1 Dummy Index 11
1.3.2 Free Index 11
1.4 Kronecker Symbols 11
1.5 Linear Equations 14
1.6 Results on Matrices and Determinants of Systems 15
1.7 Differentiation of a Determinant 18
1.8 Examples 19
1.9 Exercises 23
2 Tensor Algebra 25
2.1 Introduction 25
2.2 Scope of Tensor Analysis 25
2.2.1 n-Dimensional Space 26
2.3 Transformation of Coordinates in Sn 27
2.3.1 Properties of Admissible Transformation
of Coordinates 30
2.4 Transformation by Invariance 31
2.5 Transformation by Covariant Tensor
and Contravariant Tensor 32
2.6 The Tensor Concept: Contravariant and Covariant Tensors 34
2.6.1 Covariant Tensors 34
vii
viii Contents
2.6.2 Contravariant Vectors 35

2.6.3 Tensor of Higher Order 40
2.6.3.1 Contravariant Tensors of Order Two 40
2.6.3.2 Covariant Tensor of Order Two 41
2.6.3.3 Mixed Tensors of Order Two 42
2.7 Algebra of Tensors 43
2.7.1 Equality of Two Tensors of Same Type 45
2.8 Symmetric and Skew-Symmetric Tensors 45
2.8.1 Symmetric Tensors 45
2.8.2 Skew-Symmetric Tensors 46
2.9 Outer Multiplication and Contraction 51
2.9.1 Outer Multiplication 51
2.9.2 Contraction of a Tensor 53
2.9.3 Inner Product of Two Tensors 54
2.10 Quotient Law of Tensors 56
2.11 Reciprocal Tensor of a Tensor 58
2.12 Relative Tensor, Cartesian Tensor, Affine Tensor,
and Isotropic Tensors 60
2.12.1 Relative Tensors 60
2.12.2 Cartesian Tensors 63
2.12.3 Affine Tensors 63
2.12.4 Isotropic Tensor 64
2.12.5 Pseudo-Tensors 64
2.13 Examples 65
2.14 Exercises 71
3 Riemannian Metric 73
3.1 Introduction 73
3.2 The Metric Tensor 74
3.3 Conjugate Tensor 75
3.4 Associated Tensors 77
3.5 Length of a Vector 84
3.5.1 Length of Vector 84
3.5.2 Unit Vector 85
3.5.3 Null Vector 86
3.6 Angle Between Two Vectors 86
3.6.1 Orthogonality of Two Vectors 87
3.7 Hypersurface 88
3.8 Angle Between Two Coordinate Hypersurfaces 89
3.9 Exercises 95
Contents ix
4 Tensor Calculus 97
4.1 Introduction 97
4.2 Christoffel Symbols 97
4.2.1 Properties of Christoffel Symbols 98
4.3 Transformation of Christoffel Symbols 110
4.3.1 Law of Transformation of Christoffel Symbols
of 1st Kind 110
of 2nd Kind 111
4.4 Covariant Differentiation of Tensor 113
4.4.1 Covariant Derivative of Covariant Tensor 114
4.4.2 Covariant Derivative of Contravariant Tensor 115
4.4.3 Covariant Derivative of Tensors of Type (0,2) 116
4.4.4 Covariant Derivative of Tensors of Type (2,0) 118
4.4.5 Covariant Derivative of Mixed Tensor
of Type (s, r) 120
4.4.6 Covariant Derivatives of Fundamental Tensors
and the Kronecker Delta 120
4.4.7 Formulas for Covariant Differentiation 122
4.4.8 Covariant Differentiation of Relative Tensors 123
4.5 Gradient, Divergence, and Curl 129
4.5.1 Gradient 130
4.5.2 Divergence 130
4.5.2.1 Divergence of a Mixed Tensor (1,1) 132
4.5.3 Laplacian of an Invariant 136
4.5.4 Curl of a Covariant Vector 137
4.6 Exercises 141
5 Riemannian Geometry 143
5.1 Introduction 143
5.2 Riemannian-Christoffel Tensor 143
5.3 Properties of Riemann-Christoffel Tensors 150
5.3.1 Space of Constant Curvature 158
5.4 Ricci Tensor, Bianchi Identities, Einstein Tensors 159
5.4.1 Ricci Tensor 159
5.4.2 Bianchi Identity 160
5.4.3 Einstein Tensor 166
5.5 Einstein Space 170
5.6 Riemannian and Euclidean Spaces 171
5.6.1 Riemannian Spaces 171
x Contents
5.6.2 Euclidean Spaces 174

5.7 Exercises 175
6 The e-Systems and the Generalized Kronecker Deltas 177
6.2 e-Systems 177
6.3 Generalized Kronecker Delta 181
6.4 Contraction of ijk 183
6.5 Application of e-Systems to Determinants and Tensor
Characters of Generalized Kronecker Deltas 185
6.5.1 Curl of Covariant Vector 189
6.5.2 Vector Product of Two Covariant Vectors 190
6.6 Exercises 192
Part II: Differential Geometry 193

7 Curvilinear Coordinates in Space 195
7.2 Length of Arc 195
7.3 Curvilinear Coordinates in E3 200
7.3.1 Coordinate Surfaces 201
7.3.2 Coordinate Curves 202
7.3.3 Line Element 205
7.3.4 Length of a Vector 206
7.3.5 Angle Between Two Vectors 207
7.4 Reciprocal Base Systems 210
7.5 Partial Derivative 216
7.6 Exercises 219
8 Curves in Space 221
8.2 Intrinsic Differentiation 221
8.3 Parallel Vector Fields 226
8.4 Geometry of Space Curves 228
8.4.1 Plane 231
8.5 Serret-Frenet Formula 233
8.5.1 Bertrand Curves 235
8.6 Equations of a Straight Line 252
8.7 Helix 254
8.7.1 Cylindrical Helix 256
8.7.2 Circular Helix 258
8.8 Exercises 262
Contents xi
9 Intrinsic Geometry of Surfaces 265

9.2 Curvilinear Coordinates on a Surface 265
9.3 Intrinsic Geometry: First Fundamental Quadratic Form 267
9.3.1 Contravariant Metric Tensor 270
9.4 Angle Between Two Intersecting Curves on a Surface 272
9.4.1 Pictorial Interpretation 274
9.5 Geodesic in Rn 277
9.6 Geodesic Coordinates 289
9.7 Parallel Vectors on a Surface 291
9.8 Isometric Surface 292
9.8.1 Developable 293
9.9 The Riemannian–Christoffel Tensor and Gaussian
Curvature 294
9.9.1 Einstein Curvature 296
9.10 The Geodesic Curvature 308
9.11 Exercises 319
10 Surfaces in Space 321
10.2 The Tangent Vector 321
10.3 The Normal Line to the Surface 324
10.4 Tensor Derivatives 329
10.5 Second Fundamental Form of a Surface 332
10.5.1 Equivalence of Definition of Tensor bαβ 333
10.6 The Integrability Condition 334
10.7 Formulas of Weingarten 337
10.7.1 Third Fundamental Form 338
10.8 Equations of Gauss and Codazzi 339
10.9 Mean and Total Curvatures of a Surface 341
10.10 Exercises 347
11 Curves on a Surface 349
11.2 Curve on a Surface: Theorem of Meusnier 350
11.2.1 Theorem of Meusnier 353
11.3 The Principal Curvatures of a Surface 358
11.3.1 Umbillic Point 360
11.3.2 Lines of Curvature 361
11.3.3 Asymptotic Lines 362
11.4 Rodrigue’s Formula 376
11.5 Exercises 379
xii Contents
12 Curvature of Surface 381

12.2 Surface of Positive and Negative Curvatures 381
12.3 Parallel Surfaces 383
12.3.1 Computation of aαβ and bαβ 383
12.4 The Gauss-Bonnet Theorem 387
12.5 The n-Dimensional Manifolds 391
12.6 Hypersurfaces 394
12.7 Exercises 395
Part III: Analytical Mechanics 397

13 Classical Mechanics 399
13.2 Newtonian Laws of Motion 399
13.3 Equations of Motion of Particles 401
13.4 Conservative Force Field 403
13.5 Lagrangean Equations of Motion 405
13.6 Applications of Lagrangean Equations 411
13.7 Himilton’s Principle 423
13.8 Principle of Least Action 427
13.9 Generalized Coordinates 430
13.10 Lagrangean Equations in Generalized Coordinates 432
13.11 Divergence Theorem, Green’s Theorem, Laplacian
Operator, and Stoke’s Theorem in Tensor Notation 438
13.12 Hamilton’s Canonical Equations 442
13.12.1 Generalized Momenta 443
13.13 Exercises 444
14 Newtonian Law of Gravitations 447
14.2 Newtonian Laws of Gravitation 447
14.3 Theorem of Gauss 451
14.4 Poisson’s Equation 453
14.5 Solution of Poisson’s Equation 454
14.6 The Problem of Two Bodies 456
14.7 The Problem of Three Bodies 462
14.8 Exercises 467
Contents xiii
Appendix A: Answers to Even-Numbered Exercises 469

References 473
Index 475
Preface
Differential Geometry is the study of geometric properties of curves and

surfaces and their higher dimensional analogues using the methods of ten-
sor calculus. It has a long and rich history and, in addition to its intrin-
sic mathematical value and important connections with various other
branches of mathematics, it has many applications in various physical
sciences. Differential Geometry is a vast subject. A comprehensive intro-
duction would require prerequisites in several related subjects. In this
elementary introductory course, we discuss tensor calculus and its appli-
cations in details and many of the basic concepts of Differential Geometry
in the simpler context of curves and surfaces in ordinary 3-dimensional
Euclidean spaces. Our aim is to build both solid mathematical understand-
ing of the fundamental notions of Differential Geometry and sufficient
visual and geometric intuition of the subject.
We hope that this study is in the interest of researchers from a variety of
mathematics, science, and engineering backgrounds and that Master level
students will also be able to readily study more advanced concepts, such as
properties of curves and surfaces, geometry of abstract manifolds, tensor
analysis, and general relativity.
This book has three parts. The first part contains six chapters dealing
with the calculus of tensors. In the first chapter, we deal with some prelim-
inaries necessary for treatment of the materials in the succeeding chapters.
In the second and third chapters, we discuss the algebra of tensors and
metric tensors. Tensor calculus with Christoffel’s symbols and Riemannian
Geometry are dealt with in the remaining chapters of this section. Part
II contains six chapters discussing the applications of tensor calculus to
geometry, mainly curves and surfaces in three-dimensional Euclidean
space. The remaining two chapters in Part III mainly study Analytical
Mechanics with tensor notation. At the end of each chapter, a large number
of problems have been completely solved and exercises containing care-
fully motivated examples have been incorporated.
xv
xvi Preface
I would like to thank Professor Dr. Hanaa Hachimi and Dr. G. Suseedhran
who gave me the scope and encouragement, with technical support, to
write this book. I am also thankful to Professor Arindam Bhattacherya
from Jadavpur University, India for offering some valuable suggestions for
the improvement of this book.
Lastly, I wish to record my appreciation to the publishers and printers
for their care and effort in bringing out this study in book form.
Dr. Dipankar De
Agartala, India
20th April 2021
About the Book
An Introduction to Differential Geometry with Tensor Applications

discusses the theory of tensor, curves, and surfaces and its application
in Newtonian Mechanics. Since tensor analysis deals with entities and
properties that are independent of choice of reference frames, it forms an
ideal tool for the study of Differential Geometry and classical and celestial
mechanics. This book provides a profound introduction to the basic the-
ory of Differential Geometry: curves and surfaces. The author has tried to
keep the treatment of the advanced material as lucid and comprehensive
as possible mainly by including the utmost detailed calculations, numer-
ous illustrative examples, and a wealth of complementing exercises with
complete solutions making the book easily accessible even to beginners in
the field. Differential Geometry, from the beginning on, characterizes the
author’s utmost efficient didactical approach. This book is one of the out-
standing standard primers of modern Differential Geometry and a basic
source for a profound introductory course or as a highly recommendable
reference for effective self-study of the subject.
xvii
Introduction
Euclidean Geometry was one of the most important branches of mathe-

matics for the last 2000 years. It was also the main tool used by scientists
for the development of astronomy. In 300 B.C., Euclid treated geometry as
a deductive system. In 1621, Sir Henry Savile raised some questions con-
cerning what he called “two blemishes” in geometry: the theory of pro-
portion and the theory of parallels. Euclid’s axiom of parallels (postulate
V in the first book of Elements) propagates that any two given lines in a
plane, when produced indefinitely, will intersect if the sum of two inte-
rior angles made by a transversal with these lines is less than two right
angles. In 1826, a Russian mathematician, Nicolai Lobachevski, presented
a paper to the mathematician faculty of the University of Kazan based on
the assumption that it is possible to draw through any point in a plane two
lines parallel to a given line. Hungarian mathematician John Bolyai pub-
lished some results in 1831, which, conceptually, have little difference from
those of Lobachevski and perhaps it contains a deeper appreciation of the
metric properties of space. However, it was only after Riemann’s profound
dissertation on the hypotheses that the underlying foundations of geome-
try appeared posthumously in 1867, showing the importance of the metric
concepts in geometry.
Riemann appeared to have been unaware of the work of Lobachevski
and Bolyai, although it was well known to Gauss. Later, Beltrami pub-
lished his classical paper on the interpretation of non-Euclidean geome-
tries (1868) in which he analyzed the work of Lobachevski, Bolyai, and
Riemann and stressed the fact that the metric properties of space are mere
definitions. It appears that three consistent geometries are possible on sur-
faces of constant curvatures: Lobachevskian on a surface of constant neg-
ative curvature, Riemannian on a surface of constant positive curvature,
and Euclidean on a surface of zero curvature. These geometries are called
hyperbolic, elliptic, and parabolic, respectively.
Tensor calculus is concerned with the study of abstract objects, called
tensors, which are independent of frames of reference used to describe
them. In the early part of the 17th century, geometry was developed by
French mathematician Descartes. The great physicist W. Voigi discovered
tensors and gave them this name in his remarkable book A Textbook of
Dipankar De. Introduction to Differential Geometry with Tensor Applications, (1–6) © 2022
Scrivener Publishing LLC
1
2 Introduction to Differential Geometry with Tensor
Crystal Physics published in 1910. Ricci and his student, Levi-cita, (1901)
have been developing the subject of tensors, but it is well known that
Einstein’s use of tensors as a tool in his general theory of relativity (1914)
was mainly responsible for the sudden emergence of tensor calculus as a
popular field of mathematical activities. The application of tensors in the
field of mathematics and physics was mainly accelerated after the publica-
tion of Einstein’s famous paper The General Theory of Relativity.
Tensor is a generalization of the term vector and Tensor Calculus is a
generalization of vector analysis. The concept of invariance of mathemat-
ical objects, under coordinate transformations, permeates the structure of
tensor analysis to such an extent that it is important to get at the outset a
clear notion of the particular brand of invariance we have in mind. In the
given reference frame, a point P is determined by a set of coordinates, xi.
If the coordinate system is changed, point P is described by a new set of
coordinates, but the transformation of the coordinates does nothing to the
point itself.
Here, we discuss how the term tensor may be considered as a general-
ization of the term vector.
We start with a two-dimensional Euclidean space, E2, provided with a
system of rectangular Cartesian coordinates. Let P and Q be two points of
E2, with (x1, x2) and (y1, y2) as their respective coordinates. Then, the coor-
dinates of the directed line segment PQ are (x1 − y1, x2 – y2).
Denoting x1 − y1 and x2 – y2 by z1 and z2 respectively, the coordinates of
PQ can be expressed as
zi = xi − yi , i = 1,2. (0.1)

Next, we consider an orthogonal transformation of coordinate axes
given by
x1 = a11 x1 + a12 x 2 + b1
(0.2)
x 2 = a21 x1 + a22 x 2 + b2 ,

 a11 a12 
where   is orthogonal, with its determinant equal to 1.
 a21 a22 
We can express (0.2) as follows:
∑ ∑
2 2
xi = xi = aij x j + bi , (0.3)
j =1 j =1

Introduction 3
denoted by (x1, x2) and (y1, y2), the coordinates of Q and P in the new coor-
dinates system. Then, the coordinates of vector PQ in the new coordinate
system are ( x1 − y1 , x 2 – y 2 )1) and are denoted by ( z1 , z 2 ) . Then PQ can be
expressed as
zi = x i − yi (0.4)
2 2
x i − yi = ∑(a x + b ) − ∑(a y + b )
j =1
ij j i
j =1
ij j i
∑ ∑
2 2
= aij ( x j − y j ) = aij z j
j =1 j =1

Hence, Equation (0.4) can be written as
∑
2
zi = aij z j (0.5)
j =1

∂ xi
∑
2
or zi = zj (0.6)

j =1 ∂xj
∂ xi
[from Equation (0.3) we get aij =
∂xj
The above equation shows that the coordinates of a vector PQ of E2
transform according to a certain law in Equation (0.6) when referring to a
new coordinate system.
This was first pointed out by Felix Klein in 1872.
Similarly, if we consider n vectors, it can be shown that there exists an
object with components C j1 j2 .. jn in a coordinate system, according to
∂ y i1 ∂ y i2 ∂ y in j1 j2 .. jn

C j1 j2 .. jn = ∑ ∂ x j1 ∂ x j2

∂ x jn
C

(0.7)
Hence, we conclude that a tensor of E2 may be regarded as generaliza-

tion of a vector of E2 defined from the transformation law. Similarly, a ten-
sor of En can be obtained as a generalization of a vector of En. A tensor
obtained from the orthogonal transformation of a rectangular Cartesian
coordinate system is called a Cartesian Tensor and a tensor from a general

transformation coordinate system is simply called a tensor.
The flourishing of the subjects of tensors and Differential Geometry and
Mechanics is due to Einstein and Grassman. Then, many mathematicians
and researchers developed Differential Geometry with tensor applications.
Sasaki and Hsu defined and studied almost all contact structures and
their integrability conditions. In 1970, Yano and Okumura studied struc-
ture manifold and Walker, A.G. (1955) studied the properties of the
manifold (λu, v,) with an almost product structure in which there exists
a (1,1) tensor field, f, whose square is unity. K. Yano (1963) generalized
the concept of an almost complex structure and defined f-structures as a
(1,1) tensor field f (satisfying f3 + f = 0). In 1972, K. Kenmotsu studied a
certain class of an almost contact manifold. Janssen and Vanhecke (1981)
named this structure a Kenmotsu structure and the differentiable manifold
equipped with this structure is called a Kenmotsu manifold. Many authors
have studied slant immersions in almost Hermitian manifolds. The study
of Differential Geometry of tangent and cotangent bundles was started by
Sasaki (1958) and then Yano and Davies and Ledger. The theory of subman-
ifolds as a field of Differential Geometry is as old as Differential Geometry
itself. A study of the submanifolds of a manifold is a very interesting field
of Differential Geometry. In 1981, B.Y. Chen, D.E. Blair, A. Bejancu, M.H.
Sahid (1994-95), and some others studied different properties of subman-
ifolds. Sasaki (1960) and others studied differentiable manifolds in detail.
Differential Geometry is the study of geometric properties of curves,
surfaces, and their higher dimensional analogues using the methods of
Tensor Calculus. For the study of curve by this method of calculus, its
parametric representation is a covariant and discuss tangent and normal
and binormal, which is of fundamental importance to the theory of the
curve. We will study the geometric properties of surface imbedded in the
three-dimensional Euclidean space by means of Differential Geometry,
termed as intrinsic properties and intrinsic geometry of surface. The
study of the geometry of surfaces was carried out from the point of view
of a two-dimensional being whose universe is determined by the surface
parameters u1 and u2 and it was based entirely on the study of the first qua-
dratic differential form.
Differential Geometry has a long and rich history and, in addition to
its intrinsic mathematical value and important connections with various
other branches of mathematics, it has many applications in various physi-
cal sciences, e.g., solid mechanics, computer tomography, and general rel-
ativity. Differential Geometry is a building block in Physics and Classical
Mechanics, which was developed extensively by Newton. It deals with the
Introduction 5
motion of particles in a fixed frame of reference. Within those frames, other

coordinate systems may be used so long as the metric remains Euclidean.
The reference system generally used in astronomy is determined by “fixed
stars”. It is termed as the primary inertial system. The motion of the earth
relative to its primary inertial system is so negligible that Newtonian laws
which can be applied without modification to the study of motion of parti-
cles is referred to as a system of axes fixed in the earth.
Part I
TENSOR THEORY
1
Preliminaries
1.1 Introduction
Some quantities are associated with their magnitude and direction, but cer-
tain quantities are associated with two or more directions. Such a quantity
is called a tensor, e.g., the stress at a point of an elastic solid is an example
of a tensor which depends on two directions: one is normal and the other
is that of force on the area. Tensor comes from the word tension.
In this chapter, we discuss the notation of systems of different orders,
which are applied in the theory of determinants, symbols, and summa-
tion conventions. Also, results on some matrices and determinants are
discussed because they will be used frequently later on.
1.2 Systems of Different Orders

Let us consider the two quantities, a1, a1 or a1, a2, which are represented by
ai or ai, respectively, for i = 1, 2. In such cases, the expressions ai, ai, ai j, ai j,
and aij are called systems. In each value of ai and ai are called systems of first
order and each value of ai j, ai j, and aij is called a double system or system
of second order, of which a12, a22a23, a13, and a43 are called their respective
components. Similarly, we have systems of the third order that depend on
three indices shown as ai jk, aikl, ai jm, ai jn, and aijk and each number of their
respective components are 8.
In a system of order zero, it is shown that the quantity has no index,
such as a. The upper and lower indices of a system are called its indices of
contravariance and covariance, respectively. For a system of Akij , i and j are
indices of a contravariant and k is of covariance. Accordingly, the system
Aij is called a contravariant system, Aklm is called a covariant system, and is
Aij called a mixed system.
9
1.3 Summation Convention Certain Index

If in some expressions a certain index occurs twice, this means that this
expression is summed with respect to that index for all admissible values
of the index.
∑
4
Thus, the linear form ai xi has an index, i, occurring in it twice.
i =1
We will omit the summation symbol Σ and write aixi to mean a1x1 + a2x2 +
a3x3 + a4x4. In order to avoid Σ, we shall make use of a convention used by
A. Einstein which is accordingly called the Einstein Summation Convention
or Summation Convention.
Of course, the range of admissible values of the index, 1 to 4 in this case,
must be specified. If the symbol i has a range of values from 1 to 3 and j
ranges from 1 to 4, the expression
aijxj, (i = 1,2,3 and j = 1,2,3,4) (1.1)
represents three linear forms:
a11x1 + a12x2 + a13x3 + a14x4
a21x1 + a22x2 + a23x3 + a24x4
a31x1 + a32x2 + a33x3 + a34x4 (1.2)
Here, index i is the identifying (free) index and since index j, occurs
twice, it is the summation index.
We shall adopt this convention throughout the chapters and take the
sum whenever a letter appears in a term once in a subscript and once in
superscript or if the same two indices are in subscript or are in superscript.
∑ ∑ a uv .
3 3
i j
Example 1.3.1. Express the sum ij
i =1 j =1
= ∑ ∑ a u v = ∑ (a u v + a u v
3 3 3
Solution: aij ui v j ij
i j
i1
i 1
i2
i 2
+ ai 3ui v 3 )
i =1 j =1 i =1
= ∑ a u v +∑ a u v +∑ a u v
3 3 3
i 1 i 2 i 3
i1 i2 i3
i =1 i =1 i =1
= (a11u1v1 + a21u2v1 + a31u3v1) + (a12u1v2 + a22u2v2 + a32u3v2)

+ (a13u1v3 + a23u2v3 + a33u3v3)
Preliminaries 11
1.3.1 Dummy Index

The summation (or dummy) index can be changed at will. Thus, Equation
(1.1) can be written in the form aikxk if k has the same range of values as j.
We will assume that the summation and identifying indices have ranges
of value from 1 to n.
Thus, aixi will represent a linear form
a1x1 + a2x2 + a3x3 + … + anxn
∑ ∑
n n
For example, aik x i x k can be written as aikxixk and here, i
i =1 k =1
and k both are dummy indexes.
So, any dummy index can be replaced by any other index with a range
of the same numbers.
1.3.2 Free Index

If in an expression an index is not a dummy, i.e., it is not repeated twice,
then it is called a free index. For example, for ai jxj, the index j is dummy,
but index i is free.
1.4 Kronecker Symbols

A particular system of second order denoted by δ ij i, j = 1,2,n , is
defined as
δ ij = 1 for i = j
(1.3)
= 0 for i ≠ j

Such a system is called a Kronecker symbol or Kronecker delta.

For example, δ ij , by summation convention is expressed as
δ ii = δ i1 + δ i2 + . + δ in = (δ 11 ) + (δ 22 ) +  + δ nn = 1 + 1 +  + 1 = n +
We shall now consider some properties of this system.

Property 1.4.1. If x1, x2, … xn are independent variables, then
∂x i
= 1 for i = j
∂x j
= 0 for i ≠ j
∂x i i
Hence, j = δj (1.4)
∂x
Property 1.4.2. From the summation convention, we get
δ ii = δ 11 + δ 22 + .. + δ nn = 1 + 1 +  + 1 = n
Similarly, δii = δii = n
Property 1.4.3. From the definition of δi j, taken as an element of unit

matrix I, we have
1 0 .. 0
ij
0 1 .. 0
I ( )
.. .. .. ..
0 0 .. 1
1 0 .. 0
ij
0 1 .. 0
nt is |
Its determinan | 1
.. .. .. ..
0 0 .. 1

Property 1.4.4.
j
Again, δ ij a = δ1i a1 + δ 2i a 2 +  + δ ii ai + .. + δ ni an
= 0 + 0 + 1.ai + 0 +  + .. = ai (1.5)
Preliminaries 13
Similarly, δ ij ai = a j (1.6)
j
δ ij aik = δ 1j a1k + δ 2j a2k + δ j a jk
+ . + δ nj ank = a jk
= 0 + 0…+ 1.a jk + 0.. + 0 = a jk

Similarly, δ ij a jk = aik (1.7)

Property 1.4.5.
δ ijδ kj = δ 1iδ k1 + δ 2i δ k2 + …+ δ niδ kn = 1.δ k1 + δ k2 +  + δ kn = δ ki

∂x i ∂x j ∂x i
Also, by definition, δ ijδ kj = = = δ l1δ ki
∂x j ∂x k ∂x k
In particular, when i = k, we get δ ijδ ij = δ ii = n
Remark 1.4.1. If we multiply xk by δ ki , we simply replace index k of xk with

index i and for this reason, δ ki is called a substitution factor.
Example 1.4.1. Evaluate (a) δ ijδ lj and (b) δ ijδ ljδ kl where the indices take
all values from 1 to n.
Solution: (a) We have δ ijδ lj = δ 1iδ l1 + δ 2i δ l2 +  + δ niδ ln (1.8a)

Now, δ 1iδ l1 = δ 11δ l1 + δ 12δ l1 + . + δ 1nδ l1 = 1.δ l1 = δ l1
Similarly, other terms of i are δ l2 , δ l3 ….δ ln .
∴δ ijδ lj = δ 1iδ l1 + δ 2i δ l2 +  + δ niδ ln = δ l1 + δ l2 + δ l3 + . + δ ln = δ li (1.8b)
(b) δ ijδ ljδ kl = δ liδ kl = δ ki by 1.8b
Example 1.4.2. If xi and yi are independent coordinates of a point, it is

shown that
∂x j ∂ y k
= δ ij
∂ y k ∂x i

Solution: The partial derivative of ϕ in two coordinate systems are different

and are connected by the following formula of Differential Calculus:
∂ y ∂φ ∂ y 1 ∂φ ∂ y 2 ∂φ ∂ y n ∂φ ∂ y k
= 1 i + 2 i + . + n =
∂x ∂ y ∂x ∂ y ∂x ∂ y ∂x i ∂ y k ∂x i
∂x j ∂x j ∂ y k
In particular, when φ = x j , we have = (1.9a)
∂x i ∂ y k ∂x i

∂x j
Since xj is independent of xj, = 0 when j ≠ i
∂x i
= 1 for i = j (1.9b)
Hence, the result follows from (1.9a) and (1.9b).
1.5 Linear Equations

Let us consider n linear equations such that
aijxj = bi, (1.10a)
where x1, x2, …. xn are n unknown variables.

Let us consider:
For the expansion of det |ai j| in terms of cofactors we have
aij A jk = aδ ik (1.10b)
where a = |ai j| and the cofactor of ai j is Ai j.

We can derive Cramer’s Rule for the solution of the system of n linear
equations:
Now, multiplying both sides of (1.10a) by Ai j, we get
Ai jai jxj = biAi j
by (1.10b), we get, axj = biAi j.

From here, we can easily get
Preliminaries 15
bi Aij
xj = , where a = aij ≠ 0.
a
Example 1.5.1. Show that aij Akj = aδ ik , where a is a determinant ai jie a =
|ai j| of order 3 and Ai j are cofactors of ai j.
Solution: By expansion of determinants, we have:
a11A11 + a12A12 + a13A13 = a

a11A2j + a12A2j + a13A2j = 0
a11A31 + a12A32 + a13A33 = 0
Which can be written as a1jA1j = a a1jA2j = 0 and a1jA3j = 0 [we know
aijAij = a].
Similarly, we have a2jA1j = 0 a2jA2j = a and a2jA3j = 0
a3jA1j = 0 a3jA2j = 0 and a3jA3j = a.
Using Kronecker Delta Notation, these can be combined into a single

equation:
a1 j Akj = aδ 1k a2 j Akj = aδ 2k a3 j Akj = aδ 3k [ δ ij = 1, when i = j

= 0 , when j ≠ i ].

All nine of these equations can be combined into aij Akj = aδ ik.
1.6 Results on Matrices and Determinants of Systems

It is known that if the range of the indices of a system of second order are
from 1 to n, the number of components is n2. Systems of second order are
organized into three types: ai j, ai j, aij and their matrices,
(ai j), (ai j), (aij ) :
 a11 a12 … a1n   a11 a12 … a1n   a11 a12  . an1 

 a 21 a 22 … a 2n  ,  a21 a22 …  a2 a2  . a2 
a2n  and  n1 2 n

 an1 an 2 … ann   a a … ann  n n
   n1 n2  a1 a2  . an 
each of which is an n × n matrix.
We shall now establish the following results:
Property 1.6.1. If aij bpj = c ip , then (aij )(bpj ) = (c ip ) and aij bpj = c ip
.
Proof: We shall prove this result by taking the range of the indices from 1
to 2, but the results hold, in general, when they range from 1 to n.
We get aij bpj = a1i b1p + a2i bp2 . Hence, c ip = a1i b1p + a2i bp2 .
 c11 c12 
i 1 i 2
∴  = ( a1 )(bp ) + ( a2 )(bp )
 c12 c22 
 a11 a11   b1 0   a12 a12   b12 0 

1
=   + 2 2  
 a12 a12   0 b21   a2 a2   0 b22 

 a11b11 a11b21   a12b12 a12b22  a11b11 + a12b12 a11b21 + a12b22
=  2 1 2 1 +  2 2 2 2 =  2 1 2 2 2 1 2 2
 a1 b1 a1 b2  a2 b1 a2 b2  a1 b1 + a2 b1 a1 b2 + a2 b2
 a1 a 2  b1 b1

1 1 1 2 i j
=  1   2 2 = (a j )(bp )
 a2 a2 
2 b1 b2

Taking the determinant of both sides, we get aij bpj = c ip , as we know
|AB| = |A||B|.
ik k
Property 1.6.2. If aij bik = c kj , then, (bik )T (aij ) = (c kj ) and aij b = c j ,
k
where (bik)T is the transpose of (c j )
Proof: We have aijbik = a1jbik + a2jbik, hence, c kj = a1 j b1k + a2 j b 2 k .

Therefore,
 a b11 + a b 21 a b11 + a b 21 
 c1 c1 
1 2  11 21 12 22 
 2 2  = 
 c1 c2  12 22 12 22
 a11 b + a 21 b a12 b + a 22 b 
 b11 b 21   a11 a12  ik T

=  12 22    = (b ) (aij)
a
 b b   21 22  a

T
Taking determinants of both sides, we get c kj = bik aij = bik aij (since
│AT│ = │A│).
Preliminaries 17
Property 1.6.3. Let the cofactor of the element aij in the determinant aij
be denoted by Aij. . Then, by summation convention we have
aij Akj = a1i Ak1. + a2i Ak2. +  + ani Akn = δik aij = δik a and
aij A kj = a1j A1k + a 2j A2.k +  + anj Ank = δik aij = δ kj a, where a = aij .

If the cofactor of aij is represented by Akj, it is expressed by the equation:
kj k
aij A = aδ i .
If we divide the cofactor Akj of the element of akj by the value a of the
determinant, we form the normalized cofactor, represented by:
1
b kj = Akj.
a
The above equation becomes
aijb kj = δik
Property 1.6.4. Let us consider a system of n linear equations:
aij x j = bi , i, j = 1,2,n

for n unknown xi, where aij ≠ 0
Aik aij x j = bi Aik , where Aik is cofactor of aij .
bi Aik
ax j = bi Aik ∴ x j = , which is called Cramer’s Rule, for the solution
a
of n linear equations.
Property 1.6.5. Considering the transformation zi = zi(yk) and yi = yi(xk),

∂ zi
let N function zi(yk) be of independent N variables of yk so that ≠ 0.
∂ yk
Here, N equation zi = zi(yk) is solvable for the z’s in terms terms of yi’s.
∂ yi
Similarly, yi = yi(xk) is a solution of yi in terms of xi’s so that ≠ 0.
Now, we have by the chain rule of differentiation that ∂ xk
∂ z i ∂ z i ∂ y1 ∂ z i ∂ y 2 ∂ zi ∂ y N ∂ zi ∂ y j
= + +  . + = .

∂ x k ∂ y1 ∂ x k ∂ y 2 ∂ x k ∂ yN ∂ xk ∂ y j ∂ xk
Taking the determinant, we get
∂z i ∂z i ∂ y j ∂z i ∂ y i (1.11)
= =
∂x k ∂ y j ∂x k ∂ y j ∂x j

Considering a particular case in which zi = xi, Equation (1.5) becomes
∂x i ∂x i ∂ y i ∂x i ∂ y i
= or δik =
∂x k ∂ y j ∂x j ∂ y j ∂x j

Or
xi yi
1
yj xj
xi 1
.
yj yi
xj

This implies that the Jacobian of Direct Transformation is the reciprocal

of the Jacobian of Inverse Transformation.
1.7 Differentiation of a Determinant

Consider the determinant aij = a and let the element aij be a function of
x1, x2 … xn, etc. Let Aij be the cofactor of aij of det a.
Then, the derivative of a with respect to x1 is given by
Preliminaries 19
∂a11 ∂a12 ∂a1 a11 a12 ………an1 a12 a22 ………an2

…… n
∂ x1 ∂ x1 ∂ x1 ∂a12 ∂a22 ∂a 2 a12 a22 ………an2
∂a …… n
= a12 a22 ………an2 + ∂ x1 ∂ x1 ∂ x1 +… ………………. .
∂ x1
………………. ………………. ∂a1n ∂a2n ∂an
…… n
a1n a2n ………ann n n n
a1 a2 ………an ∂ x1 ∂ x1 ∂ x1

a11 1 a21 2 a1 a12 1 a22 2 a2

A1 A1  2 A1n A2 A2  n A2n
x1 x1 x1 x1 x1 x1

a1n 1 a2n 2 an
 An An  n Ann
x1 x1 x1

∂ai1 i ∂ai2 i ∂an ∂a j

= A1 + A2 +…+ . i Ani = i Aij
∂ x1 ∂ x1 ∂ x1 ∂ x1
∂a ∂a j
Therefore, in general, we can write = i Aij .
∂x p ∂x p
1.8 Examples
Example 1.8.1. Write the terms contained in S = aijxixj taking n = 3.
Solution: Since the index i (or j) occurs both in subscript and superscript,
we first sum on i from 1 to 3, then on j from 1 to 3.
S a xx
1j
1 j
a xx2j
2 j
a xx
3j
3 j
(a x x
11
1 1
a xx12
1 2
a x x ) (a x x a x x a x x )
13
1 3
21
2 1
22
2 2
23
2 3
(a x x
31
3 1
a xx32
3 2
a xx)
33
3 3
a (x ) 1 2
a (x ) 2 2
a (x ) (a a )x x (a a )x x (a
3 2 1 2 1 3
a )x x2 3

11 22 33 12 21 13 31 32 23
Example 1.8.2. Express the sum of ∑i3=1 ∑3j =1 ∑3k =1 aijk x i x j x k .

Solution: Here, the number of terms is 33 = 27.
Since the index i (or j or k) occurs both in subscript and superscript, we
first sum on i from 1 to 3, then on each term of its 3 terms we sum j from 1
to 3. This results in 9 terms. Then, on each of the 9 terms we sum k from 1
to 3, which results in 27 terms. Like the last example, we sum
S = (a1jkx1xjxk + a2jkx2xjxk +a3jkx3xjxk))
= (a11k x 1 x 1 x k + a12 k x 1 x 2 x k + a13k x 1 x 3 x k )

+ + (a31k x 3 x 1 x k + a32 k x 3 x 2 x k + a33k x 3 x 3 x k )
= [(a111 x 1 x 1 x 1 + a122 x 1 x 2 x 2 + a133 x 1 x 3 x 3 ) + (a121 x 1 x 2 x 1 + a122 x 1 x 2 x 2 + a123 x 1 x 2 x 3 .)]
+. + ( + a32 k x 3 x 2 x k + [a331 x 3 x 3 x 1 + a332 x 3 x 3 x 2 + a333 x 3 x 3 x 3 )
= [a111 ( x 1 )3 + a222 ( x 2 )3 + a333 ( x 3 )3 + [a123 + a132 + a231 + a213 + a321 + a312 ]x 1 x 2 x 3
+ [a112 + a121 + a211 ]( x 1 )2 x 2 + [a113 + a131 + a311 ]( x 1 )2 x 3 + [a221 + ..]( x 2 )2 x 1
2 2 3 3 2 1 3 2 2
+ [a223 + .]( x ) x + [a331 + .]( x ) x + [a332 + .]( x ) x .
Example 1.8.3. If f is a function of n variables xi, write the differential of f.

Solution: Since f = f(x1, x2, … xn),
∂f ∂y ∂y ∂f
from calculus, we have df = 1 dx 1 + 2 d x 2 +  + n d x n = i d x i .
∂x ∂x ∂x ∂x
Example 1.8.4. (a) If apqxpxq = 0 for all values of the independent variables
x1, x2, … xn and apq‘s are constant, show that aij + aji = 0.
(b) If apqrxpxqxr = 0 for all values of the independent variables x1, x2, … xn
and apqr‘s are constant, show that akij + akji + aikj + ajki + aijk + ajik = 0.
Solution: Differentiating: apq xp xq = 0 (1.12a)
with respect to xi
∂ p ∂
a pq x q p
i ( x ) + a pq x
q
i (x ) = 0
∂x ∂x
q p
or a pq x δ i + a pq x pδ iq = 0 (1.12b)
or a pi x p + aiq x q = 0

Preliminaries 21
Differentiating (1.12b), with respect to xj, we get
∂ p ∂
a pi j ( x ) + aiq (x q ) = 0
∂x ∂xj
or a piδ jp + aiqδ qj = 0
or a ji + aij = 0.

(b) Differentiating apqrxpxqxr = 0

with respect to xi
∂ p r ∂ p q ∂
a pqr x q x r p
i ( x ) + a pqr x x
q
i ( x ) + a pqr x x (x r ) = 0
∂x ∂x ∂ xi
or , a pqr x q x rδ ip + a pqr x p x rδ iq + a pqr x p x qδ ir = 0
or , aiqr x q x r + a pir x p x r + a pqi x p x q = 0

Differentiating with respect to xj, we get
∂ q p ∂ ∂
aiqr r
j ( x )x + a pir x
r
j ( x ) + a pqi ( x p )x q
∂x ∂x ∂xj
∂ ∂ ∂
+ aiqr ( x q ) j ( x r ) + a pir p r
j ( x )( x ) + a pqi ( x )
p
(x q ) = 0
∂x ∂x ∂xj
or aiqr δ qj x r + a pir x pδ rj + a pqiδ jp x q + aiqr ( x q )δ rj + a pir δ jp ( x r ) + a pqi ( x p )δ qj = 0
or aijr x r + a pij x p + a jqi x q + aiqj ( x q ) + a jir ( x r ) + a pji ( x p ) = 0

Differentiating in the same way, with respect to xk we get
or aijk + akij + ajki + aikj + ajik + akji = 0.

Example 1.8.5. If aij is a double system such that aki a kj = δ ij , show that
aij = ±1 .
Solution: We have aki a kj = δ ij , taking determinant aki a kj = δ ij ,
a11 a12  . .an1 a11 a12  . .an1

i k i a12 a22  n a12 a22  n
Or , a a
k j j or ,
 
a1n a2n  . .ann . a1n a2n  . .ann .
1 0 00
0 1 00 i
[sin ce det.Of j is unit det I
 . .
0 0 0 1

2
a11a12  . . an1
a12a22  n 2
Or = 1, ⇒ aij = 1 ⇒ aij = ±1

a1na2n  . . ann

i k i
Example 1.8.6. If aij is a double system such that ak a j = δ j , show that
i i
either ak − δ k = 0 or aki + δ ki = 0.
2
Solution: From above result aij = 1
2 2 2
⇒ aij = δ ij ⇒ aij − δ ij = 0
⇒ ( aij − δ ij )( aij + δ ij ) = 0 ⇒ ( aij − δ ij )( aij + δ ij ) = 0
⇒ either aij − δ ij = 0 or aij + δ ij = 0

Preliminaries 23
a11 a12 a31 b11 a12 b31

i 2 2 2
Example 1.8.7. If a j = a1 a2 a3 and bij = b12 b22 b32 , show that
a13 a23 a33 b13 b23 b33
i i p
aij bmk = c ij where cm = a pam (1.13a)
Solution: aij bmk = aij ekmsb1kb2mb3s [ we know, amk = eijk a1i a2j a3k ] …(α)
1.13b
= (ekms aij )b1kb2mb3s

= (eijt aki amjast )b1kb2mbss [ aij e pqr = eijk aipaqj ark
= eijt (akib1k )(amjb2m )(ast b3s )

= eijt c1i c2j c3t = c ij [ by (i ) and by (α )]
[By (1.13a) and by (1.13b)]

i k i p
The above result can be stated as a j bm = a p am . It is the result of the
multiplication of two determinants of the third order.
1.9 Exercises
1. Write out in full the following expression.
∂ xi j i
b (f) δ ijδ ljδ klδ ik
(a) δ ij a j (b) δijxixj (c) δijδjk (d) δ ii (e) a =
∂ yj
2. Expand the following using the summation convention.
∂
a) AijBjk b) aij x j c) k
( gb k )
3. Prove the following. ∂x
a) δijajk = aik b) δijaij = aik c) δijδjkanm = aim d) δ ijδ jk = δ ik .
∂
4. Show that (a p x p ) = ai for all values of independent
∂ xi
variables, x1, x2, … .xn, and where xp’s are constants.
5. Calculate
∂
a) (aij x j )
∂ xk
∂ i j k
b) p (aijk x x x )
∂x
∂ xr r
6. Using the relation δ s , show that
∂ xs
∂ i j i
p (aij x x ) = (aip + a pi )x
∂x
7. Express each of the following sums using the summation
convention:
(a) ∑ni =1 ∑nj =1 ui v j
(b) ∑ni =1 ∑nj =1 ∑nk =1 aijk x i x j x k
8. Evaluate each of the following (range of indices 1 to n):
(a) δ ij Aikl (b) δ ijδ lk B jl (c) ai a jδ ij (d) δ ijδ ljδ klδ ik
i i q i i q p
9. (a) If x i = aip y p and y = bq z , show that x = aqbp z .
i i q
(b) If x i = aip y p and z = bq z , show that z i = aqpbip y q .
10. If yi are n independent functions of variables xi and zi are
∂ xi
n independent functions of yi and if ui = v j j and
∂ y
∂ yi
v i = w j j (i , j = 1,2,n), then show that ui = w j ∂ j x i .
∂z ∂z
11. If a = aij and bij is a−1 times the cofactor of aij in the determi-
nant of aij , show that bij ali = δ lj .
12. Prove that aij blk = c ij where cli = aipblp.
2
Tensor Algebra
2.1 Introduction
Tensor Analysis is concerned with the study of abstract objects, called ten-
sors, whose properties are independent of the reference frames used to
describe the objects. In a given reference frame, a point is determined by a
set of coordinate systems. If the coordinate system is changed, the point is
described by the new set of coordinates, but the transformation of coordi-
nates does nothing to the point itself.
This chapter is devoted to a discussion of the transformation of coor-
dinate systems. We discuss the concepts of tensors and their algebra
with respect to a certain transformation of coordinates and its law of
transformations.
2.2 Scope of Tensor Analysis

In a given reference frame, a vector, A, is determined uniquely by a set of
components, Ai. If a new coordinate system is introduced, the same vector,
A, is determined by a set of components, Bi, and these new components are
related in a definite way to the old ones. It is the law of transformation of
components of a vector that is the essence of vector ideas and the same is
true of tensors.
Since tensor analysis deals with entities and properties that are indepen-
dent of the choice of reference frame it forms an ideal tool for the study of
natural laws.
The concept of invariance of mathematical objects, under coordinate
transformations, permeates the structure of tensor analysis to such an
extent that it is important to get at outset a clear notion of the particular
brand of invariance we have in mind. We shall suppose that a point is an
invariant. In a given reference frame, a point, P, is determined by a set of
coordinates, xi. If the coordinate system is changed, point P is described
25
by a new set of coordinates, yi, but the transformation of coordinates does

nothing to the point itself. Again, a pair of points (P1, P2.) determines a
vector P1 P2.. This vector, in a particular reference frame, is uniquely deter-
mined by a set of components, Ai. A transformation coordinate does noth-
ing to vector P1 P2., but in the new reference frame P1P2. is characterized by
a different set of components, Bi.
A set of points, such as those forming a curve or surface, is also invari-
ant. The curve may be described in a given coordinate system by an equa-
tion which usually changes its form when the coordinates are changed,
but the curve itself remains unaltered. In general, an object, whatever it
nature, is an invariant provided that it is not altered by a transformation of
coordinates.
2.2.1 n-Dimensional Space

An ordered set of n-real numbers x1, x2, …xn is called an n-tuple of real
numbers and is denoted by (x1, x2, …xn). The set of all n-tuples of real
numbers is said to form an n-dimensional arithmetic continuum and
each n-tuple is called a point of this continuum. Such a continuum
shall be denoted by Sn. Sometimes an Sn is called an n-dimensional space
because it can be endowed with the structure of an n-dimensional linear
space.
The coordinates (x1, x2, …xn) can be assigned to every point in Sn with
respect to a chosen coordinate system establishing a 1-1 correspondence
between the points of Sn and set of all coordinates like (x1, x2, …xn). If (x1,
x2, …xn) are the coordinates of a point P, we shall write xi as the coordinates
of P. The corresponding coordinate system shall be denoted by (xi).
Definition 2.2.1. A space (or manifold) of N dimensions is defined as any

set of objects that can be placed in a 1-1 correspondence with the totality
of ordered sets of N (real or complex) numbers x1, x2, …xn, such that |xi −
Ai| < ki (i = 1,2 … N), where A1 …..AN are constants and k1, k2 …. kN are
real numbers.
If the number xi is real, the N dimensional space is real.
We denote the space of N-dimensions with the symbol VN and we use
the term ‘points’ to denote objects.
Alternately, a set of ‘points’ M is defined to be manifold if
i) each point of M has an open neighborhood and

ii) has a continuous 1 − 1 map onto an open set of n-tuples of
real numbers.
Tensor Algebra 27
There is no implication in these definitions that the concept of distance

between the points has any meaning. If a rule is specified for the measure-
ment of the distance between points, the space VN is called metric.
2.3 Transformation of Coordinates in Sn

Let xi be the coordinates of a point, P, with respect to a coordinate system
and yi be the coordinates of the same point with respect to another coordi-
nate system and let the two systems be related by the equations:
T: yi = yi(x1, x2, … xn), i = 1,2 … n , (2.1)
where values of yi are single valued continuous functions of x1, x2, … xn

denoted by yi(x1, x2, … xn) and have continuous partial derivatives up to
any desired order and further the determinant
∂ y1 ∂ y1 ∂y
1
, 2 ⋅⋅ n
∂x . ∂x . ∂x .
2 2
∂y ∂y ∂y2
,  ⋅⋅
∂ x .1 ∂ x .2 ∂ x .n ≠0 (2.1a)

∂ yn ∂ yn ∂ yn
,  ⋅⋅
∂ x .1 ∂ x .2 ∂ x .n

This determinant is called the Jacobian of transformation (2.1) and is
∂ y .i ∂ y .i
denoted by and the Determinant (Jacobian) J ≡ | does not
∂x . j ∂x . j
vanish at any point of region R, i.e.,
∂φ i
≠ 0.
∂ x. j
In virtue of Equation (2.1), functions yi are independent and Equation
(2.1) can be solved for xi as functions of yi, giving
T−1: x.i = xi(y.1 y.2 … y.n), i = 1,2..n, (2.2)

where the functions1 xi(y) are single valued.
1
we use the notation xi(y) and f(x) as xi(y1,y2,….yn) and f(x1,x2,…xn) respectively.
It would follow then that not only a single valued inverse2 of Equation (2.2)
exists, but the functions xi(y.i) in (2.2) are also of class C1 in some neighborhood
of the point under consideration. We shall refer to a class of coordinate trans-
formations with the properties, i.e., when the class of functions are continuous
together with their first n partial derivatives, as admissible transformations.
Relations (2.1) and (2.2) are called formulas of transformations of coor-
dinates of Sn. They help to determine the coordinates of any point of Sn with
respect to one coordinate system when the coordinates of the same point,
with respect to another coordinate system, are known.
Example 2.3.1. We consider a system of equations specifying the relation

between the spherical polar coordinates xi and the rectangular Cartesian
coordinates yi:
 y 1 = x 1sinx 2cosx 3

T :  y 2 = x 1sinx 2 sinx 3 .

 y 3 = x 1cosx 2

The Jacobian of this transformation is
y1 y1 y1
x1 x 2 x 3 sinx 2cosx 3 x1cos x 2cosx 3 x1sinx 2sinx 3
y2 y2 y2
sinx 2sinx 3 x1cosx 2sinx 3 x1sinx 2cosx 3 (x1 )2 sin x 2 0
x1 x 2 x 3 2 1 2
cosx x sinx 0
y3 y3 y3
os x 3 x1 xx2sin
x1sin 2
x 33
3 1 2 3 1 2 2
inx x sinx cosx (x ) sin x ≠ 00

inx 2 0
If we suppose that x1 > 0, 0 < x2 < π, and 0 ≤ x3 < 2π, then j ≠ 0 and the
inverse transformation is given by
2
we use Cn to denote the class of functions which are continuous and with their first n par-
tial derivatives.
Tensor Algebra 29
x 1 = ( y 1 )2 + ( y 2 )2 + ( y 3 )2

( y 1 )2 + ( y 2 )2
T −1 : x 2 = .
y3
y2
x 3 = tan −1
y1

Example 2.3.2. Show that the equation x1 = 4cosx2 in spherical coordinates

represents a sphere.
Solution: Let yi be catrtesian coordinates and xi be spherical coordinates.
Then,
y 1 = x 1sinx 2cosx 3 y 2 = x 1sinx 2 sinx 3 and y 3 = x 1cosx 2

∴ ( y 1 )2 + ( y 2 )2 + ( y 3 )2 = ( x 1sinx 2cosx 3 )2 + ( x 1 sinx 2 sinx 3 )2 + ( x 1cosx 2 )2
= ( x 1sinx 2 )2 {(sinx 3 )2 + (cosx 3 )2 } + ( x 1cosx 2 )2
= ( x 1sinx 2 )2 + ( x 1cos x 2 )2 = ( x 1 )2
⇒ ( x 1 )2 = ( y 1 )2 + ( y 2 )2 + ( y 3 )2
∴ x 1 = ( y 1 )2 + ( y 2 )2 + ( y 3 )2
y3 y3
cosx 2 = =
x1 ( y 1 )2 + ( y 2 )2 + ( y 3 )2

Putting these values in the given equation, we get
y3
( y 1 )2 + ( y 2 )2 + ( y 3 )2 = 4
( y 1 )2 + ( y 2 )2 + ( y 3 )2
( y 1 )2 + ( y 2 )2 + ( y 3 )2 = 4y 3
( y1 )2 + ( y 2 )2 + ( y 3 − 2)2 = 4,

which is the equation of a sphere in Cartesian coordinates.

2.3.1 Properties of Admissible Transformation of Coordinates

It will be shown that all admissible transformations of coordinates from a
group and, hence, every coordinate system in the family can be obtained
from a particular transformation by admissible transformations.
There is a relation between the Jacobians of Equations (2.1) and (2.2),
the exact nature of which is given below:
Theorem 2.3.1. If J and j′ are the Jacobians of transformations (2.1) and

(2.2), then JJ′ = 1.
Proof: We insert the values of xi from (2.2) in (2.1) and obtain a set of
identities, yi
∴ yi = yi[x1(y1, y2,…yn) … …xn(y1, y2,…yn)].
The differentiation, with respect to yi, gives
∂ yi i ∂ y i ∂x α
= δ j = , α = 1, 2 ,….n,
∂y j ∂x α ∂ y j

∂ y i ∂x α ∂ y i ∂x i
but = ⋅ = j . j ′ . Here, let j and j′ are the Jacobian of T
∂x α ∂ y j ∂x k ∂ y k
and T−1, respectively.
Since δ ij = 1 , we see that j.j′ = 1 and we know that j ≠ 0 in R.
Theorem 2.3.2. The Jacobian of the product transformation is equal to the

product of the Jacobians of transformations entering in the product.
The following are admissible transformations T1 and T2
T1: yi = yi(x1, x2,…xn) and
T2: zi = zi(y1, y2,…yn), i = 1,2, …n
Let transformation T3: zi = zi(y1, y2, …yn) = zi[yi(x1, x2, …xn)] = zi[y1(x1,
x , …xn)] …yn(x1, x2, …xn)], which is the product of T1 and T2.
2
T3 = T1T2
Tensor Algebra 31
∂z i ∂ y ∂ ∂z i ∂ y i
If the Jacobian of T3 is j3, we can write j3 = = = j2 j1,
∂ yα ∂x j ∂ y j ∂x j
where j2 and j1 are the Jacobian of T2 and T1, respectively.
Theorem 2.3.3. The set of all admissible transformations of coordinates

forms a group.
Proof: It is obvious.
i) The product of two admissible transformations is also a trans-

formation belonging to a set of admissible transformations.
Therefore, the closure property of transformation holds.
ii) The product transformation possesses an inverse.
iii) The identity transformation i.e., xi = yi exists.
iv) The associative property is also obvious:
T3(T2 T1) = (T3 T2) T1.
Therefore, the admissible transformation forms a group.
2.4 Transformation by Invariance

Let F(A) be a function of a point, A, in a coordinate system (xi) in n –
dimensional manifold Vn and suppose F(A) is a continuous function. In
reference frame (xi), F(A) is assumed to be form f(x1, x2, …xn). Value F(A)
depends on A, but not the coordinate system used to represent A, so F(A)
is called the scalar function. We introduce a new reference frame (yi) by
means of transformation T:
T: xi = xi(y1, y2, …yn) (2.3)
Function F(A) in (yi) is
f [x1(y1, y2, … yn) … …xn(y1, y2, …yn)] ≡ g(y1, y2, … yn) (2.4)
The value of f(x1, x2, …xn) at A(x1, x2, …xn) is the same e3 as g(y1, y2,
… yn) at A(y1, y2, … yn). Then, f is called an invariant of Vn with respect
In specific case, F(A) may represent the water level of the sea and f(x) is the function of it in
3
the x-reference frame. g(y) is the representation of F(A) in the y-reference frame.
to transformation T if f ≡ g and we call this substitution transformation,

G0: f(x(y)) = g(y), the transformation by invariance.
2.5 Transformation by Covariant Tensor

and Contravariant Tensor
Here, we discuss the Law of Transformation of Entities determined by the
sets of partial derivatives of Scalar. We consider a continuously differentia-
ble function, f(x1, x2, … xn), representing the scalar f(A) and a transforma-
tion of coordinates.
T: xi = xi(y1, y2, …yn) (2.5)
If we form a set of n particle derivatives:
∂f ∂f ∂f
1 , 2 ⋅⋅ (2.6)
∂x ∂x ∂x n
Now, when we substitute into each function f xi(x1, x2, …xn), the values
of x from Equation (2.5), we get
g1(y1, y2, … yn), g2(y1, y2, … yn), ….gn(y1, y2, … yn). (2.7)
It should be noted that the set of functions in (2.6) are not the same as
the functions as (2.7). The partial derivatives
∂f ∂f ∂f
1, 2 ⋅⋅
∂y ∂y ∂ yn

are computed by the rule for the differentiation of composite functions:
∂f ∂ f ∂x α
G1 : = (i,α = 1, 2, … n) (2.8)
∂ y i ∂x α ∂ y i

Tensor Algebra 33
By transformation T1 of function f(x1, x2, …xn), where T1: xi = xi(z1, z2,

…zn), the set of functions corresponding to (2.6) is determined by law G1
∂f ∂ f ∂x α
= .
∂z i ∂x α ∂z i
If we have a set of n functions, A1(x),….An(x), associated with X-

coordinate system, and agree to calculate the corresponding quantities
B1(y),….Bn(y) in y-Systems by means of covariant law G1,
∂x α
Bi ( y ) = Aα ( x ) (2.9)
∂ yi

Set {Ai(x)} represents the component of covariant vector in X-coordinate

system and set {Bi(y)} is represented by the same covariant vector in the
Y-system.
The displacement vectors in the X-system have its components
dx1, dx2 …dxn (2.10)
and the same displacement vector when refers to the Y-system, having
components
dy1, dy2 …dyn, (2.11)

∂ yi α
where G 2 : dy i = dx (i,α = 1, 2 … n).
∂x α
If we have a set of quantities A1(x),…An(x), then the law G2, determining
the corresponding quantities B1(y),….Bn(y), is
∂ yi
Bi ( y ) = α Aα ( x ) (2.12)
∂x
Law G2 is called the contravariant law and the sets of quantities trans-
forms in accordance with components of a contravariant vector.
We now consider covariants, contravariants, and mixed tensors.
2.6 The Tensor Concept: Contravariant

and Covariant Tensors
2.6.1 Covariant Tensors
Definition 2.6.1. Let Ai be a set of n functions of n coordinates, xi, in a
given coordinate system (xi). Then, Ai are said to form the components of a
covariant vector if these components transform according to the following
rule on change of coordinate systems from (xi) to another system (yi):
∂x j (2.13)
Bi = Aj
∂ yi

∂A
It is obvious that the gradient of a scalar A, defined , is an example
of a covariant vector. ∂x i
An example of the gradient of ϕ is denoted by
grad i j k .
x y z xi

From Equation (2.13) we can obtain that the expression of Ai in a coor-
dinate system (xi) in terms of coordinate system (yi) is as follows:
∂ yi
Multiplying both sides of (2.13) by , we obtain
∂x k
yi yi x j
Bi Aj
xk x k yi
i yi x j j
k Ai since k i k
x y
Ak .

∂ yi
Thus, Ak = Bi (Covariant Law) (2.14)
∂x k
Example 2.6.1. If ϕ is a scalar function of coordinates xi, this shows that
∂φ
is a covariant vector.
∂x i
Let us consider the coordinate transformation from xi to yi.
Tensor Algebra 35
i.e., yi = yi(x1, x2, … … xn).
Therefore, ϕ is a function of coordinates xi (i.e.,, x1, x2 …xn) in the first

system and a function of coordinates yi in the second system.
By the Chain Rule:
∂φ ∂φ ∂ x 1 ∂φ ∂ x 2 ∂φ ∂ x n
= + + ⋅ +
∂ y i ∂x1 ∂ y i ∂x 2 ∂ y i ∂x n ∂ y i
∂φ ∂ x j
=
∂x j ∂ y i

∂φ
Here, forms the component of a covariant vector.
∂x j
The covariant vector is also called a covariant tensor of rank one.
A subscript is always used to indicate covariant components and super-
script is always used to indicate contravariant vectors.
2.6.2 Contravariant Vectors

Definition 2.6.2. Let Ai be a set of n functions of n coordinates xi in a given
coordinate system (xi). Then, Ai are said to form the components of a con-
travariant vector if these components transform according to the following
rule on change of a coordinate system from (xi) to another system (yi).
∂ yi j
Bi = A (Contravariant Law)
∂x j
n
∂yi i ∂yi j
i
Therefore, dy = ∑ j =1
∂x j
dx = j dx .
∂x
According to the definition, the quantities of dxi, considered above,
form the components of the contravariant vector.
Example 2.6.2. A covariant tensor has components xy, 2y – z2, and xz

in rectangular coordinates. Find its covariant components in spherical
coordinates.
Solution: Here, x1 = x, x2 = y, x3 = z (Cartesian Coordinate System)
x 1 = r x 2 = θ , x = θ , x 3 = φ (Spherical Coordinate System)

A1 = xy , A2 = 2y – z2 , A3 = xz
According to the law of transformation, we have
xj
Ai A j , (i 1,2 ,3)
xi
and we like to evaluate A1 , A2 , A3 , where A1, A2, and A3, are known.
We know that x = r cos θ sin ϕ
y = r sin θ sin ϕ
z = r sin θ.
∂x1 ∂x 2 ∂x 3
Now, A1 = A 1 + A2 + A3
∂x 1 ∂x 1 ∂x 1
∂x ∂y ∂x 3
= xy + (2 y − z 2 ) + 1 xz
∂r ∂r ∂x
∂ ∂ ∂
= ( xy ) (r cosθ sinφ ) + (2 y − z 2 ) (r sinθ sinφ ) + ( xz ) (r cosθ )
∂r ∂r ∂r

= (rcosθsinϕ.rcosθsinϕ)(cosθsinϕ) + [2r sin θ sin ϕ – (r cos θ)2]
(r sin θ sin ϕ) + (r cos θ sin ϕ)r cos θ . cos θ
== r2 cos.3 θsin3ϕ + 2r2 sin2 θ sin2 ϕ – r3 cos2 θ sin θ sin ϕ + r2cos3
θ sin ϕ
∂x1 ∂x 2 ∂x 3
A2 = A + 2 A2 + 2 A3 =
2 1
∂x ∂x ∂x
r sinθ sinφ cosθ cos φ + 2r 2 sinθ cosθ sin 2 φ − r 3 cos3 θ sinφ − r 3 sin 2 θ cosθ cosφ − .-
3 2
∂x1 ∂x 2 ∂x 3
A3 = A + 3 A2 + . 3 A3 .
3 1
∂x ∂x ∂x

r 3 sin3 sin2 cos r sin cos (2r sin sin r 2 cos2 ) 0

Example 2.6.3. If Ai and Bi are contravariant and covariant vectors, respec-

tively, then the sum of AiBi is an invariant.
∂ x .i j xk
Proof: We have Ai = j A and, Bi Bk ,
∂x xi
Tensor Algebra 37
xi j xk
Hence, Ai Bi A Bk
xj xi
x i xk k
i
Bk A j j Bk A j Ak Bk Ai Bi , since i
j aj ai .
xj x
This implies the proof.
Example 2.6.4. Prove that there is no distinction between contravariant and

covariant vectors if the transformation law is of the form y i = ami x m + bi ,
where a′s and b′s are constants, such that ari ami = δ mr .
Solution: y i = ami x m + bi (2.15a)

i
∂y
We get = ami
∂x m
From (2.15a) ari y i = ari ami x m + ari bi
= δ mr x m + ari bi = x r + ari bi
∴ x r = ari y i − ari bi
∂x r i
∴ i = ar
∂y
∂ yi m i m
Now, the Contravariant Transformation is Ai = A = am A
∂x m
(2.15b)
∂x m m i
Now, the Covariant Transformation is Ai = A = am Am (2.15c)
∂ yi

From (2.15b) and (2.15c), it is shown that there is no distinction between

contravariant and covariant tensor by this law of transformation.
∂ Ai
Example 2.6.5. If Ai is a covariant vector, is determines whether is
a tensor. ∂x
Solution: Since Ai is a covariant vector, we have
∂x k
Ai = Ak (2.16a)
∂ yi

Differentiating both sides with respect to yj,
∂A i ∂2 x k  ∂x k ∂x l  ∂x k ∂x k ∂x l ∂x k ∂2 x k
= A k + = +
 ∂ x l ∂ y j  ∂ y j ∂ y j ∂ y j ∂ x l ∂ y j ∂ y i A k .
∂y j ∂y j ∂y i

(2.16b)
∂ Ai
From (2.16b), it follows that is not a tensor due to the presence of
∂x
a second term on the right hand side of (2.16b).
Example 2.6.6. If aijuiuj = 0 is an invariant for an arbitrary contravariant,

vector ui, show that aij + aji = 0.
Solution: Since aijuiuj is an invariant, we have
aiju iu j aijuaijiuujiu j aijuiu j

x i p x j xq i p x j q
aij u aijq u p u u
yp y y yq
x i x j p qx i x j p q
aij uaij u p q u u
y p yq y y
p q
x x i jx p x q i j
a pqq aupqqu i uu
yi y j y yj
x p x q i xj p x q i j
aij a pq aij a pqu u i 0 j u u 0
yi y j y y

Or, Aiju iu j Or,
0 Aiju iu j 0 (2.17a)

∂x p ∂x q
where Aij = ( aij − a pq ) i (2.17b)
∂y ∂y j
Tensor Algebra 39
and u i is an arbitray, following that
Aij + A ji = 0.
(2.17c)
By (2.17c), we may write (2.17b) as
∂x p ∂x q ∂x p ∂x q
aij − a pq i + a ji − a pq =0
∂y ∂y j ∂ y j ∂ yi

∂x p ∂x q ∂x p ∂x q
aij + a ji = a pq + a pq
∂ y j ∂ yi ∂ yi ∂ y j

(Replacing the dummy indices (p, q) in the second term of the right
hand side by (q, p), respectively)
∂x p ∂x q
= (a pq + aqp ) (2.17d)
∂ yi ∂ y j

From (2.17d), it follows that (apq + aqp) is a covariant tensor of order 2.
Example 2.6.7. Let S be the set of transformation of covariant vectors and T,

T′ be two transformations from system (xi) to system (yi) and from (yi) to
(zi), given by
∂ yi
T : Ai = A p (2.18a)
∂x p
∂z i
T ′: Ai = A p (2.18b)
∂ yr

Then, product transformation T.T′ is
∂ y r ∂z i
T.T ′: Ai = A p , by (2.18a) and (2.18b)
∂x p ∂ y r
∂z i (2.18c)
= Ap
∂x p
It follows that T.T′ ∈ S (2.18d)

Let I be the transformation given by
∂x p
I : Ai = Ai . (2.18e)
∂x p
Then, IT = TI = T
⇒ I is the identity transformation and
I ∈ S (2.18f)
∂x p
From (2.18a), we get A p = Ai (2.18g)
∂ yi
and its transformation is denoted by T1.

Since T1 represents the transformation from (yi) to (xi), it is an inverse
transformation.
T1 ∈ S (2.18h)
Now, we can show easily that if T″ represents a transformation from (zi)

to (yi), then
(T″.T′). T = T″.(T′.T) (2.18i)
Hence, by (2.18d), (2.18f), (2.18h), and (2.18i), it follows that S forms a

group.
2.6.3 Tensor of Higher Order

2.6.3.1 Contravariant Tensors of Order Two
Let Aij be a set of n2 functions of n coordinates, xi is in a given system
of coordinates (xi). Then, Aij are said to form the components of a con-
travariant tensor of order two (or rank 2) if these components transform
according to the following rule on change of coordinate system from (xi)
to another system ( x i ).
ij∂ y i ∂ y j kp (2.19)
B = k A
∂x ∂x p
Tensor Algebra 41
The formula expressing components Aij in a coordinate system (xi), in

terms of those in another system ( x i ), is obtained as follows:
∂x r ∂x s
Multiplying both sides of (2.12) by . . , we get
∂ yi ∂ y j
∂ x r ∂ x s ij ∂ x r ∂ x s ∂ y i ∂ y j kp
. . B = . . A
∂ yi ∂ y j ∂ y i ∂ y j ∂x k ∂x p
∂ x r ∂ y i ∂ x s ∂ y j kp
= . . A
∂ y i ∂x k ∂ y j ∂x p
= δ krδ ps Akp
= Ars .

r s
∂ x ∂ x ij
Hence, Ars = . . B (2.20)
∂x i ∂x j
The following example shows the existence of a contravariant tensor of
second order.
∂ x .i ∂x . j p ∂x i ∂x j p k
We have Ai = k Ak and B j = B , hence A i j
B = B A .
∂x ∂x p ∂x k ∂x p
Now, we write BpAk = Ckp and Ai B j = C ij..
The relation can be written as
∂ x i ∂ x j kp
ij
C = k C
∂x ∂x p
By virtue of (2.20), it follows that Cij or AiB j are the components of a
contravariant tensor of Order 2.
A contravariant tensor of order r may be similarly defined by consider-
ing a system of order r of type Ai1i2……irr..
2.6.3.2 Covariant Tensor of Order Two

Let Aij be a set of n2 functions of n coordinates xi in a given system of coordi-
nates (xi). Then, Aij are said to form the components of a covariant tensor of
order two (or rank 2) if these components transform according to the follow-
ing rule on change of coordinates system from (xi) to another system ( x i ) :
∂x k ∂x p
Bij = i Akp . (2.21)
∂y ∂y j

The existence of such a tensor may be shown by taking two covariant

vectors, Ai and Bj, and proceeding as the above case.
The formula expressing components Aij in a coordinate system (xi), in
terms of those of another system ( x i ), is as follows:
∂x r ∂x s
Aij = Brs (2.22)
∂x i ∂x j
A covariant tensor of order r may be similarly defined by considering a

system of order r of type Ai1i2……ir ..
2.6.3.3 Mixed Tensors of Order Two

Definition 2.5.3. Let Aij be a set of n2 functions of n coordinates xi in a
given system of coordinates (xi). Then, Aij are said to form the components
of a mixed tensor of order two (or rank 2) if these components transform
according to the following rule on a change of coordinates system from (xi)
to another system (yi):
i∂ y i ∂x p k
B = k
j Ap (2.23)
∂x ∂ y j

A mixed tensor of order two sometimes said a mixed tensor of second
order with a first order of contravariance and first order of covariance.
Similarly, Aij in a system (xi), in terms of those in another system (yi),
is as follows:
∂x i ∂ y s r
Aij = As
∂ y r ∂x j

Example 2.6.8. Show that the Kronecker delta is a mixed tensor of order
two.
Solution: If δ ij transforms to j i in coordinate system (yi) by the law for
y j xl m y j xm y j j
mixed tensors of order two, then j i m i l m i i ,
x y x y i
y
hence δ ij is a mixed tensor of order two, having the same components in
every coordinate system.
Tensor Algebra 43
Definition 2.6.4. The sets of nr+s quantities, in the x-coordinate system by

the expressions Ai1j1i2j2........ js
ir ( x ) is a mixed tensor, covariant rank r, and contra-
variant rank s, provided the corresponding quantities Bi1j1i2j2........ js
ir ( y ) in the
y-coordinate system are given by the law:
x 1 x 2 x r y j1 y j1 y j2 y js 1 2 ..... s
Bi1j1i2j2........
ir
js
 . A
y i1 y i2 y ir x 1 x 1 x 2 x s 1 2.  r

(2.24)
From the above structure of formulas, we can deduce an important

theorem:
If all components of a tensor vanish in one coordinate system, then they
necessarily vanish in all other admissible coordinate systems.
Definition 2.6.5. A tensor whose components are all zero in a coordinate

system is called a zero tensor, so we can define the zero tensor as a tensor
whose components are all zero in every coordinate system.
2.7 Algebra of Tensors

The addition and subtraction of two tensors of the same type is a tensor of
the same type.
Theorem 2.7.1. If the components of tensor are all zero in one coordinate
system, then they are also zero in every other coordinate system.
Proof: Let the components Aij11ij22..... ip
..... jq of a tensor of type (p, q) be all zero in
coordinate system (x ) and its components in another coordinate system
i
(yi) by Aij11ij22.....ip
..... jq . Then, by (2.24), we get
i .t
i1i2 .....i p y i1 y i2 y p x t1 x t2 x q r1r2 .....rp
A j1 j2 ..... jq   At t .....t , (2.25)
x r1 x r2 r j
xp y1 y2
j j
y q 12 q

but Atr11tr22 .....

.....rp i1i2 .....i p
tq = 0 and it follows from (2.25) that A j1 j2 ..... jq = 0, hence proved.
A tensor whose components are all zero in every coordinate system is
called a zero tensor.
i i .....i i i .....i
Theorem 2.7.2. If A j11 j22 ..... pjq and B j11 j22 ..... pjq are components of two tensors of
type (p, q), then Aij11ij22.....ip i1i2 .....i p
..... jq + B j1 j2 ..... jq …are
, the components of another tensor
of type (p, q).
Proof: Let Aij11ij22.....ip i1i2 .....i p
..... jq and B j1 j2 ..... jq be two tensors of type (p, q) in coordinate
system (xi) and Aij11ij22.....ip i1i2 .....i p

..... jq and B j1 j2 ..... jq be their components in another coor-
dinate system (yi).
By (2.25) we can write
i .t
i i .....i i i .....i y i1 y i2 y p x t1 x t2 x q i1i2 .....i p i i .....i
A j11 j22 ..... pjq B j11 j22 ..... jpq   A j1 j2 ..... jq B j11 j22 ..... pjq .
x r1 x r2 r j
x p x j1 y 2
j
yq
In (xi) system, Aij11ij22.....ip i1i2 .....i p

..... jq + B j1 j2 ..... jq is the tensor of (p, q), hence proved.
i i .....i i i .....i
Similarly, it can be proven that A j11 j22 ..... pjq − B j11 j22 ..... pjq are another tensor of
type (p, q).
Theorem 2.7.3. If Aij11ij22.....ip

..... jq are the components of a tensor of type (p, q)
and ϕ is a scalar, then φ Aij11ij22..... ip

..... jq are the components of another tensor of
type (p, q).
Proof: Let Aij11ij22.....ip
..... jq be the components of a tensor of type (p, q) in (x ) and
i
i i .....i
A j11 j22 ..... pjq be their components in another coordinate system (yi). Let ϕ and
φ be a scalar in the respective system of coordinates.
Then, we have
i .t
i i .....i y i1 y i2 y p x t1 x t2 x q r1r2 .....rp
A j11 j22 ..... jpq At t .....t
x r1 x r2 r j
xp y1 y2
j j
y q 12 q
i .t
y i1 y i2 y p x t1 x t2 x q r1r2 .....rp
At t .....t , since
x r1 x r2 r j
xp y1 y2
j j
y q 12 q
i t
y i1 y i2 y p x t1 x t2 x q r1r2 .....rp
At1t2 .....tq
x r1 x r2 r j
xp y1 y2
j j
yq
Tensor Algebra 45
i i .....i
It follows that φA j11 j22 ..... pjq are the components of tensor type (p, q).
2.7.1 Equality of Two Tensors of Same Type

Two tensors of the same type are said to be equal in the same coordinate
system if their components in this system are equal to each.
If Aij11ij22.....ip i1i2 .....i p
..... jq and B j1 j2 ..... jq are components of two equal tensors in the same
coordinate system, then we must have Aij11ij22..... ip i1i2 .....i p
..... jq = B j1 j2 ..... jq .
If two tensors are equal in a certain coordinate system, then they are
equal in every coordinate system.
Remark 2.7.1.
(i) If aij AiB j = 0 for two distinct arbitrary vectors, Ai and B j,

then aij = 0.
(ii) The two tensors are said to be equal in a certain coordinate
system if they are of the same type and their correspond-
ing components in the coordinate system are equal.
(iii) The order of the indices of a tensor are important. Consider
tensors Aij and Aji. Aij − Aji is a tensor, but not generally a
zero tensor.
Cases i) Aij = Aji and ii) Aij = −Aji are discussed in the next section.
2.8 Symmetric and Skew-Symmetric Tensors

2.8.1 Symmetric Tensors
Definition 2.8.1. A tensor is said to be symmetric with respect to two con-
travariant (or two covariant) indices if its components remain unchanged
on an interchange of the two indices.
Thus, tensor Aijk is symmetric if Aijk = Ajik = Aikj = Akji for every pair of indices.
The definition of symmetry of tensors obviously would not be satisfac-
tory if the symmetry of its components are not persevered under the trans-
formation of coordinates.
So, it can be easily shown that if a tensor is symmetric with respect to
two covariant or contravariant indices in any coordinate system, then it
remains unchanged in any other coordinate system.
That is, the symmetric property remains unchanged under coordinate

transformation.
2.8.2 Skew-Symmetric Tensors

Definition 2.8.2. A tensor is said to be skew-symmetric (or anti-symmetric)
with respect to a pair of contravariant (or covariant) indices if the compo-
nents change sign on an interchange of the pair of indices.
Thus, tensor Aij is skew-symmetric if Aij = −Aji for every i and j.
ijk jik
In general, the tensor is Alm Alm .
It should be noted that skew-symmetry cannot be defined for a tensor
with respect to two indices of which one is contravariant and the other is
covariant.
This also happens in the case of symmetric tensors, but in an excep-
tional case it is provided.
We have δ ij = δ ij .
Theorem 2.8.1. The components of a tensor of type (0,2) can be expressed as

the sum of a symmetric tensor and a skew-symmetric tensor of the same type.
Proof: Let aij be the components of a tensor of type (0,2).

We can express aij as follows:
1 1
aij = (aij + a ji ) + (aij − a ji )
2 2
= Aij + Bij ,

1 1
where Aij = (aij + a ji ) and Bij = (aij − a ji ).
2 2
Since aij is a tensor of type (0,2), aji is also a tensor of type (0,2). Hence,
both Aij and Bij are tensor of type (0,2). Since Aij = Aji and Bij = −Bji, tensor
Aji is symmetric and Bij is skew-symmetric, hence proved.
Tensor Algebra 47
This is also true for a (2,0) type tensor, i.e., contravariant tensor of the
second order.
Example 2.8.1. Show that:
1
(i) A symmetric tensor of the second order has only n(n + 1)
different components. 2
(ii) A skew symmetric tensor of the second order has only
1
n(n − 1) different non-zero components.
2
Solution:
(i) Let Aij be a symmetric tensor of order two so that Aij = Aji.
If each of the indices i and j take values 1 to n, then Aij will
have n2 components. Out of these n2 components, n com-
ponents A11, A22,… …Ann, and A22 …..Ann are independent.
Thus, the remaining components are (n2 – n), which can be taken in
pairs, since, A12 = A21 , A31 = A13, etc.
Hence, the total number of independent components are

1 1
n + (n 2 − n) = n(n + 1).
2 2
(ii) Let Aij be a skew-symmetric tensor of order two so that Aij
= −Aji. As above, Aij have n2 components. Out of these, n
components A11, A22, … … Ann are all zero. [A11 = −A11]
Omitting these, there are (n2 − n) components which are independent

and can be taken pair wise (ignoring the sign).
Hence, the total number of independent non-zero components is

1 2 1
(n − n) = n(n − 1).
2 2
Example 2.8.2. If Ti is the component of a covariant vector, show that

 ∂Ti ∂Tj 
 j − i  are components of a skew-symmetric covariant tensor of
∂x ∂x
order two.
Solution: Since Ti is a covariant vector, by the covariant law of transfor-
mation,
∂x j
Ti = Tj .
∂ yi

Partially differentiating with respect to yi, we get
Ti xj
Tj
yk yk yi
xj
2
x j Tj (2.26a)
Tj
y k yi yi y k
xj
2
x j x l Tj
Tj
y k yi yi y k xl
∂Tk ∂2 x j ∂ x j ∂ x l ∂Tj
Similarly, = T j + (interchanging i and k).
∂ yi ∂ y k ∂ yi ∂ y k ∂ y i ∂x l
Now interchanging j and l in the 2nd term,
∂Tk ∂2 x j ∂ x j ∂ x l ∂Tl
= Tj + k i (2.26b)
∂ yi ∂ y k ∂ yi ∂ y ∂ y ∂x j

Subtracting (2.26b) from (2.26a),
∂Tl ∂Tk  ∂2 x j ∂ x j ∂ x l ∂Tj 

− =  k i Tj + i
∂ y k ∂ yi  ∂y ∂y ∂ y ∂ y k ∂ x l 
 ∂2 x j ∂ x j ∂ x l ∂T 
−  k i Tj + k i lj 
 ∂y ∂y ∂ y ∂ y ∂x 
∂ x j ∂ x l ∂Tj ∂ x j ∂ x l ∂Tl
= −
∂ y i ∂ y k ∂x l ∂ y k ∂ y i ∂x j
∂ x j ∂ x l  ∂Tj ∂Tl 
= − ,
∂ y i ∂ y k  ∂ x l ∂ x j 
Tensor Algebra 49
which obeys the Law of Covariant Tensor of rank 2.

∂ Ti ∂ Tk
Therefore, − is a covariant tensor of order 2.
∂ y k ∂ yi
∂ Ti ∂ Tk
Now, we have to show − is a skew-symmetric tensor.
∂ Tk ∂ Tk ∂ y k ∂ yi
Let Cik = i − i
∂y ∂y
∂ Tk ∂ Tk  ∂ Tk ∂ Tk 
∴ Cki = i − i = − i − i = −Cik .
∂y ∂y  ∂ y ∂ y 

∂ Tk ∂ Tk
Hence, i − i is a component of a skew-symmetric tensor of 2
nd
order. ∂ y ∂ y
Example 2.8.3. If a tensor Tijk is symmetric in the first two indices from
the left and skew-symmetric in the second and third indices from the left,
show that Tijk = 0.
Solution: Here, Tijk = Tjik. Since it is symmetric, with respect to is two indi-
ces from the left and i and j
= −Tjki, it is skew-symmetric with respect to k& i
= −Tkji, it is symmetric with respect to j &k.
= −(−Tkij), it is skew-symmetric with respect to j& i
= Tkij
= Tikj, it is symmetric with respect to k & i
= −Tijk, it is skew-symmetric with respect to k & j.
∴ Tijk = −Tijk
or 2Tijk = 0
⇒ Tijk = 0
Example 2.8.4.
∂
(a) If aij are constants, calculate (aij x i x j ).
∂ xk
(b) If aij is symmetric or skew symmetric, calculate it.
∂
(c) Find k l (aij x i x j ) if aij is symmetric.
∂x ∂x
∂ ∂ ∂ ∂
Solution: k
(aij x i x j ) = aij k ( x i x j ) = aij x i k ( x j ) + aij x j k ( x i )
∂x ∂x ∂x ∂x
= aij x δ k + aij x δ k = aik x + akj x = a jk x + akj x . (since j is a dummy index)
i j j i i i j j
∂
(b) If ajk is symmetric, then ajk = akj; (aij x i x j ) = 2a jk x j
∂ xk
∂
If ajk is skew symmetric, then ajk = −akj; k
(aij x i x j ) = 0
∂x
∂
(c) From (b), if aij is symmetric, (aij x i x j ) = 2a jk x j .
∂ xk
∂ ∂ ∂
Now, k l
(aij x i x j ) = l (2a jk x j ) = l (2alk x l ) and i is a dummy
index. ∂ x ∂ x ∂ x ∂ x
=2alk
Example 2.8.5. If Tijkl is a tensor which satisfies the relations
Tijkl + Tijlk = 0; ….(i) Tijkl + Tjikl = 0;…..(ii) and Tijkl + Tiklj + Tiljk = 0, …
(iii)
then show that Tijkl = Tklij.

Solution: Putting i = l, l = k, k = j and j = l in (iii), we get
Tlijk + Tljki + Tlkij = 0 (iv)
Now, putting, i = k, j = l, and k = j in (iii), we have
Tkijl + Tkjli + Tklij = 0 (v)
Lastly, putting i = j and j = l in (iii), we get
Tjikl + Tjkli + Tjlik = 0, (vi)
Now, adding (iii), (iv), (v), and (vi), we have by (i) and (ii),
2Tiklj + 2Tljki = 0
or Tiklj = −Tljki
=Tljik (vii)
Now, putting k = j, l = k, j = l in (vii), we get
Tijkl = Tklij.
Tensor Algebra 51
Example 2.8.6. If aij (≠ 0) are the components of a covariant tensor of order

2, such that baij + caji = 0, where b and c are non-zero scalars, show that
either b = c and aij is skew-symmetric or that b = −c and aij are symmetric.
Solution: Given baij + caji = 0, we have baij = −caji ………………(i)
Multiplying both sides of (i) by b, we get
b2aij = −bcaji = −c(baji)
= −c(−aij)
= c2aij
Or (b2 – c2)aij = 0
⇒ b2 – c2 = 0, because aij ≠ 0
or b = ±c.
Case I: When b = c from (i),
baij = −caji = caij

⇒ aji = −aij,
which implies that aji is skew symmetric tensor.
Case II: When b = −c from (i),
baij = −caji = −caij
⇒ aji = aij,
which implies that aji is a symmetric tensor.
2.9 Outer Multiplication and Contraction

2.9.1 Outer Multiplication
If Aij is a contravariant tensor of order two and Bkl is a covariant tensor of
order two, then their product is a mixed tensor Cklij of order four such that
xi xj xr x s
Cklij Aij Bkl A pq Brs
xp xq xk xl
xi xj xr x s pq
A Brs
xp xq xk xl
xi xj xr x s pq
Crs ,
xp xq xk xl
but this is the law of transformation of a mixed tensor of order four.
Therefore, Cklij is a mixed tensor of order four. Such products are called
outer products of two tensors.
Theorem 2.9.1. If Aij11ij22..... ip

..... jq and Bl1l2 .....ls
k1k2 ..... kr
are components of two tensors
of type (p, q) and (r, s), respectively (r and s not being zero), then the
quantities Aij11ij22.....ip k1k2 ..... kr
..... jq and Bl1l2 .....ls are the components of a tensor of type
(p + r, q + s).
The proof can be completed by using (2.25) and proceeding exactly in
the same manner as in the case of the above result.
Example 2.9.1. Show that the product of two tensors Pji and Qtrs are ten-
sors of order five (tensor of type (3,2)).
Solution: By law of covariant and contravariant tensor transformation, we
get
∂ y i ∂x α β ∂ y r ∂ y s ∂ x ρ γδ
Pjl = Pα and Qt
rs
= Qρ
∂x β ∂ y j ∂x γ ∂x δ ∂ y t

∂ y i ∂ x α ∂ y r ∂ y s ∂ x ρ β γδ
l rs
PQ = β
j t Pα Qρ .
∂x ∂ y j ∂x γ ∂x δ ∂ y t

This is the law of transformation of tensor order five, so the product of

the given tensor is of order five.
Example 2.9.2. If Pi and Q j are two contravariant vectors, then prove that
their outer product PiQ j is a tensor of order two, but that the converse is
not true.
Tensor Algebra 53
Solution: We have
∂ yi
Pi = Pk
∂x k
∂ yi
Q j = Ql .
∂x l
Taking on the outer product of Pi, Qj,
∂ yi ∂ yi k l (i)
P iQ j PQ.
∂x k ∂x l
Let Aij = P i Q j and Akl = PkQl,

∂ y i ∂ y i kl
Hence (i) can be written as Aij = A .
∂x k ∂x l
⇒Akl is a contravariant tensor of order 2.
Conversely, let us consider Euclidean space E2 and Aij, defined as
Aij = 1 if i = j
= 0, if i ≠ j.
If possible, let the outer product of Ci and Dj be Aij = CiDj,
where C1D1 = A11 = 1, Þ C1 ≠ 0 (let)
C1D2 = A12 = 0, Þ D2 = 0
C2D2 = A22 = 1,
since D2 = 0⇒C2D2 = 0, but C2D2 = 1,

which contradicts our assumption. Therefore, the converse is not always true.
2.9.2 Contraction of a Tensor

∂x i ∂x j ∂x k ∂x s
Consider a mixed tensor Alijk = .
∂x p ∂x q ∂x r ∂x l
In this, the covariant index l is a covariant index i, so that
∂x i
ijk ∂ x j ∂ x k ∂ x s pqr ∂ x j ∂ x k ∂ x s pqr
A = p
l As = q r As
∂x ∂x q ∂x r ∂x i ∂x ∂x . ∂x p
∂x j ∂ x k s pqr ∂ x j ∂ x k pqr
= q δ p As = q r Ap .
∂x ∂x r ∂x ∂x

This shows that Alijk is a contravariant tensor of order two.

This process of getting a tensor of lower order (reduced by 2) by putting
a covariant index equal to a contravariant index and performing the sum-
mation indicated is known as contraction.
The type of resulting tensor is reduced by 2 and type (3,1) is reduced to
(3-1,1-1)=(2,0), i.e., a contravariant of order 2.
i i .....i
Theorem 2.9.2. If A j11 j22 ..... pjq are the components of a tensor of type (p,q),
p ≠ 0, and q ≠ 0, then the quantities obtained by replacing any upper index
ip and lower index jq by the same index ip and performing summation over
ip are the components of a tensor of type (p – 1, q – 1).
The proof is similar to that of the above result of the theorem.
If, in a mixed tensor of contravariant of order s and covariant of order
r, we equate a covariant and a contravariant index and sum with respect to
that index, then the resulting set of nr+s−2 sums is a mixed tensor covariant
of order r − 1 and contravariant of order s − 1.
2.9.3 Inner Product of Two Tensors

Given the tensors Akij and Bqrp , if we first form their outer product Akij Bqrp
and contract this by putting p = k, then the result is Akij and Bqrk ,, which is
also a tensor called the inner product of the given tensors.
Hence, the inner product of two tensors is obtained by first taking their
outer product and then by contracting it.
Example 2.9.3. Show that any inner product of tensors C ij and Dtrs is a
tensor of rank three.
Solution: Using the transformation laws of tensors for C ij and Dtrs ,
∂ y i ∂x p k ∂ y r ∂ y s ∂ x n lm
C ij = C p (i) and Dt
rs
= Dn . (ii)
∂x k ∂ y j ∂x l ∂x m ∂ y t

The inner product of Cri and Dtrs is
Tensor Algebra 55
∂ y i ∂ x p ∂ y r ∂ y s ∂ x n lm k
Cri Dtrs = Dn C p
∂x k ∂ y r ∂x l ∂x m ∂ y t
∂ y i ∂ y s ∂ x n p lm k
= δ l Dn C p
∂x k ∂x m ∂ y t
∂ y i ∂ y s ∂ x n lm k
= Dn Cl .
∂x k ∂x m ∂ y t

Hence, the inner product of tensors Cri and Dtrs is a tensor of rank 3.
Similarly, putting i = t in the product of (i) and (ii) we get that Cri Dirs is
found to be a tensor of rank 3.
Similarly, in other cases in the same process, we get the tensor of rank 3
by product of these two tensors.
Example 2.9.4. If Pi and Qj are the the components of two contravariant

tensors of rank one respectively, then the n2 quantites PiQ j are the compo-
nents of a contravariant tensor of order two.
Solution: Pi and Q j are the components of a contravariant tensor of rank
one, respectively, by law of transformation
∂ yi α j ∂y j β
Pi = P and Q = Q .
∂x α ∂x β
Multiplying these, we get
∂ y i α ∂ y j β ∂ y i ∂ y j α β ∂ y i ∂ y j αβ
P iQ j = P Q = α P Q = α R ,
∂x α ∂x β ∂x ∂x β ∂x ∂x β
writing PαQβ = Rαβ and P iQ j = R αβ
∂ y i ∂ y j αβ
∴ R αβ = R
∂x α ∂x β
Hence, in virtue of (2.19) it follows that PαQβ or Rαβ are the components
of a contravariant tensor of second order.
2.10 Quotient Law of Tensors

We have seen that the product of two tensors is a tensor. Let us assume that
the product of two quantities is a tensor and if one of them is a tensor, is
the other quantity a tensor?
A simple test is provided by the Quotient Law, which states that if the
inner product of a set of functions with an arbitrary tensor is a tensor, then
these sets of functions are the components of a tensor.
Quotient Law plays an important role in Tensor Calculus and its appli-
cation. The name quotient law is, in a certain sense, appropriate because
the application of this law produces a tensor from two tensors, just as the
operation of the division of two numbers produces a number, namely their
quotient.
Quotient Law for Tensors:
i) For Tensors of type (0,1), if, relative to every system of

coordinates there is a set of functions Bi(i = 1,2, … n), such
that AiBi is an invariant for any tensor Ai of type (1,0), then
Bi will form the components of a tensor of type (0,1).
ii) For tensors of type (1,0), if, relative to every system of
coordinates there is a set of functions Ai(i = 1,2, … n), such
that AiBi is an invariant for any tensor Bi of type (0,1), then
Ai will form the components of a tensor of type (1,0).
iii) For tensors of type (0,2), if, relative to every system of
coordinates there is a set of functions a ik (i, k = 1,2, …
n), such that a ikAiBk is an invariant for any contravariant
tensors Ai and Bk, then a ik will form the components of a
tensor of type (0,2).
iv) For tensors of type (0,2), if relative to every system of coor-
dinates there is a set of functions aik (i, k = 1,2, … n), such
that aik Ai Bk is an invariant for any covariant tensors Ai
and Bk, then aik will form the components of a tensor of
type (2,0).
i i .....i
v) Quotient Law in General Form: Let A j11 j22 ..... pjq be np+q quanti-
ties in a certain reference frame. If Ai11 , Ai22 …. Airr B1j1 B2j2 …..Bsjs
are the components of arbitrary covariant and contravari-
ant vectors, respectively, such that
i1i2⋅⋅⋅⋅⋅⋅ir ir +1⋅⋅i
A j1 j2 ⋅⋅⋅⋅⋅ js js +1 pjq Ai11 Ai22 …. Airr B1j1 B2j2 …..Bsjs (r ≤ p,s ≤ q)

Tensor Algebra 57
i1i2⋅⋅⋅⋅⋅⋅i
are components of a tensor of type (p – r, q – s) and A j1 j2 ⋅⋅⋅⋅⋅ jpq
will form the components of a tensor of type (p, q).
Example 2.10.1. Show that the expression P(i, j, k) is a covariant tensor of

rank three if P(i, j, k)Qk is a covariant tensor of rank two and Qk is contra-
variant vector.
Solution: Let P(i, j, k) be in the coordinate system (xi) and it transformed
into coordinate system (yi).
Given that P(i, j, k)Qk is a covariant tensor of rank two, then
∂x α ∂x β
P (i , j, k)Q k = P(α , β ,γ )Qγ
∂ yi ∂ y j

Since Qk is a contravariant vector,
yk
then Q k Q
x
or Q x
Qk .
yk
Substitute it in the above expression and
∂x α ∂x β ∂x γ k ∂x α ∂x β ∂x γ
P (i , j, k)Q k = P (α , β , γ ) Q = P(α , β ,γ )Q k .
∂ yi ∂ y j ∂yk ∂ yi ∂ y j ∂ y k

∂x α ∂x β ∂x γ
P (i , j, k) = P(α , β ,γ ) since Q k is arbitrary,
∂ yi ∂ y j ∂ y k
hence P(i, j, k) is a covariant tensor of rank three.
Example 2.10.2. Show that the expression P(i, j, k) is a covariant tensor if

its inner product with an arbitrary tensor Qkjl is a tensor.
Solution: Let P(i, j,k )Qkjl = Ti l (i)
If P(i, j, k) is a (0,3) tensor, we consider by contraction and inner prod-

uct, the new tensor to be a (1,1) type.
In the coordinate system (yi), (i) is transformed to become:
P ( p, q, r )Qrqs = Tps (ii)
By Law of Tensor Transformation, we get from (i) and (ii): (expressing

Qrqs and Tps with the Law of Tensor Transformation)
∂ y q ∂ y s ∂ x k jl ∂ y s ∂ x l l
P ( p, q, r ) Qk = i Ti = (iii)
∂x j ∂x l ∂ y r ∂x ∂ y p

s i
∂ y ∂x
Multiplying (i) by and subtracting from (iii), we get
∂x l ∂ y p
 ∂ y q ∂ y s ∂x k ∂ y s ∂ x i  jl
 P ( p,q,r ) j − P(i , j, k) l Qk = 0.
 ∂x ∂x l ∂ y r ∂x ∂ y p 

Since Qkjl is an arbitrary tensor, the expression within the bracket must
be zero,
∂ y q ∂ y s ∂x k ∂ y s ∂x i
implying that P ( p,q,r ) = P(i , j , k)
∂x j ∂x l ∂ y r ∂x l ∂ y p
xi x j yr
P ( p ,q ,r ) P(i , j, k).
y p yq xk
This is the Law of Tensor of Order 3 Transformation, hence P(i, j, k) is a
tensor of order 3 (i, j as covariant indices and k as a contravariant index.).
2.11 Reciprocal Tensor of a Tensor

Let aij be a symmetric tensor of type (0,2), satisfying the condition |aij| ≠ 0.
We denote by bij the cofactor of aij in |aij|, divided by |aij|,
cofactor of aij in |aij |
i.e., bij = (we know aij Aij = |aij| = a).
|aij |
It is known from the Theory of Determinants that
aijbik = 1, when k = j
= 0, when k ≠ j,
hence aij bik = δ jk .

Tensor Algebra 59
Definition 2.11.1. If bij is the reciprocal tensor of aij, then aij is the reciprocal
(or conjugate) tensor of bij. Tensors aij and bij of type (0,2) and (2,0), respec-
tively, are called mutually reciprocal tensors or mutually conjugate tensors if
aij b ki = δ ik .
If bij is the reciprocal tensor of tensor aij, then aij is the reciprocal tensor
of tensor bij.
Since aij b ki = δ ik , it follows that
∴ |aij ||b ki | = |δ jk | ≠ 0.

We define another tensor Gij so that
cofactor of bij in |bij |

Gij = .
|bij |

We know from the property of determinants that Gij bik = δ jk .

Multiplying by apk, we get a pkGij bik = δ jk a pk
i
or Gij p a pj
(2.27)
G pj a pj

Since Gij is the reciprocal tensor of bij and from (2.27) we have to write
aij as the reciprocal tensor of bij, if bij is the reciprocal tensor of aij, then aij is
the reciprocal tensor of bij.
ik k
If the relation aij b = δ j is satisfied, then aij and bij are reciprocal tensors
to each other.
Example 2.11.1. If aij and aij are reciprocal symmetric tensors of order 2,
show that
∂aij ∂ aij
aij + aij = 0.
∂x ∂x
∂(log a) ∂ aij
Hence, show that = −aij .
∂x ∂x
Solution: We have aijaij = n.

Differentiating with respect to xk, we get
∂aij ∂ aij
aij + aij = 0.
∂x ∂x
∂aij ∂ aij
From above we get aij = −aij . (i)
∂x ∂x
If a = |aij| and by definition of reciprocal of a, we get
 1  ∂a cofactor of aij of |aij |  ∂aij 

=  k 
 a  ∂x k a ∂x
∂aij
= aij k , by definition of a
ij
∂x
∂(log a) ∂ aij
or = −aij .
∂x ∂x

2.12 Relative Tensor, Cartesian Tensor, Affine Tensor,

and Isotropic Tensors
2.12.1 Relative Tensors
Definition 2.12.1. A set of np+q components of C ij11ij22….....i pjq ,,in a coordinate system
(xi), transform according to the following formula when referring to another
coordinate system (yi):
i l
i1i2 ...i p ∂ y i1 ∂ y i2 ∂ y p ∂ x l1 ∂ x l2
ω ∂ x q k1k2…k p
C j1 j2 ..... jq =J … ⋅ …. j2 Cl1l2 .....lq (2.28)
∂ x k1 ∂ x k2 ∂ x kp ∂ y j1 ∂ y j2 ∂y

is called a relative tensor of order p+q and weight ω, where J is the Jacobian
∂x
of the transformation: J = .
∂y
If ω = 1, the relative tensor is called a tensor density. If ω = 0, then a ten-
sor is said to be absolute. In the case of a relative on two sides, the equations
must be of the same weight.
A relative tensor of order zero is called a relative scalar. A relative scalar
of weight one is called a scalar density and a relative scalar of weight zero
is called an absolute scalar.
Tensor Algebra 61
Some operations of relative tensors:
(a) Relative tensors of the same type and weight may be added
and the sum is a relative tensor of the same type and weight.
(b) Relative tensors may be multiplied by the weight of the
product, being the sum of the weights of tensors entering
in the product.
(c) The operation of contraction on a relative tensor yields a
relative tensor of the same weight as the original tensor.
We distinguish the mixed tensor from relative tensors and the term
absolute tensor is frequently used to designate the mixed tensor.
Example 2.12.1. Prove that the scalar product of a relative covariant vector
of weight ω1 and a contavariant vector of weight ω2 is a relative scalar vector
of weight ω1 + ω2.
Solution: Let Ai be the components of relative contravariant vector of
weight ω1.
i ω1
r ∂ y ∂x ∂ yi
i
Then, A = A r = Ar r j ω1 .
∂x ∂ y ∂x
When Bi are the components of a relative covariant vector of weight ω2,
then
xs x xs 2
2
Bi Bs Bs j .
yi y yi

Now,
∂ y i ∂ x s ω1 ω 2 ∂x s
Ai Bi = Ar Bs r i j j = Ar Bs r j ω1 +ω 2 = Ar Bsδ rs j ω1 +ω 2 = Ar Br j ω1 +ω 2 .
∂x ∂ y ∂x
It follows that the product is a relative vector of weight ω1 + ω2.
Example 2.12.2. If Aij and Aij are components of relative tensors of weight
w, show that
w −2
ij ∂x
ij
|A |=|A |
∂y
w +2
∂x
|Aij |=|Aij | .
∂y

Solution: By the definition of a relative tensor, we have
w
ij ∂ y i ∂ y j ∂x
pq
A =A (i)
∂x p ∂x ∂ y

w
∂x p ∂x q ∂x
Aij =Apq . (ii)
∂ yi ∂ y j ∂ y

∂ yi ∂y j ∂y
From (i) and (ii), p = = , and taking determinants on both
∂x ∂ xq ∂ x
sides, we get
w
ij ∂ y ∂ y ∂x
pq
|A |=|A |
∂x ∂x ∂ y

w
∂x ∂x ∂x
|Aij |=|Apq | ,
∂y ∂y ∂y

−1
∂ y ∂x
but = and |Apq |=|Aij |.
∂x ∂ y
w−2
∂x
Hence, |Aij |=|Aij | and
∂y
w +2
∂x
|Aij |=|Aij | .
∂y

Example 2.12.3. If aij is a covariant tensor of order 2 and |aij| = a, then
show that a is a relative tensor of order 0 and weight 1.
Solution: By Transformation Law, we get
∂x k ∂x p
aij = akp .
∂ yi ∂ y j

Tensor Algebra 63
Taking determinants on both sides of it, we get
∂x k ∂x p
|aij | = |akp |
∂ yi ∂ y j

2
∂x
a =a
∂y
or
∂x
∴ a= a , (taking the square root on both sides)
∂y

⇒ a is a relative tensor of order 0 and weight 1.
2.12.2 Cartesian Tensors

Definition 2.12.2. A tensor of Euclidian Space En, obtained by orthogo-
nal transformation of coordinate axes, is called a Cartesian Tensor. Thus, a
Cartesian Tensor of rank p in a three-dimensional Euclidean space is a set
of 3p components which transform by the rule
i l
i1i2 ...i p ∂ y i1 ∂ y i2 ∂ y p ∂ x l1 ∂ x l2 ∂ x p k1k2…kp (2.29)
P j1 j2 ..... j p = k1 k2 … kp ⋅ j1 …. j p Pl1l2 .....l p

∂x ∂x ∂ x ∂ y ∂ y j2 ∂y
Under orthogonal transformations,

T : y i = aij x j and (aij ) is orthogonal so that aij ≠ 0.
2.12.3 Affine Tensor

Definition 2.12.3. Tensors corresponding to admissible coordinate changes
by transformation
T : y i = aij x j , (2.30)

i
where a j ≠ 0. A transformation that takes rectangular coordinates (xi) to
an oblique axes system (yi) is called an affine tensor.
Thus, affine tensors are defined in the class of all such oblique coordi-
nate systems.
 ∂ yi 
The Jacobian matrices of T and T−1 are J =  j  = aij rr and
 ∂x i   ∂ x rr
J =  j  = bij rr .
 ∂ y rr
The laws for affine tensors are:
Contravariant tensor: Ai = ali Al , Aij = ali amj Alm

Covariant tensor: Ai = bil Al , Aij = bilbmj Alm (2.31)
Mixed tensor: Aij = ali amj Aml , Aijk = ali amj akn Aml

2.12.4 Isotropic Tensor

Definition 2.12.4. A Cartesian Tensor whose components remain
unchanged under rotation of axes is called an isotropic tensor.
Let ω = (ω1, ω2, ω3) be a vector and P = (bij) an arbitrary orthogonal
transformation.
Let ω = (ω 1 ,ω 2 ,ω 3 ) be the transformed vector under P.
Then, ω = P ω .
Since ω is an isotopic tensor, ω = ω .
If we substitute this in the above expression, we get Pω = ω.
We get Pω – ω = 0 (2.32)
⇒ (P – 1) ω = 0, where 0 is a null vector.
It is clear from Equation (2.32) that it has only one solution: ω = 0. Thus,
there is no isotropic tensor of rank one except the null vector.
2.12.5 Pseudo-Tensor
Definition 2.12.5. Pseudo-tensors are usually discussed in terms of
mechanics.
 a b c d 
In V4 we consider two vectors:  . From these we can form 6
determinants as: p q r s
 
Tensor Algebra 65
a b
aq bp
p q
or cij aib j a jbi .

It is shown in V4 that there are six independent components (determi-
nants) of tensors cij. If we transform it to any coordinate system, consider
the law of transformation of this set of 6 components to the corresponding
set of components in the new system of coordinates. These six components
form a pseudo-tensor.
2.13 Examples
Example 2.13.1. A Covariant Vector has components 2x, y2 – z, and z2 in
rectangular coordinates. Determine its covariant components in cylindri-
cal coordinates.
Solution: Let the rectangular coordinates be (xi) as x1 = x, x2 = y, and
x3 = z.
In (xi) the covariant vector components are A1 = 2x = 2x1, A2 = y2 – z =
(x2)2 – x3, and A3 = z2 = (x3)2.
We know the cylindrical coordinates related with rectangular coordi-
nates as
x1 = y1cosy2, x2 = y1siny2, and x3 = y3, where y1 = r, y2 = θ, and y3 = z

(let) [x1 = rcosθ, x2 = rsinθ, and x3 = z]
∂x j
Now apply the Law of Covariant Transformation: Ai = A j , i = 1, 2 , 3
∂ yi
xj x1 x2 x3
For i 1, Ai Aj A1 A2 A3
y1 y1 y1 y1
2x1cosy 2 {(xx 2 )2 x 3 }siny 2 0 2rcos 2 sin {r 2 sin2 z}

∂x j ∂x1 ∂x 2 ∂x 3
For i = 2, A2 = A j = A1 + A2 + A3
∂y2 ∂y2 ∂y2 ∂y2
= − y 1siny 2 (2rcosθ ) + y 1cosy 2 {(rsinθ )2 − z} + 0
= −rsin θ (2rcosθ ) + rcosθ (r 2 sin 2θ − z )
∂x j ∂x1 ∂x 2 ∂x 3
For i = 3 , A3 = A j = A1 + A2 + A3 = 0 + 0 + 1.z 2 = z 2 .
∂ y3 ∂ y3 ∂ y3 ∂ y3
Example 2.13.2. If aij is a skew-symmetric tensor, prove that

(δ ijδ lk + δ liδ jk )aik = 0.
Solution: Here, aij is a skew symmetric tensor, so aij = −aji.
Now, (δ ijδ lk + δ liδ jk )aik = δ ijδ lk aik + δ liδ jk aik

= δ ij ail + δ li aij
= a jl + alj = a jl + (−a jl ) = 0

= 0 (since aij is a skew-symmetric tensor)
Example 2.13.3. If Bij are components of a Covariant Tensor of second

order and Ci and Dj are components of two contravariant vectors, show
that BijCiDj is an invariant.
Solution: Since Bij are components of a Covariant Tensor of second order
and Ci and D j are components of two contravariant vectors:
∂x k ∂x p i ∂ yi r j ∂ y j s
We have Bij = i Bkp and C = C ,D = s D .
∂y ∂y j ∂x r ∂y
i ∂x k ∂x p
j ∂ yi r ∂ y j s
Now, BijC D = i Bkp r C D
∂y ∂y j ∂x ∂ys
∂x k ∂x p ∂ y i ∂ y j
= BkpC r D s
∂ y i ∂ y j ∂x r ∂ y s
∂x k ∂ y i ∂x p ∂ y j
= i r j B CrDs
s kp
∂ y ∂x ∂ y ∂ y
= δ rkδ sp BkpC r D s
= BrpC r D p = BijC i D j .

It follows that BijCiD j is an invariant.
Example 2.13.4. Show that the determinant of a tensor of type (1,1) is an

invariant.
Solution: Let Aij be a tensor of type (1,1).
Tensor Algebra 67
∂ y i ∂x p k
Then, Aij = Ap .
∂x k ∂ y j

taking matrix on both sides, we get
 ∂ y i  k  ∂x p 
Hence, A = ( )
( j )  ∂x k  Ap  ∂ y j  .
i
(i)

Taking determinants on both sides of i,
i p
i∂ y i p∂ x p
ij ∂=y ∂ y∂kx ∂ x j k Apkk
i
A
= j = k∂ x k j∂ yAjp Ap
A A i
j
∂ x ∂ x∂ y ∂ y
i p
i∂ yi p∂ x p
= Apkk ,, since .jjj and = 1,
. J ′ =JJ1,′′ = 1, where =y ∂ yk and
where jj∂= =x ∂ x
JJ ′′∂=
= A=pk A p since
, since j and and where j= and
∂x k and
J ′ = ∂y j
j
∂ x k∂ x ∂ y j∂ y
= Aii
= A=ij A jj

i
⇒ A j is an invariant.
Example 2.13.5. If Aijk B jk = C i , where Ci is a contravariant vector and B jk

is an arbitrary symmetric tensor, show that Aijk + Aijk is a tensor.
Solution: Let Aijk B jk = C i (i)
∂ y j ∂ y k pq
Now, B jk = B
∂ x p. ∂ x q ..
∂ yi s
and C i = C
∂ x s ..
and (i) can be written as
∂y j ∂yk ∂ yi ∂ y i s pq
Aijk p . q .. B pq = s .. C s = s .. Apq B (since Aijk B jk = C i is given)
∂ x ∂ x ∂ x ∂ x
 i ∂ y j ∂ y k ∂ y i s  pq
or  A jk p . q .. − s .. Apq  B = 0 (ii)
∂x ∂x ∂x
Since B jk is an arbitrary symmetric tensor, let one of the components of

the tensor be non-zero, say Bmn ≠ 0 and the rest be zero. Here, Bmn = Bnm
also.
From (ii) we get
 i ∂ y j ∂ y k ∂ y i s  mn  i ∂ y j ∂ y k ∂ y i s  nm
 A jk m. n.. − s .. Amn  B +  A jk n. m.. − s .. Anm  B = 0
∂x ∂x ∂x ∂x ∂x ∂x
 ∂y j ∂yk i ∂y
j
∂yk ∂ yi s ∂ y i s  mn
or  Aijk + A jk − Amn − Amn  B = 0
 ∂ x m. ∂ x n.. ∂ x n.. ∂ x m.. ∂ x s .. ∂ x s .. 

(iii)
Since Bmn ≠ 0, from (iii) we get
∂y j ∂yk i ∂y
j
∂yk ∂ yi s ∂ yi s
Aijk + A jk = Amn + Anm .
∂ x m. ∂ x n.. ∂ x n. ∂ x m.. ∂ x s .. ∂ x s ..
Replacing the indices j and k by k and j, respectively, from the 2nd term
of left hand side, we get
∂y j ∂yk i ∂y
j
∂ y k ∂ yi s ∂ yi s
Aijk + Akj = A mn + Anm
∂ x m. ∂ x n .. ∂ x m. ∂ x n .. ∂ x s .. ∂ x s ..
i i ∂ y j ∂ y k ∂ yi s s
( A jk + Akj ) m. n .. = s .. ( Amn + Anm ) (iv)
∂x ∂x ∂x

m n
∂x ∂x
Multiplying both sides of (iv) by , we get
∂ y p. ∂ y q..
∂ y j ∂ y k ∂x m ∂x n ∂x m ∂x n ∂ y i
( Aijk + Akji ) = s
( Amn s
+ Anm )
∂ x m. ∂ x n .. ∂ y p . ∂ y q .. ∂ y p . ∂ y q . ∂ x s ..
∂x m ∂x n ∂ y i
or ( Aijk + Akji )δ pjδ qk = p. q ..
s
s .. ( Amn
s
+ Anm )
∂ y ∂ y ∂x
i ∂x m ∂x n ∂ y i
i s s (v)
or A + A = p . q .. s .. ( Amn
pq qp + Anm )
∂ y ∂ y ∂x
Tensor Algebra 69
From (v), it follows that Aijk + Akji is a tensor of type (1,2).
Example 2.13.6. If aijuiuj is an invariant for an arbitrary covariant vector

ui, show that aij + aji is a tensor.
Solution: Since aijuiuj is an invariant, we have
aiju iu j = aijuiu j , (by the property of invariant)
∂x i k ∂x j l
= aij u u (replacing dummy indices i,
∂yk ∂ yl
j, k, l by k, l, i, j, respectively)
xk i xl j
akl u u
yi yj
xk xl i j
akl uu
yi y j
xk xl i j
aij akl uu 0
yi y j
or Biju i u j 0 (i)

 ∂x ∂x 
k l
where Bij =  aij − akl i . (ii)
 ∂ y ∂ y j 

Since u i ′ s are arbitrary, then we can write Bij + B ji = 0 (iii)

In virtue of (ii), (iii) can be expressed as
∂x k ∂x l ∂ x k ∂ xl
aij − akl i + a ji − akl j i = 0
∂y ∂y j ∂y ∂y
∂x k ∂x l ∂x k ∂x l
or aij + a ji = akl + akl (changing indices k and l in the
∂ yi ∂ y j ∂ y j ∂ yi 2nd term)
∂x k ∂x l ∂x k ∂x l
= akl + alk
∂ yi ∂ y j ∂ yi ∂ y j
∂x k ∂x l
= (akl + alk ) i
∂y ∂y j
It follows that aij + aji is a covariant tensor of order 2.
Example 2.13.7. If the relations Aijk uiujuk = 0 hold for any arbitrary contra-
variant vector ui, show that Aijk + Ajki + Akij + Ajik + Aikj + Akji = 0
where Aijk are constants for all i, j, and k.
Solution: Let P = Aijk uiujuk = 0.
It may also be written as P = Apqr upuqur = 0.
Since ur is an arbitrary contravariant vector, we get:
∂ q r p r q p
i P = Aiqr u u + Apir u u + Apqi u u =
0
∂u
∂2
j i
P = Aijr ur + Aiqjuq + A jir ur + Apiju p + A jqiuq + Apjiu p = 0.
∂u ∂u

Again, differentiating with respect to uk, we get
3
P Aijk Aikj A jik Akij A jki Akji 0
u k u j ui
Hence Þ Aijk + Aikj + Ajik + Akij + Ajki + Akji = 0

Tensor Algebra 71
2.14 Exercises
1. iscuss the transformation in which the coordinates yi are rect-
D
angular Cartesian:
y1 = x1cosx2
(a) y2 = x1sinx2
y3 = x3
1 1 2 2 1 3 1 1 1 2 1 3 1 1 1 3
(b) y1 6
x
6
x
6
x , y2
2
x
3
x
3
x , y3
2
x
2
x
2. If f is a scalar function of coordinates xj, then prove that:

∂f
i) is a covariant vector
∂x j
ii) dxj is a contavariant vector
∂ Ai
3. Show that if Ai is a covariant vector, then is not a tensor.
i
∂ xj
4. If the relation x j Ai = 0 holds for an arbitrary covariant vector
Ai, show that x ij = 0.
5. If C ijk is an an arbitrary mixed tensor and B(i, j,k )C ijk an invari-
ant, prove that B(i, j, k) ias a tensor of type Bijk .
6. If Aijk B jk = C i ,, where C i is a contravariant vector and B jk is an arbi-
trary symmetric tensor, show that Aijk + Akji is a tensor. Hence,
deduce that if Aijk is symmetric in j and k, then Aijk is a tensor.
7. If Aijk B jk = C i , where Ci is a contravariant vector and B jk is an
arbitrary skew- symmetric tensor, show that Aijk + Akji is a ten-
sor. Hence, deduce that if Aijk is skew-symmetric in j and k, then
Aijk is a tensor.
8. If aij is a tensor show that Aij, the cofactor of aij in |aij| divided by
|aij| ≠ 0, is a tensor.
9. If the relation aijvivj = 0 holds for an arbitrary covariant vector vi,
show that aij + aji = 0.
10. If aij is a skew-symmetric tensor, prove that (δ ijδ lk + δ liδ jk )aik = 0.
11. If vi ≠ 0 are the components of tensor of type (0,2) and if the
equation
fvij + gvij = 0 holds,

then prove that either f = g and vij is skew-symmetric or f = −g

and vij is symmetric.
 x1 x 2 
12. If a covariant vector has components  2 , 1  in rectangu-
x x 
lar Cartesian coordinates (x1, x2), find its components in polar
coordinates (r, θ).
13. If Bij are components of a Contravariant tensor of second order
and Ci , Dj are components of two covariant vectors, show that
BijCi Dj is an invariant.
14. If Tijkl AiB jAkBl = 0 holds for arbitrary contravariant vectors Ai, B j,
prove that Tijkl + Tkjil + Tilkj + Tklij = 0.
15. If Aklij is skew-symmetric with respect to k and l and if Bij is
defined by the equation Bij = Aklij C kl as a tensor for arbitrary
skew-symmetric tensor Ckl, prove that Aklij is a tensor.
16. If aij is a contravariant tensor such that |aij| ≠ 0, show that |aij| is
a relative invariant of weight −2.
17. If Aij is a skew-symmetric tensor and Bi is a contravariant vector,
then show that Aij Bi B j = 0.
18. If aij is a tensor such that |aij| ≠ 0 and bij is the cofactor of aij in
|aij|, examine whether bij is a relative tensor.
19. Prove that AijBiCj is an invariant if Bi and Cj are vectors and Aij is
a tensor of order 2.
3
Riemannian Metric
3.1 Introduction
In an n-dimensional Euclidean space, En, with a rectangular Cartesian
coordinate system, we can express the square of distance, ds, between the
two points xi and xi + dxi. We have seen that the expression of distance (dxi)
between two points is different for different coordinate systems. On 10th
June 1854, Riemann (1826-1866) submitted his thesis on the hypothesis
of the underlying foundations of geometry for print (published posthu-
mously in 1867), so that the mathematical world could recognize the role
played by the metric (distance) concepts of geometry. This idea of Metric
Geometry (which was later named after himself) generalized all ideas by
assuming that the distance between two neighboring points is independent
of the coordinate system.
In the preceding chapter we discussed algebraic operations on tensors
in Sn. Each of these operations on tensors produce a tensor, but the par-
tial differentiation of a covariant vector does not give a tensor. That is, the
operation of partial differentiation does not always produce a tensor. A
question therefore arises for whether a new type of differentiation can be
defined in Sn, which when applied to a tensor produces another tensor. The
answer to this question is not affirmative unless additional features can be
built into the structure of Sn.
A space which admits an object called an affine connection possesses
a sufficient structure to permit the operation of tensor calculus within
it. Riemannian space is necessarily endowed with an affine connection.
Therefore, for the development of Tensor Calculus, we can consider a
Riemannian space. In Riemannian spaces a new type of differentiation can
be defined simply and in this space, Tensor Calculus has important appli-
cations for physics and engineering, especially in the theory of relativity.
We suppose in the remainder of this chapter that our tensors are defined
in metric manifolds and that the element of arc (ds) is given by the qua-
dratic form ds2 = gij(x)dxidxi, where g ij′ s are functions belonging to C1.
73
We also assume that the symmetric tensor gij(x) is such that |gij| ≠ 0 at any
point of the region under discussion, but do not assume that our manifold
is necessarily Euclidean.
In this chapter we mainly discuss metric tensors or Riemannian metrics
and n-dimensional spaces characterized by this metric, called Riemannian
spaces. We also discuss curvilinear coordinates.
3.2 The Metric Tensor

We introduced the idea of n-dimensional space, En, by extending our
familiar concepts of ordinary Euclidean Geometry. We used the general-
ized formula of Pythagoras, |x| = ( x i x i ), where xi are the components
of vector x, referred to as a set of orthogonal Cartesian axes. Let us con-
sider a displacement (dxi) between the two pair of points A(xi) and B(xi +
dxi), where the coordinates are orthogonal Cartesian, therefore applying
Pythagoras’s formula for the square of the distance AB.
In the expression
ds2 = dxidxi, (3.1)
ds is the element of arc (AB) in En.

Change the coordinate to a new coordinate (yi) by transformation:
xi = xi(y1, … yn). (3.2)
(3.1) can be written as
2∂ xi ∂ xi α β  i ∂ xi α 
∴ ds = α dy dy  since dx = α dy  . (3.3)
∂ y ∂ yβ  ∂y 

Thus, we can write the square of the element arc ds in y – reference

frame as a quadratic form
ds2 = gαβ dyαdyβ, (3.4)
∂ xi ∂ xi
where g αβ ( y ) = α (3.5)

∂ y ∂ yβ
Riemannian Metric 75
This is a function of (yi) and they are obviously symmetric with respect
to indices α and β.
Since the arc distance (ds) is an invariant and product of contravariant
vector dyα and dyβ (each has order one) and by quotient law can be a con-
travariant of order two, the set of functions gαβ(y) is a symmetric tensor.
This is called a metric tensor or first fundamental tensor.
All essential metric properties of a Euclidean space are completely
determined by this tensor.
The space, Sn, endowed with such a structure is called a Riemannian
space1 and is denoted by Vn.
*
Next, we suppose that the tensors are defined in metric manifolds and
the element of arc (ds) is given by the quadratic form ds2 = gij(x) dxidxj,
where the gij are functions belonging to C1.
We also assume that gij(x) is such that |gij| ≠ 0 at any point of the consid-
ered region.
3.3 Conjugate Tensor

Let g be the determinant |gij| and Gij be the cofactor of gij in g.
Define the function of gij by the relation
G ij
g ij = .
g

Since gij and Gij are symmetric in subscripts, the functions gij will be
symmetric in superscripts.
G ij g
g ij g ij g ij 1
g g
lj
G
Here, g ij g lj g ij 1 when l i
g
= 0 and when l ≠ i.
*
1
German mathematician Bernhard Riemann (1826-1866), along with Weierstrass, laid the
foundations of complex analysis. Riemann introduced the concept of integration and
made basic contributions to number theory and mathematical analysis. He developed
Riemannian Geometry which formed the mathematical base for Einstein’s Relativity
Theory.
g lj g ij = δ il
(3.6)
If uj is an arbitrary contravariant tensor, then the inner product with the

tensor gij will be an arbitrary covariant tensor due to contraction,
i.e., gijuj = vi (3.7)
∴ gljvl = gljgljuj = uj,
which is a contravariant tensor of order one.

Therefore, by quotient law, gij are the components of a contravariant ten-
sor of order two.
Hence, gij is a symmetric contravariant tensor which is called the conju-
gate tensor or second fundamental tensor.
In view of (3.6), the relation between gij and gij is reciprocal. As such, first
and second fundamental tensors are also called reciprocal tensors.
Example 3.1.1. Find the conjugate metric tensor in a Riemannian space

V3, in which the distance (ds) is given by
ds2 = 5(dx1)2 + 3(dx2)2 + 4(dx3)2 – 3dx1dx2 – 3dx2dx1 + 2dx2dx3 + 2dx3dx2.
Solution: Here, g11 = 5, g22 = 3, g33 = 4, g12 = g21 = −3, g32 = g23 = 2, g13 = g31 = 0,
5 −3 0
therefore g = −3 3 2 =4
0 2 4
cofactor of g11 in g 8 6 3
g 11 = = = 2 , similarly, g 33 = = ,
g 4 4 2
5 13 3
g 12 = 3, g 23 = − g =−
2 2
cofactor g 22 in g 20
g 22 = = =5
g 4

3
2 3 −
2
5
g ij = 3 5 − .
2
3 5 3
− −
2 2 2

3.4 Associated Tensors

We have the fundamental tenors gij and gij, which allow us a number of new
combinations.
Let uj and vj be a contravariant and a covariant tensor. We now define
two tensors ui and vi as follows:
ui = gij uj (3.8)
vi = gijvj (3.9)
The inner product of tensor uj with fundamental tensor gij is another

covariant tensor u i, which is called the associated tensor of uj.
Similarly, we have vi = gijvj.
Hence, vi is the associated tensor of vj.
Thus, the indices of any tensor can be lowered or raised by forming its
inner product with either of the fundamental tensors gij or gij.
Thus, gijuj = gij gjl ul = δ li ul = ui (3.10)
Now, from (3.10), it follows that the associate to ui is ui.

Thus, if ui is the associate to ui, then ui is the associate to ui. Hence, ui and
ui are mutually associative and are associate tensors.
Definition 3.4.1 A tensor obtained by the process of inner multiplication

of any tensor Aij11…ir
… js with either of the fundamental tensors gij or g is called
ij
a tensor associated with the given tensor.

In Sn, the scalar product of two vectors of opposite variances could only
be formed as AiBi, but in Vn it is possible to extend as
AiBi = gij AjBi = AjBj = gijAiBj.
The above formula enables us to introduce the notion of length or mag-

nitude of a vector in Vn.
The procedure of raising and lowering indices is clearly reversible. The
position occupied by the raised (or lowered) index is indicated by a dot. In
general, such systems as g α i A jα = Aij . and g iα Aα j = A.ij are different. They
are identical when Aij = Aji because they are symmetric tensors.
Theorem 3.4.1. The Metric tensor gij is a covariant symmetric tensor of

rank two.
Proof: The metric is given by
ds2 = gij(x) dxidxj (3.11a)
Let xi be the coordinates in X-coordinate system and yi be the coordi-

nates in Y-coordinate system.
The metric ds2 = gij dxidxj transforms to ds 2 = g ij dy i dy j and since dis-
tance is invariant,
ds 2 = g ij dx i dx j = g ij dy i dy j . (3.11b)

Step 1. We have to show that dxi is a covariant vector.
If yi = yi(x1, x2, …xn)
∂∂ yyii 11 ∂∂ yyii 22 ∂∂ yyii nn

∴ dyiii == dy
∴dy dyii((xx11,x …xxnn))==
,x22,,… dx
dx +
+ dx
dx +
+ 
..++ dx
dx
∂∂xx11 ∂∂xx22 ∂∂xxnn
∂ yi k
= dx , which is the Law of Contravariant Vectors.
∂x k
Step 2. We have to show that gij is a covariant tensor of rank two.
∂ yi k j ∂yj l
dy i = dx and dy = dx
∂ xk ∂ xl
∂ yi k ∂ y j l
From (3.11b), g ij dx dx = g ij
i j
dx dx
∂ xk ∂ xl
∂ yi ∂ y j k l
= g ij k dx dx
∂ x ∂ xl
∂ yi ∂ y j k l
g kl dx k dx l = g ij k dx dx
∂ x ∂ xl
j
 ∂y ∂y  k l
i
or  g kl − g ij k  dx dx = 0
∂ x ∂ xl 
∂ yi ∂ y j
or g kl − g ij k
k l
l = 0, since dx and dx are arbitrary,
∂x ∂x
or ∂ yi ∂ y j
g kl − g ij .
∂ xk ∂ xl
∂ xk ∂ xl
Therefore, g ij = g kl and gij is a covariant tensor of rank two.
∂ yi ∂ y j
Step 3. We have to show that gij is a symmetric tensor.
Let gij = Aij + Bij.
1 1
Let Aij = (aij + a ji ) and hij = (aij − a ji ).
2 2
Here, Aij = Aji as a symmetric tensor and Bij = −Bji as a skew-symmetric
tensor.
Since Bij is a skew-symmetric tensor, implying that Bij = 0, therefore, gij
is a symmetric tensor.
Hence, metric tensor gij is a covariant symmetry tensor of rank two.
Theorem 3.4.2. Prove that gij dxidxj is invariant.

Proof: Let xi be the coordinates in X-coordinate system and yi be the coor-
dinates in Y-coordinate system of a point.
Since gij is a covariant tensor of rank two,
∂x k ∂x l
then g ij = g kl
∂ yi ∂ y j

k l
∂x ∂x
or g ij − g kl = 0,
∂ yi ∂ y j

 ∂x k ∂x l 
or  g ij − g kl dx i dx j = 0 (since dx i , dx j are arbitrary).
 ∂ y i ∂ y j 
∂x k ∂x l i j ∂x k i ∂x l j
g ij dx i dx j = g kl i j dx dx = g kl i d
x k
j dx = g kl dx dx
l
∂y ∂y ∂y ∂y

This implies that gij dxidxj is invariant.
Example 3.4.1. Find the components of the first and second fundamental
tensors in spherical coordinates.
Solution: Let (x1, x2, x3) and ( x 1 ,x 2 ,x 3 ) be the rectangular and spherical
coordinates of a point, respectively,
i.e., x1 = x , x2 = y , and x3 = z,
where x = r sinθcosϕ, y = rsinθsinϕ, and z = rcosθ.

Let gpq and gij be the metric tensors in Cartesian and spherical coordi-
nates, respectively.
Then, (ds)2 = (dx1)2 + (dx2)2 + (dx3)2
= gpqdxpdxq
g11 = 1 = g22· = g33 ; g12 = g13 = g32 = 0
On transformation,
∂x p ∂x q ∂x p ∂x q ∂x1 ∂x1 ∂x 2 ∂x 2 ∂x 3 ∂x 3
g ij = = g pq = g 11 + g 22 + g 33 .
∂x l ∂x j ∂x i ∂x j ∂x i ∂x j ∂x i ∂x j ∂x i ∂x j
Putting i = j = 1,
∂ x p ∂ xq
=g 11 = g pq
∂xi ∂x j
∂ x1 ∂ x1 ∂ x2 ∂ x2 ∂ x3 ∂ x3
= 1 1 g 11 + 1 g 22 + g 33
∂x ∂x ∂ x ∂ x1 ∂ x1 ∂ x1
2 2 2
 ∂ x1   ∂ x 2   ∂ x 3 
= + +
 ∂ r   ∂ r   ∂ r 
= (sinθ cosφ )2 + (sinθ sinφ )2 + (cosθ )2
= sin 2θ + cos 2θ = 1.

Putting i = j = 2,
∂x p ∂x q
g 22 = g pq
∂x i ∂x j
∂x1 ∂x1 ∂x 2 ∂x 2 ∂x 3 ∂x 3
= 2 2 g11 + 2 2 g 22 + 2 2 g 33
∂x ∂x ∂x ∂x ∂x ∂x
2 2 2
 ∂x1   ∂x 2   ∂x 3 
= + +
 ∂θ   ∂θ   ∂θ 
= (rcosθ cosφ )2 + (rcosθ sinφ )2 + (−rsinθ )2
= r 2cos 2θ + r 2 sin 2θ = r 2
∂x p ∂x q
g 33 = g pq
∂x i ∂x j
∂x1 ∂x1 ∂x 2 ∂x 2 ∂x 3 ∂x 3
= 3 3 g11 + 3 3 g 22 + 3 3 g 33
∂x ∂x ∂x ∂x ∂x ∂x
2 2 2
 ∂x1   ∂x 2   ∂x 3 
= + +
 ∂φ   ∂φ   ∂φ 
= (rsinθ . − sinφ )2 + (r sinθ cosφ )2 + 0
= r 2 sin2θ .

Hence, the first fundamental tensor, written in matrix form, is
 1 0 0 
 
 0 r2 0 
 0 0 r sin 2θ
2 
 
 1 0 0 
 
g=  0 r2 0 4 2
 = r sin θ .
 0 0 r sin 2θ
2 
 
The cofactors in g are
Gij
G11 = r 4 sin 2θ ,G22 = r 2 sin 2θ ,G33 = r 2 ,Gij = 0 for i ≠ j and also g ij = .
g

Hence, the second fundamental tensor in matrix form is

 1 0 0 
 
 1 
0 0
 r2 .
 1 
 0 0 2 2 
 r sin θ 
Example 3.4.2. Find the components of the metric tensor and the conju-
gate tensor in cylindrical coordinates.
Solution: Here, let (x1, x2, x3) and ( x 1 ,x 2 ,x 3 ) be the rectangular and cylin-
drical coordinates of a point, respectively,
i.e., x1 = x , x2 = y , x3 = z, and x 1 = ,,x 2 = φ , x 3 = z,
where x = ρ cos ϕ, y = ρ sinϕ, and z = z.

Let gpq and g ij be the metric tensors in Cartesian and spherical coordi-
nates, respectively.
g11 = g22 = g33 = 1, gij = 0 for i ≠ j

g11 = 1, g22 = 2, g33 = 1, gij = 0 for i ≠ j.
1 0 0
2
Metric tensor of first fundamental is 0 p 0 and g = ρ2.
0 0 1
Also, cofactors of g are given by G11 = ρ2, G22 = 1, G33 = ρ2, G12 = G13 =
⋯. = G32 = 0. Gij
ij
The components of the conjugate tensor are given by g = .
g
 1 0 0 
 
1
Hence, the second fundamental metric tensor is  0 0 .
 ρ2 
 
 0 0 1 
Alternatively,
y1 = x1 cos x2
y2 = x1 sin x2
y3 = x3
(dy1) = (cos x2 dx1 – x1sinx2dx2)
(dy1) = (sin x2 dx1 + x1cosx2dx2)

dy3 = dx3
ds2 = (dy1)2 + (dy2)2 + (dy3)2 = (cos x2 dx1 – x1sinx2dx2)2
+ (sin x2 dx1 + x1cosx2dx2)2 + (dx3)2
= (dx1)2 + (x1)2(dx2)2 + (dx3)2.
g11 = 1 g22 = (x1)2, g33 = 1, gij = 0 for i ≠ j
1 0 0
g = |g ij | = 0 ( x 1 )2 0 = ( x 1 )2
0 0 1

cofactor of g 11in g ( x 1 )2
g 11 = = 1 2 =1
g (x )

1
g 22 = ,
( x 1 )2
( x 1 )2
g 33 = = 1 g ij = 0 for i ≠ j
( x 1 )2  1 0 0 
 1 
The second fundamental metric tensor is  0 0  (here,
 ( x 1 )2 
ρ = x1).
 0 0 1 
3.5 Length of a Vector

3.5.1 Length of Vector
The square of the magnitude of a covariant vector Ai is defined by gijAi Aj.
The square of the magnitude of a contravariant vector Aj is defined by
gij AiAj.
Since
AiBi = gij AjBi = AjBj = gijAiBj, (3.12)
it follows that the associate vectors Ai and Ai are of the same length.
In details, by the inner product rule, we have from (3.12)
Ai Ai = gij AjAi = AjAj = gijAi Aj.
If the vector with Ai and Ai as its covariant and contravariant compo-

nents is denoted by vector A and |A| denotes its magnitude, then
1 1
the length of the vector is A = ( A. A) 2 = ( g ij A j Ai ) 2 .
Hence, the magnitude or length of the vector is
1 1 1
A = ( g ij A j A ) = ( g ij Ai A j ) = ( Ai A ) .
i 2 2 i 2
For example, the length of a vector (A1, 0,0) in 3-dimensions is

g 11 A1 A1 = g 11 A1 . Similarly, the length of a vector (0, A2, 0) is g 22 A2
and the length of a vector 0, (0, 0, A3) is g 33 A3 . Hence, the physical com-
ponents of vector Ai are ( g 11 A1 , g 22 A2 , g 33 A3 ).
3.5.2 Unit Vector

A vector Ai is said to be a unit vector or a covariant vector of unit length if
gijAi Aj = 1 (3.13)
while A vector Ai is said to be a unit vector or a covariant vector of unit

length if
gij AjAi = 1 (3.14)
Example 3.5.1. In V4, with line element
ds2 = −(dx1)2 – (dx2)2 – (dx3)2 + c2(dx4)2,

 3
show that the vector  2 , 0 , 0  is a unit vector.
 c 
Solution: Here, g11 = −1, g22 = −1, g33 = −1, g44 = c2, and gij = 0 for other gij.
 3
Let Ai denote the components of the vector  2 , 0 , 0 .
 c 
4 3
Then, A1 = 2 , A2 = 0, A3 = 0, and A = ,
c
so g ij Ai A j = g 11 ( A1 )2 + g 22 ( A2 )2 + g 33 ( A3 )2 + g 44 ( A4 )2
2
2  32
= −1.(Ö2) + 0 + 0 + c 
 c 
3
= −2 + c 2 ⋅ 2 = −2 + 3 = 1.
c
Therefore, the length of A is 1 and A is a unit vector.

The V4 with the above metric is called Minkowski Space-Time of the
Special Theory of Relativity.
3.5.3 Null Vector

A vector Ai is said to be a null vector or a covariant vector of unit length if
gijAi Aj = 0, (3.15)
while A vector Ai is said to be a unit vector or a covariant vector of unit

length if
gij AjAi = 0. (3.16)
Since the components of a zero vector are all zero, it follows that zero
vectors are a null vector, but conversely is not necessarily true.
Example 3.5.2. Show that in V4 with the line element given by
ds2 = −(dx1)2 – (dx2)2 – (dx3)2 + c2(dx4)2,

 3
the vector  −1, −1, −1,  is a null vector in the space.
 c 
Solution: Here, g11 = −1, g22 = −1, g33 = −1, g44 = c2, and gij = 0 for other gij.
 3
Let Ai denote the components of the vector  −1, −1, −1, 
 c 
3
and A1 = −1, A2 = −1, A3 = −1, A4 = .
c
Then,
g ij Ai A j g 11( A1 )2 g 22 ( A2 )2 g 33 ( A3 )2 g 44 ( A 4 )2 1( 1)2
3 3
1( 1)2 1( 1)2 c 2 1 1 1 c2 3 3 0.
c c2
This shows that it is not necessary that the components of a null vector
are all zero components.
3.6 Angle Between Two Vectors

The angle θ between two non-null covariant vectors, Ai and Bi is defined by
g ij Ai B j
cos θ = . (3.17)
g ij Ai A j g ij Bi B j

The angle θ between two non-null covariant vectors, Ai and Bi, is given
by
g ij Ai B j 1
cosθ = . (3.18)

It is to be noted that if two vectors are such that one of them is a null
vector or both of them are null vectors, then the angle between them is not
defined.
If A and B are two non-null vectors and Ai, Ai; Bi, Bi, respectively, are
contravariant and covariant components, then the angle θ between A and
B is given by
Ai Bi
cosθ = . (3.19)
A j A j B k Bk

3.6.1 Orthogonality of Two Vectors

Two vectors Ai and Bi are said to be orthogonal if
gij Ai B j = 0, (3.20)
while two vectors, Ai and Bi, are said to be orthogonal if
gijAiBj = 0. (3.21)
It follows from (3.20) and (3.21) that the angle between two non-null
π
orthogonal vectors, Ai and Bi, is . It also follows from (3.20) and (3.21)
2
that a null vector, Ai or Ai, is self-orthogonal.
Example 3.6.1. If aij is a symmetric tensor of type (0,2) and Ai and Bi are
unit vectors orthogonal to a vector Ci satisfying the conditions
aij Ai − ωgijAi + σgijCi = 0

and aijBi − ωʹgijBi + σʹgijCi = 0, where ω ≠ωʹ, show that Ai and Bi are orthog-
onal and aij AiB j = 0.
Solution: According to the given conditions:
gij AiAj = 1 (i) gijBiB j = 1 (ii)
gij AiCj = 0 (iii) gijBiCj = 0 (iv)
aij Ai – ωgij Ai + σgijCi = 0, multiplying by B j, we get

aij AiB j – ωgij AiB j + σgijCiB j = 0
or aij AiB j – ωgij AiB j = 0 (v)
Similarly, multiplying the relations
a ij Bi − ωʹgijBi + σʹgijCi = 0 by Aj,

we get
aijBi Aj − ωʹgijBi Aj + σ ʹ gijCi Aj = 0,
or aijBi Aj − ωʹgijBi Aj = 0,
or aij AiB j − ωʹgij AiB j = 0 (since aij and gij are symmetric). (vi)
Subtracting (vi) from (v), we get (ωʹ − ω)gijAiBj = 0 or gij AiB j = 0 (since

ω ≠ ωʹ).
Hence, Ai and Bi are orthogonal. From (vi), using this result, we get
aij AiB j = 0.
3.7 Hypersurface
Let u1, u2, …. um be parameters and n equations
xi = xi(u1, u2, ….um); i = 1,2 … n , n > m (3.22)
define m-dimensional subspace Vm of Vn. If we eliminate the m parameters

u1u2, …, um from these n equations, we will get (n − m) equations in x i ′ s,
which represent the m-dimensional curve in Vn.
Similarly, the n equations xi = xi (u1, u2) represent a two-dimensional
subspace of Vn. If we eliminate the parameters u1 and u2, we get n − 2
equations in x i ′ s, which represent a two-dimensional curve in Vn. This

two-dimensional curve defines a subspace denoted by V2 of Vn.
Definition 3.7.1 xi = xi(u1u2, …, un−1) represent the (n − 1) dimensional

subspace Vn−1 of Vn. If we eliminate the parameters u1u2, …, un−1, we get
only one equation in x i ′ s which represent. the n − 1 dimensional curve in
Vn. Thus, this particular curve is called a hypersurface in Vn.
Let φ be a scalar function of coordinates xi. Then,
φ(xi) = φ(xi, xi, …. xi.) = constant (3.23)
determines a family of hypersurfaces of Vn.

A parametric hypersurface is a hypersurface on which one particu-
lar coordinate xi is constant, while the others vary. Let us call it the xi-
hypersurface with equation xi = c = constant.
Definition 3.7.2 If , in a Vn, there are n families of hypersurfaces such that, at

every point, each hypersurface is orthogonal to the n − 1 hypersurface of
the other families which pass through that point, they are said to form an
n −polyorthogonal system of hypersurfaces.
Definition 3.7.3 A family of curves, one of which passes through each

point of Vn, is called a Congruence of Curves.
Definition 3.7.4 An orthogonal ennuple in a Riemannian Vn consists of an

n mutually orthogonal congruence of curves.
3.8 Angle Between Two Coordinate Hypersurfaces

Let θ(xi) = constant (i)
and φ(xi) = constant (ii)
represent families of hypersurfaces.

∂θ ∂θ
Differentiating (i) with respect to x i : i dx i = 0 implies that is
∂x ∂ xi
orthogonal to dxi. Since dxi is tangential to hypersurface (i), we conclude
∂θ
that is normal to θ(xi) = constant.
∂ xi
∂ϕ
Similarly, is normal to φ(xi) = constant.
∂ xi
If ω is the angle between the hypersurfaces, then ω is the angle between

their respective normals also.
∂θ ∂ϕ
g ij i j
Hence, cosω = ∂x ∂x .
ij ∂θ ∂θ ij ∂ϕ ∂ϕ
g g
∂ xi ∂ x j ∂ xi ∂ x j
If we take
θ = xr = C1 (iii)
φ = xs = C2 (iv)
∂θ ∂ϕ ∂x r ∂x s
g ij g ij
∴ cosω = ∂x i ∂x j = ∂x i ∂x j (v)
ij ∂θ ∂θ ij ∂ϕ ∂ϕ
r r s s
ij ∂ x ∂ x ij ∂ x ∂ x
g g g g
∂x i ∂x j ∂x i ∂x j ∂x i ∂x j ∂x i ∂x j
g ijδ irδ js g rs
= cosω =
g ijδ irδ rj g ijδ isδ js g rr g ss

If ωij is the angle between the hypersurfaces of parameters xi and xj, then
by (v),
g ij
cosω ij = . (vi)
g ii g jj

π
If the hypersurfaces are orthogonal, then ω ij = .
π 2
From (vi), we get cosω ij = cos = 0,
2
g ij
0 g ij 0.
ii jj
g g

If the hypersurfaces of parameters of xi and xj are orthogonal, then

g = 0.
ij
Example 3.8.1. Show that the angle between the vectors (1,0,0,0) and
 3
 2 ,0,0  , with c being a constant, in a space with line element given
c 
by ds2 = −(dx1)2 – (dx2)2 – (dx3)2 + c2(dx4)2, is not real.
Solution: ds2 = gijdxidxj,
where g11 = −1, g22 = −1, g33 = −1, g44 = c2
A1 = 1, A2 = 0, A3 = 0, A4 = 0
3
B1 = 2 , B 2 = 0 , B3 = 0 , B 4 = .
c
g ij Ai B j
We know cosθ = .
Now, gijAi Aj = (−1). 12 + (−1). 0 + (−1). 0 + c2.0 = −1

2
i j
2  3 2
g ij B B = (−1).( 2) + (−1).0 + (−1).0 + c .  = −2 + 3 = 1
 c 

 3
g ij Ai B j = (−1).1. 2 + (−1).0.0.0 + (−1).0 + c 2 .0.  =− 2
 c 

− 2
cosθ = , which not a real number.
−1 −1
Hence, the angle between the two vectors is not real.
Example 3.8.2. In E3 (xi) are orthogonal Cartesian coordinates and con-

sider a transformation
x1 = y1siny2cosy3
x2 = y1siny2siny3
x3 = y1cosy2,
where (yi) are spherical polar coordinates (y1 = r, y2 = θ, y1 = φ). What are
the metric coefficients of gij(y)?
Solution: Now
x1 = dy1(siny2cosy3) + y1(cosy2cosy3dy2 – siny2siny3dy3)
dx2 = dy1(siny2siny3) + y1(cosy2siny3dy2 + siny2cosy3dy3)
dx3 = dy1cosy2 – y1siny2dy2.
Since xi is an orthogonal Cartesian coordinate system,
(ds)2 = (dx1)2 + (dx2)2 + (dx3)2

= {dy1(siny2cosy3) + y1(cosy2cosy3dy2 – siny2siny3dy3)}2
+{dy1(siny2siny3) + y1(cosy2siny3dy2 + siny2cosy3dy3) }2
+ {dy1cosy2–y1siny2dy2}2
= (dy1)2Sin2y2 + (y1)2{cos2y2(dy2)2 + sin2y2(dy3)2 + cos2y2(dy1)2

+ (y1)2Sin2y2(dy2)2 – 2y1cosy2siny2dy1dy2
= (dy1)2 + (y1)2(dy2)2 + (y1)2sin2y2(dy3)2 – 2y1cosy2siny2dy1dy2
Here, g11= 1, g22 = (y1)2 = r2, g33 = (y1)2sin2y2 = r2sin2θ, g12 = −2y1cosy2siny2 =
−2rcosθsinθ and g13 = g21 = g23 = g31 = g32 = 0.
Example 3.8.3. Let gij and gij be the fundamental metric tensors and
∂ g ij ∂ g ij
reciprocal tensors, respectively. Show that g ij k + g ij = 0 and
∂x ∂ xk
ij ∂ g ij
ij
∂ log g ∂g
k = g k
= − g ij , where g = |gij|.
∂x ∂x ∂ xk
Solution:
We know g ij g jk = δ ik , so gij gij = n.
Partially differentiating
∂ g ij ∂ g ij
g ij + g ij = 0, (i)
∂ xk ∂ xk
1+ 2
g 21 g 23
we get a cofactor of g 12 = (− ) = G12
g 31 g 33
g11G11g12G12 + g13G13 = g
gijGij = g (ii)
∂g
or = G ij (iii)

∂ g ij
from (ii), multiplying gik
gikgijGij = ggik,
or δ jkG ij = gg ik ,

or Gik = ggik (iv)

∂g ∂ g ∂ g ij ∂ g ij
Now, k
= k
= G ij k and by (iii),
∂x ∂ g ij ∂ x ∂x
∂g ∂ g ij
= gg ij k by (iv)
∂ xk ∂x
∂ g ij 1 ∂ g ∂ (log g )
g ij = =
∂ xk g ∂ xk ∂ xk
∂ (log g ) ∂ g ij
= − g ij
∂ xk ∂ xk
log g g ij g ij
g ij g ij .
xk xk xk
ii 1
Example 3.8.4. If gij = 0 for i ≠ j , prove that g = (no summation
g ii
on i).
g 11 0 0 ……0
Solution: Here, |g ij | = 0 g 22 0 … . .0 = g 11 g 22  g nn
0 0 0 …… g nn ..
cofactor of g iiin g g11 g 22 … g i −1i −1 g i +1i +1 ….. g nn 1

g ii = = = .
|g ij | g 11 g 22 … g ii … g nn g ii

Example 3.8.5. If ui and vi are two orthogonal vectors, show that (gljgki −
glkgji)ulviujvk = 1
Solution: By condition, we get
gijuiuj = 1, (i)

gijvivj = 1, (ii)
and orthogonality gijuivj = 0 (iii)
Now, we have (glj gki – glk gji)ulviujvk
= glj gkiulviujvk − glk gjiulviujvk
= gljulujgkivivk − glkulvkgjiujvi
= 1.1 – 0.0 = 1.
Example 3.8.6. If ai and bi are two non-null vectors such that gijuiuj = gijvivj,
where ui = ai + bi and vi = ai − bi, show that ai and bi are orthogonal.
Solution: We have gijuiuj = gij(ai + bi)(aj + b j)
= gijaiai + gijbibi + 2gijaib j (gij is symmetric) (i)
Similarly,
gijvivj = gijaiai + gijbibi – 2gijaib j (ii)
Since gijuiuj = gijvivj are given,

From (i) and (ii) we get 4gijaib j = 0
Or gijaib j = 0, imply that ai and bi are orthogonal.
3.9 Exercises
1. If the metric is given by
ds2 = 5(dx1)2 + 3(dx2)2 + 4(dx3)2 – 6dx1dx2 + 4dx2dx3,
evaluate g and gij.
2. Let E3 be covered by orthogonal Cartesian coordinates xi and let
x1 = y1cosy2,
x2 = y1siny2, and
x 3 = y3 represent a transformation to cylindrical coordinates yi. Find

the expression for ds2 in cylindrical coordinates.
3. Determine the metric tensor and the conjugate metric tensor in
cylindrical coordinates.
4. If the metric is given by
ds2 = (dx1)2 – 2(dx2)2 + 3(dx3)2 – 8dx2dx3

(i)
dr 2
(ii) ds 2 = + r 2 (dθ 2 + sin 2θ dϕ 2 ),
r2
1− 2
a
evaluate g and gij.

5. In E3 is covered by orthogonal Cartesian coordinates xi, let
x1 = y1y2cosy3
x2 = y1y2siny3, and
x 3 = ( y 1 )2 − ( y 2 )2

represent a transformation to parabolic coordinates yi. Find the

expression for ds2.
6. Prove that gij is a symmetrical contravariant tensor of type (2,0).
7. Show that in V4 with line element ds2 = −(dx1)2 – (dx2)2 – (dx3)2 +
c2(dx4)2
each of the following vectors is a unit vector:
 3
(i)  1,1, 0 , 
c
(ii) (1,0,0,0)
8. F ind the conjugate metric tensor in a Riemannian metric space,

V2, in which the distance (ds) is given by ds2 = (dx1)2 + 2cosα(dx1)
(dx2) + (dx2)2.
9. Show that ds2 = (du)2 + [u2 + (a2)](du)2, where y1 = ucosv and y2 =
usinv, y3 = av.
10. If ui = ai + bi, where ai and bi are two orthogonal unit vectors,
show that the square of the length of vector ui is 2.
4
Tensor Calculus
4.1 Introduction
We mentioned that in Sn partial differentiation of a tensor does not, in gen-
eral, produce a tensor. With the introduction of a metric in Sn, the envi-
ronment has changed and the question arises as to whether in Vn a new
operation of differentiation can be introduced so that when applied to a
tensor it produces another tensor. The answer is affirmative, but in achiev-
ing its affirmation the fundamental tensors are again essential.
With the motivation to build up expressions involving the derivatives
of a tensor which again produce components of a tensor, in 1869 E.B.
Christoffel introduced certain combinations of partial derivatives of the
fundamental tensor gij, which proved useful in the development of the
Calculus of Tensors. A new operation of differentiation may be introduced
with the help of two functions formed in terms of the partial derivatives
of the components of the fundamental tensor. They are Christoffel symbols
l
of the first and second kind denoted respectively by [i, j, k] and . We
i j
introduce in this chapter the combinations of partial derivatives of the fun-
damental tensor gij(x), which will prove useful in the development of the
calculus of tensors.
4.2 Christoffel Symbols

Let us construct a set of functions denoted by symbol
1 g ik g jk g ij
[ij , k] (i , j , k 1,2 ,3) (4.1)
2 xj xi xk
and call them Christoffel (3-index) symbols of the first kind.
97
k
The set of functions g k [ij , ], (4.2)
i j

where gkα is the contravariant tensor constructed with g ij′ s are called
Christoffel symbols of the second kind.
No summation is indicated in the Christoffel symbols of the first kind,
but summation is to be made over l in the Christoffel symbols of the sec-
ond kind.
4.2.1 Properties of Christoffel Symbols

In this section the following properties of the symbols will be proved.
1 g ik g jk g ij
Property 4.2.1. The Christoffel symbols [ij , k] j
2 x xi xk
k
and g k [ij , ] are symmetric with respect to indices i and j.
i j
Proof:
(i) Interchanging the indices i and j in (4.1),
1 g jk g ik g ji
[ ji , k]
2 xi xj xk
1 g ik g jk g ij
(since g ij is a symmetric tensor)
2 xj xi xk
[ij , k] (4.3)
k
(ii) g k [ ji , ] g k [ij , ] by (i)
j i
k
(4.4)
i j

Tensor Calculus 99
Property 4.2.2.
k
i)[ij ,h] g kh (4.5)
i j

g ij (4.6)
ii) g lj g li ;
xk l l
i k j k

g im (4.7)
iii) g ij g km
xl m i
l j k l

g ij
iv) [ik , j] [ jk ,i] (4.8)
xk
Proof:
k
(i) By definition, g k [ij , ].
i j
Multiplying both sides by gkh, we get
k
g kh g kh g k [ij , ] h [ij , ] [ij ,h]
i j

∂ g ij
and (ii) by (iv), we have = [ik, j] + [ jk,i]
∂x k
l l
g lj g li (using 2nd Christoffel symbol).
i k j k

(iii) Since g lj g ij = δ li , differentiating with respect to xk, we get
g lj ij g ij
g g lj 0.
xk xk
Multiplying by glm, we get

g lj g ij
g lm g ij g lm g lj 0
xk xk
g ij g lj
g lm g lj g lm g ij
xk xk
m g ij g lj
j g lm g ij g lm g ij {[lk , j] [ jk ,l]}
xk xk
g lm { g ij[lk , j]} g ij { g lm[ jk ,l]}
g im i m
o, g lm g ij .
xk l k j k

g ij i j
Interchanging m and j, we get g lj g im
xk l k m k

g ij i j
g jl g im [ since g lj g jl ].
xk l k m k

1 g ij g kj g ik
(iv) [ik , j]
2 xk xi xj
1 g ji g ki g jk
and [ jk ,i] .
2 xk xj xi
Adding these two relations,
1 g ij g kj g ik 1 g ji g ki g jk
[ik , j] [ jk ,i]
2 xk x i
xj 2 xk xj xi
1 g ij g ij
2 k .
2 x xk
i
Property 4.2.3. If g= |gij| ≠ 0, then (log g ).
i j xj
Tensor Calculus 101
Proof: By the definition of reciprocal tensor

cofactor of g ijin |g ij | Gij
we know, g ij = = , (i)
|g ik | g

g 11 g 12 … … g 1n
g 21 g 22 … … g 2n
where Gij is a cofactor of gij and g = |g ij | = .
…………………
We know gij g = 1.
ij
(ii)
g n1 g n 2 … … g nn
From (i) and (ii), it is implied that g = gijGij.(iii)
Since the derivative of the determinant is obtained by differentiating
each row of it separately, keeping the other rows the same, and summing
the resulting all determinants, thus
∂ g 11 ∂ g 12 ∂ g 1n g 11 g 12   g 1n
k k 
∂x ∂x ∂ xk g 21 g 22   g 2n
∂g g 21 g 22   g 2n
= +  + 
∂ xk
 ∂ g n1 ∂ g n 2 ∂ g nn
g n1 g n 2   g nn k k 
∂x ∂x ∂ xk
∂g ∂g ∂g 
=  11k G11 + 12k G12 + .. + 1kn G1n  + .
 ∂x ∂x ∂x 
∂g ∂g ∂g 
+  nk1 Gn1 + nk2 Gn 2 + .. + nnk Gnn 
 ∂x ∂x ∂x 
∂ g ij
= ∑ni , j =1 Gij
∂ xk
∂ g ij
= ∑ni , j =1 k gg ij (from (i))
∂x
g ij
gg ij
xk
gg ij {[ik , j] [ jk ,i]} by (4.8)
i j i
g g 2g
i k j k i k

or 1 g i
2g xk i k
i 1 g
(log g ). (4.9)
i j 2g xj xj

Example 4.2.1. If (ds)2 = (dr)2 + r2(dθ)2 + r2sin2θ(dϕ)2, find the values of
1 3
(i)[22,1] and [13,3] and and
2 2 13

Solution: Here x1 = r, x2 = θ, x3 = ϕ,
g11 = 1, g22 = r2, g33 = r2sin2θ and gij = 0, when i ≠ j
1 0 0
Det( g ij ) = 0 r2 0 = r 4 sin 2θ .
0 0 r sin 2θ
2

11 cofactor of g ij in |g 11| r 4 sin 2θ

Here, g ≠ 0, ∴ g = = 4 2 = 1,
|g ik | r sin θ
cofactor of g ijin |g 22| r 4 sin 2θ 1

, g 22 = = 4 2 = 2
|g ik | r sin θ r

cofactor of g ijin |g 33| r2 1

g 33 = = 4 2 = 2 2
|g ik | r sin θ r sin θ

cofactor of g ijin |g ij |
g ij = = 0 , since all cofactors of g ij = 0,i ≠ j .
|g ik |

Tensor Calculus 103
1 g ik g jk g ij k
(i) We know, [ij ,k] and g k [ij , ]
2 xj xi xk i j
1 g 21 g 21 g 22 1 g 22 1 r2 1
[22 ,1] 2r r
2 x2 x2 x1 2 x1 2 r 2
1 g13 g 33 g13 1 g 33 1 r 2sin2 1

[13,3] 2rsin2 rsin2 .
2 x3 x1 x3 2 x1 2 r 2

k
(ii) g k [ij , ]
i j

1
g 1 [22 , ] g 11[22 ,1] g 12[22 ,2] g 13[22 ,3]
2 2
1.( r) 0 0 r

3 1 1
and g 3 [1 3, ] g 31[1 3, 1] g 32[13,2] g 33[1 3,3] 0 0 rsin2 .
13 r 2 sin2 r

Example 4.2.2. Show that if gij = 0, i ≠ j, then
k
(a) 0, whenever i, j, k, are distinct
i j
i 1 logg ii
(b)
i i 2 xi
i 1 logg ii
(c)
i j 2 xj
i 1 g jj
(d)
j j 2 xi
1 g ik g jk g ij
Solution: We know [ij , k] (4.10)
2 xj xi xk
1 g ii
(a) When i = j = k, [ii ,i] .
2 xi 1 g ik g ik g ii
(b) When i = j ≠ k, (4.10) becomes [ii ,k] .
2 xi xi xk
1 g ik g ik g ii 1 g ii
Since gik = 0, i ≠ k, it becomes [ii ,k]
2 xi xi xk 2 xk
1 g jj
and [ jj ,i] .
2 xi
1 g ii g ji g ij
(c) When i = k ≠ j, [ij ,k]
2 xj xi xi
1 g ii
, as g ij 0.
2 xj
1 g ii g ji g ij
(d) When i ≠ k ≠ j, [ii ,k]
2 xj xi xk
= 0 , as gij = 0 , gjk = 0,
k
(i) as I, j, and k are distinct, i.e., i ≠ k ≠ j, g kl [ij ,l]
i j
= 0 as gkl = 0 , k ≠ l
i 1 g ii 1 g ii 1
(ii) g ii[ii ,i] g ii as g ii
i i 2 xi 2 g ii xi g ii
1 logg ii
2 xi
i 1 1 1 g ii
(iii) i = k ≠ j, g ii[ij ,i] [ij ,i]
i j g ii g ii 2 x j
1 logg ii
2 xj
i 1 1 g jj
(iv) j = k ≠ i g ii[ jj ,i]
j j g ii 2 xi
Tensor Calculus 105
i 1 g jj
j j 2 g ii xi

Example 4.2.3. If |gij| ≠ 0, show that
g [i k , ] ([ j , ] [ j , ]).
xj i k xj i k

Solution: Christoffel’s symbol of 2nd kind
g [ik , ]
i k

Multiplying both sides by gαβ, we get
g g g [ik , ]
i k

g [ik , ] as g g 1.
i k

Differentiating both sides with respect to xj, we get
g
g j
[ik , ].
x i k i k xj xj

g
Since [ j, ] [ j, ] , [By (4.8)]
xj
g ([ j , ] [ j , ]) [ik , ]
xj i k i k xj

g [ik , ) ([ j , ] [ j , ]).
xj i k xj i k

Example 4.2.4. Show that the only non-vanishing Christoffel symbols of

the second for V2 with line element
1
(ds)2 = (dx1)2 + sin2x1(dx2)2 are sinx1cosx1 ,
2 2
2 2
cotx1 .
12 21
1 0
Solution: g11 = 1, g22 = sin x and g12 = g21 = 0 and |g| =
2 1 = sin 2 x 1
0 sin 2 x 1
cofactor of g11 in |g| sin 2 x 1

g 11 = = 2 1 = 1.
|g| sin x

1
Similarly, g 22 = and g12 = g21 = 0.
sin 2 x 1
1 g ik g jk g ij 1 g 21 g 21 g 22
[ij , k] , we get [22 ,1]
2 xj x i
x k
2 x 2
x2 x1
1 g 22 1 sin2 x1 1
2sinx1cosx1 sinx1cosx1
2 x1 2 x 1
2
1
g 1 [22 , ] g 11[22 ,1] 0 0 1.( sinx1cosx1 ) sinx1cosx1
2 2

Tensor Calculus 107
1 g 12 g 22 g 12 1 g 22 1 sin2 x1
[12 ,2] sinx1cosx1
2 x2 x1 x2 2 x1 2 x1
2 2 1
g 2 [21, ] g 22[21,2] g 22[12 ,2] sinx1cosx1 cotx1
12 21 sin2 x1

Example 4.2.5. If aij are components of a symmetric tensor, show that
1 jk g jk
a jk[ij , k] a ,
2 xi
where gjk have their usual meaning.

g jk
Solution: We know [ ji ,k] [ki , j].
xi
Multiplying both sides by ajk, we get
g jk
a jk a jk[ ji , k] a jk[ki , j]
xi
g jk
a jk[ ji ,k] a kj[ki , j] a jk i
as a jk a kj ,
x
g jk
a jk[ ji , k] a jk[ ji ,k] a jk ,
xi
g jk
or 2a jk[ ji , k] a jk ,
xi
g jk
or 2a jk[ij , k] a jk ,
xi
1 jk g jk
orajka[ij , k] a ,
2 xi
i
Example 4.2.6. Prove that g g ij g g jk 0.
x j
j k
g g ij
Solution: g g ij g ij g
xj xj xj
g i j
g ij g g jp g it
xj p j t j
g i j
g ij g g jp g g it
x j
p j t j
g i 1 g
g ij g g jp g g it
x j
p j 2g xt
1 g ij 1 g it i
g g g g jp
2 g xi 2 g xt p j
i
g g jp
p j
i
g g ij g g jk 0
x j
p k

Example 4.2.7. Calculate the non-vanishing Christoffel symbols corre-

sponding to the metric
(ds)2 = −(dx1)2 – (dx2)2 – (dx3)2 + f(x1, x2, x3)(dx4)2.
Solution: Comparing the expression ds2 = gijdxidxj(i)

with (ds)2 = −(dx1)2 – (dx2)2 – (dx3)2 + f(x1, x2, x3)(dx4)2,
we get g11 = −1, g22 = −1, g33 = −1, and g44 = f(x1, x2, x3) = f, (let)
gij = 0 for i ≠ j
Tensor Calculus 109
−1 0 0 0
0 −1 0 0
and g = =−f.
0 0 −1 0
0 0 0 f

ij cofactor of g ij in g
We know g =
|g|
cofactor of g11 in g f 1
∴ g 11 = = = −1, g 22 = −1, g 33 = −1, g 44 = and g ij
|g| −f f
= 0 for i ≠ j .

Since g11, g22, and g33 are constants, we find that all non-vanishing
Christoffel symbols of the 1st and 2nd kinds are
1 g 44 1 f 4 4
[14 , 4] 1 1
, g 4 [14 , ] g 44[14 , 4]
2 x 2 x 1 4 41
11 f f
1
log f
f 2 x x1
1 g 44 1 f 4 4
[24 , 4] g 44[24 , 4] log f
2 x2 2 x2 2 4 4 2 x2

1 g 44 1 f 4 4
[34 , 4] g 44[34 , 4] log f
2 x3 2 x3 3 4 4 3 x3
1 g 44 1 f 1 1 f 1 f
[44 ,1] g 11[44 ,1] 1.
2 x1 2 x1 4 4 2 x1 2 x1
1 g 44 1 f 2 1 f
[44 ,2] g 22[44 ,2]
2 x2 2 x3 4 4 2 x2

1 g 44 1 f 3 1 f
[44 ,3] g 33[44 ,3]
2 x3 2 x3 4 4 2 x3

4.3 Transformation of Christoffel Symbols

The fundamental tensors gij and gij are the functions of coordinates xi and
i
[ij, k] and are also the functions of coordinates xi. Let g ij , g ij , [ij,k],
j k
l
and occur in another coordinate system, yi.
j k
4.3.1 Law of Transformation of Christoffel Symbols of 1st Kind

Let [ij, k] be the functions of coordinates xi and [ij,k] be in another coor-
dinate system, yi. Then,
1 g ik g jk g ij
[ij , k] (4.11)
2 yj yi yk

Since g ij is a covariant tensor of order 2, then
xp xq
g ij g pq (4.12)
yi yj
Differentiating it with respect to yk, we get
g ij xp xq
g pq
yk yk yi yj
xp xq xp xq g pq (4.13)
k
g pqq
y yi yj yi yj yk
2
xp xq xp 2 q
x xp xq g pq xr
g pq g pq
y k yi yj yi y j yk yi yj xr yk
Tensor Calculus 111
Interchanging i, k and p, r in the last term of (iii),
g kj 2
xp xq xp 2 q
x xr xq g rq xp
g pq g pq (4.14)
yi yi y k yj yk y j yi yi yj xp yk
Interchanging j, k and q, r in the last term of (iii),
g ik 2
xp xq xp 2 q
x xp xr g pr xq
g pq g
k j pq (4.15)
yj y j yi yk yi y y yi yj xq yk
Substituting the values of (4.13), (4.14), and (4.15) in (4.11), we get
1 2 p
x xq xr xp xq g pr g qr g pq
[ij , k] 2 j i g pq
2 y y yk yi yj yk xq xp xr
2
xp xq xp xq xr 1 g pr g qr g pq
g pq
y j yi yk yi yj yk 2 xq xp xr
2
xp xq xp xq xr (4.16)
g pq [ pq ,r ]
yi y j yk yi yj yk

This is a law of transformation of Christoffel’s symbol of the 1st kind, but

it is not the transformation of any tensor due to the presence of the 1st term
of (4.16), so Christoffel’s symbol of the 1st kind is not a tensor.
The 1st term will vanish identically if the coordinate transformation is
affine, that is if y i = c ij x j and c ij ¢s are constants.

of 2nd Kind
k k
Let g kl [ij ,k] be a function of coordinate xi and g kl [ij ,l] be
i j i j
a function of coordinate yi. Then, from (4.16) we have

2
xp xq xp xq xr
[ij ,l] g pq [ pq ,r ].
yi y j yl yi yj yl
Since, gkl is a contravariant tensor of order 2,
yk y l st (4.17)
g kl g
xs xt
y k y l st 2 x p x q y k y l st x p x q x r
Now, g kl[ij ,l] g g pq g [ pq ,r ]
x s xt yi y j yl x s xt yi y j yl
yk y l y q st 2 x p yk yl xr x p x q st
g g pq g [ pq ,r ]
xs xt yl yi y j xs xt yl yi y j
yk q xp
2
yk r x p x q st
t g st g pq t g [ pq ,r ]
xs yi y j xs yi y j
y k sq 2 x p y k x p x q sr
g g pq g [ pq ,r ]
xs yi y j x s yi y j
q
as t g st g sq and t
r
g st g sr
y k 2 x p sq y k x p x q sr
g kl ij ,l g g pq g [ pq ,r ] ,
x s yi y j x s yi y j
s
as g sq g pq s sr
p and g [ pq ,r ] p q
y k 2x p s yk x p xq s
x s yi y j
p
x s yi y j p q
(4.18)
k s
y k 2x s yk x p xq
i j x s yi y j s
x y yi j p q
This is the Law of Transformation of Christoffel’s Symbol of the 2nd

Kind, but it is not the Law of Transformation of any tensor, so Christoffel’s
symbol of the 2nd kind is not a tensor unless the coordinate transformation
is affine.
xm
Multiplying (4.18) by , summing with respect to the common
value k = γ, we obtain yk
Tensor Calculus 113
xm xm y 2 s
x xm y xp xq s
.
l j y y xs i j
yy y xs yi yj p q

m m
x m x m
Since s and p , the above expression is
xs xp
2 m
x xm xp xq m
(4.19)
yi y j l j y yi yj p q

Obviously y and x can be interchanged and from (4.19) we get
y 2 m
ym y p yq m (4.20)
.
xi x j l j x xi x j p q

These formulas (4.19) and (4.20) were deduced in different ways by

Christoffel in a memoir concerned with the study of quadratic differential
forms. We will make use of these formulas to define the operation of ten-
sorial differentiation.
4.4 Covariant Differentiation of Tensor

f
We know the set of partial derivatives, , of a scalar function f(x1, x1, …
xi
f f xk
x1) represent a covariant vector since , but if we form the
f yi x k yi f
set of partial derivatives j i
of the covariant vector , we get
y y yi
2
f f xk
yi y j yj xk yi
2
f xl xk f 2 k
x
k l
,
x x yj yi xk y yj
i
2
f
which shows that the set of does not transform according to a
y yj
i
2 k
f x
tensorial derivative because of the presence of . That is, the
k
x y yji
set of partial derivatives of a covariant vector, in general, are not a tensor.
4.4.1 Covariant Derivative of Covariant Tensor

If we have a covariant vector Ak(x), then
xk
Bi ( y ) Ak .
yi
Bi xk xl Ak 2 k
x
Differentiating partially, Ak , (4.21)
yj yi yj xl y yj
i

which shows that the derivative of a vector does not form a tensor unless
the coordinate transformation xi = xi(y) is affine.
2 k
x
If we insert from (4.19) into (4.21), we get
y yj
i
Bi xk xl Ak xk xp xq k
Ak .
yj yi yj xl l j y yi yj p q

xk
Since, Ak B , we have on rearranging
y
Bi xk xl Ak xp xq k
B Ak
yj yi yj xl l j yi yj p q

Bi A x x (4.22)
or B A
yj l j x yi yj

Tensor Calculus 115
Ai
from which it is clear that the set of n2 functions A obeys the
xj i j
Law of Transformation of a Covariant Tensor of Rank 2.
Ai
Definition 4.4.1. The set of n2 functions A defines covariant
xj i j
xj as a derivative of a covariant tensor Ai (with respect to gij).
We denote the covariant xj as a derivative of Ai by the symbol Ai,j.
Ai
Ai , j A (4.23)
xj i j

4.4.2 Covariant Derivative of Contravariant Tensor

If we start with a contravariant vector Aα and differentiate the relation
yi
Bi ( y ) A (x ), we obtain
x
Bi yi x A 2 i
y x
A
yj x yj x x x yj
and using (4.20), we get
Bi l A yi x .
B A (4.24)
yj J x x yj

Ai i
Thus, the set of n2 functions Aα forms a mixed tensor of
rank 2. xj j
Ai i
Definition 4.4.2. The set of n2 functions A represents the
x j
j
covariant xj derivative (with respect to gij) of the contravariant tensor Ai.
Thus,
Ai i
A,ij A (4.25)
xj j

It should be noted that the covariant derivative of a tensor of type (1,0)

is a tensor of type (1,1).
The equation of (4.24) can be written as
yi x
A,ij A,ij (4.26)
x yj
From the above relation, it follows that the covariant derivative of tensor
of type (1,0) is a tensor of type (1,1).
4.4.3 Covariant Derivative of Tensors of Type (0,2)

Let Aij be the components of a tensor of type (0,2).
xp xq
Then, Aij Apq . (4.27)
yi yj
Differentiating (4.27) with respect to yk, we get
Aij xp xq Apq xp xq
Apq
yk yi yj yk y k
yi yj
x p
xq
Apq xp
2
xq xp 2 q
x
Apq Apq . (4.28)
yi yj yk y k yi yj yi y k
yj
We know ym
2
ym yp y q m from (4.20) and using
xxi j
l J x xi xj p q
this in the 2nd term of (4.28), we get
Tensor Calculus 117
2
x p xq xq h yl y p yr l
Apq Alq
y k yi y j yj l k xh xi x j p r
h yl xq xq y p yr l
Alq Alq
l k xh y j y j xi x j p r
x p xq
2 h xl xq xq x p xr l
Apq Alq Alq (4.29)
y yi y j
k
l k yh y j y j yi y j p r
h xq x p xr l
Ahj Alq
l k y j yi y j p r
xl xq
(since Ahj Alq , by (4.27))
yh yj
and
2 q
x xp 2 l
x xp xp h xl xq xr l
Apq k i
Apl Apl
y y yj k
y y i
yj yj J k yh y j yk q r

2 q
x xp h xp
x q
xr l
Apq k i
Aih Apl
y y yj J k yi yj k
y q r
(4.30)
Substituting the value of (4.29) and (4.28) in equation (4.28), we get
Aij xp xq Apq h xq xp xr l h xp xq xr l
Ahj Alq Aih Apl
yk yi yj yk i k yj yi yj p r j k yi yj yk q r
xq xp xr Apq l h h xp xq xr l
Alq Ahj Aih Apl
yj yi yj xr p r l k j k yi yj yk q r
Aij h h Apq l l xq xp xr
Ahj Aih Alq Apl .
yk l k j k xr p r q r yj yi yk

We denote
Apq l A l xq xp xr
Apq ,r lq Apl Aij ,k Apq ,r .
xr p r q r yi yj yk

(4.31)
Aij l l
Definition 4.4.3. The set of functions Alj Ail rep-
x k
i k j k
resents the covariant xk derivative (with respect to gij) of the covariant ten-
sor Aij. Thus,
Aij l l
Aij ,k Alj Ail (4.32)
xk i k j k

It is to be noted that the covariant derivative of a tensor of type (0,2) is

a tensor of type (0,3).
4.4.4 Covariant Derivative of Tensors of Type (2,0)

Let Aij be the components of a tensor of type (2,0).
Then,
y y ij
A A (4.33)
xi xj
Differentiating (4.33) with respect to yγ, we get
Aij xk y y 2
y xk y y 2
y xk
A Aij Aij
y xk y xi xj x k
x i
y xj xi x k
x j
y
Aij y y xk xk y r y y y
Aij
xk xi xj y y xj k i xr xi xk
y xk r y y y
Aij
xi y k j xr xj xk
Aij y y xk y xk y r
Aij
xk xi xj y xr y j
x k i
y y xk y y y xk r
Aij Aij
xi xk y xj xi xr y k j
y y y xk
Aij
xi xj xk y
Tensor Calculus 119
Aij y y xk y xk y i
Arj
xk xi xj y xi y j
x k r
y y y y xk j
Aij Air
xi xj xi xj y k r
y y
Aij
xi xj

(interchanging dummy index r and i in the 2 term and index r and j in nd
the 4th term)
Aij i j y y xk
Arj Air
xk k r k r xi xj y
y y y y
Aij Aij
xi xj xi xj

A
or, A A
y
Aij i j y y xk
Arj Air . (4.34)
xk k r k r xi xj y

Aij i j
It follows that Arj Air are components of tensors
xk k r k r
of type (2,1).
Aij i j
Definition 4.4.4. The set of functions Arj Air rep-
x k
k r k r
resents the covariant xk derivative (with respect to gij) of the contravariant
tensor Aij. Thus,
Aij i j
A,ijk Arj Air (4.35)
xk k r k r

It is to be noted that the covariant derivative of a tensor of type (2,0) is
a tensor of type (2,1).
y y xk (4.36)
(4.34) can be written as A,ijk A ij
,k .
xi xj y
Thus, the covariant derivative of tensor type (p, q) is a tensor of (p, q + 1).
4.4.5 Covariant Derivative of Mixed Tensor of Type (s, r)

Relation (4.34) can be extended in an obvious way to mixed tensors. Thus,
we define the covariant xk derivative (with respect to gij) of the mixed ten-
j j …j
sor Ai11i21....ir s by the formula
Ai1j1i2j1....irjs j1 jl j j2
Ai1j1i2j1....irjs,k k
Ailj1i22 ....jisr Ai11i2j....
3
ir
s
x k l k l
js l
 Ai1j1i2j2....irjs 1l Alij11j....3 ir js
k l i1 k
l l
Ai1j1lij32.....ijrs  Ai1j1i2j2....irjs 1l .
i2 k ir k

It can be easily shown that Ai1j1i2j1....…irj,sk are the components of a tensor of
type (s,r+1).
If A is a tensor of rank zero, we define its covariant derivative to be the
A
ordinary derivative. Thus, A,l . We also note that if gij’s are constants,
xl
the Christoffel symbols vanish identically, hence the covariant derivative
reduces to ordinary derivatives.
Aij i l
I Itt follows,
follows for Type
a Type
of ((1,1) tensor: Aij ,k
1,1) tensor: Alj Ali
xk k l j k
(4.38)
4.4.6 Covariant Derivatives of Fundamental Tensors

and the Kronecker Delta
The following theorem shows the nature of the behavior of the tensors gij,
gij, and δ ij under covariant differentiation.
Ricci’s Theorem 4.4.1. Fundamental tensors and Kronecker Deltas behave
in covariant differentiation as though they were constants, i.e., gij,k = 0,
g ,kij = 0 , and δ ij ,k = 0.
Tensor Calculus 121
Proof:
g ij l l
(i) We have g ij ,k g lj g il (using 4.32)
xk i k j k

g ij
[ik , j] [ jk ,i]
xk
g ij 1 g ij g kj g ik g ij g ki g jk
k k
x 2 x xi xj xk xj xi
g ij 1 g
k
2 ijk 0
x 2 x
(ii) Now, we use the formula for covariant derivatives of contravariant ten-
sors of order 2
g ij i j
g ij,k g rj g ir (by 4.35).
xk k r k r

We know, g ij g jk = δ ik
( g ij g jk )
0
xm
g ij jk g jk
or g g ij 0.
xm xm
Multiplying both sides by gil,
g ij jk il g jk
g g g ij g il 0,
xm xm
g jk g ij
or jl m
g jk g il m ,
x x
g lk
g
or m
g jk g il mij g jk g il {[im, j] [ jm,i]}
x x
k l
g il g jk ,
i m j m

g ij i j
or, g rj g ri (changing l to j, k to i, m to k, and change
x k
r k r k
the dummy index on the right hand side r for i and 2nd term r for j)
g ij i j
g rj g ri 0
xk r k r k

Hence, g ,kij = 0.
i j
k
l l
i l
i i
(iii) k, j l 0 using (4.38).k 0
j k
x
j
j l j k j k
It is known that gijAi = Aj where Ai is a contravariant vector. Therefore,
(gij Ai),k must give the same result as Aj,k. To decide whether this equality is
ensured, it is necessary to know the behavior of a product of two tensors
under covariant differentiation. We now consider the sum, difference, and
product of the tensors.
4.4.7 Formulas for Covariant Differentiation

It is easy to deduce from the structure of Formula (4.37) that the rules for
covariant differentiation of sums and products of tensors are identical with
ordinary differentiation. If we consider that
Ai1j1i2j2.....…irjs ( x ) and Bi1j1i2j2.....…irjs ( x ) are two tensors, then the formula

(A j1 j2… js
i1i2 .....ir + Bi1j1i2j2.....…irjs ) ,l
= Ai1j1i2j2.....…irjs,l + Bi1j1i2j2.....…irjs,l

follows directly from (4.37).

Now, the derivative of the outer and inner products are given by the
rules:

(A j1 j2… js
i1i2 .....ir
… jv
Birjs−−11.....iw ) ,l
= Ai1j1i2j2.....…irjs Birjs++11.....
… jv j1 j2… js js +1… jv
iw,l + Ai1i2 .....ir ,l Bir +1 .....iw,

Tensor Calculus 123

(A j1 j2 … js −1α
i1i2 .....ir
… jv
Birjs−−11.....iw −1α ) ,l
= Ai1j1i2j2.....…irjs,l−1α Birjs++11irj+s+22.....…iwjv−1α + Ai1j1i2j2.....…irjs−1α Birjs++11irj+s+22.....…iwjv−1α ,l

Now that we consider A j1 j2 Bi1i2 = Ci1j1i2j2 , we have
Ci1j1i2j2 j1 j2
Ci1j1i2j2,l l
C j1ij22 Ci1j1 j2 Ci1i2j2 Ci1j1i2
x i1 l i2 l l l
Bi1i2
A j1 j2 B i2 Bi1
xl i1 l i2 l
A j1 j2 j1 j2
j2
Bi1i2 A A j1
xl l l
A j1 j2 Bi1i2 ,l Bi1i2 A,jl1 j2
A j1 j2 Bi1i2 A j1 j2 Bi1i2 ,l Bi1i2 A,jl1 j2 (4.39)
,l
4.4.8 Covariant Differentiation of Relative Tensors

The covariant xl derivative of relative tensors are defined as follows. If f(x)
W
xi
is a relative tensor scalar of weight W such that g ( y ) f (x ) , then
yj
f
f ,l Wf (4.40)
xl l

This set of functions represents a relative vector of weight W.
Ai1j1i2j2.....
 js
ir
If Ai1j1i2j2.....
 js
ir ,l WAi1j1i2j2.....ijrs A j1ij22.....
 js
ir  Ai1j1i2j2.....
 js
xl i l i1 l ir l
j1 js
Ai1i2j2......ijrs .. Ai1j1i2j1.....
j2
ir .
l l

Example 4.4.1. Show that (gij Ai),k = Aj,k.

Solution: We have ( g ij Ai ),k = g ij ,k Ai + g ij A,ki (using Equation 4.39)
g ij A,ik , since g ij ,k 0
Ai i
g ij At (i)
xk t k

Aj ( g ij Ai ) g ij i Ai
Now, A g ij
xk xk xk xk
Ai
[ik , j]Ai [ jk ,i]Ai g ij
xk
Ai Aj
g ij [ik , j]Ai [ jk ,i]Ai (ii)
xk x k

By (i) and (ii) we get
Aj i
( g ij Ai ),k [ik , j]Ai [ jk ,i]Ai g ij At
xk t k
Aj
[ik , j]Ai [ jk ,i]Ai [tk , j]At
xk
Aj
[ jk ,i]Ai
xk
Aj
k
g li Al[ jk ,i]
x
Aj l
k
Al A j ,k , hence proved.
x j k

Example 4.4.2. Proved that if Aij is a symmetric tensor, then
1 Aij g 1 jk g jk
Aij, j j
A .
g x 2 xi

Tensor Calculus 125
Solution: Since Aij is a symmetric tensor, we get Aij = Aji.

Aij j l
We know that Aij,k Ail Alj .
xk l j i k
Aij j l
Putting k j, we get Aij, j Ail Alj
xj l j i j
Aij log g
Ail Alj g hl [ij ,h]
xj xl
Aij 1 g log g 1 (logg ) 1 g
Aij A jh[ij ,h]
xj 2g xj x j
2 xj 2g xj
1 Aij 1 g
g Aij A jh[ij ,h]
g xj 2 g xj
1
j
g Aij A jh[ij ,h]
g x
1
Aij, j j
g Aij A jh[ij ,h],
g x

1 jk g ik g jk g ij
but A jk[ij , k] A
2 xj xi xk
1 jk g ik g jk g ij
A A jk A jk
2 xj xi xk
1 jk g jk
A
2 xi
1 1 jk g jk
Aij, j g Aij A .
g xj 2 xi

k k
Example 4.4.3. Prove that i j i j are components of a tensor of
a b
k k
rank 3 where and are the Christoffel symbols formed from
i j i j
a b
symmetric tensors aij and bij.
Solution: We know from (4.19)
2 m
x k xm xp xq m
yi y j i J yk yi yj p q

k xm 2 m
x xp xq m
i J yk yi y j yi yj p q

k 2 m
x xp xq m yk
or
i J yi y j yi yj p q x m.

Using this equation, we can write
k 2 m
x xp xq m yk
i J yi y j yi yj p q xm
a a
k 2 m
x xp xq m yk
.
i J yi y j yi yj p q xm
b h
Subtracting, we get
k k m m xp xq yk
i J i J p q p q yi yj xm
a b a b
Tensor Calculus 127
m m
Putting Ampq , we can write the equation
p q p q
a b
xp xq yk
Aijk Ampq .
yi yj xm
k k
i j i j are the components of a tensor of order 3.
a b
Example 4.4.4. If Aij is a skew-symmetric tensor, show that
1
A,iji g Aij .
g xi

Solution: We know
Aij i j
A,ijk Arj Air .
xk k r k r

Putting k = i, we get
Aij i j
A,iji Arj Air
xi i r i r
Aij i i
Arj (If Aij is skew symmetric tensor, then A jk 0)
xi i r j k
Aij
i
Arj r log g
x x
A ij
1 g)
i
Arj
x g xr
1 Aij g
g i
Aij (replacing dummy index r with i)
g) x xi
1
i
Aij g
g x

1 1
Example 4.4.5. If Aijk is a skew-symmetric tensor, show that i
g Aijk
g x
is a tensor.
Solution: We have
Aijk i j k
A,ijk
l Arjk Airk Aijr ,
xl r l r l r l

Aijl i j l
hence A,ijll Arjl Airl Aijr .
xl r l r l r l
 j 
We know Airl  =0
 r l 
Aijl l l
A,ijll Aijr , Also log g
xl r l r l xr
Aijl
Aijr log g
xl xr
Aijl 1 ijr
A g
xl g xr
1 1
r
Aijr g k
Aijk g .
g x g x

1
Since A,lijl is a tensor, the right hand side k
Aijk g is also a
tensor. g x
Example 4.4.6. If Aij is a skew-symmetric tensor of rank two, show that
Aij A jk Aki
Aij ,k A jk ,i Aki , j .
xk xi xj
Solution: We know that

Tensor Calculus 129
Aij l l
Aij ,k Alj Ail
xk i k j k
A jk l l
A jk ,i Alk A jl
xi j i j k
Aki l l
Aki ,j Ali Akl
xj k j i j
Aij l l A jk l
Aij ,k A jk ,i Aki ,j Alj Ail Alk
xk i k j k xi j i
Akii l l
Ali Akl
xj k j i j
Aij A jk Aki l l
Alj A jl
xk xi xj i k k i
l l l l
Ali Ail Alk Akl .
k j j k j i i j

l l
Since , etc.
i k k i
Aij A jk Aki l l l
( Alj A jl ) ( Ali Ail ) ( Akl Alk ),
xk xi xj i k k j i j

Since Aij is a skew-symmetric tensor of rank two, then Alj + Ajl = 0, Ali +
Ail = 0, Akl + Alk = 0
Aij A jk Aki
Aij ,k A jk ,i Aki , j .
xk xi xj
4.5 Gradient, Divergence, and Curl

The tensorial nature of partial derivatives is a very useful feature. We can
apply it to extend the scope of classical operations of vector analysis. The
equations in theoretical physics are expressed with the help of a small
number of operators called gradient, divergence, curl, and Lapacian.
4.5.1 Gradient
The partial derivative of an invariant ϕ (a scalar function of a coordinate)
is a covariant vector which is called the gradient of ϕ and is denoted by
gradϕ or ∇ϕ.
grad
xi
Indeed, the nabla operator applied to a scalar field, which is a tensor
field of type (0,0), produces a tensor field of type (0,1).
4.5.2 Divergence
Let Ai be a contravariant vector. Then, its covariant derivative, given by
i Ai i
A ,j A ,
xj j

is a tensor of type (1,1). If we contract the indices i and j, we get the tensor
A,ii , which is the tensor of type (0,0), the invariant.
The divergence of contravariant vector Ai is defined by
i Ai i
divA A , which is sometimes written as A,ii .
xi i

Now, we derive an expression for the divergence of a contravariant vector Ai
1
divAi A,ii k
g Ak , where g |g ij |. (4.41)
g x

Proof: i Ai i
A ,i A
xi i
Ai
log g A
xi x
Ak
k k
log g Ak
x x
1
k
g Ak
g x

Tensor Calculus 131
Divergence of a Covariant Vector: Let us consider an arbitrary covari-

ant vector Ai. Then, gjkAj,k is an invariant. It is the divergence of Ai and is
denoted by divAi, therefore, div Ai = gjkAj,k.
The divergence of contravariant vector Ai is said to be the divergence of
Ai.
div Ai = g ik A j ,k = ( g ik A j ),k , as g ,kjk = 0

= A,kk = divA k = divAi .

If Ai and Ai are the covariant and contravariant components of same
vector, then div Ai = divAi.
Theorem 4.5.1. Let A be an arbitrary vector and ϕ and φ be a scalar func-
tion of coordinates xi. Then,
(i) div(ϕA) = grad(ϕ)A + ϕdivA
(ii) ∇(ϕφ) = ϕ∇φ + φ∇ϕ
(iii) ∇2(ϕφ) = ϕ∇2φ + φ∇2ϕ + 2∇ϕ.∇φ
(iv) div(φ∇ϕ) = φ∇2ϕ + ∇ϕ.∇φ
Proof:
1 (4.42a)
(i) We know divAi A,ii i
g Ai
g x
1
div( Ai ) ( Ai ),i i
g Ai
g x
1
g Ai g Ai
g xi xi
1
Ai i i
g Ai
x g x
1
grad( )A divA since divA k
g Ak
g x

div(ϕA) = grad(ϕ)A + ϕdivA (4.42b)
(ii) By definition of gradient,
( )
( ) i
x xi xi
( )
(4.42c)
(iii) Taking the divergence on both sides in Equation (4.42c),
div∇(ϕφ) = div[ϕ∇φ + φ∇ϕ]
or ∇2(ϕφ) = div(ϕ∇φ) + div(φ∇ϕ)
= ∇(ϕ)∇φ + ϕdiv∇φ + ∇(φ)∇ϕ + φdiv∇ϕ (by 4.42b)
= φdiv∇ϕ + ϕdiv∇φ + 2∇(φ)∇ϕ
(iv) Replacing A by ∇φ in (4.42b), we get
div(ϕ∇φ) = grad(ϕ)∇φ + ϕdiv(∇φ)
= ∇ϕ.∇φ + φ∇2ϕ.
4.5.2.1 Divergence of a Mixed Tensor (1,1)

Let Aij be a arbitrary tensor of type (1,1). Then, the div of Aij
divAij = Aij ,i .

Tensor Calculus 133
Aij i l
We know Aij ,k Alj Ali (using 4.38).
xk k l j k
Putting k = i, we get,
Aiij i l
Aiij ,i A ji Allj i Alii l
A j ,i x i A j i l Al j i
x i l j i
1 Aj i
g l i
1 g Aiji Aiij g Alii l i log g
g g x i A j x ii Al j i i l x ll log g
g x x j i i l x
Aiij log g l
A ji Aiij log i g Alii l ,
xi Aj xi Al j i , changing dummy index i by l from the
x x j i
2nd term,
1 g Aij l
Ali .
g x i
j i
1 j
Example 4.5.1. Prove that divAij j A,ijj g Aij Ai .
g x j
a k
Find the expression for Aij when Aij is skew-symmetric.

Solution: We know
Aij i j
A,ijk Arj Air .
xk k r k r

Putting j = k, we get
Aij i j
A,ijj Arj Air
xj k r j r
Aij i l
Arj Air log g using log g
xj k r x r
r l xr
Aij Air i
g Arj
xj g x r
k r
A ij
A ij i
g A jk replacing dummy index r by k.
xj g x j
k j
1 i
Aij g A jk (i)
g k j

2nd part: If Aij is skew-symmetric, then Aij = −Aji.
Interchanging indices j and k, we get
i i i i i
A jk Akj A jk Akj A jk 0.
k j j k j k k j k j

1
Putting this value in (i), we get divA
ij
A,ijj Aij g .
g xj
Example 4.5.2. If the components A1, A2, and A3 of a vector in cylindrical
2
coordinates, x1, x2, and x3 are x1, x3 sinx2, and e x cosx 3 , prove that
x3 2 x2 3
divAi = 2 + 1 2 cosx − e sinx .
(x )
Solution: (ds)2 = (dx1)2 + (x1)2 (dx2)2 + (dx3)2
Then, g11 = 1, g22 = (x1)2, g33 = 1, gij = 0 when i ≠ j
1 0 0
2
∴ detg g g ij 0 x1 0 x1
0 0 1
1
We get g11 = 1, g 22 = , g33 = 1, and gij = 0, when i ≠ j.
( x 1 )2
2
A1 = x 1 ,A2 = x 3sinx 2 ,A3 = e x cosx 3

Tensor Calculus 135
A1 = g1i A1 = g11A1 = x1
1
A2 = g 2i Ai = g 22 A2 = 3
1 2 x sinx
2
(x )
2
A3 = g 3i Ai = g 33 A3 = e x cosx 3

1
divAi divAi k
g Ak
g x
1
g A1 g A2 g A3
x1 x 1
x 2
x 3
1 1 3 2
( x 1x 1 ) x1 x sinx 2 (x1e x cosx 3 )
x1 x 1
x 2 1 2
(x ) x 3
1 x3 2
1
2 x 1
1 2
(sinx 2 ) 3
(x1e x cosx 3 )
x x x x
3
1 x 2
1
2 x1 1
cosx 2 x1e x sinx 3 )
x x
3
x 2
2 1 2
cosx 2 e x sinx 3
(x )
Example 4.5.3. Prove that in a V2 with line element (ds)2 = (dx1)2 + (x1)2
(dx2)2, the divergence of the covariant vector with components x1cos2x2,
−(x1)2sin2x2 is zero.
Solution: g11 = 1, g22 = (x1)2, gij = 0 when i ≠ j
detg = g = |gij| = (x1)2
1
g 11 = 1, g 22 =
( x 1 )2
A1 = x1cos2x2, A2 = −(x1)2sin2x2.
A1 = g1iA1 = g11A1 = x1cos2x2

1
A2 = g 2i A1 = g 22 A2 = 1 2 2
1 2 ( −( x ) sin 2 x ) = − sin 2 x
2
(x )
1
divAi divAi k
g Ak
g x
1
1 1
g A1 2
g A2
x x x
1 1
1 1
(x1x1cos2 x 2 1 2
(x1. sin2 x 2 )
x x x x
1 1 1
1
2 x .cos2 x 2 1
( x1 ).2.cos2 x 2
x x
2.cos2 x 2 .2.cos2 x 2 0

If the divergence of a covariant vector or contravariant vector vanishes

identically, then the vector i said to be conservative.
4.5.3 Laplacian of an Invariant

Let ϕ be an invariant and A . Then, Ai is a covariant vector. Hence,
i i
x
gijAj = Ai is a contravariant vector. The divergence of this contravariant
vector is called the Laplacian of ϕ.
The Laplacian of an invariant ϕ is denoted by ∇2ϕ.
Thus, ∇2ϕ = div (gradϕ) (4.43)
An Expression for ∇2ϕ:
1
2
k
g g kj
g x xj

1
Proof: We know A,ii g Ak . (i)
g xk
Let Ai g ij A j g ij .
xj
Tensor Calculus 137
i 1
(i) takes the form A,i k
g g kj .
g x xj
Now, by definition, ∇ 2φ = A,ii
1
2
k
g g kj (4.44)
g x xj

4.5.4 Curl of a Covariant Vector

A skew–symmetric tensor of type (0,2) formed from a covariant vector,
assuming particular importance in E3, where it can be identified with a
vector. We now introduce such a skew-symmetric tensor in, which is called
the curl of the covariant vector.
Let us consider a covariant vector Ai. Then, Ai,j is a tensor of type (0,2).
Hence, Aj,i is also a tensor of type (0,2). Consequently, Ai,j – Aj,i is a tensor
of type (0,2), which is evidently skew- symmetric. This skew-symmetric
tensor is called the curl of vector Ai and is denoted by curlAi.
An Expression for the Curl of Covariant Vector Ai
Ai Aj
curl Ai Ai , j A j ,i (4.45)
xj xi
Ai r
Proof: We have Ai,j
xj i j
Aj r
and A j ,i
xi j i

Ai r Aj r
curl Ai Ai , j A j ,i
xj i j xi j i
Ai Aj r r

xj xi i j j i

Example 4.5.4. Show that curl gradϕ = 0 for invariant ϕ.

Solution: Let Ai grad .
xi
Ai xi
2
Aj xj
2
Then, and .
xj x j j
x x i
xi xi xi x j
Ai Aj 2 2
Now, curlAi 0
xj xi x j xi xi x j
or curlAi = curlgradϕ = 0.
Theorem 4.5.2. A necessary and sufficient condition is that the curl of the
vector field vanishes if the vector field is gradient.
Proof: Let Ai be a covariant vector. Let the curl of the vector Ai vanish so
that
Ai Aj
curlAi Ai , j A j,i 0.
xj xi
We have to show that Ai = ∇ϕ where ϕ is a scalar.
Ai Aj
curlAi 0 0
xj xi
Ai j Aj i
dx dx
xj xi
dAi i
( A jdx i )
x
∂ ∂

⇒ Ai =
∫ ∂ x ( A dx ) = ∂ x ∫( A dx ),
i j
i
i j
i
Tensor Calculus 139
but if ∫( A j dx i ) is a scalar quantity, let ∫( A j dx i ) = φ . Then,
Ai grad .
xi
⇒ the vector Ai is a gradient.
Conversely, let vector Ai be a gradient such that Ai = gradϕ,
i.e., we have to show that curlAi = curlgradϕ = 0.
Conversely,
Ai Ai
xi
2 2
Ai
xj x j xi xi x j
Ai Aj 2 2
0
xj xi x j xi xi x j
Ai Aj
or curl Ai Ai , j A j ,i 0
xj xi
If the vector field is gradient, the curl of the vector field vanishes.
Example 4.5.5. If Aij are the components of the curl of a covariant vector,
prove that
Aij,k + Aki,j + Ajk,i = 0.

Solution: Let Aij be the components of the curl of a covariant vector Bi. We
have
Aij = curlBi = Bi,j – Bj,i.

Interchanging indices i and j, Aji = Bj,i – Bi,j = (Bi,j – Bj,i) = −Aij.

⇒ Aij is skew-symmetric.
Bi Bj
Aij curlBi Bi , j B j ,i
xj xi
2
Aij Bi 2
Bj
Aij ,k (i)
xk x xjk
x xi
k

Aij l l
Now, Aij ,k Alj Ail (ii)
xk i k j k

2
Bi
2
Bj l l
Alj Ail (by (i)) (iii)
x xj
k
x xi
k
i k j k

Putting i = j, j = k, k = i in (iii), we get
2
Bj 2
Bk l l
A jk ,i Alk A jl (iv)
x xk
i
x xj
i
j i k i

Putting i = j, j = k, k = i in (iv), we get
2
Bk 2
Bi l l
Aki ,j Ali Akl (v)
x xi
j
x xi
j
k j i j

Adding (ii), (iv), and (v), we get
Aij ,k Aki , j A jk ,i
2
Bi Bj 2
l l
Alj Ail
x xj
k
x xi k
i k j k
2
Bj 2
Bk l l
Alk A jl
x xk
i
x xj
i
j i k i
Bk2
Bi 2 l l
Ali Akl 0.
x xi
j
x xij
k j i j

Tensor Calculus 141
The result is equivalently written as
𝛻k Aij + 𝛻i Ajk + 𝛻j Aki = 0.
4.6 Exercises
1. rove that all Christoffel symbols vanish at a point, if and only if
P
gij’s are all constant at the point.
i
2. If Aij is a skew-symmetric tensor, show that Aik 0.
j k
1 qm g qm
3. If Apq is a symmetric tensor, show that Aqm[i ,q ,m] A .
2 xi
4. Show that the number of independent components of the
1
Christoffel symbols in a Vn is, at most, n 2 (n + 1).
2
5. Calculate the non-zero Christoffel symbols corresponding to
metrics
(i) (ds)2 = (dx1)2 + (x1)2(dx2)2 + (dx3)2
(i) (ds)2 = (dx1)2 + f(x1, x2)(dx2)2 + (dx3)2
(ii) (ds)2 = a2[(sinhy1)2 + (siny2)2]{(dy1)2 + (dy2)2] + (dy3)2
6. alculate the Christoffel symbols of the 2nd kind in (i) rectangu-

C
lar, (ii) cylindrical, and (iii) spherical coordinates.
7. Show that A,i = ij 1 (
∂ Aij g )
, where Ai is a skew-symmetric
j
i
tensor. g ∂ x
8. (i) If Aij is a symmetric tensor, show that Aij,k is symmetric in i
and j.
(ii) If Aij is a symmetric tensor such that Aij,k = Aik,j, show that Aij,k is a
symmetric tensor.
9. If Ai is a contravariant vector such that A,ki = ak Ai where ak is a
covariant vector, show that ak is gradient.
r
r
Aijk r
10. Prove that Aijk,l Ar jk Air k Aijr Aijk
x i l j l k l l
is a tensor.
11. I f the contravariant components Ai of a vector in spherical coor-

dinates r, θ, and ϕ are r, 2cosθ, and – ϕ, find curlAi.
If ϕ is scalar, then gijϕ,ij is a scalar and is equal to
12.
1 ∂  kj ∂φ 
 gg .
g ∂x k  ∂x j 
5
Riemannian Geometry
5.1 Introduction
The covariant derivative of a tensor is, in general, a tensor. If the resulting
tensor is again subjected to covariant differentiation, then again we get a
tensor and so on. In the case of an invariant, the operation of covariant
differentiation is commutative, but a tensor of order greater than or equal
to one of the operation is not commutative. This is due to a peculiarity in
which the operation is undertaken, namely a Riemannian space.
Characteristic peculiarity of such a space in a certain tensor is called
the curvature tensor whose components can be expressed with the help of
components of fundamental tensors. The operation of covariant differenti-
ation is not, in general, commutative in a Riemannian space.
We discuss in this chapter Riemannian-Christoffel tensors, Ricci’s ten-
sor and its properties, and discuss Riemannian and Euclidean spaces.
5.2 Riemannian-Christoffel Tensor ∂ 2u

A sufficient condition for the equality of mixed partial derivatives
∂x∂ y
∂ 2u
and , of a function u(x,y), is that u(x,y) be of class C2, but this is not
∂ y∂ x
sufficient to ensure the equality of mixed covariant derivatives. Indeed, it
is shown that if the order of covariant differentiation is to be immaterial,
our tensors must be defined over a particular metric manifold X, for which
a certain tensor of rank four, made up entirely of the gij’s, vanishes. This
tensor is known as the Riemann-Christoffel tensor. It plays a basic role in
many investigations of Differential Geometry and other branches of engi-
neering sciences.
The covariant derivative of a tensor is a tensor; hence it can be differenti-
ated again to obtain a new tensor which is called the 2nd covariant derivative
of the given tensor.
143
Consider the covariant xj derivative of Ai (with respect to gij)
∂ Ai  α  (5.1)
Ai , j = −   Aα
∂ x j i j 
If aij = Ai,j, then aij,k = Ai,jk.

Now, differentiate (5.1) covariantly with respect to xk, their results
should be a tensor.
∂ Ai , j  α  α 
Ai , jk = k −  Aα , j −   Ai ,α
∂x i k   j k
∂  ∂ A  α    α  ∂ A  β  
= k  ij −   Aα  −    αj −   Aβ 
∂ x  ∂ x i j   i k   ∂ x α j  
 α  ∂ A  γ  
−    αi −   Aγ  (5.2)
 j k  ∂ x i α  
and
∂  ∂ Ai  α    α   ∂ Ai  β  
Ai ,kj = −   Aα −   −  Aβ
∂ x j  ∂ x k i k   i j   ∂ x k α k  
 α  ∂ A  γ  
−    αi −   Aγ  (5.3)
k j   ∂ x i α  
Now (5.2) and (5.3) can be differentiated as indicated:
α 
∂ 
2
∂ A i j  α ∂ A
Ai , jk = k i j −  k  Aα −   αk
∂x ∂x ∂x i j  ∂ x
 α ∂ A  α  β 
−   αj +     Aβ
i k  ∂ x i k  α j 
 α  ∂ A  α  γ 
−   αi +     Aγ (5.4)
 j k∂ x  j k  i α 
Riemannian Geometry 145
α 
∂ 
i k  A −  α  ∂ Aα
2
∂ Ai
Ai ,kj = j k − α  
∂x ∂x ∂xj i k  ∂ x
j
 α ∂ A  α  β 
−   αk +     Aβ
i j  ∂ x i j  α k 
 α  ∂ A  α  γ 
−   αi +     Aγ (5.5)
k j  ∂ x k j  i α 
If we subtract (5.5) from (5.4), we get
i j i k
Ai , jk Ai ,kj A A A A
i k j x k
i j k xj

and the interchanging of α and β in the first terms of each preceding line
gives
i k i j
Ai , jk Ai ,kj A .
xj xk i k j i j k

(5.6)
Since Aα is a covariant tensor of rank 1 and on right hand side the dif-
ference of two Ai,jk − Ai,kj is a covariant tensor of rank 3, then by Quotient
Law we conclude that within in the bracket of (5.6), there is a mixed tensor
of rank 4, i.e.,
α  α 
∂  ∂ 
i k  − i j  +  β   α  −  β   α  = Rα .
      ijk

∂xj ∂ xk i k  β j  i j  β k 
If the left hand side of (5.6) is to vanish, that is, if the order of covariant
derivative is to be immaterial, then Rijk 0, but since Aα is arbitrary, in gen-
eral Rijk ¹ 0, so that the order of covariant differentiation is not immaterial.
Since Christoffel symbols are functions of gij’s, it follows that tensor Rijk
is formed exclusively from gij and its derivatives up to second order.
This tensor Rijk was first introduced by G.B. Riemann and E.D.
Christoffel and called the Riemann-Christoffel Tensor. It is also named the
Riemann-Christoffel Curvature Tensor of type (1,3) or curvature tensor.
It is clear from (5.6) that a necessary and sufficient condition for the
validity of the inversion of the order of covariant differentiation is so that
the vector R ijk vanishes identically.
∂ ∂ α  α 
   
∂ xk ∂ xl  j k  j l
Rijkl ≡ + (5.7)
 i  i  α  α 
       
 j k  j l  j k  j l

is called the mixed Riemann-Christoffel tensor or the Riemann-Christoffel

tensor of the second kind.
The associated tensor
Rijkl ≡ g ia Rαjkl (5.8)

is known as the covariant Riemannian-Christoffel tensor, or the

Riemannian-Christoffel tensor of the 1st kind.
∂ ∂  α  α 
   
Rijkl ≡ ∂ x k ∂ x l +  j k  j l (5.9)
[ jk , i][ jl , i] [ik ,α ][il ,α ]

A formula which is a special case of (5.6) was established by Ricci and

is given here.
Ai1 ....im , jk⋅ − Ai1 ....im ,kj⋅ = ∑αm=1 Ai1 ....iα −1hiα +1⋅  .im, .Rihα jk

In the special case of a tensor of rank 2:
Aij ,kl − Aij ,lk = Aiα Rαjkl + Aα j Rikl

α
.

Next we consider a contravariant vector Ai
i∂ Ai  i  α
Then, A =,j + A (5.10)
∂ x j α j 
(aij = A,ij , then aij ,k = A,ijk . from (5.1) we get the following )

Now,
∂ aij  i  α 
i
(A ) = a
, j ,k
i
j ,k = k + aαj   − ali  
∂x k a   j k
∂  ∂Αi  i  α α 
i  i 
α 
= +  A  + A, j   − A,α  
∂ x k  ∂ x j α j  k α   j k
∂  ∂Αi  i  α   ∂Αα  α  α   i 
= + A + + A  
∂ x k  ∂ x j α j    ∂ x j α j   k α 
 ∂Αi  i  α   α 
− α + A   
 ∂x α α    j k 
 i 
∂ 
α j  Aα +  i  ∂Α + ∂Α  i 
2 i α α
∂ (A )
A,ijk = k +   k  
∂x ∂xj ∂ xk α j  ∂ x ∂ x j k α 
 i   α  m ∂Αi  α   α   i  m
+  A − α   −   A (5.11)
k α  m j  ∂ x  j k   j k  α m 

Interchanging j and k, we get from (5.11)
 i 
∂ 
α k  Aα +  i  ∂Α
2 i α
∂ (A )
A,ikj = +   k
∂ x j ∂ xk ∂xj α k  ∂ x
∂Αα  i   i   α  m
+ k +   A
∂ x  j α   j α  m k 
∂Αi  α   α  i  m
− α   −   A (5.12)
∂x k j  k j  α m 
Using Young’s Theorem for calculus of several variables, we get
∂ 2 ( Ai ) ∂ 2 ( Ai )
=
∂ xk ∂ x j ∂ x j ∂ xk
  i   i 
 ∂ α j  ∂ α k  
∴ A,ijk − A,ikj =   k  −  j   Aα
 ∂x ∂x 
 
  m   i   m   i  α
+  −    A
 α j  m k  k α   j m 

(interchanging α and m in the last two terms of (5.12))
  i   i  
 ∂   ∂  
 α j α k    m   i   m   i   α
=  − +   −    A
 ∂ x k ∂ x j   α j  m k  k α   j m  
  
 
i
= Rakj Aα
i
= − Rajk Aα

  l   l  
 ∂   ∂    
i k i j   m  l   m   l  
l
Now, g hl Rijk = g hl   j  −  k   +     −   
 ∂ x ∂ x   i k  m j   j i  k m  
  
  
∂   l    l  ∂ g hl ∂   l 
= j  g hl    −   j − k  g hl   
∂ x  i k   i k  ∂ x ∂ x  i j  
 l ∂ g  m  l   m  l 
+   hlk + g hl     − g hl    
i j  ∂ x i k  m j   j i  k m 
∂   l  ∂   l   l 
=  g hl    −  g hl    −   ([hj, l] + [l j,h])
∂ x j  i k   ∂ x k  i j   i k 
 l  m  m 
+   ([hk, l] + [l k,h]) + [mj, h]  − [km, h] 
i j  i k   j i
∂ ∂  l 
= j [ik , h] − −
k [ij , h]   ([hj, l] + [l j,h])
∂x ∂x i k 
 l   l   l 
+   ([hk, l] + [l k,h]) + [l j,h]  − [kl , h] 
i k  i k   j i
∂ ∂  l   l  (5.13)
= j [ik , h]
− k [ij , h] +   ([hk, l] −   ([hj, l]
∂x ∂x i j  i k 

The tensor of type (0,4) is defined as
l
Rhijk = g hl Rijk .
(5.14)
Equation (5.13) is called Riemann-Christoffel Curvature Tensor of type

(0,4).
If we multiply this equation by ghβ and sum, we get
g hβ Rhijk = g hβ g hl Rijk
l
= δ lβ Rijk
l β
= Rijk (5.15)

The Riemann-Christoffel tensor of the second kind is obtained by rais-

ing the first covariant index in the tensor Rhijk.
5.3 Properties of Riemann-Christoffel Tensors

If we determine the properties of the set of functions defining the Riemann-
Christoffel tensor of the first kind, we expand the determinants in (5.9) and
insert Christoffel’s symbols into it. We get
∂ ∂  α  α 
  
Rijkl = ∂ xk ∂ xl +  j k j l
[ jk , i][ jl , i] [ik ,α ][il ,α ]
∂ ∂ α  α 
= k[
jl,i] − l [ jk,i] +  [il ,α ] −  [ik ,α ]
∂x ∂x  j k  j l
∂ 1  ∂ g li ∂ g ji ∂ g jl  ∂ 1  ∂ g ki ∂ g ji ∂ g jk 
= + − − + −
∂ x k 2  ∂ x j ∂ x l ∂ x i  ∂ x l 2  ∂ x j ∂ x k ∂ x i 
 α  p   p  α 
+ g pa     − g pa    
 j k  i l  i k   j l 
1  ∂2 g ∂ 2 g ji ∂ 2 g jl ∂2 g ∂ 2 g ji ∂ 2 g jk 
=  k li j + k l − k i − l kt j − l k + l i 
2 ∂x ∂x ∂x ∂x ∂x ∂x ∂x ∂x ∂x ∂x ∂x ∂x 
 α  p   p  α 
+ g pa     − g pa    
 j k  i l  i k   j l 
1  ∂2 g ∂ 2 g jl ∂2 g ∂ 2 g jk   α  p 
=  k li j − k i − l ki j + l i  + g pa    
2 ∂x ∂x ∂x ∂x ∂x ∂x ∂x ∂x   j k  i l 
 p  α 
− g pa     (5.16)
i k   j l 
We get the following equations from (5.16):

(a) Rjikl = −Rijkl
(b) Rijlk = −Rijkl
(c) Rklij = Rijkl
(d) Rijkl + Riklj + Riljk = 0
(e) Rijkl + Rklj
i i
+ Rljk =0
The Riemann-Christoffel tensor of (a) and (b) are skew-symmetric

with respect to the first two and last two indices, respectively, and (c) is
symmetric with respect to a group of two indices. It follows from these

identities that distinct, nonvanishing components of Rjikl are of three types:
(i) Symbols with two distinct identities, i.e., Rijij
(ii) Symbols with three distinct identities, i.e., Rijik
(iii) Symbols with four distinct identities, i.e., Rijkl.
This is to easily verify that the total number of n distinct non-vanishing
components of
n n(n 1)(n 2)(n 3)

Rijkl .
4 12

Property 5.3.1. The number of distinct non-vanishing components of the
1
covariant curvature tensor does not exceed n 2 (n 2 − 1).
12
Proof: We know that Riemann-Christoffel curvature tensor satisfies 4 of
the following properties:
i) Rijkl = − Rijkl 
(Skew − Symmetry Property)
ii) Rijkl = − Rijlk 
iii) Rijkl = Rklij (Block Symmetry Property)
iv) Rijkl + Riklj + Riljk = 0, (Cyclic Property )

Due to the above properties, all n4 components of Rijkl are not all non-
vanishing independent components. The following 4 cases may arise:
Case 1. All 4 indices are the same, i.e., Riiii

Case 2. There are only 2 distinct indices, i.e., Rijij
Case 3. Three indices of Rijkl are distinct, i.e., Rijik
Case 4. When all indices i,j,k,l all are different
Case 1. We consider Riiii. it is self-skew symmetric, i.e., Riiii = − Riiii [by i)

and ii)].
⇒2Riiii = 0
⇒Riiii = 0
⇒Riiii has no non-vanishing component.

Case 2. The two indices, i and j, can be selected for arrangement. For
n n 1
I, n ways and for j, (n 1) ways, so the total number of
1 1
selections = n(n − 1).
Since Rijkl is skew-symmetric in the first two and last two indices, i.e.
Rijij = − Rjiij =−(−Rjiji) = Rjiji.
⇒If two indices are interchanged, these two components are the same,
i.e., the number of components would be half.
For symmetry properties, no change of components means no reduction.
Moreover, for cyclic
Rijij + Riijj + Rijji = Rijij + 0 – Rijij = 0 (by (ii) and (iii)),
so, due to the cyclic property, there is no deduction.

1
The total number of components = n(n − 1).
2
Case 3. We consider that there are 3 distinct indices i, j, and k.
As in Case 2, the number of selections of components
 n   n −1   n − 2 
= = n(n − 1)(n − 2).
 1   1   1 
Due to the symmetric property, Rijik = Rikij the number of selections is

1
reduced to n(n − 1)(n − 2).
2
Due to skew-symmetric property there is no reduction. Since Riijk = 0,
Rijik = 0 also occur for the cyclic property, there is no reduction (since
Bianchi identity is self-satisfied).
Case 4. When i, j, k, and l have four distinct indices, i can be chosen by
 n   n −1   n − 2   n − 3 
= n(n − 1)(n − 2)(n − 3) ways.
 1   1   1   1 
By skew symmetric property, we have
Rijkl = −Rjikl and Rijkl = −Rijlk.

The number of independent non vanishing components reduces to

1
n(n − 1)(n − 2)(n − 3).
4
Also for the symmetry property, i.e. (Rijkl = Rklij), the number of non-
vanishing components reduces to
1
n(n − 1)(n − 2)(n − 3)
8
and for cyclic property, we get Rijkl + Riklj + Riljk = 0 or Rijkl = −(Riklj + Riljk).
Therefore, one among every three can be expressed in terms of the
2
remaining two terms. Hence, it would be reduced by th number of terms,
so the total number of components 3
2 1
= ⋅ n(n − 1)(n − 2)(n − 3)
3 8
1
= n(n − 1)(n − 2)(n − 3).
12
Combining all cases, the total number of Components of Curvature

Tensor of order 4
1 1 1
= 0 + n(n − 1) + n(n − 1)(n − 2) + n(n − 1)(n − 2)(n − 3)
2 2 12
1
{ 1
= n(n − 1) 1 + (n − 2) + (n − 2)(n − 3)
2 6 }
1 1
= n(n − 1) (6n − 6 + n 2 − 5n + 6)
2 6
1 1
= n(n − 1)(n 2 + n) = n 2 (n 2 − 1).
12 12
Theorem 5.3.1
(α) Rijkl = − Rijlk
(β) Rijkl + Rklj
i i
+ Rljk =0
i
(γ) Rikl = 0
(δ) Rijkl = −Rjikl
Proof: (α) We know
∂ ∂  i  i 
   
∂ xk ∂ xl  j k  j l
Rijkl = +
 i  i  α  α 
       
 j k  j l  j k  j l
∂ i  ∂  i   i  α   i  α 
=  −   +    −   
∂ xk  j l ∂ xl  j k  j k j l  j l j k

Interchanging k and l we get from above equation
∂  i  ∂  i   i  a   i  α 
Rijlk =  − k   +    −   
∂ xl j k ∂ x  j l  j l j k  j k j l
= − Rijkl .

(β) We know
∂  i  ∂  i   i  α   i  α 
Rijkl =  −   +     −    . (i)

Similarly,
i
i i i i
Rklj …․ (ii)
x l
k j x j
k l k l j l k j k l

and
i
i i i i
Rljk (iii)
x j
l k x k
l j l j j l l k l j

i i i
Adding (i), (ii), and (iii), we get R jkl + Rklj + Rljk = 0
∂  i  ∂  i   i  α   i  α 
γ) We know Rijkl =  −   +     −    .
Putting j = i, we get from above
i ∂  i  ∂  i   i  α   i  α 
Rikl =  −  +   −   
∂ x k i l  ∂ x k i l  i l  i k  i l  i k 
[in 2nd term interchange dummy indices k and l]

=0
δ) We know,
1 ∂2g ∂ 2 g jl ∂2g ∂ 2 g jk 
Rijkl =  k li j − k i − l ki j + l i 
2 ∂x ∂x ∂x ∂x ∂x ∂x ∂x ∂x 
 α  p   p  α 
+ g pα     − g pα     (from 5.16)
 j k  i l  i k   j l 

Interchanging i and j, we get
1  ∂ 2 g lj ∂ 2 g il ∂ 2 g kj ∂2g 
R jikl =  k i − k j − l i + l ik j 
2 ∂x ∂x ∂x ∂x ∂x ∂x ∂x ∂x 
 α  p   p α 
+ g pα     − g pα    
i k   j l   j k  i l 
(interchanging dummy index p and α)
1  ∂ 2 g lj ∂2g ∂ 2 g kj ∂2g 
=  k i − k il j − l i + l ik j 
2 ∂x ∂x ∂x ∂x ∂x ∂x ∂x ∂x 
 p  α   α  p 
+ g α p     − g α p     = − Rijkl
i k   j l   j k  i l 
Similarly, we can easily prove by using the above equations.

Theorem 5.3.2
(i) Rklij = Rijkl
(ii) Rijkl + Riklj + Riljk = 0
Proof: (i) We know,
1  ∂ 2 g li ∂ 2 g jl ∂ 2 g ki ∂ 2 g jk 
Rijkl =  k j − k i − l j + l i 
2 ∂x ∂x ∂x ∂x ∂x ∂x ∂x ∂x 
 α  p   p  α 
+ g pα     − g pα     ... (i)
 j k  i l  i k   j l 
Interchange i and k in (i) and we get
1 ∂2g ∂ 2 g jl ∂2g ∂ 2 g ji 
Rkjil =  i lk j − i k − l ik j + l k 
2 ∂x ∂x ∂x ∂x ∂x ∂x ∂x ∂x 
 α  p   p  α 
+ g pα     − g pα    
 j i  k l  k i   j l 
Now, interchange j and l and we get
1  ∂ 2 g jk ∂ 2 g lj ∂2g ∂ 2 g ji 
Rklij =  i l − i k − j ik l + j k 
2 ∂x ∂x ∂x ∂x ∂x ∂x ∂x ∂x 
α  p   p  α 
+ g pα     − g pα    
l i   k j   k i  l j 
Since the symmetric property of gij and 2nd Christoffel’s symbols,
1  ∂ 2 g li ∂ 2 g jl ∂ 2 g ki ∂ 2 g jk 
Rklij =  k j − k i − l j + l i 
2 ∂x ∂x ∂x ∂x ∂x ∂x ∂x ∂x 
 α  p   p  α 
+ g pα     − g pα    
 j k  i l  i k   j l 
= Rijkl, which is known as the block symmetric property.

(ii) From (i),
1 ∂2g ∂ 2 g jl ∂2g ∂ 2 g jk 
Rijkl =  k li j − k i − l ki j + l i 
2 ∂x ∂x ∂x ∂x ∂x ∂x ∂x ∂x 
 α  p   p  α 
+ g pα     − g pα    
 j k  i l  i k   j l 
1  ∂ 2 g ji ∂ 2 g kj ∂2g ∂2g 
Riklj =  l k − l i − j li k + j kl i 
2 ∂x ∂x ∂x ∂x ∂x ∂x ∂x ∂x 
 α  p   p  α 
+ g pα     − g pα    
k l  i j  i l  k j 
1 ∂2g ∂2g ∂ 2 g ji ∂ 2 g lj 
Riljk =  j ki l − j lk i − k l + k i 
2 ∂x ∂x ∂x ∂x ∂x ∂x ∂x ∂x 
 α  p   p  α 
+ g pα     − g pα    
l j  i k  i j  l k 

Adding these three equations, we get

Rijkl + Riklj + Riljk = 0, which is known as the cyclic property.
Example 5.3.1. Calculate the components of Rijkl of the Riemannian tensor

for the metric
ds2 = (dx1)2 − (x2)−2(dx2)2.
Solution: Here, g11 = 1, g22 = −(x2)−2,g12 = g21 = 0, and
1 0
g 2 2
(x 2 ) 2
0 (x )

g11 = 1, g12 = g21 = 0, g22 = (x2)2
The only non-vanishing Christoffel symbol of 2nd kind is
 2  22 2 2 1 ∂ 2 −2
 22  = g [22,2] = ( x ) 2 ( −( x ) )
  2 ∂ x

1
= ( x 2 )2 (−2)( x 2 )−3 = −( x 2 )−1
2
α
Using equation (5.6), Riemannian Curvature Tensor Rijk
1 ∂  1  ∂  1 
R212 =  + 1  2 2 
∂ x 2  2 1  ∂ x  
 1   β   1   β 
+  −  β 2  =0
 β 1   2 2     2 1 

2 ∂  2  ∂  2 
R212 =  + 1  2 2 
∂ x 2  2 1  ∂ x  
 2   β   2   β 
+  −  β 2   = 0.
β1
   2 2     2 1 
We know Rijkl g i R jkl
α 1 2
R1212 = g 1α R212 = g 11 R212 + g 12 R212 = 0.

5.3.1 Space of Constant Curvature

An n-dimensional Riemannian space Vn is said to be a space of constant
curvature if its Riemann-Christoffel curvature tensor can be expressed as
Rijkl = ρ(δ li g jk − δ ki g jl ), where ρ is a constant.

and Rijkl = ρ(gilgjk − gik gjl).
5.4 Ricci Tensor, Bianchi Identities, Einstein Tensors

5.4.1 Ricci Tensor
α
We define the Ricci Tensor as Rij = Rijα .
We have
∂ ∂  l   l 
   
l
∂ x j ∂ xk α j α k 
Rijk = +
 l  l  α  α 
      
i j  i k  i j  i k 
∂  l  ∂  l   l  α   l  α 
=  − k  +    −    .
∂xj i k  ∂ x i j  α j  i k  α k  i j 

Putting k = l, we get
l l l l
Rijll
xj i l xl i j j i l l i j
l l
j i
log g l
x x x i l j i l
l
log g [ log g .]
x i j i l xi

Therefore,
∂ 2 log g ∂  α   α   β   β  ∂ 
Rij = i −  +     −    β log g 

∂x ∂xj
∂ xα i j  β j  i α  i j  ∂ x
(5.17)
(changing dummy index α to β and l to α).

Now, interchanging i and j
∂ 2 log g ∂  α   α   β   β  ∂ 
R ji = j −   +    −    β log g  .

i
∂x ∂x ∂ xα  j i  β i   j α   j i  ∂ x
Interchanging α and β in the third term,
∂ 2 log g ∂  α   β   α   β  ∂ 
R ji = j −   +    −    β log g  .

i
∂x ∂x ∂ xα  j i  α i   j β   j i  ∂ x
Therefore,
Rij = Rji (5.18)
Tensor Rij is symmetric since Rij = Rji is one of the distinct components
1
of Rij is n(n + 1) . In a four dimensional manifold n = 4, so that, if we
2
set Rij = 0, we obtain 10 partial differential equations, which Einstein has
adopted as his equations of the gravitational field in free space in the gen-
eral theory of relativity.
5.4.2 Bianchi Identity

In the development of that theory, another tensor introduced by Einstein
plays an important role. The tensor is obtained from an identity the called
Bianchi Identity.
Theorem 5.4.1. Rijkl ,m + Rijlm ,k + Rijkm ,l = 0
We know
i i i i
Rijkl
xk j l xl j k j k j l j l j k
i i i s i s
Rijkl
x m
xm xk j l xl j k j k j l j l j k
i i s i i s
xm xk j l xm xl j k j l xm j k j l xm j k
s i i s
j k x m
j l j l x m
j k

From the covariant derivative of a mixed tensor, we have
∂ i i 
α  i  α  i α  α  i 
Rijkl ,m = m R jkl − Rα kl   − R jα l   − R jkα   − R jkl  
∂x  j m k m  l m  α m 
∂2  i  ∂2  i   s  ∂  i 
=   − l  +   m  
∂ xm ∂ x k m
 j l ∂ x ∂ x  j k  j l ∂ x  j k
 i  ∂  s   s  ∂  i   i  ∂  s 
+   m  −   m  −   m  
 j l ∂ x  j k  j k ∂ x  j l  j l ∂ x  j k
 α  i  α  i α  α  i 
− Rαi kl   − R jα l   − R jkα   − R jkl  
 j m k m  l m  α m 
Similarly, by changing k, l, and m cyclically and adding the resulting

equations, we get
Rijkl ,m + Rijlm ,k + Rijkm ,l = 0. (5.19)

This equation is known as ‘Bianchi’s Second identity’.

i i
Multiplying R jkl ,m by ghi ie, g hi R jkl ,m = Rhjkl ,m .
Equation (5.19) becomes
Rhjkl,m + Rhjlm,k + Rhjmk,l = 0. (5.20)
Flat Space
i
A Riemannian space whose Riemannian-Christoffel tensor R jkl vanishes
identically is called a flat space. Otherwise, it is called non-flat.
Example 5.4.1. The Minkowski Space-Time with metric
ds2 = (dx1)2 + (dx2)2 + (dx3)2 − c2(dx4)2
is a flat.
Solution: Here, ds2 = (dx1)2 + (dx2)2 + (dx3)2 − c2(dx4)2/
We know ds2 = gijdxidxj
∴g11 = g22 = g33 = 1 , g44 = −c2. Here, c is the speed of light
and gij = 0 when i ≠ j.
∂ g ij
Hence, = 0 for all i , j , k = 1,2,3,4.
∂ xk
Its implied that all Christoffel symbols are zero and therefore the
Riemann-Christoffel curvature tensor also vanishes identically.
⇒Minkowski Space-Time is a flat.
Example 5.4.2. In a V2, the components of the Ricci tensor are propor-
tional to the components of the metric tensor.
The components of the Ricci tensor are R11, R12, R21, and R22 and com-
ponents of metric tensor are g11, g12, g21, and g22. Both are symmetric with
respect to indices.
We know Rij = gklRlijk = gklRilkj (by the property of Ricci tensor)
∴Rij = g11Ri11j + g12Ri12j + g21Ri21j + g22Ri22j
∴R11 = g11R1111 + g12R1121 + g21R1211 + g22R1221
= 0 + 0 + 0 +g22R1221 [here R1111 = 0 and R1121 = −R1211]

g 11 cofactor of g ij of g
= (− R1212 ) , where g ij =
g det g

R11 R
∴ = − 1212
g 11 g

Similarly, R22 = g 11 R2112 + g 12 R2122 + g 21 R2212 + g 22 R2222
g
= g 11 R2112 + 0 + 0 + 0 = − 22 R1212
g
R22 R
= − 1212
g 22 g
Lastly, R12 = g 11 R1112 + g 12 R1122 + g 21 R1212 + g 22 R1222

g
= g 21 R1212 = − 12
g
R12 R
= − 1212
g 12 g
Since both Rij and gij are symmetric and from above relation we get
R12 R
= − 1212 .
g 12 g

R11 R22 R12 R12 R1212
Combining all the cases,
g 11 g 22 g 12 g 12 g
∴The components of the Ricci tensor are proportional to the compo-
nents of the metric tensor.
Example 5.4.3. Prove that in a V2
R
Rijkl ( g ik g jl g il g jk ).
2
Solution: We know that in V2 the components of the Ricci tensor
are proportional to the components of the metric tensor.
R11 R22 R12 R12 R1212

= = = −
g 11 g 22 g 12 g 12 g

R1212
∴ or Rij = − g ij
g
We know that the covariant curvature tensor vanishes if its three or
more indices are same.
∴R1111 = R2222 = R1112 = R1121 = R1211 = R1121 = R2111 = R1211 = R1222 = R2111 = 0
R
Now, we have to show that the given relation Rijkl = − ( g ik g jl − g il g jk )
follows easily. 2
The non-vanishing components are R1212, R2121, R1221, and R2112.
Now, we have R = g ij Rij

R1212
= g ij (− g ij )
g
1
= −2 R1212 (i , j = 1,2),
g
R
so R1212 = − g
2
R g 11 g 12
=−
2 g 21 g 22
R
= − ( g 11 g 22 − g 21 g 12 ).
2
Changing indices with symmetric property of gij, we get
R
R2121 = − ( g 22 g 11 − g 12 g 21 )
2
R
R1221 = − ( g 12 g 21 − g 11 g 22 )
2
R
R2112 = − ( g 21 g 12 − g 22 g 11 ).
2
R
Combine these relations and we get Rijkl = − ( g ik g jl − g il g jk ).
2
Example 5.4.4. Prove that in a Riemannian space div(Rijkl ) R jl ,k R jk ,l .
Solution: We know the Bianchi Identity
Rijkl ,m + Rijlm ,k + Rijkm ,l = 0

Contracting in the above relation, we get
Rijkl ,i + Rijli ,k + Rijki ,l = 0

Rijkl ,i = − Rijli ,k − R jk ,l
or
= R jl ,k − R jk ,l
∴ div( Rijkl ) = R jl ,k − R jk ,l

∂
Example 5.4.5. If Rij,k = 2BkRij + BiRkj + BjRik, prove that Bk = k (log R )
Solution: We have Rij,k = 2BkRij + BiRkj + BjRik. ∂x
Using the Associate Tensor Formula, B j = gijBi, and the Scalar Curvature.
R = gijRij, we get
gijRij,k = 2Bk gijRij + gijBiRkj + gijBjRik
= 2BkR + B jRkj + BiRik
R,k = 2BkR + BiRki + BiRik = 2BkR + 2BiRik ….(i) [ Rik = Rki ]
From the given condition, we get
Rij,k − Rik,j = BkRij − BjRik,
or gijRij,k − gijRik,j = Bk gijRij − Bj gijRik,
1
or R,k − Rkj , j = Bk R − Bi Rik ,
2
1
or R,k R,k Bk R Bi Rik
2
or R,k = 2BkR − 2BiRik (ii)
From (i) and (ii) it follows that BiRik = 0

and by this value, from the equations, we get
R,k = 2BkR
1 1 ∂R ∂
∴ Bk = R,k = k = (log R ).
2R 2R ∂ x ∂ xk
5.4.3 Einstein Tensor

1 i
Theorem 5.4.2. Prove that the tensor Rij j R is divergence free.
2
Proof: We know Equation (5.20), Bianchi’s Second identity
Rijkl,m + Rijlm,k + Rijmk,l = 0.

Multiply it by gilgjk and we get
gilgjkRijkl,m + gilgjkRijlm,k + gilgjkRijmk,l = 0,
or gilgjkRijkl,m + gilgjkRijlm,k + gilgjkRijmk,l = 0,
or gjkRjk,m + gilgjk (−Rijml,k) + gilgjk (−Rjimk,l) = 0.

Since Rijlm = −Rijml, Rijmk = −Rjimk,
∴ gjkRjk,m − gjkRjm,k −gilRim,l = 0,
k l
or R,m Rm,k Rm,l 0  g jk R jk R

or R,m − Rmk ,k − Rmk ,k = 0
∴ R,m − 2 Rmk ,k = 0
1
∴ Rmk ,k − R ,m = 0,
2
k 1 k k
or Rm,k m R,k 0,  R,m m R,k ,
2
1
or  Rmk − δ mk R  = 0, (5.21)
 2  ,k
k jk
where Rm = g R jm .
k 1 k
⇒ Rm m R is divergence free.
2
The tensor of type (1,1) is defined by
1
Rij − δ ij R ≡ G ij and (5.22)
2
is called the Einstein Tensor.

Scalar Curvature: The Scalar Curvature R of a Riemannian space is defined
by
R = gijRij.
Example 5.4.6. In a V2 where g11 = g22 = h > 0 and g12 = g21 = 0 are functions
of x1 and x1, show that
R
Rij = g ij ,
2
where Rij is the Ricci Tensor and R is the Scalar Curvature.

Solution: We know Rij = gklRlijk
= g11R1ij1 + g12R1ij2 + g21R2ij1 + g22R2ij2 (i)
h 0
detg ij = = h2
0 h

cofactor of g ij in g cofactor of g ij in g
g ij = =
g h2

h 1 12
g 11 = g 22 = 2 = , g = g 21 = 0
h h
1
Rij (R1ij1 R2ij 2 )
h
1 1
R11 (R1111 R2112 ) R2112
h h
1 1
R22 (R1221 R2222 ) R2112
h h
1
R12 (R1121 R2122 ) 0
h
1
R21 (R1211 R2212 ) 0
h
Now, we know R g ij Rij
g 11R11 g 12 R12 g 21R21 g 22 R22

1 11 2
(R11 R22 ) 2R2112 R2112
h hh h2
1 1R 2 R R
R11 = R1221 = h = h = g 11
h h2 2 2
1 1R 2 R R R R
R22 R2112 h h g 22 and R12 g 12 0 and R21 g 21 0
h h2 2 2 2 2
R
Rij g ij .
2
1
Example 5.4.7. If in a Vn (n > 2), Rij − g ij R = 0 , show that Rij = 0.
2
1
Solution: We have Rij − g ij R = 0 (i)
2
Multiplying the given relation by gij, we get
1
g ij Rij − g ij g ij R = 0,
2
1
or R − n R = 0,
2
2−n
or R = 0 . Hence, R = 0.
2
From (i), we get for R = 0,
Rij = 0.
Example 5.4.8. If g ik Rkj = Rij and g ij Rij = R,

1 ∂R
show that Rij ,i = .
2 ∂x
Solution: Using covariant differentiation on both sides of g ik Rkj = Rij , we
get
g ik Rkj ,l Rij ,l [ g ,ikl 0].

Hence, Rij ,i g ik Rkj ,i g ik Rkjl

l l
,i [ Rkjl Rkj ]
g ik ( g pl R pkjl ), i [ g pl R pkjl l
Rkjl ] (i)
g ik g pl R pkjl ,i[ g ,pli 0]
Using a Bianchi Identity, we can write Rpkjl,j + Rpkli,j + Rpkij,l = 0. (ii)

Using (ii), we can write (i)
Rij,i g ik g pl R pkjl,i

= −gikgpl(Rpkli,j + Rpkij,l)
= −gpl(gikRpkli,j + gikRpkij,l)
( )
= − g pl − g ki Rkpli, j + g ik Rkpji,l   R pkli, j = − Rkpli, j and R pkij,l = Rkpji,l 
g pl Ripli, j Ripji,l

= −gpl (−Rpl,j + Rpj,l)

= (gplRpl),j − gplRpj,l
= gplRpl,j − gplRpj,l
R, j Rij,l (replacing dummy index l by i)

Hence, 2 Rij ,i = R , j
or Rij ,i = 1 R, j = 1 ∂ R .
j
2 2 ∂x
5.5 Einstein Space

A Riemannian space Vn is said to be an Einstein Space if its Ricci Tensor can
be expressed as
Rij = λgij, (5.23)
where λ is a scalar.
Now, using inner multiplication by gij, we get
gijRij = λgijgij.
Since R = gijRij and gijgij = n
∴R = nλ
R
or λ = ,
n
from (5.23) we get
R
Rij = g ij . (5.24)
n
R R ij
The equation Rij = g ij = ρ g ij, where ρ = and R = g Rij, is known
n n
as the Einstein Gravitational Equation at points where matter is present. It
corresponds to the Poisson Equation, ∇2V = ρ, in the Newtonian Theory
of Gravitation.
Theorem 5.5.1. Prove that the Scalar Curvature of an Einstein Space is
constant.
Proof: We know from Einstein Space that
R
Rij = g ij .
n
1
Covariant differentiating, we get Rij,k R,k g ij [ g ij,k 0].
n
1
Inner multiplying by gil, we get g il Rij,k R,k g ij g il
n
1 l
or R lj,k j R,k .
n
1 l
Now, contracting l and k, we get R lj,l j R,l
1 1 n
or R, j = R, j .
2 n
1 1
∴( − )R, j = 0,
n 2
∴R,j = 0
Hence, for Einstein Spaces, the scalar curvature R is constant.
5.6 Riemannian and Euclidean Spaces

5.6.1 Riemannian Spaces
Let n-dimensional space Vn be in an x-coordinate system and metrize it by
prescribing the element of arc ds, such that
ds2 = gij dxi dxj (5.25)
There is a positive definite quadratic form in the differential dxi. We

assume that the functions gij(x) are to be of class C1 in Vn.
The space Vn so metrized is called a Riemannian n-dimensional Space Rn.
Let a coordinate system Y, defined by the Y-frame, are given by
T: yi = yi (x1,…..,xn).
Apply a symmetric tensor gij(x) so that gij(x) has a constant component,

hij, throughout in Rn.
We note first that the components of gij(x), when reffered to the Y-frame,
are given by
∂ xα ∂ x β
hij = g αβ . (5.26)

∂ yi ∂ y j
 k 
If hij’s are constants, then Christoffel symbols   vanish iden-
 i j  y

hij ∂ hij
tically, hij,l , and, since hij,l = 0, by Ricci theorem, we have =0
y l
∂ yl
in Rn.
Theorem 5.6.1. A necessary and sufficient condition that the metric coef-
ficients gij(x) reduce to constants hij in some reference y-frame is that
 k 
Christoffel symbols   vanish identically.
 i j  y
From this theorem we can deduce a system of differential equations
that must be satisfied by the functions yi(x1,… . .,xn), if there is to be a y-
coordinate system in which the hij’s are constants.
We know the Law of Transformation is
∂ y α ∂ y β  m  ∂ 2 y m  γ  ∂ y m
−   = i j −  γ
∂ x i ∂ x j  α β  y ∂ x x  i j ∂x

 m 
and, since  = 0 , we have the system of equations
α β 
 y
∂ 2 y m  γ  ∂ y m
−  =0 (5.27)
∂ xi x j  i j  ∂ xγ
 x
System (5.27) of second order partial differential equations can be

rewritten as a system of 1st order partial differential equations.
y
ui (i 1, 2,  n)
xi (5.28)
ui
u( 1, 2,. . n)
xi i j

This system, in general, will be incompatible. For existence of the solu-

tion, we form these conditions in a symmetric form and consider the
system:
∂ fα α 1 2 m 1 2 n
i = Fi ( f , f , . f , x , x , x ). (α = 1,2.. m; i = 1,2, n ),
∂x
(5.29)
where Fi are known functions of f ‘s and x’s. Equation (5.29) is special-

ization of (5.28) if we set f 1 = y, f 2 = u1, … f m = un,. Function Fi is defined
over the n-dimensional region R. Let us refer to the region of definition
of functions as R'. This region consistsof R of variables xi and −∞ < f i < ∞.
We will suppose that the functions Fi are of class C1 in R'. Since R' is
∂ Fiα
open, we will assume that are bounded in R’.
∂ fi
Since Fi are of class C1 in R' , it follows that f α's are of class C2 and hence,
∂2 f α ∂2 f α
= . (5.30)
∂ xix j ∂ x jxi
This is the necessary condition for the integrability of system (5.29).

Now we obtain
∂ 2 f α ∂ Fiα ∂ Fiα ∂ f β
= +
∂ xix j ∂ x j ∂ f β ∂ x j
∂ Fiα ∂ Fiα β
= + Fj , from (5.29).
∂xj ∂ f β
From (5.30), we get a necessary condition of integrability, the set of

equations
∂ Fiα ∂ Fiα β ∂ Fjα ∂ Fjα β

+ Fj , = + Fi , (α , β = 1,2.. m;i, j = 1, 2 ,n).
∂ xj ∂ f β ∂ xi ∂ f β

(5.31)
The dependent variables in (5.28) are y, u1, …. un, and in (5.29) are f1, …fm,
i.e., f 1 = y, f 2 = u1, …… f n+1 = un,. Then, (5.28) gives
∂ f1 1
i = Fi = ui (i = 1, 2, n )
∂ x
∂ fα α  γ   α = 2,3n + 1
and = Fi =  uγ  .
∂ xi α − 1i   i ,γ = 1,2,n 
If we substitute the values of Fi in (5.31), we get
u u (5.32)
i j j i
Rkiju 0

The first sets of equations are satisfied identically for the symmetric
property of the Christoffel symbol and states that the set equations of (5.28)
i
will have a solution if the Riemann-Christoffel tensor Rkij vanishes iden-
tically, since this tensor vanishes when metric coefficients are constants.
Theorem 5.6.2. A necessary and sufficient condition that a symmet-

ric tensor gij with g ij ≠ 0 will reduce under a suitable transformation of
coordinates to a tensor hij, where the hij’s are constants, is that the Riemann-
Christoffel tensor formed from the gij’s be a zero tensor.
5.6.2 Euclidean Spaces

If the quadratic form Q = hij yiyj is a positive definite, there exists a nonsin-
gular linear transformation reducing Q to the canonical form Q = (y1)2
+⋯.+(yn)2.
Thus, if gij(x) are the coefficients in the positive definite quadratic dif-
ferential form
ds2 = gijdxidxj, (5.33)
characterizing metric properties of there exists a real functional

transformation
T: yi = yi(x), which reduces it to the form
ds2 = (dy1)2 +⋯..+(dyn)2, (5.34)

i
provided that R jkl vanishes identically in Rn.
A metric manifold Rn in which it is possible to affect the reduction of
ds2 = gij dxi dxj to ds2 = (dy1)2 +⋯..+(dyn)2 is called an Euclidean n-
dimensional manifold (En) and we see that
Rn is Euclidean if, and only if, the Riemann Tensor of the manifold is a
zero tensor.
5.7 Exercises
∂ ∂ α  α 
1. Show that Rijkl = k [ jl , i] − l [ jk , i] +  [il,α ] −  [ik,α ].
2. Show that ∂x ∂x  j k  j l
1  ∂ 2 g li ∂ 2 g jl ∂ 2 g ki ∂ 2 g jk 
Rijkl = Rijkl =  k j − k i − l j + l i 
2 ∂x ∂x ∂x ∂x ∂x ∂x ∂x ∂x 
+ g αβ ([jk, β][il, α] − [jl, β][ik, α]).

3. Show that Rαα jk = 0.
R
4. If Rij = ρgij, then ρ = where R = gijRij.
n
R R R R R
(i) If n = 2, show that 11 = 22 = 12 = 12 = − 1212 .
g 11 g 22 g 12 g 12 g
(ii) If n = 3, the tensor Rijkl has six distinct components and there are
six equations f or Rjk = gilRijkl. Prove that the solutions of these equa-
tions for Rijkl are given by
R
Rijkl = g il R jk + g jk Ril − g ik R jl − g jl Ril + ( g ik g jl − g il g jk ).
2
α
5. Find the expression for div. Rijk .
6. If in a Vn(n > 2)Rij,k + Rki,j + Rjk,i = 0, prove that the scalar curvature is
constant.
i i i
7. If A j = R j + δ j (aR +b), where a and b are constants and notations carry
i
their usual meanings, determine the value of a for which A j,i 0 .
R2
8. If in a Vn, Rij Rij ,
n
where R = g g Rpq and R = gijRij,
ij ip jq
R
prove that Rij = g ij.
n
9. If Ai is a covariant vector such that
r
Ai,j +Aj,j = 0, show that Ai , jk = − Ar Rkij .
10. If gklRij − gjkRil + gilRjk – gijRkl = 0, show that the space is an Einstein
Space.
11. If in a Riemannian space of dimension n(n>2),

h
Rijk ( )
= ρ δ kh g ij − δ hj g ik ,
where ρ is a constant, prove that the space is an Einstein Space.

6
The e-Systems and the Generalized
Kronecker Deltas
6.1 Introduction
The concept of symmetry and skew-symmetry with respect to pairs of
indices can be extended to cover the sets of quantities that are symmetric
or skew-symmetric with respect to more than two indices. We consider in
this chapter the sets of quantities Ai1ik or Ai1ik , depending on k indices,
written as subscripts or superscripts, although the quantities of A may not
represent tensors.
6.2 e-Systems
Definition 6.2.1. The system of quantities Ai1ik (or Ai1ik ) , depending on
k indicies, is said to be completely symmetric if the value of symbol A is
unchanged by any permutation of the indices.
Definition 6.2.2. The system of quantities Ai1ik (or Ai1ik ) , depending on

k indicies, is said to be completely skew-symmetric if the value of symbol A
is unchanged by any even permutation of the indices and A merely changes
signs after an odd permutations of indices.
For a skew-symmetric system it follows from Definition 6.2.2 that the
term containing two of the same indices is not necessarily zero. Thus, if
one has a skew-symmetric system of quantities Aijk where i, j, and k assume
values 1, 2, and 3, then A122 = 0, A123 = −A213, A312 = A123, etc.
In general, the components Aijk of a skew-symmetric system satisfy the
relations Aijk = −Aikj = −Ajik = Ajki = Akij = −Akji.
i i
Definition 6.2.3. If the values of A 1 n (or Ai1in ) are +1 when i1 … in is
an even permutation of the numbers 1,2,…n and -1 when i1 … in is an odd
177
permutation of numbers 1,2,…n and if it is zero in all other cases, then the
i i
system A 1 n (or Ai1in ) is called an e-system.
We consider four particular systems, two of which are of second order
and another two are of third order. They are represented by eij, eij and eijk,
eikj. The number of components of the first two are each 4 and last two are
27 each.
Now, the components of eij, eij are:
e11 = 0, e22 = 0, e12 = 1, e21 = −1
e11 = 0, e22 = 0, e12 = 1, e21 = −1 (6.1)
e123 = e231 = e312 = 1;
e213 = e321 = e132 = −1
and the remaining 21 components are zero.

e123 = e231 = e312 = 1 (if ijk is an even permutation of 123)
e213 = e321 = e132 = −1 (if ijk is an odd permutation of 123)
and remaining 21 components are zero.
These systems are called e–systems of the second and third order,
respectively.
The covariant ε tensor of second order is defined by
11 0,, g , g , 0, g g ij (6.2)

This tensor is skew-symmetric. According to the Tensor law of

Transformation,
u p uq u u u u
 ij i j
 pq or  ij i j
 i

u u u u u uj
The E-Systems and Generalized Kronecker Deltas 179
u1 u2
g ui ui as  22 0
11
u1 u2
uj uj
 11  22 0
(u1 , u 2 ), (u1 , u 2 ),
 22 g g and  21 g g.
(u1 , u 2 ) (u 2 , u1 )
Let us consider ϵij,

ϵij = ϵrs girgjs
which is called the contravariant ε tensor of second order.
Since ϵ11 = ϵ22 = 0, we get  ij rs g ir g js g g i1 g j 2 g i 2 g j1
∴ ϵ = ϵ = 0 and
11 22
1 1
 12 g [ g 11 g 22 g 12 g 21] g .
g g

1 1
and  21 g [ g 21 g 12 g 22 g 11] g .
g g
1 ij
Hence, we can write  ij e ,ij g eij (6.3)
g
The third order e-system eijk, eijk: If the e-system depends on three indi-
ces, ijk, then
eijk = 0 if any two indices are alike,
eijk = e123 = 1, if ijk is an even permutation of 123 and
eijk = –e123 = –1, if ijk is an odd permutation of 123.
e123 = e231 = e312 = 1 ; e213 = e321 = e132 = –1 (6.4)
e123 = e231 = e312 = 1 ; e213 = e321 = e132 = –1

ijk 1 ijk
It can be written that  e ; ijk g eijk , (6.5)
g
which are respectively contravariant and covariant tensors called permuta-

tion tensors in a 3-dimensional space. The formula of permutation tensor
for ϵijk or ϵijk are
+1, when i, j, k are in even permutation of 123.

eijk 0, when any two of indices i, j,k are alike (6.6)
1, when i,, j,k are in odd permutation of 123

We shall now establish some results using e-Systems of 2nd and 3rd order.
By means of these systems and Kronecker Delta, proofs of a number of
properties of determinants are considerably simplified. They are also useful
for writing down briefly different expressions important in the theory of
determinants.
Property 6.2.1 e-Systems of Second Order
a11 a12
Let aij a11a22 a12a12 e12a11a22 e21a12a12 [e12 1, e21 1]
2 2
a 1 a2
eija1ia2j (6.7)

Similarly, it can be shown that
aij e ijai1a2j . (6.8)
Let us now consider the expression eij aip aqj , where the indices p and q are
free and can be assigned values 1 and 2 at will.
Now we have eij a1i a2j = e12a11a22 + e21a12a12

= e12a11a22 − e12a12a12

i
= e12 a . j
i j i
Thus, eij a2a2 = e21 a j .
i j i
Similarly, eij a2a1 = e21 a j .
From these relations, we get eij aip aqj = e pq aij .
⇒ aij e pq = eij aipaqj (6.9)

i pq
Similarly, it can be shown that a j e e ijaipaqj . (6.10)
The above definitions of e-systems of second order can obviously be

i i
extended to define e-systems of nth order e 1 n and ei1in involving n
a11 a12 . . an1

a12 a22 . . an2
indices and aij .
.. .. .. ..
a1n a2n . . ann
6.3 Generalized Kronecker Delta

i ……i
Definition 6.3.1. A symbol δ j11…… jkk depending on k superscripts and k sub-
scripts, each of which runs from 1 to n, is called a Generalized Kronecker
Delta provided that
(i) it is completely skew-symmetric in superscripts and

subscripts;
(ii) if the superscripts are distinct from each other and the
subscripts are the same set of numbers as the superscripts,
the value of the symbol is +1 or −1 according to an even
or odd number of transpositions required to arrange the
superscripts in the same order as the subscripts;
(iii) in all other cases the value of the symbol is zero.
If we consider δ klij , it follows from definition that if i = j or k = l or if the

set i, j is not the set k, l, then δ klij = 0. In all other cases, δ klij equals to +1 or
−1 according to whether kl is an even or odd permutation of ij, i.e.,
δ kl11 = δ pq
22
= δ 1323 =  = 0 (except odd or even permutation)
and δ 1222 = δ 21
21 31
= δ 31 = δ 2323
= 1 (even permutation)
and δ 22
12
= δ 1331 = δ 31
13
= δ 1221 = −1 (odd permutation)
From Definition 6.2.3, it follows that the direct product e i1in ei1in of
the two systems e i1in and ei1in is a generalized Kronecker Delta.
For example, eαβγ eijk has the following values:
(a) 0, if two or more subscripts or superscripts are the same;

(b) +1, if the difference in the number of transpositions of αβγ and ijk
from 123 is an even number.
(c) −1, if the difference in the number of transpositions of αβγ and ijk
from 123 is an odd number.
(b) and (c) can be write another way
(b') +1, if even number of transpositions is required to arrange the
subscripts in the same order as the superscripts.
(c') +1 if an odd number of transpositions is required to arrange the
subscripts in the same order as the superscripts.
αβγ
We can write eαβγ eijk = δ ijk .
It implies that the e-symbols can be defined in terms of the Kronecker
Deltas.
e i1in = δ 12i1in
and δ i12n
1in = ei1in
n

Since e = + or −1, when the set of distinct integers i1 … in is obtained

from the set 1, 2, … n by an even or an odd permutation, e = 0 in all other
cases. The e–systems and generealized Kronecker Deltas prove useful in
calculations involving alternating sets of quantities.
Example 6.3.1. Evaluate eijeik, in e-systems of second order if i, j = 1,2.

Solution: By the summation convention, eijeik = e1je1k + e2je2k.
When j = k = 1 or 2, then
ei1ei1 = e11e11 + e21e21 = 1.1 + 0.0 = 1
ei2ei2 = e12e12 + e22e22 = 0.0 + 1.1 = 1.
Thus, when j = k, then eijeik = 1.

Again, when j ≠ k, say j = 1, k = 2 or j = 2, k = 1, we get
ei1ei2 = e11e12 + e21e22 = 1.0 + 0.1 = 0
ei2ei1 = e12e11 + e22e21 = 0.1 + 1.0 = 0.
Thus, if j ≠ k, then eijeik = 0.

Therefore, in general, eij e ik = δ jk for i, j = 1,2 …. n.
For particular, when j = k
eij e ik = δ ii = δ 11 + δ 22 + . + δ nn = 1 + 1 + .. + 1 = n.
6.4 Contraction of δ αijkβγ

ijk
Let us contract δ αβγ on k and γ. The result for n = 3 is
ijk ij1 ij 2 ij 3 ij
δ αβ k = δ αβ 1 + δ αβ 2 + δ αβ 3 ≡ δ αβ .

This expression vanishes if i = j or α = β. If we set i = 1 or j = 2, we
123 12 12
get δ αβ 3 = δ αβ and hence, δ αβ = 0 unless αβ is a permutation of 12, and
12 12
δ αβ = 1 if αβ is an even permutation of 12, δ αβ = −1 if αβ is an odd permu-
tation of 12.
ij
If we contract δ αβ , we obtain a system depending on two indices (first
1
contract it and the multiply the result by ).
2
i 1 ij 1 i1 i2 i3
j ( 1 2 3 )
2 2
1 1 1
If δ αi ≡ δ αij j = (δ αi11 + δ αi 22 + δ αi 33 ) in δ αi , we get δ α1 = (δ α122 + δ α133 ). It
2 2 2
vanishes unless α = 1, in which δ 11 = 1.
i
Similar results can be obtained by setting i = 2 or i = 3. Thus, δ α has
values
(a) 0 if i ≠ α
(b) 1, if i = α.
By counting the number of terms appearing in the sums, it is not diffi-

cult to show that, in general,
1 ij
δ αi = δ α j and δ ijij = n(n − 1) (6.11)
n −1
We can also deduce that
(n − k )! i1ir ir +1ik
δ ij11ir
 jr = δ j  j i i (6.12)
(n − r )! 1 r r +1 k
n!
And δ ii11
ir
ir = n(n − 1)(n − 2).(n − r + 1) = (6.13)
(n − r)!
As a special case of (6.13), we have the formula
e i1in ei1in = n! (6.14)
and from (6.12) we can deduce that
 jr
e i1ir ir +1in e j1 jr ir +1ik = (n − r )!δ ij11ir
(6.15)
i i
Example 6.4.1. If a set of np + q quantities A j11 jpq , i, j 1, 2,..n, and it sym-
metric in two or more indices (superscripts or subscripts), then show that
j j r r
δ i11iq q A j11 jpq = 0.

Solution: Suppose that Aij11ip

 jq is symmetric in j1 and j2, then
j1 jq r r j1 j2 jq r r j2 j1 jq r r

i1iq A j11 jp2 jq i1iq A j12 j1p jq i1iiq A j12 j1p jq .

However, j1 and j2 are the dummy indices, hence
j j r r r r
δ i11iq q A j11 j2p jq = −δ i1j1iqjq A j11 j2p jq .

Therefore, δ i1j1iqjq Arj11 rp

j2  jq = 0.
6.5 Application of e-Systems to Determinants and

Tensor Characters of Generalized Kronecker
Deltas i
The determinant aij of nth order, with elements a j , consists of the sum
of the products of the elements where each term in the sum contains one
and only one element from each row and each column of the determi-
nant. The sign of each term in the sum is determined by the character
of permutation of the indices. Thus, if the superscripts in the product
ai11 ai12 .ainn are arranged in the normal order 1, 2, … n, then the product
will sign if the number of transpositions necessary to arrange the sub-
scripts in the normal order is even. The sign is if the required number of
transpositions is odd.
Since e i1i2 in = δ 12i1in
n and ei1i2 in = δ i1in ,
12n
a11 a12 . . an1

a12 a22 . . an2
if we take aij a, we get the following results
.. .. .. ..
a1n a2n . . ann
a = aij = ei1i2 in a1i1 a2i2  .anin (by similar arguments as in the cases of (6.7),
(6.8), (6.9), and (6.10)
aij e i1i2in ai11ai22 .ainn

An example consider:
a11 a12 a31
i
We take a j = a12 a22 a32 ,
3 3 3
a1 a 2 a3
If this determinant is expanded to a = aij = ∑ a1i a2j a3k, where ijk is a permu-
tation of 1, 2, 3 the + or –sign is assigned to the term a1i a2j a3k according to
whether this permutation is even or odd. Hence, the determinant can be
written as aij = eijk aαi aβj aγk .
Consider next terms eijk aαi aβj aγk (i, j, k; α, β, γ = 1, 2, 3).
We will observe that this system is completely skew-symmetric in αβγ.

Since ijk are dummy indices, we can change them at will and write
eijk aαi aβj aγk = eijk aαk aβj aγi = eijk aγi aβj aαk .

If k and i are interchanged, this e–symbol will be changed
eijk aαi aβj aγk = −eijk aγi aβj aα.k .

It implies that an interchange of α and γ changes the sign, so that the

system under consideration is skew-symmetric in α and γ and from this
study we can write
eijk aαi aβj aγk = aij eαβγ .. (6.16)

Similarly, we can show that
e ijk aiα a βj akγ = aij eαβγ . (6.17)
It follows at once from these expressions that an interchange of two col-

umns (or two rows) of the determinant aij changes its sign and if two
columns (or two rows) in it are identical, then its value is zero.
This result can be generalized to determinants of the nth order, so that
for any permutation of rows we can write
eαβ…..γ aij = e ij……k aiα a βj  . .akγ (6.18)

and for any permutation of columns
eij…k aij = eαβ…..γ aiα a βj  . .akγ (6.19)

The expression of the determinant in terms of the elements of the first

column and their cofactors can be written as
aij = a1i1 ei1i2…in a2i1 …..ani1

= a1α Aα1, (6.20)
where Aα1 = ei1i2…in a2i1 …..ani1 is the cofactor of the element a1α .
Property 6.5.1 Product of Two Determinant

a11 a12 . . an1 b11 b21 . . bn1
i a12 a22 . . an2 b12 b22 . . bn2
If a j and bij ,
.. .. .. .. .. .. .. ..
a1n a2n . . ann b1n b2n . . bnn

then aij bij = c ij , where c ij = aαi bαj .

Proof: Since bij = eijkb1i b2j  .bnk,
we know aij bij = aij = eijkb1i b2j .bnk (since eijk aij = eαβ…..γ aiα a βj ..akγ )
= ( aij eijk )(b1ib2j .bnk )

(e .. ai a j .ak )(b1ib2j .bnk )

(
= eαβ…..γ )(aiα b1i )(a βj b2j )..(akγ bnk )
= eαβ…..γ )(c1α )(c2β )..(cγn ( )
= c ij ,

where c ij = aαi bαj = a1ib1j + a2i b 2j .. + anibnj .
Property 6.5.2 Partial Derivative of a Determinant

∂ a ∂ aβα β
Show that = Aα ,,
∂x ∂xj
i j
where a = aij and a j are the functions of the variables x1, x2, … xn and Ai
is the cofactor of element aij of = aij .
Solution: We know
a = ei1i2 in a1i1 a2i 2 .anin .
Differentiating this expression, we get
∂a ∂ ai1 ∂ ai 2 ∂ ain
= (ei1i2 in )( 1 a2i2 anin + a1i1 2 .anin  + a1i1 a2i2  n )
∂x ∂x ∂x ∂x
i1 i2 in
∂a ∂a ∂a
= ( 1 Ai11 + 2 Ai21 +  + n Ain1 ) [using Aα1 ] = ei1i2 in a2i1 ..ani1 ]
∂x ∂x ∂x
α
∂ aβ β
= Aα .
∂xj
Property 6.5.3 Permutation of ei1i2 in and e i1i2 in is relative tensors of
weight −1 and +1, respectively.
Proof: Let us consider an admissible transformation
T: yi = yi(x1, x2, … xn)

∂y ∂ yi
and its Jacobian j = . If we put aij = in e i1i2 in aij = eα1α 2 α n aiα1 1 aiα2 2 …..aiαn n
∂x ∂xj
1 ∂x
i1
1
ai2 2 …..aiαn n and consider = , we obtain
j ∂y
∂ x α1α 2α n i1 i2
e i1i2 in = e aα1 aα 2 …..aαinn
∂y
∂ x α1α 2αn ∂ y i1 ∂ y i2 ∂ y in (6.21)
= e ….. ,
∂y ∂ x α1 ∂ x α 2 ∂ x αn

which is the Law of Transformation of relative tensors of weight +1. In an

entirely similar way, we deduce that
−1
∂x ∂ x α1 ∂ x α 2 ∂ x αn
ei1i2 in = eα1α 2 α n ….. in , (6.22)
∂y ∂ y i1 ∂ y i2 ∂y

so that ei1i2 in is a relative tensor of weight −1.
6.5.1 Curl of Covariant Vector

Let Ai be a covariant vector. Then, Ai,k is a tensor of type (0,2). The product
ϵiklAl,k is then a tensor of type (1,0). If we denote this vector by Bi, then Bi is
called the curl of vector Ai.
Thus, in a V3,
1 ikl
Curl Ai = Bi = ϵiklAl,k = e Al ,k
g
1 ikl
Bi e Al ,k ,
g

1 1kl 1 123 1 132
or B1 e Al ,k e A3,2 e A2,3 (other eikl 0)
g g g
1
( A3,2 A2,3 ),
g
or B 2 = 1 e 2 kl Al ,k = 1 e 213 A3,1 + 1 e 231 A1,3 (other e ikl = 0)

g g g
1
= ( A1,3 − A3,1 )
g
1
and B3 = ( A2,1 − A1,2 ) .
g
Thus, components of curl Ai (components of Bi)

1 1 1
= ( A3,2 − A2,3 ), ( A1,3 − A3,1 ), ( A2,1 − A1,2 ).
g g g
Ai Aj
Since Ai , j A j,i ,
xj xi
A3 A2 A1 A3 A2 A1
A3,2 A2,3 , A1,3 A3,1 , A2,1 A1,2 ,
x2 x3 x3 x1 x1 x2
the components of curl Ai are
1  ∂ A3 ∂ A2  1  ∂ A1 ∂ A3  1  ∂ A2 ∂ A1 
 2 − 3  ,  − ,  − 
g ∂x ∂x g  ∂ x 3 ∂ x1  g  ∂ x1 ∂ x 2 

 1  ∂ A3 ∂ A2  1  ∂ A1 ∂ A3  1  ∂ A2 ∂ A1  
CurlAi =   2 − 3  ,  3 − 1  ,  1 − 2   .
 g ∂ x ∂ x g ∂ x ∂ x g ∂x ∂x 

6.5.2 Vector Product of Two Covariant Vectors

Let Ai and Bi be two covariant vectors of V3. Then, the product ϵijkAjBk is
a tensor of type (1,0). If we denote it by Ci, then Ci = ϵijkAjBk is called the
vector product (cross product) of vectors Ai and Bi.
Ci = ϵijkAjBk
C1 = ϵ1jkAjBk
= ϵ123A2B3 + ϵ132A3B2 others ϵ1jk = 0
1 ijk 1 ijk 1 132 1

= ( A2 B3 − A3 B2 ) (  e  123  )
g g g g

C2 = ϵ213A1B3 + ϵ231A3B1
1
= (− A1 B3 + A3 B1 )
g

and
1
C3 = ϵ312A1B2 + ϵ321A2 B1 = ( A1 B2 − A2 B1 ).
g

The components of the vector product of two vectors, Ai and Bi, are
1 1 1
( A2 B3 − A3 B2 ), (− A1 B3 + A3 B1 ), ( A1 B2 − A2 B1 ).
g g g

Hence,
 1 1 1 
A× B=  ( A2 B3 − A3 B2 ), (− A1 B3 + A3 B1 ), ( A1 B2 − A2 B1 ) .
 g g g 
Example 6.5.1. Prove that in a V3,
curl(ϕA) = ϕcurlA – A × gradϕ, where ϕ is an invariant.
Solution:
We have curlA = eijkAk,j.
Now, if ϕAk = Bk,
then Bk,j = ϕ,j Ak + ϕAk,j. (i)
Hence, curl(ϕA) = curlBk = eijkBk,j
= eijk(ϕ,j Ak + ϕAk,j)
= eijkϕAk,j − eikjϕ,j Ak [eijk = −eikj]

= ϕeijkAk,j − eikjAkϕ,j
=ϕcurlA – A × gradϕ [∵ A × B = eijkAjBk].
6.6 Exercises
ijk αβγ
1. Verify that δ αβγ a = aijk − aikj + a jki − a jik + a kij + a kji .
2. Show that
(a) eij e ik = δ jk
(b) eijk e irs = δ rj δ ks − δ krδ js
(c) erij erkl = δ ikδ jl − δ ilδ jk
3. Prove that ijk g is a covariant tensor of rank three where ϵijk is the
usual permutation tensor.
4. If in a V3, Ci = ϵipqApq, where Apq is a skew-symmetric tensor, show that
2Apq = ϵijtCt.
5. If Aij is a skew-symmetric tensor in V3, show that ( g A23 , g A23 , and
g A23 ) are the components of a covariant vector. Show further that if Aij
1 1 1
is a skew-symmetric tensor in V3, then ( A23 , A31 , and A12 )
g g g
are the components of a contravariant vector.
i i i
6. Show that the permutation symbols ei1i2 in and e 1 2 n are relative tensors
of weight −1 and +1, respectively, and also show that they are associated.
7. Show that the Kroneker Deltas behave as constants in a covariant
differentiation.
8. Show that if a set of quantities Ai1i2 ik is skew-symmetric in the sub-
scripts (k in number), then δ ij11ij22jjkk Ai1i2 ik = k ! A j1 j2  jk .
9. Show that the value of ei1i2…ir ir +1…in e j1 j2… jr jr +1 jn is n!.
10. Prove that ϵij is a covariant tensor of rank 2.
δ js δ tj
11. Show that∈
ϵijk
ijk
ϵistist =
∈ .
δ ks δ kt
Hence, deduce that
(i) ijk ijt 2 t

k (ii) ijk ijk 3!
12. Show that eimneirsamn = ars − asr.
13. Show that eijkeijk, where eijk and eijk are e –systems of third order are 6.
ij
δ αi δ βi ijk
δ αi δ βi δ γi
14. Show that δ αβ = and δ αβ γ = .
δ αj δ βj δ αj δ βj δ γk
Part II
DIFFERENTIAL GEOMETRY
7
Curvilinear Coordinates in Space
7.1 Introduction
What is the pure idea of coordinates? The pure idea is in representing points
of space by triple numbers. This means that we should have a one to one
map P(x1, x2, x3)↔(y1, y2, y3) in the whole space, or at least in some domain,
where we are going to use coordinates. In Cartesian coordinates, this map
is constructed by means of vectors and bases. The triplets of numbers (x1,
x2, and x3) are curvilinear coordinates of point P in R. Arranging other
coordinate systems, one can use other methods. For example, in spherical
coordinates, y1 = r is a distance from the point P to the center of a sphere
and, y2 = θ, y3 = ϕ are two angles. The discussion of this chapter is confined
mainly in curvilinear coordinates in E3 and studies reciprocal base systems
and the meaning of covariant derivatives.
7.2 Length of Arc

Consider that n−dimensional space R can be covered by an x-coordinate
system and that a one-dimensional subspace of R is determined by a curve,
C, so that
C: xi =: xi(t), i = 1,2 … n, (7.1)
where t is a real parameter in t1 ≤ t ≤ t2. The one-dimensional manifold C

is called an arc of a curve.
We deal in this book only with those curves for which x i(t) and
dx i
x i(t ) ≡ are continuous functions in t1 ≤ t ≤ t2.
dt
Let F( x 1 ,x 2 ,…..x n ,x 1 ,x 2 ,…..x n ) be a continuous function in the inter-
val t1 ≤ t ≤ t2. We suppose that F(x ,x ) > 0 unless every x = 0 and for every
positive number k,
195
n
1 2
1  2 n 1 2 n
1  2 n
F ( x ,x ,…..x ,kx ,kx ,…..kx ) = k F( x ,x ,…..x ,x ,x ,…..x ).
t2

The integral s =
∫t1
F(x ,x )dt

(7.2)
is called the length of C and space R is called metrized by (7.2).

We choose to define the length of arc by the formula
t2
dx α dx β

s=
∫t1
g αβ (x)
dt dt
dt , (α , β = 1, 2,…n,)

(7.3)
where gαβ(x) is a positive definite quadratic form in variable x α .

This resulting geometry is called Riemannian geometry and the space
is R-metrized and, in this way, is the Riemannian n‑dimensional space Rn.
Consider the admissible transformation of coordinates T: yi = yi(, x2,…..
xn), such that the square of arc element ds
ds2 = gαβdx αdxβ (7.4)
can be reduced to the form
ds2 =dyidyi. (7.5)
Then, the Riemannian manifold Rn is called an n −dimensional Euclidean

manifold En and the reference frame Y in which arc ds in En is given by (7.5)
is an orthogonal Cartesian reference frame. Here, En is the generalization
of a Euclidian plane determined by the totality of the pairs of real values
(y1, y2). The square of the element of arc ds assumes the familiar form
ds2 = (dy1)2 + (dy2)2.

 ) = kF(x ,x ) for every
A function F(x, x ), satisfying the condition F(x ,kx
k > 0, is called positively homogenous of degree 1 in the x i .
Curvilinear Coordinates in Space 197
 ) = kF(x ,x )
Theorem 7.2.1. A function, F(x ,x ) satisfies the condition F(x ,kx
for every k > 0. This condition is both necessary and sufficient to ensure
t2
independence of the value of ’ integral s =
∫ t1
F(x ,x )dt of a particular mode
of parametrization of C. Thus, if t in C : xi =: x i(t) is replaced by some func-
tion, t = ϕ(s), and we denote x i(ϕ(s)) by ξi(λ) so that: x i(t) = ξi(s) and we have
t2 t2
¢
equality
∫
F(x ,x )dt =
t1 ∫
F(ξ ,ξ )ds ,
t1
i
dx
¢i
where ¢i and t1 = ϕ(s1) and t2 = ϕ(s2).
ds
Proof: Suppose that k is an arbitrary positive number and put t = ks so that
t1 = ks1 and t2 = ks2.
Then, by (7.1), x i(t) = x i(ks)
dx i dx i(ks) dx i(ks) dt dt
and ¢ i kx i(ks).  k
ds ds dt ds ds
t2
Putting these values in s =
∫ t1
F(x ,x )dt, we get
ks2 ks2
s F[x(ks), x (ks)]dt F[x(ks), x (ks)]kds  dt kds
ks1 ks1
s2 i
(s), ¢ (s)]ds.
i i
F[  x(ks) (s )
s1
We must have the relation F( , ) F(x ,kx  ) kF(x , x )

Conversely, if this relation is true for every line element of C and each
k > 0 , then the equality of integrals is assured for every choice of parame-
ters t = ϕ(s), ϕ′(s) > 0, and s1 ≤ s ≤ s2 with t1 = ϕ(s1) and t2 = ϕ(s2).
Example 7.2.1. Consider a sphere, S, of radius a, immersed in a

3-dimensional Euclidean manifold E3, with a center at origin (0,0,0) of a
set of orthogonal Cartesian axes 0 – X1X2X3.
Solution: Let T be a plane tangent to S at (0,0, −a) and let the points on
this plane be referred to as a set of orthogonal Cartesian axes O′−Y1Y 2, as
shown in Figure (7.1).
If we draw from O(0,0,0) a radial line OM, intersect sphere S at M(x1, x2,
x3), and the plane T at O(y1, y 2, −a), then the points M on the lower half of
the sphere S are in 1 − 1 correspondence with points (y1, y2) of the tangent
plane T.
If M(x1, x2, x3) is any point on a radial line OM, then the symmetric
equation of this line is:
x1 − 0 x 2 − 0 x 3 − 0
= = =λ
y1 − 0 y 2 − 0 −a − 0

or x1 = λy1, x2 = λy2, x3 = −λa (7.6)
These are general points of sphere, so it satisfies the sphere with equation:
(x1)2 + (x2)2 + (x3)2 = a2,
or (λy1)2 + (λy2)2 + (−λa)2 = a2,
X3
O X2
M2
M1
C1
O’ Y2
X1
N2
Y1
C2
N1
Figure 7.1
or λ[(y1)2 + (y2)2 + (a)2] = a2
a2 a
∴λ 2 = 1 2 2 2 2 ⇒ λ=
( y ) + ( y ) + (a) ( y ) + ( y 2 )2 + (a)2
1 2

1 ay1 2 ay 2
x ,x , x3
1 2 2 2 2 1 2 2 2 2
(y ) (y ) (a) (y ) (y ) (a)
2
a
.
( y1 )2 ( y 2 )2 (a)2 (7.7)

These are the equations giving the analytical 1 − 1 correspondence of the

points of N on T and S.
Let M1(x1, x2, x3) and M2(x1 + dx1, x2 + dx2, x3 + dx3) be two close points
on curve C lying on S.
The Euclidean distance M1M2 , along C is given by
ds2 = dx idx i (i = 1,2,3) (7.8)

i
i x
and dx dy ( 1,2).
y
∂x i α ∂x i β ∂x i ∂x i α β
becomes ds 2 = dx i dx i =
7.8 becomes
(7.8) dy dy = dy dy
∂yα ∂yβ ∂yα ∂yβ
ds 2 = g αβ ( y )dy α dy β ,
xi xi
where g ( y ) are functions of y computed from (7.7).
y y
If the image C2 of C on T is given by
y1 y1(t )
C2 : t1 t t 2 ,
y2 y 2 (t )

then the length of C can be computed from integral

t2 t2

s=
∫t1
g αβ ( y )dy α dy β =
∫t1
g αβ ( y ) y α y β dt

1 1
(dy1 )2 dy 2 )2 (y y 2 y 1 y1 )
ds 2 a2 .
1
1 2 [( y1 )2 ( y 2 )2 ]
a (7.9)
1 1 2
t2 (dy 1 )2 + dy 2 )2 +  1 1
2 (y y − y y )
s=
∫ t1
1+
1 1 2
a
2 2
2 [( y ) + ( y ) ]
dt
We see that the resulting formulas refer to a 2-dimensional mani-

fold determined by the variables (y1, y2) in a Cartesian plane T and that
the geometry of the surface of the sphere imbedded in a 3-dimensional
Euclidean manifold can be visualized on a 2-dimensional manifold R2 with
a metric determined by (7.9). 1
If the radius of S is very large, in (7.9), implying 2 → 0 and the geom-
a
etry of the surface of the sphere is then approximated by the Euclidean
metric
ds2 = (dy1)2 + dy2)2. (7.10)
For a large value of radius a, metric properties of the sphere S are indis-
tinguishable from those of the Euclidean plane and the sum of the angles
of the curvilinear triangle drawn on S will be nearly equal to 180° by
Euclidean geometry.
The main point of this example is to indicate that the geometry of a
sphere, imbedded in a Euclidean 3-dimensional space with the element of
arc in the form of (7.8) is indistinguishable from the Riemannian geometry
of a 2-dimensional manifold R2 with metric (7.9).
The latter manifold, referred to as Cartesian frame Y, is not Euclidean
since (7.9) cannot be reduced by admissible transformation to (7.10).
7.3 Curvilinear Coordinates in E3

The apparatus of tensor analysis was developed initially as a tool of
the analytic study of geometries. Since dynamics, mechanics of its
continuous media, and relativity lean heavily on geometrical properties

of the 3-dimensional space of physical experience, we devote most of the
chapters to an investigation of properties of curves and surfaces imbed-
ded in E3.
Let P(y) be the point in a Euclidean 3-space E3, referred to a set of
orthogonal Cartesian axes Y. Consider a coordinate transformation
T: xi = xi(y1, y2, y3), (i = 1,2,3)

∂ xi
such that the xi are of class C1 and J = ≠ 0 in the same region R of E3.
∂ yj
The inverse transformation T−1: yi = yi(x1, x2, x3), (i = 1,2,3) will be single
valued and the transformation T and T−1 establishes a 1 − 1 correspon-
dence between the sets of values (x1, x2, x3) and (y1, y2, y3).
The triplets of numbers (x1, x2, x3) are called curvilinear coordinates of
the points P in R.
7.3.1 Coordinate Surfaces

Let x1 be kept fixed, i.e.
x1 = constant in T, then
x1 = x1(y1, y2, y3) = constant, (7.11)
defines a surface. If the constant is now to assume different values, we get a

one-parameter family of surfaces. Similarly, other families
x2 = x2(y1, y2, y3) = constant and x3 = x3(y1, y2, y3) = constant.
The condition is that the Jacobian, j ≠ 0, in the region under consider-

ation that the surfaces
x1 = c1, x2 = c1, x3 = c1, (7.12)
intersects at one, and only one, point. The surfaces of (7.12) are called
coordinate surfaces and their intersection pair by pair are called coordi‑
nate lines because, along this line, the variable x3 is the only one that is
changing.
7.3.2 Coordinate Curves

Let two of the coordinates x1, x2, x3 be kept fixed in T, i.e. x2 = c2, constant
and x3 = c3, constant in T, and x1 be allowed to vary. Then, P(y1, y2y3) will
satisfy the relations:
yi = yi(x1, c2, c3).
Since y1, y2, and y3 are the functions of one variable, it follows that P(y1,
y y ) will lie on a curve, called a coordinate curve. This coordinate curve
2 3
x2 = c2, x3 = c3
is called the x1 −curve. Thus, the line of intersection of x2 = c2, x3 = c3 is the

x1 −coordinate line. Therefore, the x1 −curve lies on both the surfaces of
x2 = c2, x3 = c3. Similarly, we can define the x2 – and x3 −curves. It should
also be noted that through a given point P(y1, y2y3), 3 coordinate curves
corresponding to fixed values c1, c2, and c3 are passed through.
Example 7.3.1. Consider a coordinate system defined by the transformation
y1 = x1sinx2cosx3
y2 = x1sinx2sinx3
y3 = x1cosx2.
The surface x1 = constant are spheres, x2 = constant are circular cones,

and x3 = constant are planes passing through the Y 3-axis (Figure 7.2).
Solution: Squaring and adding the equation, we get
(y1)2 + (y2)2 + (y3)2 = (x1sinx2cosx3)2 + (x1sinx2sinx3)2 + (x1cosx2)2
= (x1sinx2)2 + (x1cosx2)2 = (x1)2 (i)
Squaring and adding its two equations, (y1)2 + (y2)2 = (x1sinx2)2
or x 1sinx 2 = ( y 1 )2 + ( y 2 )2 (ii)

Y3
X1
X2
O Y2
X3
Y1
Figure 7.2
( y1 )2 ( y 2 )2
Dividing (ii) by the last given equation, we get tanx 2
y3
( y 1 )2 + ( y 2 )2
(ii) by the last given equation, we get tanx 2 = (iii)
y3

Dividing the first two equations, we get
y2
tanx 3 = .
y1
(iv)
From (i), x 1 = ( y 1 )2 + ( y 2 )2 + ( y 3 )2 ,
2 −1 ( y 1 )2 + ( y 2 )2
from (iii) x = tan ,
y3
3 −1 y2
and from (iv) x = tan .
y1
So, the inverse transformation is given by the above equations.
If x1 > 0,0 < x2 < π, 0 ≤ x3 < 2π, this is the familiar spherical coordinate
system.
Example 7.3.2. The transformation
y1 = x1cosx2
y2 = x1sinx2
y3 = x3
defines a cylindrical coordinate system.
Solution: The Jacobian of transformation is given by
cos x 2 − x 1sinx 2 0
J= sinx 2 x 1cosx 2 0 = x 1 ≠ 0.
0 0 1

Hence, the inverse transformation exists and is given by
T −1 : x 1 = ( y 1 )2 + ( y 2 )2

y22 2 −1 y
2
tan x = 1 or x = tan
y y1

and x3 = y3
Y3
X 3 = const.
X 1 = const.
O Y2
X2 X 2 = const.
Y1
Figure 7.3
if x1 > 0,0 ≤ x2 < 2π, −∞ < x3 < ∞, this coordinate system defines a cylindri-
cal coordinate system, as shown in Figure (7.3).
The coordinates of the surface are
x1 constant c1 ( y1 )2 ( y 2 )2 c1 ,

which is circle.
y2
x 2 tan 1 1 c2 y 2 y1tanc2 , which is a straight line.
y
and y3 = c3, which is a plane parallel to the y1y2 – plane
7.3.3 Line Element

Here, we have to obtain the line element of E3 in curvilinear coordinates.
Let P(y1, y2, y3) and Q(y1 + dy1, y2 + dy2, y3 + dy3) be two neighboring points
in R. The Euclidean distance between these points is determined by the
quadratic form
ds2 = (dy1)2 + dy2)2 + dy3)2
= dyidyi
yi
and since dy i dx , we get
x
2 i i yi yi
ds dy dy dx dx
x x
yi yi y y
dx dx i j
dx idx j
x x x x
g ijdx idx j ,
(7.13)
y y
where g ij i
( , 1,2 ,3).
x xj
Obviously, gij is symmetric. Moreover, it is a tensor, since ds2 is an invari-
ant and the vector dxi is arbitray.
Denote by g the determinant |gij|; this is positive in R. Since gijdxidxj is a.
positive definite form, we can introduce the conjugate symmetric tensor gij,
ij G ij
defined by g = , where Gij is the cofactor of the element gij in g.
g
7.3.4 Length of a Vector

Consider a contravariant vector Ai in a curvilinear coordinate system.
Now, we form the invariant
1
i j 2
A ( g ij A A ) (7.14)

In orthogonal Cartesian coordinate gij = δij and when i = j, gij = 1,
1
i i 2
A = (A A ) .
Therefore, in the orthogonal Cartesian frame, (7.14) assumes the form
A = [( A1 )2 + ( A2 )2 + ( A3 )2 ]2 .
We see that A represents the length of the vector Ai.

Similarly, the length of the covariant vector Ai is defined by the formula
1
ij
A = ( g Ai A j ) , 2 (7.15)

a vector whose length is 1 and is called a unit vector. From (7.13) we get
ds2 = gijdxidxj
dx i dx j
or 1 = g ij ,
ds ds
dx i
so that ≡ λ i is a unit vector. It is a contravariant vector.
ds
If xi = yi and the coordinate system is Cartesian,
dx 1 dx 2 dx 3 3
then ≡ λ1 , ≡ λ2 , λ are direction cosines of the displacement
ds ds ds
vector (dx1, dx2, dx3). Accordingly, we take this vector λi to define the direc-
tion in space relative to a curvilinear coordinate system X.
7.3.5 Angle Between Two Vectors

Consider two unit vectors of different directions λi and μi at same point
P, as shown in Figure (7.4). Since the manifold under consideration is
Euclidean, the cosine law with Pythagoras’s formula gives
ST 2 RS 2 RT 2 2RS RT Cos

and since λi and μi are unit vectors, RT = RS = 1, and hence
2
ST 2(1 cos ) (7.16)

and the position vector ST = λ i − µ i .

2
Using the formula (7.15), we get ST = g ij (λ i − µ i )(λ j − µ j )
= gijλiλj + gij μi μj – 2gijλi μj
= 1 + 1 − 2gijλi μj
= 2(1 − gijλi μj) (7.17)
From (7.16) and (7.17), we get
cosθ = gijλi μj (7.18)
λ–µ
θ µ
R
Figure 7.4
We can use this formula to define the angle θ between two directions λi
and μj.
If Ai and Bi are any two vectors, then from definition of the length of a
vector, it is clear
g ij Ai B j
cosθ = .

This leads to the formula ABcosθ = gij AiB j, defining an invariant which
is a “scalar product” AB of elementary vector analysis.
Example 7.3.3. Prove that the angles θ12, θ23, and θ31 between the coordi-
nate curves in a three-dimensional coordinate system are given by
g 12 g 23 g 31
cosθ12 = , cosθ 23 = , and cosθ 31 = .
g 11 g 22 g 22 g 33 g 33 g11

Solution:
We know the expression
ds2 = gijdxidxj
for the square of the element of arc (ds) between P1 (x1, x2, x3) and
P2(x1 + dx1, x2 + dx2, x3 + dx3) and the lengths of the elements of arc
measured along the coordinate lines of a curvilinear system X are
2
ds(1) = g 11dx 1dx 1 = g 11 (dx 1 )2 ⇒ ds(1) = g 11 (dx 1 ), where s(1) denotes the arc
length along the x1 curve.
2
Similarly, ds(2) = g 22 (dx 2 ) , ds(3)
2
= g 33 (dx 3 ). (7.19)

Thus, the length of displacement vector (dx1, 0,0) is given by g 11 (dx 1 )

and that of (0, dx2, 0) is g 22 (dx 2 ) and the length of vector (0,0, dx1) is
given by g 33 (dx 3 ), as shown in Figure (7.5).
X3
P2(x + dx)
ds(3) = g33 dx3 ds X2

θ23
θ13 θ12
ds(2) = g22 dx2
P1(x)
ds(1) = g11 dx1
X1
Figure 7.5
Angles θ12, θ23, and θ31 are between the coordinate curves
in a three-dimensional coordinate system. Its cosines are
g12dx1dx 2 g12 g 22
cos 12 1 2
, cos 23 ,
g 11 (dx ) g 22 (dx ) g11 g 22 g 22 g 33
g 31
and cos 31 .,
g 33 g 11
Theorem 7.3.1. A necessary and sufficient condition that a given curvilin-
ear coordinate system X be orthogonal is that gij = 0 for i ≠ j at every point
of the region R.
Proof: We know that if θij is the angle between λi and μj, they are displace-
ment vectors along coordinate lines and
g ij λ i µ j
cosθ ij = .
g ij λ i λ i g ij µ j µ j

Angles θ12, θ23, and θ31 are between the coordinate curves in a
three-dimensional coordinate system. Its cosines are
g 12dx 1dx 2 g 12 g 22
cosθ12 = 1 2
= , cosθ 23 = , and cosθ 31
g 11 (dx ) g 22 (dx ) g 11 g 22 g 22 g 33
g 31
= ,
g 33 g 11

since curvilinear coordinate system X is orthogonal, i.e., θij = 90°.
g ij
Therefore, = 0 ⇒gij = 0.
g ii g jj
g ij
Conversely, let g ij = 0 ⇒ =0
g ii g jj
∴cosθij = 0.
Therefore, curvilinear coordinate system X is orthogonal.
7.4 Reciprocal Base Systems

Here, we interpret the main results of curvilinear coordinate systems
in notation of elementary vector analysis, as shown Figure (7.6). Let a
Cartesian coordinate system be determined by a set of orthonormal base
vectors b1, b2, and b3. Then, the position vector r of any point P(y1, y2, y3)
can be represented in the form
r = biyi (i = 1,2,3). (7.20)
Since the base vectors bi are independent of the position of the point,
dr = bidyi. (7.21)
Y3 X3
a3
a2
X2
P
a1
r X1
b3
Y2
O b2
b1
Y1
Figure 7.6
By definition, the square of the length of the arc element (ds) is between
points
P(y1, y2, y3) and Q(y1 + dy1, y2 + dy2, y3 + dy3)
ds2 = dr.dr (7.22)
= bi.bjdyidyj
= δ ij dy i dy j = dy i dy i ,

a familiar expression for the square of the element of arc in orthogonal

Cartesian coordinates.
Let a set of equations of transformation
xi = xi(y1, y2y3) (i = 1,2,3)
define a curvilinear coordinate system X. The position vector r can now be

regarded as a function of coordinates xi and we can write
r i
dr dx (7.23)
xi
r r
and ds 2 dr .dr dx idx j
xi xj
g ijdx idx j ,

∂r ∂r
where g ij =
∂x i ∂x j (7.24)
r
The geometrical interpretation of is a base vector directed tangen-
xi r
tially to the X coordinate curve, so that we denote
i
i
ai..
(7.22) and (7.23) become x
dr = aidxi (7.25)
gij = ai. aj.

We observe that the base vector ai is no longer independent of the coor-
dinates (x1, x2, x3).
Result 7.4.1. Here dr = aidxi and dr = bidyi and we get

∂ yi
a j dx j = bi dy i = bi j dx j ,
∂x
∂ yi
⇒ a j = bi j (since dx j are arbitrary )
∂x
We see that the base vector aj transforms according to the Law of
Transformation of components of covariant vectors.
Result 7.4.2. The components of base vectors ai, when referred to in Xi

coordinate systems, are
a1: (a1, 0,0) a2: (0, a2, 0), and a3: (0,0, a3).
They are not necessarily unit vectors because
g11 = a1. a1 = (a1)2 ≠ 1, similarly g22 = a2. a2 ≠ 1 , g33 = a3. a3 ≠ 1.

Again, g ij = ai . a j = |ai | a j cos θ = 0 if i ≠ j implies that a curvilinear
coordinate system is orthogonal if gij = 0.
Result 7.4.3. Any vector can be written in the form A = kdr.

r i
We know dr dx
xi
r i r
or A kdr k i
dx (kdx i ) ai Ai , where Ai = kdxi.
x xi
Here, Ai is the contravariant component of a vector of A.
Result 7.4.4. The vectors a1A1, a2A2, and a3A3 form the edges of the parallel
pipes, whose diagonal will be A. Since the ai are not unit vectors in general,
we see that the length of edges of Parallelepiped are
g 11 A1 , g 22 A2 , g 33 A3 since g11 = a1 . a1 = (a1 )2 .

Result 7.4.5. Consider three non-coplanar vectors
a2 × a3 2 a3 × a1 3 a1 × a2
a1 = a = a = , (7.26)
[a1a2 a3 ] [a1a2 a3 ] [a1a2 a3 ]
where a2 × a3 etc. Denote that the vector product of a2, a3 and [a1a2a3] is a
scalar triple product a1.a2 × a3.
a2 a3 [a1a2a3 ]
a1. a1 a1. 1
[a1a2a3 ] [a1a2a3 ]
a2 × a3 [a2a2 a3 ]
a1. a2 = a2 . = = 0.
[a1a2 a3 ] [a1a2 a3 ]
It is obvious that ai . a j = δ ij
1
Example 7.4.1. Show that [a1a2 a3 ] = g and [a1a 2a3 ] = ,
g
where g = |gij|.
Solution: Let the components of base vectors ai be
a1: (a1, 0,0) a2: (0, a2, 0), a3: (0,0, a3).
a1 0 0
Then, [a1a2 a3 ] = 0 a2 0 = a1a2 a3
0 0 a3
[[a1a2a3] = a1. a2 × a3 is numerically equal to the volume of parallelepiped
made by its edges a1, a2, a3,]
g 11 g 12 g 13
and g = |g ij | = g 21 g 22 g 23
g 31 g 32 g 33

g11 = a1. a1 = (a1)2.

Similarly, g22 = (a2)2 and g33 = (a3)2 and g12 = g23 = g13 = 0
(a1 )2 0 0
g 0 (a2 )2 0 (a1 )2 (a2 )2 (a3 )2
0 0 (a3 )2

or a1a2 a3 = g .

1 i i
Therefore, [a1a2 a3 ] = a1a2 a3 = g and [a1a 2 a 3 ] = , since a . a j = δ j .
g
Definition 7.4.1. The triple products [a1a2a3] and [a1a2a3] are reciprocally
related. Moreover,
a2 × a3 a 3 × a1 a1 × a 2 (7.27)
a1 = , a2 = , and a3 = ,
[a1a 2 a 3 ] [a1a 2 a 3 ] [a1a 2 a 3 ]
The system of vectors a1, a2, a3 are called the reciprocal base system.
Hence, if the vectors a1, a2, and a3 are unit vectors associated with
orthogonal Cartesian, coordinates then the reciprocal system of the vec-
tors defines the same system of coordinates.
The differential of a vector r in the reciprocal base system is dr = aidxi,
where dxi are the components of dr. Then,
ds2 = dr.dr = (aidxi)(ajdxj)
= aiajdxidxj
∴ ds2 = gijdxidxj,
where gij = aiaj = gji. (7.28)
The system of base vectors determined by Equation (7.25) can be used

to represent an arbitrary vector A in the form of A = aiAi, where Ai are the
covariant components of A.
The scalar product of the vector aiAi, with the base vector aj, noted that
the latter is directed along the Xj-coordinate line and we get
a j .a i Ai = δ ij Ai = A j .

Aj
Thus, is the length of the orthogonal projection of vector A on the
g jj
Aj
tangent to the xj coordinate curve at P, whereas is the length of the
g jj
edge of the parallel pipes whose diagonal is vector A.
A3
x3 A
g33
A2
g22
x2
a3
a2
a1 A1
g11
P
x1
Figure 7.7
Since A = ai Ai =. aiAi,
or ai. ajAi =. ai. ajAi,

or g ij Ai = δ ij .Ai = A j,,

we see that the vector obtained by lowering the index in Ai is precisely

the covariant vector Ai and thus, it is seen to represent the same vector A
referred to by two different base vectors. Hence, the distinction between
the covariant and contrvariant components of A disappears whenever the
base vectors are orthogonal.
7.5 Partial Derivative A

Theorem 7.5.1. If A is a vector along the curve in E3, prove that A, j a .
xj
Proof: We suppose that the components of A are continuously differentia-
ble functions of yi in R and, if we introduce a curvilinear coordinate system
X by means of transformation
T: xi = xi(y1, y2, y3),
then, the corresponding components Ai(x) will be continuously differ-

entiable functions of point (x1 x2, x3) determined by the position vector
r(x1 x2, x3).
r
A vector A can be expressed by base vector ai as
xi
A = Aiai, (7.29)
where Ai is the component of A,

Change ΔA in A at point P(x1 x2, x3) assumes a different position
P′(x1 + Δ x1, x2 + Δ x2, x3 + Δ x3).
Then, ΔA = (Ai + ΔAi)(ai + Δai) − Aiai
= ΔAiai + ΔaiAi + ΔAiΔai
In ordinary calculus, the principal part of the change by dA
dA = dAiai + daiAi (7.30)
(neglecting ΔAiΔai).
This formula states that the change in A arises from two sources:
(i) change in the components Ai as the values (x1 x2, x3) are
changed,
(ii) change in the base vectors ai as the position of point (x1 x2,
x3) is altered.
The partial derivative of A, with respect to xi, is defined as the limit
DA A
lim j
Dxi
0 Dx xj
A Ai ai i
j
a
j i
A. (7.31)
x x xj
A
Now, we have to show that is a covariant derivative of Ai.
xj
ai a
Now , we establish a . (7.32)
xj i j

We know gij = ai. aj, differentiating with respect to xk
ai aj
k
g ij .a j . ai .
x xk xk
∂ ∂ ai ∂ ak
Similarly, j g ik = j . ak + . ai
∂x ∂x ∂xj
aj ak
i
g jk . ak . aj
x xi xi
∂ ai ∂ a j
and we assume T is of class C2, therefore, =
∂ x j ∂ xi
1
[ij ,k] g ik g jk g ij ,
2 xj xi xk
ai
or [ij , k] . ak , (7.33)
xj
ai
or [ij , k]a k ,
xj
ai
or .a [ij , k]a k . a
xj
 α 
= [ij,k]g kα =  ,
 i j 
∂ ai  a 
implying that =  aα . This is established in (7.31).
∂xj  i j 
 
Inserting these values in (7.30), we get
A Ai ai i Ai
ai A ai a Ai
xj xj xj xj i j
A
Ai a .
xj i j

The expression in the bracket is A,i j , thus
∂A
= A,jα ,aα . (7.34)
∂xj
A
Theorem 7.5.2. If A is a vector along the curve in E3, prove that A j ,k a j .
xk
Proof: We know A = Aiai
A Ai i ai
a Ai (i)
xk xk xk
ai a j = δ ij

Differentiating with respect to xk, we get

∂ ai ∂aj
k
. a j + k . ai = 0
∂x ∂x
ai aj i
or . aj .a
xk xk
a
ai a . (from 7.31)
j k

Since a i aα = δ αi ,
∂ ai  a   i 
therefore, . a j = −δ αi  = − 
∂ xk  j k   j k 
∂ ai  i 
j
= −  a . (7.35)
∂ xk  j k 

Substituting this value in (i), we get
A Ai i ai Ai i i
a Ai a a j Ai
xk xk xk xk k j
Aj j i
a a j Ai
xk j k
Aj i
Ai a j A j ,k a j .
xk j k

7.6 Exercises
1. If Ai = gij Aj, show that Ai ,k g i A,k . Ai , j Ai
2. Prove that if A is the magnitude of A , then A j =
i
.
A
3. If y are the rectangular Cartesian coordinates, show that in E3
i
2 i
y yi 2 i
y y
[ , ] and .
x x x x x xi

4. Show that the area of the parallelogram constructed on the base

vectors a2, and a3 is gg 11 , where gij and gij are the conjugate and
metric tensors in a curvilinear coordinate system and g = |gij|.
d( g ij Ai )
5. Show that k
Ai ,k Bi Bi ,k Ai .
dx
8
Curves in Space
8.1 Introduction
In this chapter we describe the geometry of space curve and the Serret-
Frenet formulas analogous to the derivatives of the tangent, normal, and
binormal unit vectors in terms each other. The formulas are named after
two French mathematicians who independently discovered them: J. F.
Frenet, in his thesis of 1847 and J. A. Serret in 1851. A set of three remark-
able formulas are generally known as Frenet’s formulas, which character-
ized, in small, all essential geometric properties of space curves. We also
deal with Helix, an equation of straight lines, and intrinsic differentiation.
8.2 Intrinsic Differentiation

We are now going to introduce another kind of differentiation which may
be regarded as a generalization of ordinary differentiation in Euclidean
spaces with rectangular Cartesian coordinates.
Let a vector field A(x) be defined in some region of E3 and let
C: xi = xi(t) , t1 ≤ t ≤ t2
be a curve in that region. The vector A(x) is defined over the one dimen-
sional manifold C, depending on the parameter t, and if A(x) is a differen-
tiable vector and the xi(t) belong to the class C1, then
dA ∂ A dx j
=
dt ∂ x j dt
dx j
= A,αj aα .
dt

221
We know
A A
A, j a Ai a ,
xj xj

i j
so
dA A dx j
Ai a
dt xj dt

i j
dA dx j
Ai a .
dt dt
i j

δ Aα dAα  α  i dx j
The formula ≡ + A (α = 1, 2 , 3)
δt dt  i j  dt
(8.1)
is called the absolute or intrinsic derivative of Aα, with respect to parameter

δ Aα
t, and is denoted by .
δt
We observe the following properties of intrinsic derivatives.
1. We will make free use of intrinsic differentiation in the treat-

ment of geometry of curves and surfaces.
2. It also follows the familiar rules for differentiation of sums,
products, etc. and remains valid for the process of intrinsic
differentiation.
3. If the vector field Aα is defined in the neighborhood of C, as
well as on C, we can write
δ Aα α dA
β
= A,β .
δt dt
Curves in Space 223
α α
4. If A is scalar, then obviously, δ A = dA .
δ dx l δt dt
g
5. δ t ij = g ij ,l = 0 since g = 0. Hence, show that the intrin-
dt ij,l
sic derivative of gij = 0.

Also, the intrinsic derivative of δ ij = 0.w
With the extension of the process of intrinsic differentiation to tensors
of rank greater than one, we can write
(i) If Ai is a covariant vector,
Ai dAi dx
A
t dt i dt

Aij dAij i dx j dx
(ii) A j
Ai
t dt dt dt
Aij dAij i dx dx
(iii) Aj Ai
t dt dt i dt
(iv)
Aijk dAijk i dx i dx dx
A jk Ai k Aij
t dt dt j dt k dt

δ
We observe that g ij = 0 and the fundamental tensors gij and gij can be
δt
taken outside the sign of intrinsic differentiation.
d( g ij Ai A j ) δ Aj
Example 8.2.1. Prove that = 2 g ij Ai .
dt δt
Solution: Since gijA A is scalar,
i j
i j i j
then d( g ij A A ) = δ ( g ij A A )
dt δt
δ ( Ai A j )
= g ij ,
δt
and since gij is independent of t and gij, AiAj is invariant.
 δ Ai δ Aj 
= g ij  A j + Ai .
 δt δ t 
Interchanging i and j in first term, we get
d( g ij Ai A j )  δ Aj δ Aj  δ Aj
= g ij  Ai + Ai  = 2 g ij Ai (since gij is symmetric)
dt  δt δt  δt
δ g ij
Example 8.2.2. If gij are components of metric tensors, show that = 0.
δt
Solution: The intrinsic derivative of gij is
g ij dg ij dx dx
g j gi
t dt i dt j dt
g ij dx dx dx
g j gi
x dt i dt j dt
g ij dx
g j gi .
x i j dt
g ij g ij dx
[i , j] [ j ,ii] , (as g j [i , j]
t x dt i
and gi [ j , i ])
j

g ij
but [i , j] [ j ,i]
x
g ij
⇒ 0
t
dx i d2xi i j
i dx dx
k
Example 8.2.3. Show that A .
i t dt dt 2 j k dt dt
dx
We know that are the components of a contravaiant vector.
dt
Curves in Space 225
  dx i  
 ∂    j 
δ  dx   dx   dx 
i i
 dt   i k
 dx dx k
Now   =   =  +  
δ t  dt   dt  ,k  dt   ∂ x k  j k
 dt  dt

 dx i 
∂
 dt  dx k  i  dx j dx k
= + 
∂ xk dt  j k  dt dt
d 2 x i  i  dx j dx k
= 2 +  .

dt  j k  dt dt
Example 8.2.4. Show that the intrinsic derivative of a tensor of type (p, q)
is again a tensor of type (p, q).
i ...i
Solution: Let us consider a tensor of type (p, q) whose components A j11 .... pjq .
are functions of a parameter, t. Then, the intrinsic derivative of the tensor,
with respect to parameter t, is denoted by δ Aij11.......i pjq . and is defined as
δt
i ...i i ...i dx l
A j11 .... pjq . A j11 .... pjq ,l. ,
t dt
where a comma (,) denotes covariant differentiation.

dx l
Since is a contravariant vector, a tensor of type (1,0) is along the
dt i ...i
curve x = ϕ (t) and A j11 .... pjq ,l. is a tensor of type (p, q + 1) along the same curve
i i
i1 ...i p dx l i ...i
and A j1 .... jq ,l . is a tensor of type (p,q) along the curve, i.e. A j11 .... pjq . is a
dt t
tensor of type (p, q) along the curve.
Thus, the intrinsic derivative of a tensor of type (p, q) is again a tensor
of type (p, q).
Example 8.2.5. If the intrinsic derivative of a contravariant vector vanishes

identically, show that its magnitude is constant.
Solution: Let Ai be the components of a cotravariant vector such that
δ Ai
= 0.
δt
If the square of magnitude of vector Ai is gijAiAj,
δ ( g ij Ai A j ) δ Aj
then = 2 g ij Ai = 0.
δt δt
⇒ gijAiAj is constant
⇒The square of magnitude of the vector Ai is constant.
8.3 Parallel Vector Fields

Consider a curve, as shown in Figure (8.1).
C: xi = xi(t), t1 ≤ t ≤ t2 and i = 1,2,3,
Y3 X3
P X2
A X1
Y2
O
Y1
Figure 8.1
in some region of E3 with a vector A at point P of C. We suppose that the

functions xi(t) are of class C1. If we construct at every point of C a vector
equal to A in magnitude and parallel to it in direction, we obtain what is
known as a parallel field of vectors along curve C.
Curves in Space 227
If A is a parallel field along C, then the vectors of A do not change along

dA
the curve and we can write = 0.
dt i
It follows that components A of A satisfy a set of simultaneous differen-
i
tial equations δ A = 0
δt
dAi  i  α dx β
or + A =0 (8.2)
dt  α β  dt

Every solution of Equation (8.2) satisfies the initial conditions and must
form a parallel field along C.
Example 8.3.1. If Ai and Bi are two vectors of constant magnitudes and
undergo parallel displacements along a given curve, then show that they
are inclined at a constant angle.
Let Ai(t) and Bi(t) be any two solutions of (8.2). We verify that the
lengths of these two vectors indeed do not change as we move along the
curve. Moreover, the angle θ between vectors Ai and Bi remains fixed as
parameter t is allowed to change.
Solution: prove this
We have A.B = AB cosθ = gijAiB j and, if gijAiB j is to remain constant along
d
C, then g ij Ai B j = 0,.
dt
But gijAiB j is an invariant, and, since gij behave like constant in the pro-
cess of covariant differentiation. We can write
d δ δ δ
( g ij Ai B j ) = ( g ij Ai B j ) = g ij ( Ai )B j + g ij ( B j )Ai .
dt δt δt δt
δ Ai δ Bi
Since the field Ai and B j satisfy (8.2), = 0 and = 0 , we conclude
that gijAiB j is constant along C. δt δt
d
( g ij Ai B j ) 0, or
dt
d
( AB cos ) 0
dt
∴ cosθ = constant.
If Ai = Bi, then gijAiBj = A2 is a constant along C and this implies that θ =
constant.
Since vectors Ai are defined at every point (xi) of the manifold, we can
write
dAi ∂ Ai dx k
= ,
dt ∂ x k dt
so that Equation (8.2) assumes the form
 ∂ Ai  i  α  dx k
 k +α k A  = 0.
 ∂ x    dt

The parallel vector field in E3 satisfies the system of equations
∂ Ai  i  α i
+  A = 0 or A,k = 0.
∂ x k  α k 

Similarly, the condition for parallel displacement of covariant vector

Ai is
∂ Ai  i 
+  Aα = 0 or Ai,k = 0.
∂ x k  α k 

8.4 Geometry of Space Curves

Let the parametric equations of the curve C in E3 be
C: xi = xi(t), t1 ≤ t2 and i = 1,2,3.
The square of the length of an element of C is given by
ds2 = gijdxidxj (8.3)
and the length of arc s of C is defined by the integral

Curves in Space 229
t2
dx i dx j

s=
∫ t1
g ij
dt dt
dt (8.4)
dx i dx j (8.5)
From (8.3), we get 1 = g ij
ds ds
dx i
and if we put = λ i , Equation (8.5) can be written as
ds
gijλiλj = 1. (8.6)
Thus, the vector λ with components λi is a unit vector. Moreover, λ is

tangent to C since its components λi, when the curve C i is referred to a
dy
rectangular Cartesian reference frame Y, become λ i = . These are the
direction cosines of the tangent vector to curve C. ds
Consider a pair of unit vectors λ and μ with components λi and μi at any
point P of C, as shown in Figure (8.2). We suppose that λ is tangent to C at
P. The cosine of the angle θ between λ and μ is given by
cos θ = gijλiμj (8.7)
and if λ and μ are orthogonal, we have
gijλiμj = 0 (8.8)
λ + dλ
λ
Q(x + dx) C
P(x) r + dr
Figure 8.2
Any vector μ satisfying Equation (8.7) is said to be normal to C at P.

If we take the intrinsic derivative with respect to the arc parameter s of
quadratic relation (8.6), gij behaves in covariant differentiation like con-
stants and we obtain
δλ i j δλ j i
g ij λ + g ij λ = 0.
δs δs
δλ i j
Since gij is symmetric, g ij λ = 0.
j δs
δλ
Here, the vector either vanishes or is normal to C (for orthogonal-
δs
ity) and if it does not vanish, we denote that the unit vector is co-directional
j
with δλ by μj and write
δs
j 1 δλ j ,
µ = (8.9)
χ δs
where χ > 0 is so chosen to make μj a unit vector.

The vector μj, determined by Formula (8.9), is called the principal nor-
mal vector to curve C at point P and χ is the curvature of C at P.
The plane determined by the tangent vector λ and the principal normal
vector μ is called the osculating plane to curve C at P.
Since μ is a unit vector
gijμiμj = 1. (8.10)
Differentiating intrinsically the orthogonality relation (8.10), we get
δλ i j δµ j i
g ij µ + g ij λ =0
δs δs
δµ j i δλ i j
g ij λ = − g ij µ
δs δs
j δλ j
= −gijχμjμj [since χµ = and
δs
gijμ μ = 1 from (8.9) & (8.10)]
i j
= −χ.1 = −χ
Curves in Space 231
δµ j i
Therefore, g ij λ = −χ (8.11)
δs
and since gijλiλj = 1 and we multiply gijλiλj on the right hand side of Equation
(8.11),
δµ j i
g ij λ = − χ g ij λ i λ j .
δs
 δµ j 
g ij λ i  + χλ j  = 0
 δs 

δµ j j
shows that vector δ s + χλ is orthogonal to λi.
We define a unit vector ν, with components νj, by formula
1  δµ j 
νj =  + χλj  , (8.12)
τ  δs 

where τ > 0 is chosen to make νj a unit vector.

The vector ν will be orthogonal to both λ and μ.
The number τ in (8.12) is called the torsion of C at P and the vector ν is
binormal.
8.4.1 Plane
If a curve lies on a plane, then it is called a plane curve, otherwise it is called
a skew curve.
The plane determined by λ and μ is called the osculating plane at P ( x 0i )
and its equation is given by
g ij (x i x0i ) j
0,
where xi is any point on the plane.
The plane determined by ν and μ is called the normal plane at P ( x 0i ) and

its equation is given by
g ij ( x i − x 0i )λ j = 0

and that of the plane determined by ν and λ is called the rectifying plane at
P ( x 0i ) and its equation is given by
g ij ( x i − x 0i )µ j = 0.

Result 8.4.1. Choose the sign of τ in such a way that
g eijk λ i µ j ν k = 1, (8.13)

so that the triad of unit vectors λ, μ, ν and forms, at each point P of C, a

right–handed system of axes.
Result 8.4.2. From the application of e on determinant and triple scalar

products, we get
λ1 µ1 ν1
1
eijk λ i µ j ν k = λ2 µ2 ν2 = λ .( µ × ν ).
g
λ3 µ3 ν3

2
∂ yi
Result 8.4.3. Since eijk is a relative tensor of weight −1 and g = , it
∂x j
follows that
1 ijk
ijk = g eijk and  ijk = e
g

is an absolute tensor, the left hand side of (8.13) is an invariant, and vk in

(8.13) is determined by the formula
Curves in Space 233
v k = ijk λi µ j , (8.14)
where λi and μj are associated vectors giaλα and giaμα.

The number τ in Equation (8.12) is called the torsion of C at P and the
vector ν is binormal.
∂A
We recall formula i = A,iα aα .
∂x
If vector field A is defined along C, we can write
∂ A dx i α dx
i
= A,i aα (8.15)
∂ x i ds ds
δ Aα dAi
Using intrinsic derivative = A,iα , we can write the above equa-
tion as δs ds
dA δ Aα
= aα .
ds ds
Let r be the position vector of point P on C, then tangent vector λ is

determined by
dr
= λ i ai = λ .
ds
From (8.15) and the above equation, we get
d 2r d λ δλ α
= = aα = c ,
ds 2 ds δs
where c is perpendicular to λ.
8.5 Serret-Frenet Formula

The set of three formulas known as Serret-Frenet formulas characterize all
essential geometry properties of space curves. Two of these formulas have
already been derived. These are
1 δλ i (8.9)
(i) µ i = χ >0
χ δs
1  δµ i  δµ i
(ii) ν i =  + χλ i  , or = τν i − χλ i . (8.12)

τ δs  δs

The third is expressed as
δν i
(iii) = −τµ i . (8.16)
δs
The first gives the rate of turning of tangent vector λ as the points move
along the curve, the second gives that of the principal normal μ, and the
third one specifies the rate of turning of the binormal ν as point P moves
along the curve.
We know, ν k = ijk λi µ j ,
1 ijk
where λi and μj are the associated vectors giαμα and giαλα and ijk ≡ e is
an absolute tensor. g
If we differentiate the above equation intrinsically, we get
δν k ijk δλi δµ j (8.17)

= µ j + ijk λi ,
δs δs δs
since the covariant derivatives of ijk are zero.

δλi δµ .i
Using (8.9) as = χµ i and (8.12) as = τν i − χλi and putting them
δs δs
in (8.17),
δν k ijk
=  χµi µ j + ijk λi (τν j − χλ j ).
δs
Here, ijk χµi µ j = 0 and ijk χλi λ j = 0 since ϵijk is skew symmetric and
ijk
 λiν j = − µk
Curves in Space 235
k
ijk i j
s
= −τ μk, established in (8.16).
The Serret-Frenet formula can be written in Christoffel symbols as
d λ i  i  dx k
j
+ λ = χµ i
ds  j k  ds
 
d µ i  i  dx k
j
+ µ = τν i − χλ i (8.18)
ds  j k  ds
 
dν i  i  dx k
j
+ ν = τµ i
ds  j k  ds
 
Equation (8.18) can be written in matrix form as
 λi   0 χ 0   λi 
δ  i    i 
µ
δ s  i  =  −χ 0 τ   µ  (8.19)
  0 −τ 0   νi 
 ν    

8.5.1 Bertrand Curves

The curves whose principal normals are also the principal normals of
another curve are called Bertrand curves.
If the two curves are. C1 and C2:
C1 : x i = x i (s ) and C2 : x i = x i (s )

have common principal normals at any of their points and one is called the
Bertrand of another or one is called Bertrand associate of another.
Example 8.5.1. Show that a twisted curve (τ ≠ 0) is a Bertrand curve if, and
only if, its curvature χ and torsion τ are connected by a linear equation of
the form aχ + bτ = 1, where a and b are non-zero constants, hence showing
that a circular helix is a Bertrand curve.
Solution: Let Г: xi = xi(s)
and Γ : x i = x i (s ) = x i (s ) + a(s )µ i
(i)
Differentiating intrinsically with respect to s, we get
δ xi δ xi δµ i  δa 
= + a′(s )µ i + a(s ) Here a′ = δ s 
δs δs δs
= λ i + a′(s )µ i + a(s )(τν1i − χλ i )

i
δs (ii)
or λ i = λ i + a′(s )µ i + a(s )(τν i − χλ i )
δs
since the two curves are Bertrand, μi is parallel to µ i. Taking the inner
product of (ii), we get
δ si
µiλ i
= λ i µ i + a′(s )µ i µ i + a(s )(τν i µ i − χλ i µ i )
δ s
or 0 = 0 + a′(s). 1 + a(s)(τ.0 – χ.0)
⇒a′(s) = 0,
∴ a(s) = constant,
which is the necessary condition so that two curves are Bertrand curves.
Now, (ii) becomes
δ si
i
λ = λ i + a(s )(τν i − χλ i )
δs
δ si (iii)
or λ i = λ i (1 − a χ ) + aτν i .
δs
Curves in Space 237
For curve Γ, we have
d i i dλ i i dλ i i
(λ λ ) = λ + λ
ds ds ds
ds
= χµ i λ i + ( χµ i )λ i = 0.
ds

i i
constant
This implies that the tangents to the two curves are inclined at a constant
angle, but their principal normals coincide, and, therefore, the binormals
of these curves are inclined at the same constant angle.
Let θ be the the inclination of νi to ν i, then θ is constant.
If the curve Г is not plane, i.e., τ ≠ 0 (twisted curve), then from (iii)
ds ds
λ i = λ i (1 − a χ ) + aτν i . (iv)
ds ds
λ i lies in the plane λiνi, i.e., on the rectifying plane.
Also, we can write λ i = cos θ λ i + sinθν i . (v)

Comparing Equations (iv) and (v), we have
1 a a ds
cos sin ds
1 a a cot . (vi)

Differentiating Equation (v) with respect to s, we get
ds dθ dθ
χµ i = − sinθλ i + cos θχµ i + cos θν i − τµ i sinθ
ds ds ds
dθ
= µ i ( χ cos θ − τ sinθ ) + (cos θν i − sinθλ i ).
ds

Since the curves are Bertrand curves, µ i and μi are parallel and equal.
d i i
Therefore, (cos sin ) 0
ds
d
0 constant
ds
1 – aχ = aτ cotθ = bτ, where b = a cotθ = constant.
Hence, we get a linear equation in τ and χ with constant coefficients as
aχ + bτ = 1.
This is the necessary condition so that the two curves are Bertrand
curves when the curves are not plane.
Example 8.5.2. Find the curvature and torsion at any point of the curve:
The equations of curve C in cylindrical coordinates
x1 = a
x2 = θ(s)
x3 = 0.
Solution: This curve is a circle of radius a. The square of the elements of s

in cylindrical coordinate is
ds2 = (dx1)2 + (dx2)2 + (dx3)2
= gijdxidxj i, j = 1,2,3
where, x1 = a, x2 = θ(s), x3 = 0; y1 = x1 cos x2, y2 = x1 sin x2, y3 = x3
ds2 = (dy1)2 + (dy2)2 + (dy3)2 = (dx1)2 + (x1)2(dx2)2 + (dx3)2), so that g11 = 1,
g22 = (x1)2, g33 = 1, and gij = 0 where i ≠ j, g = (x1)2 = a2, and this verifies that
the non-vanishing Christoffel symbols are
 1  1  2  1
  = −x ,  = 1 .
 22   12  x
Curves in Space 239
The components of tangent vector λ to circle C are
dx i 1 dx 1 dx 2 dθ
λi = :λ = = 0, λ 2 = = , and λ 3 = 0.
ds ds ds ds
Since λi is a unit vector,
∴ gijλiλj = 1 at all points of C,
g11(λ1)2 + g22(λ2)2 + g33(λ3)2 = 1,

2
 dθ 
or 1.0 + ( x1 )2 + 1.0 = 1,
 ds 

2
d2 d 1
or a . 1, .
d (s d (s a
Now, using the 1st Serret-Frenet formula,
d λ i  i  j dx k
+  λ = χµ i , for i = 1
ds  j k  ds
d λ 1  1  i dx k
χµ1 = + λ
ds  j k  ds

1 2 dx 2 d
0 0 ( x1 ) 2
2 2 ds ds
2
d 1 1 1
0 ( x1 ) 2
a.( 2 2
) a. a. (8.20a)
ds a a2 a

d λ 2  2  j dx k d  dθ   2  2 dx 1
2
for i = 2, χµ = + λ = + λ
ds  j k  ds ds  ds   2 1  ds
d  1 1 dθ dx 1 11 (8.20b)
= + 1 = 0+ .0 = 0
 
ds a x ds ds aa

d λ 3  3  j dx k  3  2 dx 2
for i = 3, χµ 3 = +  λ = 0+  λ = 0 + 0 = 0.
ds  j k  ds 
2 2
 ds

(8.20c)
For μ as a unit vector, we have
∴ gijμiμj = 1 at all points of C
or g11(μ1)2 + g22(μ2)2 + g33(μ3)2 = 1 , or 1.(μ1)2 + 0 + 0 = 1
or, (μ1)2 = 1 (8.20d)
(Here,
1 g 21 g 21 g 22 1 g 22 1 (x1 )2 1
[22,1] -- -- - .2 x1 -x1 .
2 x2 x2 x1 2 x1 2 x1 2
1 1
g 1l [22 ,l] g11[22,1] .(-x1 ) -x1.
22 g 11

1 g 22 g 12 g 12 1 (x1 )2 1 1 1 2
[12, 2] .2 x x
2 x1 x2 x2 2 x1 2 21
2 1 1 1 1 1 2
g 2l [12 ,l] g 22[12, 2] (x ) x ,
12 g 22 (x1 )2 x1 22
1 1 (x1 )2
g 2l [22 ,l] g 22[22 ,2] . 0.
(x1 )2 2 x 2
Curves in Space 241
All others Christoffel symbols are zero.)

From above,
1
χµ1 = −
a
2
1 2  1
or ( χµ ) =  − 
a
1
∴ χ 2 ( µ1 )2 =
a2
2 1 1
.1
a2 a (nonnegative).
1 1
From (8.20a), χµ = −
a
1 1
∴ µ1 = −
a a
⇒μ1 = −1.
1
We get, μ1 = − 1, μ2 = μ3 = 0, and χ =
a
2nd Part : Using the 2nd Serret-Frenet formula:
dµi  i  j dx k
for, i = 1,  j k µ = τν i − χλ i ,
ds   ds
 1  1 dx 2
or 0 +  µ = τν 1 − χλ 1 ,
12
  ds
dθ 1
or 0 + 0.µ1 = τν 1 − .0
ds a
∴τν 1 = 0 (8.20e)

d µ 2  2  j dx k
for i = 2, + µ = τν 2 − χλ 2 ,
 2  1 dx 2
or 0 +  µ = τν 2 − χλ 2 ,
 1 2  ds
1 dθ 11
or 0 + 1 µ1 = τν 2 − ,
x ds aa
1 1 11
or (−1) = τν 2 −
a a aa
2
∴τν = 0 (8.20f)

For ν is a unit vector, we have
gijνiνj = 1 or g11(ν1)2 + g22(ν2)2 + g33(ν3)2 = 1
or 1. (ν1)2 + (x1)2(ν2)2 + 1.(ν3)2 = 1.
Multiplying τ2, we get τ2(ν1)2 + τ2 (x1)2(ν2)2 + τ2 . (ν3)2 = τ2
or, by (8.20e) and (8.20f), (ν3)2 = 1 (8.20g)
Using the S.F. 2nd formula,
d µ 3  3  j dx k
+ µ = τν 3 − χλ 3

3 1 3
or 0 0 .0 , 0 (8.20h)
a
By (8.20g) in (8.20h), we get τ = 0 and we can easily find νi as ν1 = ν2 = 0,

ν = 1.
3
To determine the components of ν, we know
ν i = irs λr µs .
Curves in Space 243
i = 1 gives ν 1 = 1rs λr µs = 123λ2 µ3 + 132 λ3 µ2 = 0.

i = 2 gives ν 2 =  2rs λr µs =  213λ1 µ3 +  231λ3 µ1 = 0.

i = 3 gives ν 3 = 3rs λr µs = 312 λ1 µ2 + 321λ2 µ1 = 321λ2 µ1

1 321
= e λ2 µ1 .
g

2 1
Here, 2 g 22 a2 a
a
and μ1 = g11μ1 = 1 and (−1) = −1
3 1 321 1
Therefore, v = e λ2 µ1 = (−1).a.(−1) = 1
g a
∴ v = (0,0,1)
Example 8.5.3.
(a) Find the curvature and torsion at any point of the circular
helix where the equation in cylindrical coordinates is: C: x1
= a, x2 = θ(s), x3 = kθ.
(b) Show that the tangent vector λ at every point of C makes a
constant angle with the direction of x3-axis.
Solution:
Here, ds2 = (dx1)2 + (dx2)2 + (dx3)2.
= gij dxidxj, i, j = 1,2,3.
This curve is a circle of radius a. The square of the element of ds in cylin-

drical coordinates is
ds2 = (dx1)2 + (dx2)2 + (dx3)2
= gijdxidxj i, j = 1,2,3
here, x1 = a, x2 = θ(s), x3 = kθ; y1 = x1 cos x2, y2 = x1 sin x2, y3 = x3
ds2 = (dy1)2 + (dy2)2+ (dy3)2 = (dx1)2 + (x1)2(dx2)2 + (dx3)2,
so that g11 = 1, g22 = (x1)2, g33 = 1 and gij = 0 where i≠j, and g = (x1)2 = a2,
so that g11 = 1, g22 = (x1)2, g33 = 1 and gij = 0, where i≠j,
dx 2 dθ dx 3 dθ
and λ 1 = 0 , λ 2 = = λ3 = =k .
ds ds ds ds
We know gijλi λj = 1⇒g11(λ1)2 + g22(λ2)2 + g33(λ3)2 = 1
2 2
2 dθ   dθ 
0+a . + k2 =1
 ds   ds 

2
 dθ  1
= 2 2
 ds  k +a

 dθ  1
= 2 2 = A(let ) (8.21a)
 ds  k +a

Using the 1st Serret-Frenet formula,
d λ i  i  dx k
j
χµ i = + λ
ds  j k  ds
 
for i = 1
d λ i  i  dx k
j
 1  dx 2
χµ1 = + λ = 0+  λ
2
ds  j k 
 
ds  2 2  ds
2
1 dθ
2  dθ 
= (− x )λ . = −a (8.21b)
ds  ds 

Curves in Space 245
d λ 2  2  dx k d  dθ   2  dx 2
j
χµ 2 = + λ = + λ
2
ds  j k  
ds ds ds  2 2  ds
   
 2  2 dx 1 2
 dθ  1 dθ
+  λ = 0. + .0 = 0 (8.21c)
2 1  ds  ds  a ds
 
d λ 3  3  dx k
j
 3  dx 2
χµ 3 = +  λ = 0 +  λ
3
= 0 + 0 = 0.
ds  j k ds 3 2 ds
   
(8.21d)
We know, gijμi μj = 1,
or g11μ1 μ1 + g22μ2 μ2 + g33μ3 μ3 = 1,
or χ2(μ1)2 + 0 + 1.0 = χ2,
or a2A4 = χ2
χ2 = a2A4,
a
or χ = aA2 = (8.21e)
a + k2
2
(Here,
1 g 21 g 21 g 22 1 g 22 1 (x11)2 1
[22,1] -- -- - .2 x1 -x1 .
2 x2 x2 x1 2 x1 2 x1 2
1 1
g 1l [22 ,l] g11[22,1] .(-x1 ) -x1 .
22 g11

1 g 22 g12 g12 1 (x1)2 1 1 1
[12, 2] .2 x x
2 x1 x2 x2 2 x1 2
2 2 1 1 1 1 1
g 2l [12 ,l] g 22[12, 2] (x ) x
21 12 g 22 (x1 )2 x1

All other Christoffel symbols are zero.)
−aA2
µ1 = = −1,
aA2
a
∴ µ1 = −1, µ 2 = µ 3 = 0 , χ = ,
a + k2
2
Using the 2nd Serret-Frenet formula,

d µ1  1  j dx k
+ µ = τν 1 − χλ 1 and we can easily find τ.
d µ i  i  j dx k
For i = 1, + µ = τν i − χλ i
d µ1  1  j dx k
+ µ = τν 1 − χλ 1 ,

 1  2 dx 2 ,
or τν 1 − χλ 1 = 0 +  µ
 2 2  ds
1
 1  2 dx 2 dx 2
or τν − χ .0 = 0 +  µ = a.0. =0
22 ds ds
 
1 1 g 22 2
g 11[22 ,1] g 11 x1 a,
2 2 2 x1 1 2
1 1 g 22 1
g 22[12 , 2]
a2 2 x1 a
∴τν1 = 0.
Curves in Space 247
d µ 2  2  j dx k
For i = 2, τν 2 − χλ 2 = + µ
 2  1 dx 2  2  2 dx 1
= 0+  µ  21 µ
 1 2  ds   ds
 2  3 dx 2  2  2 dx 3
+ µ + µ
 3 2  ds  2 3  ds
 2  1 dx 2  2  2 dx 1 1 dθ 1
= µ + µ = (−1) + .0.0
 1 2  ds  2 1  ds a ds a
 1  dθ 1
=−  =−
 a  ds a k 2 + a2
  2 
22 22 1  ∂ g 22 
  = g [12,2] = g  
  1 2  2  ∂x1 
1 x 1 1  2  22
= .2 2 = &   = g [32,2]
2 a a  3 2 
1  ∂ g 22 
= g 22  − 3  = 0 ]
2  ∂x 
1
τν 2 = χλ 2 −
a k + a2
2
a 1 1 k2
= − =− 3 .
a2 + k 2 k 2 + a2 a k 2 + a2 2 2 2
a(a + k )
d µ 3  3  dx k
j
For i = 3, τν 3 − χλ 3 = + µ =0
ds  j k  ds

 3  1  ∂ g 33   3  
 33
 = g [23,3] = 1.  2  = 0 &   = g 33
[13,3] = 0
 3 2
  2 ∂x  3 1  

αk
Or τν 3 = χλ 3 = 3
(a 2 + k 2 ) 2 .
We can find ν by the following relations: ν t = trs λr µs .

ν 1 = 123 λ2 µ3 + 132 λ3 µ2 = 0
1 231 dθ 1 1
ν 2 =  213 λ1 µ3 +  231λ3 µ1 = 0 + e (−1)k = 1. (−1)k 2 2
g ds g k +a
−k
=
a k 2 + a2
1 321 dθ 1 dθ
and ν 3 = 312 λ1 µ2 + 321λ2 µ1 = 0 + e (−1) = (−1) (−1)
g ds g ds
1
= .
a k + a2
2
We know gijνi νj = 1,
or g11ν1 ν1 + g22ν2 ν2 + g33ν3 ν3 = 1,
or τ2g22ν2 ν2 + τ2g33ν3 ν3 = τ2,
2 2
 k2   ak 
or a 2  − 3  +−
2
3  =τ ,
 a(a 2 + k 2 ) 2   (a 2 + k 2 ) 2 
k4 a2k 2 k 2 (a 2 + k 2 ) k2
τ2 = + = = ,
(a 2 + k 2 )3 (a 2 + k 2 )3 (a 2 + k 2 )3 (a 2 + k 2 )2
k2
or τ 2 =
(a + k 2 )2
2
k
τ= .
(a + k 2 )
2
We can verify τνi with these results.
Example 8.5.4. Using the results of Problem 1, show that the ratio of the
curvature χ to the torsion τ is constant. Show using Frenet’s formulas that
τ
whenever = constant and the coordinates are Cartesian, νi = cλi + bi,
χ
where c and bi are constants.
k a
Here, τ = and χ = 2 2 .
2 2
(a + k ) (a + k )
Curves in Space 249
k
τ (a 2 + k 2 ) k
Therefore, = = is constant
χ a a
(a 2 + k 2 )
1 2 dx 2 dθ 1 dx 3 dθ k
and λ = 0 , λ = = = 1 λ3 = =k = 1
ds ds ds ds
(a 2 + k ) 2 2
(a 2 + k 2 ) 2
−k 1
ν1 = 0 , ν 2 = 2 2
ν3 = .
a k +a a k + a2 2
νi =cλi
Example 8.5.5. Show that
1
i j 2
(i) g ij
s s
i δµ k j
(ii ) τ = ijk λ µ
δs
Solution: We know the 1st Serret-Frenet formula
i 1 δλ i δλ i
µ = or = χµ i ,
χ δs δs
where χ > 0 is chosen to make μi a unit vector.
δλ i j
g ij µ = g ij χµ i µ j
δs
=χ [ g ij µ i µ j = 1 ],

δλ i 1 δλ j
or χ = g ij δs χ δs
(as χ > 0),
2 δλ i δλ j
or χ = g ij δs δs
1
i j 2
g ij .
s s
(ii) We know from the 2nd Serret-Frenet formula,
δµ i
= τν i − χλ i
δs
δµ k
ijk λ i µ j = ijk λ i µ j (τν k − χλ k )
δs
= ijk λ i µ jτν k − ijk λ i µ j χλ k
( )
= ijk λ i µ j τν k − 0 = ν kν kτ = τ (as ν ν
k
k
)
= 1 and ijk λ i µ j λ k = 0 .

1 δλ i δ 2 λ k
Example 8.5.6. Show that τ = 2 ijk λ i 2 , where notations have
their usual meaning. χ δ s δ s
1 δλ i δλ i
Solution: We know that µ i = or, = χµ i .
χ δs δs
δ 2 λ i δχ i δµ i δχ i
2 = µ + χ= µ + (τν i − χλ i )χ (using the 2nd Serret-Frenet
δs δs δs δs
formula).
i 2 k
1 i δλ δ λ 1 i i δχ 
Now, 
2 ijk λ 2 = 2 ijk λ χµ  µ k + (τν k − χλ k )χ 
χ δs δs χ δ s 
1 δχ 
= ijk λ i µ j µ k + τχ ijk λ i µ jν k − χ 2ijk λ i µ i λ k 
χ  δ s 
1
= (0 + τχ − 0) = τ .
χ
Example 8.5.7.
δ 2 λ i dχ i
Show that 2 = µ + χ (τν i − χλ i )
δs ds
2 i
δ λ dτ i dχ i
2 = ν − ( χ 2 + τ 2 )µ i − λ
δs ds ds
δ 2ν i i i dτ i
2 = τ ( χλ − τν ) − µ,
δs ds
Curves in Space 251
where symbols have their usual meanings.

Solution: We know from the first Serret-Frenet formula
δλ i
= χµ i .
δs

δ 2λ i δ i d( χµ i )  i 
j dx
k
= ( χµ ) = +   ( χµ )
δ s2 δ s ds j k  ds
 
d( χ ) i d( µ i )  i  dx k
= µ + χ + χµ j  
ds ds  j k  ds
d( χ ) i  d( µ i )  i  dx k 
= µ +χ + µj   
ds  ds  j k  ds 
   
d( χ ) i δµ i
= µ +χ
ds δs
d( χ ) i
= µ + χ (τν i − χλ i ) (from the Serret-Frenet 2nd formula)
ds
2 i
δ λ d( χ ) i
2 = µ + χ (τν i − χλ i )
δ s ds
(ii) Using the Serret-Frenet formula,
δµ i
= τν i − χλ i
δs
δ 2 µ i δ (τν i ) δ ( χλ i )
= −
δ s2 δs δs
d(τν i )  i  dx k  d( χλ i )   k
i  i  dx
= + τν i   −  + χλ  j k 
ds  j k  ds  ds   ds 

d(τν i )  i  dx k  d( χλ i )  i  dx k 
= + τν i   − + χλ i  
ds j k  ds  ds j k  ds 
   
  
d(τ ) i d(ν i )  i  dx k  d( χ ) i d(λ i )  i  dx k 
= ν + τ + τν i   − λ + χ + χλ i  
ds ds j k  ds  ds ds j k  ds 
     
d(τ ) i  d(ν i )  i  dx k  d( χ ) i  d(λ i )  i  dx k 
= ν +τ  +νi   − λ −χ + λi   
ds  ds  j k  ds  ds  ds  j k  ds 
d(τ ) i δν i d( χ ) i δλ i
= ν +τ − λ −χ
ds δs ds δs
d(τ ) i d( χ ) i  δν i i δλ
i
i
= ν + τ (−τµ i ) − λ − χ ( χµ i )  δ s = −τµ , δ s = χµ 
ds ds  
d(τ ) i d( χ ) i
= ν − (τ 2 + χ 2 )µ i − λ.
ds ds
 d(τµ i )   k
δ 2ν i δ i i  i  dx
(iii) = (−τµ ) = −  + τµ  
δ s2 δ s ds j k  ds 
   
dτ i d µ i  i  dx k
=− µ − τ − τµ i  
ds ds  j k  ds
dτ i  dµi  i  dx k 
=− µ −τ  + µi   
ds  ds  j k  ds 
dτ i δµ i
=− µ −τ
ds δs
dτ i  δµ i 
=− µ − τ (τν i − χλ i )  Θ = (τν i − χλ i )
ds δs 
dτ
= τ ( χλ i − τν i ) − µ i .
ds
8.6 Equations of a Straight Line

Let Ai be a vector field defined along a curve C in E3, where C is given para-
metrically as
Curves in Space 253
C:xi = xi(s) , s1 ≤ s ≤ s2 , (i = 1,2,3.),
s being the arc parameter.

If the vector field Ai is parallel, then it follows from (8.3) that
δ Ai
=0
δs
dAi  i  α dx β
or + A =0 (8.22)
ds  α β  ds

We shall use this equation to obtain the equations of a straight line in

general curvilinear coordinates. The characteristic property of a straight
line is that the tangent vector λ to a straight lines is directed along the
straight line, so that the totality of tangent vectors λ forms a parallel vector
dx i
field. Thus, the field of tangent vector λ i = must satisfy (8.22) and we
have ds
δλ i d 2 x i  i  dx α dx β
= 2 +  = 0.
δs ds αβ ds ds
 
d 2 x i  i  dx α dx β
The equation +  = 0 is a straight line equation.
ds 2  α β  ds ds
In Cartesian coordinates, the Christoffel symbols vanish and we obtain
d2xi
the familiar straight line equation 2 = 0.
ds
In view of geometrical interpretation of curvature χ as a measure of the
rate of turning of the tangent vector to the curve will be zero since the
angle is constant.
Example 8.6.1. Show that a space curve is a straight line if, and only if, its
curvature is zero at all points of it.
Solution: Suppose that the curvature χ = 0 at all points of a space curve Г.
Then, by Serret-Frenet’s 1st formula,
δλ i
= χµ i
δs
δλ i
= 0 for all s , [since χ = 0]
δs
⇒λi = Constant (with respect to s)
This means that λi is fixed direction
Space curve is a straight line.
Hence Γ is a straight line
Conversely, Space curve is a straight line.
⇒ direction of the tangent vector of space is fixed.
⇒χμi = 0 (since μi = 0)
It follows that χ = 0.
d λ i  i  j dx k
or + λ =0
ds  j k  ds

d i i
, 0 , or ks c ,
ds
which is a straight line. Hence Γ is a straight line.
8.7 Helix
A space curve is called a helix if the tangent vector at every point to helix
makes a constant angle with a fixed direction.
a2
90°–α
α
P
µ
Figure 8.3
Curves in Space 255
Let C: xi = xi(s) be an equation of a helix. By definition, if ai is the fixed

direction and α is the angle between ai and the tangent vector λj, as shown
in Figure (8.3), then
π
cosα = g ij a j λ i , 0 < |α | ≤ (8.23)
2
δλ i δaj i
g ij a j + g ij λ =0
δs δs
δλ i
or g ij a j = 0,
δs
 δλ i i
or g ij a j χµ i = 0 ,   δ s = χµ 
or g ij a j µ i = 0 , as χ ≠ 0 (8.24)

This shows that aj is orthogonal to μi.
Since the curvature χ of a helix does not vanish, the principal normal μi
is perpendicular everywhere to the generators. Hence, the fixed direction
of the generators is parallel to the plane of μi and νi and, since it makes a
constant angle α with λi, it makes a constant angle of 90° − α with νi.
We know from the definition of a helix that the curvature and torsion
are in a constant ratio. Since the angle between aj and νi is 90° − α,
Cos(90° − α) = gijajνi or sinα = gijajνi (8.25)

Differentiating intrinsically (8.24) with respect to s, we get
δµ i
g ij a j = 0.
δs
Using Serret-Frenet’s 2nd formula,
gijaj(τνi − χλi) = 0.
Using (8.23) and (8.25), we have τ
τgijajνi – χgijajλi = 0
or τsinα – χCosα = 0.
χ
Therefore, = tanα = constant,
τ
but ai is parallel to the plane of λi and νi and must be parallel to the vector
χ
τλi + χνi, which is inclined to λi at an angle tan −1 . This angle is constant.
τ
Therefore, the curvature and torsion are in a constant ratio.
Conversely, we can prove that a curve whose curvature and torsion are
in a constant ratio is a helix.
We suppose that a curve different from a straight line is
χ 1
= , constant ie, τ = c χ .
τ c
Using Serret-Frenet’s 3rd formula, we get

δν i
= −τµ i = −c χµ i
δs
δλ i  δλ i 
= −c  using 1st frenet formula, = χµ i 
δs δs 
δ i
or (ν + cλ i ) = 0.
δs
Therefore, νi + cλi is a constant vector, say ai, i.e., νi + cλi = ai.
Now, gijaiλj = gij(νi + cλi)λj

= gijνiλj + cgijλiλj = 0 + c.1 = c.
This shows that the tangent vector λi makes a constant angle with a fixed
direction ai and the curve is therefore a general helix.
Theorem 8.7.1. The necessary and sufficient condition for a given curve to
be a helix is that the ratio of the curvature to the torsion is constant.
8.7.1 Cylindrical Helix

The cylindrical helix is a curve upon a cylinder which cuts the generator
of the cylinder at a constant angle α with a fixed direction. Let us consider
Curves in Space 257
x3 as the axis of the cylinder. Then, the equation of cylinder C: xi = fi(t),

i = 1,2,3 (Figure 8.4).
α
µ λ
α
α
P
α
γ
C: f t(t)
Figure 8.4
Let α be the angle at any point P on the curve between tangent a and
the generator of the cylinder passing through P. The unit vector along the
generator is
k = (0,0,1)
and the unit tangent vector to the curve at P is
δ x i δ x i /δ t δ x i /δ t
λi = = =
δs δ s /δ t |δ x i /δ f |
( f 1 , f 2 , f 3 )
= 1
{ f 1 (t )}2 + { f 2 (t )}2 + { f 3 (t )}2  2
.
( f 1 , f 2 , f 3 ).(0 ,0 ,1)
∴ cosα = g ij k j λ i = 1
{ f 1 (t )}2 + { f 2 (t )}2 + { f 3 (t )}2  2
f 3
= 1 ,
 2  2 2 
{ f (t )} + { f (t )} + { f (t )}  2
1 3 2

or ( f 3 )2 = cos 2 α { f 1 (t )}2 + { f 2 (t )}2 + { f 3 (t )}2 
( f 3 (t ))2 cos2 { f 3 (t )}2 cos2 { f 1(t )}2 { f 2 (t )}2 ,

or sin 2 α { f 3 (t )}2 = cos 2 α { f 1 (t )}2 + { f 2 (t )}2 

1
f 3 (t ) = cotα { f 1 (t )}2 + { f 2 (t )}2  2

1

3
∫
∴ f (t ) = cotα { f 1 (t )}2 + { f 2 (t )}2  2 dt + constant ,
1
or, f 3 (t ) = x 3 (t ) = cotα
∫ [{x 1}2 + {x 2 }2 ]2 dt + constant .
Therefore, the equation of the helix is
1
∫
x 1 (t ) = f 1 (t ),x 2 (t ) = f 2 (t ), x 3 (t ) = cotα [{x 1 }2 + {x 2 }2] 2 dt + constant .
(8.26)
8.7.2 Circular Helix

It follows from the definition that a helix can always be drawn on the sur-
face of a cylinder.
If a space curve Γ lies on a circular cylinder and the tangent at each point
of it makes a constant angle with the axis of the cylinder, then the curve is
a helix. Such helixes are said to be Circular Helixes.
A particular case of the cylindrical helix is the circular helix.
Let x1 = acosθ, x2 = asinθ, and
θ θ
x 3 = cotα
∫ 0
a 2 (cosθ + sin 2θ ) dα = cotα
∫ 0
a dα
= acotα .θ = bθ , where b = acotα = constant .
The parametric equation of a circular helix is
x1 = acosθ, x2 = asinθ, x3 = bθ , b ≠ 0. (8.27)

Example 8.7.1. Prove that
δλ i δ 2 λ j δ 3λ k 5 d  τ 
ijk 3 =χ .
δs δs δs 2
dx  χ 

Hence, a space curve is a helix if, and only if,
δλ i δ 2 λ j δ 3λ k
ijk = 0.
δ s δ s2 δ s3
Curves in Space 259
Solution:
Using Serret-Frenet’s 1st formula,
δλ i
= χµ i.
δs
δ 2λ i δµ i δχ
2 = χ + µi
δs δs δs
δχ
= µi + χ (τν i − χλ i ) .
δs

δ 3λ i 2
i δ χ δµ i δχ δχ δ
Now, 3
= µ 2
+ + (τν i − χλ i ) + χ (τν i − χλ i )
δs δs δs δs δs δs
i i
2
δ χ δµ δχ δχ  δν δτ δλ i δχ i 
= µi 2 + + (τν i − χλ i ) + χ τ +νi −χ − λ
δs δs δs δs  δs δs δ s δ s 
δ 2 χ δχ δτ
= µi 2
+ (τν i − χλ i ) + χτ (−τµ i ) + χν i − χ 2 ( χµ i )
δs δs δs
δχ i δχ δ 2 χ δχ
−χ λ ] + (τν i − χλ i ) = µ i 2 + (τν i − χλ i ) + χτ (−τµ i )
δs δs δs δs
 δτ δχ  δχ  δ 2χ 
+ν i  χ + τ  − χ 2 ( χµ i ) − 2 χ λ i =  2 − χτ 2 − χ 3  µ i
 δs δs  δs  δs 
 δτ δχ  δχ i .
+ χ + 2τ  ν i − 3 χ λ
 δs δs  δs

Therefore,
δλ i δ 2λ j δ 3λ k
i
δ 2 λ j δ 3 λ k ijk δλ i δ 2 λ2 j δ 3 λ3k
ijk δ s δs δs
s δ s2 δ s3 δs δ s2 δ s3
  δ 2 χ 
2
 δχ
  +δ χχ(τν jj − 2χλ jj )3 δk χ2 − χτ 22 − χ 33  µkk
2
δχ = ijk jχµii  µj jj δχ
i
µj χ=(τν
ijk χµ− χλ µ )δs + χ2(τν − χτ − χλ− χ)µ  δ s2 − χτ − χ  µ
δs  δs δ s  δ s 
δτ δχ δχ
 δτ δχ  +  χ δτδχ+ 2τk δχ  ν kk − 3 χ δχ λ kk 
+χ + 2τ  ν+k −χ3 χδ s + λ 2τ δ s  ν − 3 χ δ s λ 
 δs δ s   δ sδ s δ s  δs 
 i j δχ 2 i j 3 i j
−
 δs   δ s2
260 IntroductiontoδτDifferential
δχ  k Geometry
δχ k  with Tensor
+χ + 2τ  ν − 3 χ λ
 δs δs  δ s 
 δχ 
=  χijk µ i µ j + χ 2τ ijk µ iν j − χ 3ijk µ i λ j 
 δs 
 δ 2 χ 3 k  δτ δχ  δχ k 
2
 2 − χτ − χ  µ +  χ + 2τ  ν k − 3 χ λ
 δ s δ s δ s  δ s 
 δ 2 χ   δτ δχ  δχ k 
= (0 + χ 2τλ k − χ 3ν k )  2 − χτ 2 − χ 3  µ k +  χ + 2τ  ν k − 3 χ λ 
 δ s   δs δs  δs 
 δ 2χ   δτ δχ 
=  2 − χτ 2 − χ 3  ( χ 2τλ k µ k + χ 3ν k µ k ) +  χ + 2τ  ( χ 2τλ kν k + χ 3ν kν k )
 δs   δs δs 
δχ 2 k k
−3 χ ( χ τλ λ + χ 3ν k λ k )
δs
 δτ δχ  δχ
= 0 + χ3  χ + 2τ  − 3 χ ( χ 2τ )
 δs δs  δs
 δτ δχ  1  δτ δχ  d τ
= χ3  χ − τ  = χ5 2  χ − τ  = χ5 s   .
 δs δs  χ  δs δs  d  χ
τ
Second Part: If a curve is a helix, then is constant.
χ
d τ
Hence,   = 0.
ds  χ 
δλ i δ 2 λ j δ 3 λ k
Therefore, ijk = 0.
δ s δ s2 δ s3
δλ i δ 2 λ j δ 3 λ k
Conversely, let ijk = 0,
δ s δ s2 δ s3
or ijk ( χµ i )χ (τν j − χλ j )(− χτ 2 − χ 3 )µ k = 0 ,

or ijk ( χµ i )χ (τν j − χλ j )(τ 2 + χ 2 )µ k = 0 ,

or − ijk χ 3 µ i µ k (τν j − χλ j )(τ 2 + χ 2 ) = 0 ,

or − ijk χ 3 µ kτν j + ijk χ 3 µ k χλ j = 0, as (τ2 + χ2) ≠ 0 and μi are an

arbitrary vector,
Curves in Space 261
or − ijkτµ kν j + ijk χµ k λ j = 0 ,

or (τλi + χνi) = 0
gij(τλj + χνj) = 0.
If ai is the fixed direction, then, by hypothesis, gijaiλj = cos α and gijaiνj
= sin α.
Therefore, gijai(τλj + χνj) = 0

∴ τ cos α + χ sin α = 0
or tan constant ,
hence, the curve is a helix.
Example 8.7.2. If the tangent and the binormal to a space curve make
χ sinα dα
angles α and β with a fixed direction, then show that = − .
τ sinβ dβ
Let a be the fixed direction and the tangent vector with component λi
i
to a space curve C and make an angle α with ai. The binormal μi makes an
angle β with ai. Then, by the hypothesis
gijaiλj = cos α and gijaiνj = cosβ.
δα dα δβ d β
since α and β are invariants, = and =
δ s ds δ s ds
Differentiating intrinsically, we get
δλ j δ d dα
g ij ai = cos α = (cosα ) = − sinα
δs δs ds ds
j
δν δ d dβ
g ij ai = cos β = (cosβ ) = − sinα .
δs δs ds ds

Using Serret-Frenet formulas, we get
δλ j dα dα
g ij ai = − sinα or, g ij ai χµ j = − sinα
δs ds ds
i δν j dβ dβ
and g ij a − sinβ or, g ij ai (−τµi ) = − sinβ .
δs ds ds
Dividing these two, we get
dα
g ij ai χµ j − sinα ds
=
g ij aiτµ i dβ
sinβ
ds
sin d
or .
cos d
8.8 Exercises δ Ai δ Ai
1. If for a contravariant vector A ,i
= 0 , prove that = 0 where Ai is
δt δt
the associate tensor of A . i
d( g ij Ai B j ) δ Ai j δ Bj
2. Prove that = g ij B + g ij Ai .
dt δt δt
3. If the intrinsic derivative of a vector A along a curve C vanishes at all
points of C, show that the magnitude of A is constant along curve C.
4. If two unit vectors are such that at all points of a given curve C their
intrinsic derivatives along C are zero, show that they are inclined at a
constant angle along C.
5. Find the curvature and torsion at any point of a given curve, namely the
circle C, where
C: x1 = a, x2 = t, x3 = 0, and ds2 = (dx1)2 + (x1)2(dx2)2 + (dx3)2.

χ
6. Show from the Serret-Frenet formula that whenever = constant and
the coordinates are Cartesian, τ
νj = cλi + bi,
where c and bi are constants and the tangent vector makes a constant
angle with a fixed direction.
7. Let C be a cylindrical helix determined by
Curves in Space 263
 y 1 = φ (σ )

C:  y 2 = ϕ (σ )
 3
 y = kσ , k = constant ,

where σ is the arc parameter of the directrix curve C′ in the y1y2 −plane,
so that
(dσ)2 = (dy1)2 + (dy2)2 and (ds)2 = (1 + k2)(dσ)2 and show that
φ′ ϕ ′ k
φ ′′ ϕ ′′ 0
φ ′′′ ϕ ′′′ 0 1
τ= 2 .
χ (1 + k 2 )3
φ ′ϕ ′′ − ϕ ′φ ′′
χ= .
1 + k2

χ
Also, verify that = k.
τ
8. Show that a space curve is a straight line if, and only if, its curvature is
zero.
9. Show that a space curve is a plane curve if, and only if, its torsion is zero.
1 δλ i δ 2 λ k
10. Show that τ = 2 ijk λ i .
χ δ s δ s2
9
Intrinsic Geometry of Surfaces
9.1 Introduction
We will discuss the differential geometry of surfaces by means of Tensor
Calculus. In this chapter we study the properties of surfaces imbedded in
three-dimensional Euclidean spaces. It will be shown that the properties
can be phrased independently of the space in which the surface is immersed
and that they are concerned solely with the structure of the differential
quadratic form for the element of arc of a curve drawn on the surface.
All such properties of surfaces are termed as intrinsic properties and the
geometry based on the study of differential quadratic forms is called the
intrinsic geometry of the surface.
9.2 Curvilinear Coordinates on a Surface

We study the surface in which the surface is imbedded to a set of orthog-
onal Cartesian axes Y and the locus of points are satisfying the equation
F(y1, y1, y1) = 0 (9.1)
A surface S is defined, in general, due to Gauss as the set of points whose

coordinates are functions of two independent parameters. Thus, the equa-
tion of a surface is of the form
S:yi = yi (u1,u2), (9.2)
where u11 ≤ u1 ≤ u12 and u12 ≤ u 2 ≤ u22 and yi are real functions of class C1
in the region of definition of the independent parameters u1 and u2. If we
assign to u1 in (9.2) some fixed value u1 = c, we obtain Figure (9.1) as a locus
of the one–dimensional manifold
265
u2 = u02
P0
u1 = u01
Figure 9.1
yi = yi(c,u2), (i = 1,2,3),
which is a curve lying on the surface S defined by Equation (9.2). We shall

call it a u2−curve. Similarly, u2 = constant in (9.2) defines a u1− curve, along
which only u1 varies. By assigning to u1 and u2 a succession of fixed val-
ues, we obtain a net of curves, on the surface which are termed coordinate
curves.
The intersection of a pair of coordinate curves is obtained by setting
1 2
u u0 , u 2 u0 determines P0.
1
The variables u1and u2 determining the point P on S are called the

curvilinear or Gaussian coordinates on the surface. If one introduces a
transformation
u1 u1 u 1 , u 2
, (9.3)
u2 u2 u 1 , u 2

∂(u1 , u 2 )
where u (u , u )
α 1 2
are of class C and Jacobian J =
1
does not van-
∂(u 1 , u 2 )
i
ish in the region of variables u , then one can insert the values from (9.3)
in (9.2) to obtain a different set of parametric equations
y i = f i (u 1 , u 2 ) , i = 1,2,3. (9.4)

Intrinsic Geometry of Surfaces 267
The parametric representation of a surface is not unique and there are

infinitely many curvilinear coordinates systems which can be used to locate
points on a given surface S.
For example, the equations
x1 = u1 + u2 x1 = v1 coshv2
x2 = u1−u2 and x2 = v1 sinhv2
x3 = 4u1−u2

x3 = (v1)2
represent the same surface: (x1)2–(x2)2 = x3.

The two representations may be related by parametric transformation
v = 2 u1u 2 ,,
1
1 u1
v2 log 2 .
2 u
9.3 Intrinsic Geometry: First Fundamental

Quadratic Form
In the last section, we saw that the properties of surfaces that can be
described without reference to the space in which the surface is imbed-
ded are called intrinsic properties. A study of intrinsic properties is made
to depend on a certain quadratic differential forms describing the metric
character of the surface.
Here, we will be dealing with two distinct sets of variables: those refer-
ring to the space E3 in which the surface is imbedded and two curvilinear
coordinats u1 and u2, referring to the two dimensional manifold S. We use
Latin letters for the indices referring to space variables and Greek letters
for the surface variables.
A transformation T of space coordinates from one system (xi) to another
system ( x i ) will be written as
T : xi x i x 1, x 2 , x 3
and a transformation of Gaussian surface coordinates will be denoted by

uα = uα (u 1 , u 2 ) .
Consider a surface S defined by
yi = yi(u1,u2), (9.5)
where yi are the orthogonal Cartesian coordinates covering space E3 in

which the surface S is imbedded and a curve C on S is defined by
uα = uα (t ), t1 ≤ t ≤ t 2 , (9.6)

where uα’s are the Gausian coordinates covering S. The curve defined by
(9.6) is a curve in a three-dimensional Euclidean space and its element of
arc is given by the formula
ds2 = dyidyi. (9.7)
From (9.5), we get
∂ yi α
dy i = du (9.8)
∂uα
and from (9.6),
∂uα
α
du = dt.
∂t
In (9.7), we have
∂ yi ∂ yi α β
ds 2 = du du
∂uα ∂u β
= aαβ duα du β ,
∂ yi ∂ yi .
Where aαβ = (9.9)
∂uα ∂u β
The expression of ds2
ds 2 = aαβ duα du β (9.10)

Is the square of the linear element of C lying on the surface S and aαβ duα du β
is called the first fundamental quadratic form of the surface.
The length of the arc of the curve defined by Equation (9.6) is given by
the formula
t2 duα
∫
α
s= α β
aαβ u u dt , where 
u = .
t1 dt
From (9.10), it follows that since ds2>0, at once upon setting uα = con-
stant and u1 = constant in turn, ds2(1) = a11(du1)2 and ds2(2) = a22(du2)2. Thus,
a11 and a22 are positive functions of u1 and u2.
Example 9.3.1. Show that aαβ is a covariant metric tensor of the surface.
Consider a transformation of surface coordinates
uα = uα (u 1 , u 2 ) (9.11)
∂uα
with a non-vanishing Jacobian J = .
∂u α
We have u = u (u , u ) , that
α α 1 2
∂uα γ
duα = du .
∂u γ
We have
ds 2 a du du
u u
=a du du
u u
u u
If we set a a ,
u u
we see that the set of quantities aαβ represents a covariant tensor of rank
two.
Now, we have to show that aαβ is a symmetric tensor.
Let
1 1
aαβ =
2
( aαβ + aβα ) + ( aαβ − aβα ) = Aαβ + Bαβ. .
2

It is clear to us that Aαβ is symmetric and Bαβ is a skew-symmetric tensor.

Now, aαβ duα du β = Aαβ duα du β + Bαβ duα du β ,
or a A du du B du du , (i)
or ( aαβ − Aαβ ) duα du β = − Bβα duα du β , since Bαβ is skew-symmetric,

or ( aαβ − Aαβ ) duα du β = − Bαβ du β duα , with an interchanging dummy
index,
or a A du du B du du . (ii)

Thus, we get 2 Bαβ duα du β = 0 from (i) and (ii).

⇒ Bαβ = 0 (Since duα and du β are arbitrary)
1
Therefore, aαβ = ( aαβ + aβα ) = Aαβ
2
Hence, the set of quantities aαβ represents a symmetric covariant tensor
of rank 2.
The tensor aαβ is called the covariant metric tensor of the surface.
9.3.1 Contravariant Metric Tensor

Since ds 2 = aαβ duα du β is positive definite, the determinant
a11a12
a= > 0.
a21 a22

If we can define the reciprocal tensor by the formula

αβ cofactor of aαβ in a
a = , then aαβ aβγ = δ γα .
a
a22 12 a a
11
We have a = , a = a 21 = − 12 , and a 22 = 11
a a a
The contravariant tensor aαβ is called the contravariant metric tensor.
The direction of a linear element on the surface can be specified either
by direction cosine or direction parameters.
duα
λα = (9.12)
ds
and
dy i y i du
.
ds u ds
We define the length of the surface vector Aα. That is, the vector deter-
mined by A1(u1,u2) and A2(u1,u2), by the formula
A = aαβ Aα Aα .

We have ds 2 = aαβ duα du β
or
du du
1 a
ds ds
a

This shows that λα are components of a unit vector.

The covariant vector
λβ = aαβ λ α (9.13)

is sometimes called the direction moment and from (9.13) we get
2
a λβ = a aαβ λ = δα λ = λ and a λβ λγ = λ λγ = λ , (9.14)
γβ γβ α γ α γ γβ γ
so that
λ α λα = aαβ λ α λ β .

2
If we interchange the dummy index β and γ in (9.14), we get a βγ λγ λβ = λ .
Therefore, aγβ is a symmetric tensor.
9.4 Angle Between Two Intersecting Curves

on a Surface
We have to find the angle between two intersecting curves on a surface.
The equation of curve C drawn on the surface S can be written in the form
C : uα = uα (t ) .

Since the uα(t) are assumed to be class C2, curve C has a continuously
turning tangent. Let C1 and C2 be two such curves intersecting at point P of
S, as shown in Figure (9.2). If λα and μα are two unit vectors in the direction
d1uα α d2 µ α
of the tangents to C1 and C2 at P, respectively, then λ =
α
,µ = ,
ds1 ds2
where s1 and s2 denote the arc length of C1 and C2 respectively. We denote
the direction cosine of the tangent lines to C1 and C2 at P by ξ i and η i ,
respectively. The cosine of angle θ between C1 and C2 is
cosθ = ξ iη i , (9.15)
∂ y i d1uα i ∂ y i du β
where ξ i = , η = β and
∂uα ds1 ∂u ds2
Y3
ηi C2 i
P θ
ξi
C1
S
b3 r
O b2 Y2
b1
Y1
Figure 9.2
∂ yi α i ∂ yi β
ξi = λ , η = β µ (9.16)
∂uα ∂u
Putting these values in (9.15), we get
∂ yi ∂ yi β α
cosθ = µ λ
∂uα ∂u β
∂ yi ∂ yi
and since aαβ = ,
∂uα ∂u β
Therefore,
cosθ = aαβ µ β λ α ,
(9.17)
If curves C1 and C2 are orthogonal,
π
θ= , ⇒ cosθ = 0.
2
From (9.17), we get
aαβλαμβ = 0 (9.18)
Example 9.4.1. The coordinate curves will form an orthogonal net if a12 =
0 at every point of the surface.
Solution: The surface vectors λα and μβ are taken along the coordinate
curves, that is, if λα and μβ are orthogonal, then
1 1
λ1 = , λ 2 = 0 and µ1 = 0 , µ 2 = .
a11 a22

We know aαβ λα μβ = 0 and substituting these values, we get
a11 µ1λ 1 + a22 µ 2 λ 2 + a12 λ 1 µ 2 + a21λ 2 µ1 = 0

1 1
or 0 + 0 + a12 +0=0
a11 a22
∴ a12 = 0
Conversely, it can be easily shown.
Example 9.4.2. If θ is the angle between the parametric curves, show that
a12 a
cosθ = and sinθ = .
a11a22 a11a22
1 a12
Solution: We know cosθ = aαβ λ α µ β = aαβ δ 1α δ 2β =
a11a22 a11a22
2 2 a 212 a11a22 − a 212 a

sin θ = 1 − cos θ = 1 − = =
a11a22 a11a22 a11a22
∴ sinθ = a .
a11a22
9.4.1 Pictorial Interpretation

Here, r denotes the position vector of any point P on the surface and if bi
are the unit vectors directed along the orthogonal coordinate axes Y, then
the equation yi = yi(u1,u2) of the surface S can be written in vector form, as
r(u1,u2) = biyi(u1,u2)
We can write
∂r ∂r
ds 2 = dr .dr = duα duα
∂uα ∂uα
=aαβ duα duα,

∂r ∂r
where aαβ = .
∂uα ∂uα
Y3
∂r
∂u2 u2
P
∂r
∂u1
u1 S
r
b3
O Y2
b2
b1
Y1
Figure 9.3
∂r
Putting = aα where a1 and a2 are tangent vectors to the coordinate
∂uα
curves, we get a11 = a1.a1,a12 = a1.a2,a22 = a2.a2.
The space components of a1 and a2 are ξ i and η i , respectively
We can define an element area dσ of the surface S by the formula
d σ = a1 × a2 du1du 2
= a11a11 − a12 2 du1du 2 = adu1du 2 .
In a two-dimensional manifold, the skew-symmetric e-systems can be

defined by the formulas
e11 e22 e11 e 22 0 and e12 e 21 e12 e21 1

and since these systems are relative tensors, the expression

1
 e and  ae are absolute tensors. Using  − symbols, let
a
θ be the angle between the unit vectors λ α and µ α , then
2
i i i i i i
Si n2 1 cos 2
2 2 2
ξ 1 η1 ξ 1 η1 ξ 2 η 2
= + +
ξ 2 η 2 ξ 3 η 3 ξ 3 η 3

2
λ 1 µ1
= ( J12 + J 22 + J 32 ) = a ( λ 1 µ 2 − λ 2 µ1 )
2
2 2
λ µ

sinθ = a ( λ 1 µ 2 − λ 2 µ1 ) = aeαβ λ α µ β ,

which is numerically equal to the area of the parallelogram constructed by

unit vectors λ α and µ α .
It follows from the result that a necessary and sufficient condition for
the orthogonality of two surface unit vectors λ α and µ α is αβ λ α µ β = 1.
Example 9.4.3. Prove that the parametric curves on a surface given by

x1 = a sin u cosv, x2 = a sin u sinv, and x3 = acos u form an orthogonal system.
Solution: Here, the symmetric covariant tensors ααβ are
2 2 2
 ∂x1   ∂x 2   ∂x 3 
a11 =  + + = (a cos u cosv )2 + (a cos u sinv )2
 ∂u   ∂u   ∂u 
+ (−a sin u)2

2
= a
2 2 2
 ∂x1   ∂x 2   ∂x 3 
a22 =  + + = (−a sin u sinv )2 + (a sin u cosv )2
 ∂v   ∂v   ∂v 
= a 2 sin 2u
∂x1 ∂x1 ∂x 2 ∂x 2 ∂x 3 ∂x 3
a12 = + +
∂u ∂v ∂u ∂v ∂u ∂v
= (a cos u cosv)(−a sin u sinv) + (a cos u sinv)(a sin u cosv) = 0.
Here, a12 = 0, implying that the coordinate curves are orthogonal.

Let λα and μβ be taken along the parametric curves.
For u−curve, dv = 0 and for v− curve, du = 0.
α
Thus, unit vector λ1 along the u−curve is
 du dv   1  1 
λ1α =  ,  = ,0 = ,0
 ds1 ds2   a11   a 

 du dv   1   1 
λ2β =  , = 0, , = 0,
 ds2 ds2   a22   asinu 

∴ cosθ = aαβ λ1α λ2β = 0

⇒ cosθ = 0
π
⇒θ= .
2
Thus, the given parametric curves on a surface form an orthogonal

system.
9.5 Geodesic in Rn
We know that if a Riemannian Space Vn is in a Euclidean Space, En, a coor-
dinate system exists in which the components gij(i,j = 1,2,..n) of the metric
tensor are constant throughout the space. In this case, all the Christoffel
symbols vanish.
If Vn is not Euclidean, then the Christoffel symbols do not vanish at all
points of Vn, but it is possible to find a coordinate system. In fact infinitely
many Christoffel symbols are at a given point P of Vn. Such coordinates are
geodesic coordinates for that particular point, P. Point P is called the pole or
origin of the geodesic coordinate system.
We are now in a position to discuss the problem of finding curves of
minimum length joining a pair of given points on the surface. Such a curve
is called a geodesic through P and Q. We are going to deduce the equation
of geodesic for the case of the n-dimensional Riemannian manifolds.
Let the metric properties of the n−dimensional manifold Rn be deter-
mined by
ds2 = gij dxi dxj , (i,j = 1,2,3) (9.19)
The length of a curve C, is represented in Rn by equations
C : x i = x i (t ) , t1 ≤ t ≤ t 2 ,

given by
t2

s=
∫t1
g αβ x α x β dt, (α , β = 1,…n )

(9.20)
The extremals of the function of (9.20) are geodesics in Rn.

Let
F g x x (9.21)

and to form Euler’s equations, we compute Fx i and Fx i .
−1
1 ∂ g αβ α β
Fx j =
2
( g αβ x α x β ) 2
∂x j
x x ,

−1
Fx j = ( g αβ x α x β ) 2 g αj x α

Substituting these values in Euler’s equations gives

∂ g αβ α β
 
j x x
d g α j x α  ∂ x
− =0 (9.22)
dt  ( gαβ x α x β )  2 ( gαβ x α x β )

ds
Since = g αβ x α x β ,
dt
Equation (9.22) can be written in the form
∂ g αβ α β
d  g α j x α  ∂ x j x x
dt  ds − ds
= 0,
 2
 dt  dt
g j x d 2 s
g j ,
1 g dt 2
or g j x x x x x
x 2 xj ds
dt
1 g j g j g g j x d 2 s/dt 2
or g j x x x ,
2 x x xj ds
dt
 d 2 s ds 
g α j xα + [αβ , j ] x α x β = g α j x α  2 /  (9.23)
 dt dt 

These are the desired equations of geodesics.

If we chose parameter t to be the arc length s of the curve, we get
ds
= g αβ x α x β = 1.
dt
d 2s
Then, = 0 and from (9.23),
dt 2
g α j xα + [αβ , j ] x α x β = 0. (9.24)

Here, the symbol ‘ . ’ denotes the differentiation w.r.t arc parameter s.

If we multiply by tensor gij and sum,
g ij g α j xα + g ij [αβ , j ] x α x β = 0

i
i
or x x x 0

 i  α β
or xi +  x x = 0 (i = 1,2,…n )
αβ  (9.25)
 
(α,β=1,2,…n),
which is a simple form of geodesics in Rn.

We observe that these above equations are the same with equations of a
straight line in E3.
If the manifold Rn is Euclidean and a coordinate system exists in which
the Christoffel symbols vanish, then in (9.25), the 2nd part of it vanishes.
Therefore,
d 2xi
xi = = 0.
ds 2
The general solution of the equation is
x i = Ai s + B i ,
that is, the geodesic in En are straight lines.

If we assume a given surface S as a Riemannian two-dimensional man-
ifold R2, covered by Gaussian coordinates uα, then (9.25) takes the form
d 2u du du
0 , ,, 1, 2 . (9.26)
ds 2 ds ds

Hence, at each point of S there exists a unique geodesic with arbitrary

α duα
prescribed direction λ = . Thus, if there exists a unique solution to
ds
u (s) passing through two given points on S, then the curve of uα(s) is the
α
curve of shortest length joining these points.
Example 9.5.1. Find the differential equation for the geodesics on an arbi-
trary cylinder immersed in E3
Solution: We choose the Y3− axis parallel to the generators of the cylinder
and let the trace of the generators of the cylinder on aY1Y2− plane be given
as (Figure (9.4)).
C : y1 = ϕ(σ),
y 2 = ψ (σ ) ,

where 𝛔 is the arc length of C.
Y3
ds dy3
O Y2
dσ
C
Y1
Figure 9.4
Since (dσ )2 = (dy 1 )2 + (dy 2 )2 and the element of arc (ds) of the geodesic
is
(ds)2 = (dσ)2 + (dy3)2,
we get a11 = 1.a22 = 1, a12 = a21 = 0.

d 2uγ  γ  duα du β
2 +
We know,  =0
ds α β ds ds
 
for γ = 1, uγ = u1 = σ , and the second part of above equation is zero.
d 2σ
=0
ds 2
and γ = 2, u2 = y3
d2 y3
= 0.
ds 2
We get σ = As + B
y3 = A1s + B1.
If A≠0, we can write y3=C1σ+C2, where C1 and C2, are arbitrary constants.
Therefore, the equations of the geodesics are
y1 = ϕ(σ),
y2 = ψ(σ)
y3 = C1σ + C2.
Hence, the curve is a helix whose pitch is determined by C1 and C2 deter-

mining the origin for the arc parameter σ.
Example 9.5.2. Find the differential equation for geodesic in spherical

coordinates:
y1 = acosu1cosu2
y2 = acosu1sinu2
y3 = asinu1.
Solution: Here, (ds )2 = a 2 (du1 )2 + a 2 (cosu1 )2 (du 2 )2
a11 = a 2 , a22 = a 2 (cosu1 )2 , a12 = a21 = 0.
a11 a12 a2 0
a= = = a 4 (cosu1 )2,
a21 a22 2
0 a (cosu ) 1 2
a 2 (cosu1 )2 1 22 a2 1
a11 = 4 1 2 = 2 , a = 1 2 = 2 , a12 = a 21 = 0,
a (cosu ) a a (cosu ) a (cosu1 )2
4

u11
and s =
∫ u10
(1 + (cos 2u1 (u 2 )2 du1 .
d 2u du du
We know, 0.
ds 2 ds ds
For γ = 1,
d 2u1 1 du2 du2 1 du1 du2 1 du1 du1
0
ds 2 2 2 ds ds 1 2 ds ds 1 1 ds ds
Here,
1 1 1 a22 1
a11 22,1 2a 2cosu1sinu1 cosu1sinu1
22 a2 2 u1 2a 2

1 1
a11 12,1 0, a11 11,1 0
12 11

d 2u1 1 du2 du2 1 du1 du2 1 du1 du1

0,

2
d 2u1  du 2 
or 2 + ( −cosu1sinu1 )  + 0 + 0 = 0,
ds  ds 

2
2 1
 2
or d u2 − cosu1sinu1  du  = 0. (i)
ds  ds 

For γ=2,
d 2u 2 2 du2 du2 2 du1 du2 2 du1 du1
0
2 2 1 1
0, a 22 12, 2 . 2a 2 cosu1sinu1 tanu1
22 12 2 1 2
a (cosu ) 2

2
a22 11, 2 0
11

d 2u 2 2 du2 du2 2 du1 du2

0 0
ds 2 2 2 ds ds 1 2 ds ds

d 2u 2 du1 du 2
or 2
− tanu1 =0 (ii)
ds ds ds
(i) and (ii) are required equations of the geodesic for the surface.
Example 9.5.3. In orthogonal Cartesian frame Y, if curve C is to lie on the
sphere of radius a show that the geodesic are great circles.
Solution: We consider the function
t2

J=
∫ t1
x i x i dt (i = 1,2,3)

(i)
If C is to lie on a sphere of radius a, the constraining relations are in the

form
φ = x i x i − a 2 = 0, (ii)
when the center of the sphere is at origin, the function
G = x i x i − λ (t )( x i x i − a 2 ) (iii)
Now, the differential equation for the extremal is
dGx i
− Gx i = 0
dt
d dx i
or 2 t xi 0 i 1, 2, 3 , (iv)
dt ds
dx i
where ds = x i x i , Gx i = and Gx i = −2λ (t ) x i .
ds
Eliminating λ(t) from two sets of equations of (iv), we get
 dx 1  1  dx 
2
x 2d  − x d  =0
 ds  
ds 

dx1 dx 2
or d x2 x1 0
ds ds
 dx 1  2
1  dx 
x2  − x  = constant = C1 . (v)
 ds  
ds 
Similarly, eliminating of λ(t) from last two sets of equations of (iv), we

get
 dx 1  3
1  dx 
x3  − x  = constant = C2 . (vi)
 ds  
ds 

From (v) and (vi), we get
x 3dx 1 − x 1dx 3 x 2dx 1 − x 1dx 2

C1 = C 2
( x 1 )2 ( x 1 )2
x3 x2
or C1d C2 d .
x1 x1
Integrating both sides, we get
 x3   x2 
C1  1  + C3 = C2  1 
x  x 

or C3 x1 C2 x 2 C1x 3 0, (vii)
which is the plane passing through the origin.

Equation (vii) with equation ii shows that the geodesic on the surface of
a sphere are an arc of great circle.
Example 9.5.4. Find the geodesic on the surface
y1 = u1cosu2
y2 = u1sinu2
y3 = 0
imbedded in E3. The coordinate yi are orthogonal Cartesian.

Solution: Here, ds 2 = (du1 )2 + (u1 )2 (du 2 )2
a11 = 1, a22 = (u1 )2 , a12 = a21 = 0 ,

so that a = (u1)2 and
1
a11 = 1, a 22 = , a12 = a 21 = 0 and
(u1 )2
1 1 a22 1 1
a11 22,1 1. 2u u1
22 2 u1 2

2 2 1 1 a22 1 1 1 1
a 22 12, 2 2u ,
12 21 (u1 )2 2 u1 (u1 )2 2 u1

all others
i
0
jk

The differential equations of a geodesic for the surface are
d 2u du du
0.
ds 2 ds ds

d 2u1 1 du du
For γ = 1 0,
ds 2 ds ds
2 1 1 du2 du2
or d u2 0 ,
ds 2 2 ds ds
2 1 2 2
or d u − u1  du  = 0. (i)
2  ds 
ds
d 2u 2 2 du du
For γ = 2 0,
ds 2 ds ds
d 2u 2 1 du1 du2
or 2 0,
ds 2 1 2 ds ds

2 2 1 2
or d u2 + 2 11 du du = 0 (ii)
ds u ds ds
(i) and (ii) are the differential equations of geodesic on the surfaces.
Example 9.5.5. Find the differential equations of the geodesic for the
metric
ds 2 = (dx 1 )2 + {( x 2 )2 –( x 1 )2 }(dx 2 )2 .
2 2 1 2
Solution: a11 = 1, a22 = ( x ) – ( x ) a12 = a21 = 0,
a = ( x 2 )2 – ( x 1 )2

1
a11 = 1, a 22 = , a12 = a 21 = 0.
( x ) – ( x 1 )2
2 2

∂a22 ∂a
Thus, = −2 x 1 , 222 = 2 x 2 .
∂x1 ∂x
The non-zero Christoffel symbols are
1 1 a22 1
a11 22,1 1. 2 x1 x1
22 2 x1 2

2 2 1 22 1 g 22
a 21, 2 2 2
12 21 2 2{(x ) (x ) } x1
1 2
2 x1 x1
2{(x 2 )2 (x1 )2 (x 2 )2 (x1 )2

2 1 a22 x2
a22 22, 2
22 2{(x 2 )2 (x ) } x 2
1 2
(x 2 )2 (x1 )2 .

Hence, the desired geodesic equations are
i
xi x x 0 are:

for γ = 1
2
d 2 x1 1 dx 2
0
ds 2 22 ds

2
d 2 x1 1  dx 
2
+ x  =0
ds 2  ds  (i)

and for γ = 2
2
d2x2 2 dx 2 2 dx1 dx 2
2 0
ds 2 22 ds 1 2 ds ds

2
d 2x 2 x2  dx 2  −2 x 1 dx 1 dx 2
+   + = 0 (ii)
ds 2 ( x 2 )2 –( x 1 )2  ds  ( x 2 )2 –( x 1 )2 ds ds
(i) and (ii) are required equations of geodesic.
9.6 Geodesic Coordinates

If Riemannian space Rn is Euclidean, a coordinate system exists in
which the components gij of metric tensors are constants through-
∂ g ij
out the space, which implies that = 0. Consequently, the vanish-
∂ g ij ∂x k
ing of is equivalent to the vanishing of all Christoffel symbols,
∂x k
1  ∂g ∂ g jk ∂ g ij 
i.e. [ij , k ] =  ikj + i − k  = 0. If Rn is not Euclidean, then the
2  ∂x ∂x ∂x 
Christoffel symbols do not vanish at all points in Rn, but it is possible to
find infinitely many coordinate systems in which they vanish at any given
point of Rn. Such coordinates systems are called geodesic coordinate system.
Let us consider a surface net S whose curvilinear coordinates are and
also consider a point P (u01 , u02 ) on S. Let vα be coordinates of some other
net on S.
Then,
uα = uα ( v1 , v 2 ) (α = 1,2 ). (9.27)

The second derivative formula of the transformation of Christoffel

symbols is
∂ 2 uα  α  ∂ u β ∂ uγ  ν  ∂ uα

+  β γ  λ = λ µ  v . (9.28)
∂v λ ∂v µ   ∂v ∂v
µ
  ∂v
u  v 
If there exists a transformation of coordinates of (9.27) for which

 v 
  vanishes at P, then for that particular point,.
 λ µ 
∂2 uα  α  ∂ u β ∂ uγ
 
λ µ
+   λ µ =0 (9.29)
∂v ∂v  β γ  ∂v ∂v
u

We exhibit the next solution of Equation (9.29), yielding a transforma-

tion of (9.27) to a coordinate system vα in which Christoffel symbols vanish
at P.
Let us take a second degree polynomial
1  α  λ µ
uα = uαP + v α −   v v , (9.30)
2  λ µ 
P
 α 
α
where uP is the value of uα at P and  λ µ  are the values of the
Christoffel symbols at P.   P
For verification, if Equation (9.30) satisfies (9.29), we get
∂ uα α
 α  λ
µ =δµ −  λ µ  v (9.31)
∂v   P

and
∂ 2 uα  α 
λ µ = − λ µ 
. (9.32)
∂v ∂v 
 

P
From this computation we conclude that the coordinates are geodesic at

all points of an arbitrarily prescribed analytical curve.
9.7 Parallel Vectors on a Surface

The concept of parallel vector fields along a curve C imbedded in E3
was generalized by Levi-Ciita to curves imbedded in n− dimensional
Riemannian manifolds.
For the usefulness of this concept we consider:
A surface S imbedded in E3 and a curve C on S.
We take equations of C in the form:
C : uα = uα (t ) , t1 ≤ t ≤ t 2
and suppose that the metric properties of S are governed by the tensor aαβ.
If Aα is a surface vector field defined along C, we can calculate the sur-
face intrinsic derivative
A dA du . (9.33)
A
t dt dt

It is identical with parallel vector fields along a space curve.

Accordingly, we take the differential equation
dA du
A 0 (9.34)
dt dt

If parameter t is chosen as the arc length, s, Equation (9.34) can be

written as
dA du
A 0 (9.35)
ds ds

duα
and if Aα is taken to be a unit tangent vector to C so that Aα = ≡ λα ,
with aαβλαλβ = 1, then ds
d 2u du du
0. (9.36)
ds 2 ds ds

This is an equation of geodesic on S.
9.8 Isometric Surface

The properties of surface (i.e., the lengths of curves and angle between
intersecting curves) with which we have been concerned, depend entirely
on the study of the first fundamental quadratic form
ds2 = aαβduαduβ. (9.37)
These properties constitute a body of what is known as the Intrinsic

geometry of surfaces.
We have seen that the intrinsic property of a surface depends on the
metric tensor of the surface and its derivatives. For example, two surfaces,
a cylinder and a cone, appear to be entirely different when viewed from
the enveloping space and yet their intrinsic geometries are completely
indistinguishable since the metric properties of cylinders and cones can
be described by identical expressions for the square of the elements of
arc.
If a coordinate system exists on each of the two surfaces such that the
elements on them are characterized by the same metric coefficients, aαβ, the
surfaces are called isometric.
The surfaces of the cylinder and cone are isometric with the Euclidean
plane since these surfaces can be rolled out or developed on the plane with-
out changing the length of arc elements and the measurements of angles
and areas.
9.8.1 Developable
A surface which is isometric with a Euclidean plane E2 is called a
developable.
We introduce an important scalar invariant, known as the Gaussian
Curvature, which will enable us to determine the circumstances under
which a given surface is developable, that is, isometric with a Euclidean
plane.
A surface S is called a developable surface or simply a developable if the
Gaussian Curvature vanishes at every point of it.
Example 9.8.1. Consider Catenoid
S1:y1 = v1cosv2
y2 = v1sinv2
v1
y 3 = acosh −1
a
 y3 
obtained by revolving the catenary y 2 = cosh.   about the y3-axis.
 a
Show that surface S1 is isometric with the surface of Helicoid, defined by
S2: y1 = u1cosu2
y2 = u1sinu2
y3 = au2.
Solution: For S1: The 1st fundamental form

ds2 = dyidyj is easily found to be
(v1 )2
ds 2 = aαβ dv α dv β = 1 2
1 2 1 2 2 2
2 (dv ) + (v ) (dv ) (i)
(v ) − a
a2
[ds 2 (dy1 )2 (dy 2 )2 (dy 3 )2 (dv1 )2 (v1 )2 (dv 2 )2 (dv1 )2
(v1 )2 a2
(v1 )2 1 2 1 2 2 2 1 y3
(dv ) (v ) (dv ) and v acosh ,
(v1 )2 a2 a
 y3  1 1 2  y 
3
2 y
3
∴ dv1 = a sinh   dy 3 , dy 3 = dv 1
,sin h   = cos h −1
 a a  y3   a a
sinh  
 a
(v1 )2 (v1 )2 − a 2
= 2 −1 = ],
a a2
(v1 )2 1 2 2 ( 1 )2 − a 2  , a12 = a21 = 0.
so that a11 = 1 2 2 , a22 = (v ) ≡ a +  v
(v ) − a
For surface S2
αβ dv dv = ( du ) + (u ) + a  ( du )
2 α β 1 2  1 2 2 2 2
ds = a (ii)

a11 = 1,a12 = 0, a22 = (u1)2+a2
 (u1 )2 2
Now, if we set in (i), (v1)2 − a2 = (u1)2,v2 = u2 and (dv1 )2 = 1 2 ( du1 )  ,
we obtain, (i) to become (ii).  (v ) 
Therefore, ds 2 = ( du1 ) + (u1 ) + a 2  ( du 2 ) .
2 2 2
Since this is identical with (ii), the surfaces of S1 and S2 are isometric.
9.9 The Riemannian–Christoffel Tensor

and Gaussian Curvature
With the first fundamental quadratic form of the surface
ds2 = aαβduαduβ, (9.10)
we can form the Christoffel symbols with respect to this surface and the
corresponding Riemann tensor
∂ ∂  λ   λ 
γ    
Rαβγδ = ∂x ∂x δ + β γ β δ .
   
[ βγ ,α ][ βδ ,α ] [αγ , λ ] [αδ , λ ]

This tensor is skew-symmetric in the first two and last two indices, so
that, for the surface, S
Rααβγ = Rαβγγ = 0, R1212 = R2121 = − R2112 = − R1221 . (9.38)

Hence, all non-vanishing components of Riemann-Christoffel tensors

are equal to R1212 or its negative.
We define the quantity κ
R1212
κ= , where a = aαβ (9.39)
a
and it is called Gaussian Curvature or the total Curvature of the surface S.

If we introduce the two dimensional ϵ− tensors,
1 αβ
αβ = a eαβ  αβ = e
a
we can write (9.39) using (9.38) as
R1212 = aκ = a e12 a e12κ = 1212κ
and generalize it as
Rαβγδ = αβ γδ κ . (9.40)

Since  αβ αβ = 2,
1
[   1111  2222  1212  2121 0 0 a 1. 1
a
1
a 1. 1 2]
a
we can solve (9.40) for κ and obtain
 αβ  γδ Rαβγδ =  αβ  γδ καβ γδ = 4κ

1
κ = Rαβγδ  αβ  γδ . (9.41)
4
These equations show that Gaussian Curvature is an invariant.
Theorem 9.9.1. A necessary and sufficient condition that a surface S is iso-

metric with a Euclidean plane is that the Riemannian tensor (or Gaussian
curvature of S) be identically zero.
Proof: When a surface S is isometric with the Euclidean plane, there exists
on S a coordinate system with respect to which a11 = a22 = 1,a12 = 0. It is
obvious that in this case Rαβγδ = 0 in this particular coordinate system and
since Rαβγδ is a tensor, it must vanish in every coordinate system.
Conversely, if the Riemannian tensor vanishes at all points of the sur-
face, Theorem of §5.6.1 states that there exists a coordinate system on the
surface such that a11 = a22 = 1,a12 = 0.
9.9.1 Einstein Curvature

We consider an invariant
R = a µν, Rµν, (9.42)

where
α
Rµν = Rµνα = a λα Rλµνα (9.43)

are the Ricci tensors (introduced in § 5.4).

If we multiply (9.43) by a µν
R = a µν Rµν = a µν a λα Rλµνα (9.44)

Using (9.38), (9.44) is equivalent to
R = −2 R1212 ( a11a 22 − a12a12 ) .

a22 22 a11 12 a
Since a11 = ,a = , a = − 12 ,
a a a
a R1212
R = −2 R1212 2 = −2 (9.45)
a a
R1212
Comparing this with κ = , we get
a
R
κ=−
2
R = −2.κ
The invariant R is called the Einstein Curvature of S.

A surface which holds
Rαβγδ = ρ ( aαδ aβγ − aαγ aβδ ) ,

where ρ is scalar, is called a surface of constant curvature.

Geometrical properties which are expressible in terms of the first fun-
damental form may be called intrinsic properties. Since only metric coeffi-
cients aαβ are involved in this definition, the properties described by κ are
intrinsic properties of the surface, S.
Example 9.9.1. If the coordinate system is orthogonal, show that
1  ∂  1 ∂a22  ∂  1 ∂a11  
κ=− 1 1  +  
2 a  ∂u  a ∂u  ∂u 2  a ∂u 2  

Solution: If the system is orthogonal,
a11 0
a12 = a21 = 0 ∴ a = = a11a22
0 a22

1 1
a11 = , a 22 = .
a11 a22
We know
∂ ∂  λ   λ 
γ    
Rαβγδ = ∂x ∂x δ + β γ   β δ
 
[ βγ,α ][ βδ,α ] [αγ , λ ] [αδ , λ ]

∂ ∂  λ   λ 
   
R1212 = ∂u1 ∂u 2 +  2 1   2 2 
[21,1][22,1] [11, λ ] [12, λ ]

1 ∂a11 1 ∂a11 , 1 ∂a11 1 ∂a22

[11,1] = 1 , [12,1] = 2 [11,2 ] = − 2 , [12,2 ] =
,
2 ∂u 2 ∂u 2 ∂u 2 ∂u1
1 ∂a11 1 ∂a22 1 ∂a22

[ 21,1] = 2 , [ 22,1] = − 1 , [ 22,2 ] =
2 ∂u 2 ∂u 2 ∂u 2
11 1 a11 2 22 1 a22
: a 21, 1 2
and a 21, 2 1
21 21 2a11 u 21 2a22 u

1 11 1 a22 2 22 1 a22
: a 22, 1 and a 22, 2
22 22 2a11 u
1
2 2 2a22 u
2

∂ ∂  λ   λ 
R1212 = 1
[ 22,1] − [ 21,1] + 
2 [12, λ ] −  [11, λ ]
∂u ∂u  2 1   2 2 

1 2
22,1 21,1 12,1 12, 2
u1 u2 21 21
1 2
11,1 11, 2
22 22
1 ∂ ∂a22 1 ∂ ∂a11 1 ∂a11 1 ∂a11 1 ∂a22 1 ∂a22

=− 1 1 − 2 2 + 2 2 +
2 ∂u ∂u 2 ∂u ∂u 2a11 ∂u 2 ∂u 2a22 ∂u1 2 ∂u1
1 ∂a22 1 ∂a11 1 ∂a22 1 ∂a11
+ 1 +
2a11 ∂u 2 ∂u 2a22 ∂u 2 2 ∂u 2
1
2
1  ∂ 2 a22 ∂ 2 a11  1  ∂a11  ∂a22 ∂a11 
=−  1 2 + +   + 1 
2  (∂u ) (∂u 2 )2  4a11  ∂u 2  ∂u ∂u1 
2
1  ∂a22  ∂a22 ∂a11 
+   + 2 
4a22  ∂u1  ∂u ∂u 2 
1  ∂  1 ∂a22  ∂  1 ∂a11  
=− a11a22  1  1  + 2 2 
2  ∂u  a11a22 ∂u  ∂u  a11a22 ∂u  
1  ∂  1 ∂a22  ∂  1 ∂a11  
=− 1 + 2   [since a=a11 a22]
a 1 2 
2  ∂u  a ∂u  ∂u  a ∂u  
 1 1  ∂  1 ∂a22  ∂  1 ∂a11  
= a −   + 2 2  .
 2 a  ∂u  a ∂u  ∂u  a ∂u  
1 1

Therefore, the Gaussian Curvature is

R1212 1 1 1 a22 1 a11

1 1 2 2 .
a 2 a u a u u a u

Example 9.9.2. Calculate the total curvature (Gaussian) of the manifold
whose quadratic form is
ds2 = a2 sin2 u1(du2)2 + a2(du1)2.
Solution: Here, a11 = a2,a22 = a2sin2u1,a12 = a21 = 0
a11 0
a= = a11a22 = a 4 sin 2u1
0 a22

a 2 sin 2u1 1 1 1
a11 = 4
22
2 1 = 2 , a = = 2 2 1
a sin u a a22 a sin u
α 1 2 1
R1212 = a1α R212 = a11 R212 + a12 R212 = a11 R212 .
We know
i i
x k
x l
k l
Rijkl i i
jk jl jk jl

1
1 1
R1212 a11R212 a11 1 2
x 22 x 21
1 1 1
2 22 22 2 21

Here,
1 1 1 a22 1 2 1
a11 22,1 a 2sinu1cosu1 sin2u1
22 a2 2 u1 2a 2
2
1 1 1 a11
a11 21,1 0,
21 a 2 2 u2

2 1 1 a22 1 1 2
a22 21, 2 a 2sinu1cosu1
21 a sin u 2 u1
2 2 1 2 2 1
a sin u 2
cotu1
2
0
22

1 1 1 1
R1212 a11 sin2u
1
1
u 2 1 2 2 2 1
1 1 1 1 1
a2 1
sin2u1
u 2 11 22 12 21
1 2 1 2
21 22 22 21

1 1
a2 2cos2u1 0 0 0 sin2u1cotu1
2 2
cosu1
a2 cos2u1 sinu1cosu1
sinu1
2 2 1 2 1 2 1
a sin u cos u cos u a2 sin2u1
R1212 a 2 sin 2u1 1

κ= = = .
a 4 sin 2u1 a 4 sin 2u1 a 2
Example 9.9.3. Show a surface of revolution defined by
y1 = u1cosu2.
y2 = u1sinu2
y3 = f(u1)
f ′f ′′
κ= , when f is class C2.
u [1 + (f ′ )
1 2 2
]
Solution:
2
2 1 2 2 2 3 2 1 2 1 2 2 2 f
ds (dy ) (dy ) (dy ) (du ) (u ) (du ) (du1 )2
u1
2
f
1 (du1 )2 (u1 )2 (du2 )
u1

2
 ∂f 
a11 = 1 +  1  = 1 + ( f ′ )2 , a22 = (u1 )2 , a12 = a21 = 0
 ∂u 

a11 0 ∂f
= a11a22 = (u1 ) 1 + ( f ′ )  ,where f ′ = 1
2 2
a=
0 a22 ∂u

a22 1 1 1 1
a11 = = = 22
2 , a = = 1 2 , a12 = a 21 = 0,
a11a22 a11 1 + (f ′ ) a22 (u )
1 1
1 2
1
u 2 2 u 21
R1212 a11R212 a11
1 1
1 2 2 2 1

1 1 1 1
a11 1 2
u 2 2 u 21 11 2 2
1 1 1 2 1 2
12 21 21 2 2 2 2 21

1 1 a11 1 1 ff
a11 11,1 a11 . 2ff
11 2 u1 2 1 ( f )2 1 ( f )2

1 1 a22 u1
a11 22,1 a11 ,
2 2 2 u1 1 ( f )2
1 1 a11
a11 21,1 a11 0
21 2 u2
2 1 a22 1 1 1
a22 21, 2 a22 2 u1 = 1
21 2 u1 1 2
(u ) 2 u
1 1 1 1 1 1
R1212 a11
u1 2 2 u2 21 11 22 12 21
1 2 1 2
21 22 22 21

 
]  f ′f ′′ 1
1 1 1
∂ u u u
[
= 1 + ( f ′)
2
1 
− 2 
− 0 + − 0 + 0 +
 ∂u  1 + ( f ′ )  1 + ( f ′) 1 + ( f ′) 1 + ( f ′ ) u 
2 2 2 1
2  1  f ′f ′′u1 1 
= 1 + ( f ′ )   − 2 + 2 + 2
 1 + ( f ′ )  1 + ( f ′ )  1 + ( f ′ ) 
2
f ′f ′′u1 f ′f ′′u1
= 1 + ( f ′ ) 
2
=
2
1 + ( f ′ ) 
2
1 + ( f ′ )2 

f ′f ′′u1
R1212 1 + ( f ′ )2  f ′f ′′
κ= = 1 2 = 2.
(u ) 1 + ( f ′ )  u1 1 + ( f ′ )2 
2
a
 
Example 9.9.4. Also show that surfaces of the above problem (Example
9.8.1) are non-developable.
Solution: Now we calculate the Gaussian Curvature of S1
(v1 )2
a11 , a22 (v1 )2 and a12 = a21 = 0
(v1 )2 a2
a11 0 (v1 )4

a= = a11a22 =
0 a22 (v1 )2 − a 2

1 (v1 )2 − a 2 22 1 1
a11 = = 1 2 , a = = 1 2
a11 (v ) a22 (v )
1 1 1 1
R1212 a11
v1 22 v2 21 11 22
1 1 1 2 1 2
12 21 21 22 22 21
so
1 (v1 )2 a2 1 a11 1 (v1 )2 a2

a11 11.1 .
11 (v1 )2 2 v1 2 (v1 )2
(v1 )2 2v1 2v1 (v1 )2 a2 (v1 )2 a2 v1a2 a2
2
.
(v1 )2 a2 (v1 )2 (v1 )2 a2
2
v1 (v1 )2 a2
1 1 a22 1 (v1 )2 a2 1
a11 22,1 a11 2v
22 2 v1 2 (v1 )2
(v1 )2 a2 1
, 0
v1 21
2 1 a22 1 1 1
a22 21, 2 a22 2v1
21 2 v1 2 (v1 )2 v1

(v1 )2 (v1 )2 a2 a2 (v1 )2 a2

R1212 0
(v1 )2 a2 (v1 )2 v1 v1
2
a2 v1
(v1 )2 a2 1
0 0
v1 v1
(v1 )2 (v1 )2 a2 a2 (v1 )2 a2

(v1 )2 a2 (v1 )2 (v1 )2 (v1 )2
(v1 )2 2(v1 )2 a2 2(v1 )2 a2
(v1 )2 a2 (v1 )2 v1 )2 a2
2(v1 )2 − a 2
R (v1 )2 − a 2 2(v1 )2 − a 2
Start here κ = 1212 = = ≠ 0.
a (v1 )4 (v1 )4
(v1 )2 − a 2
Thus, S1 is not developable.

For S2 surface: a11 = 1,a12 = a21 = 0,a22 = (u1)2 + a2 and a = (u1)2+a2
1 1 1
a11 = = 1, a 22 = = 1 2 2
a11 a22 (u ) + a
1 1 1 1 1 1
R1212 a11 1 2
v 22 v 21 11 22 12 21
1 2 1 2
21 22 22 21
1 1 a22 1
a11 22,1 a11 1. 2u1 u1
22 2 u1 2

2 22 1 a22
22 1 1 1 u1
a 21, 2 a 2u
21 2 u1 (u1 )2 2
a 2 (u1 )2 a2

 u1  (u1 )2 −a 2
R1212 =  −1 − 0 + 0 − 0 + 0 + u1 1 2 2  = 1 2 2 − 1 = 1 2 2
 (u ) + a  (u ) + a (u ) + a
−a 2
R (u1 )2 + a 2 −a 2
κ = 1212 = 1 2 2 = 2 ≠ 0 , implying that the surface S2 is
a (u ) + a [(u1 )2 + a2 ]
not developable.
Example 9.9.5. Determine whether the surface with the metric
ds2 = (u2)2(du1)2 + (u1)2 du2)2
is developable or not
Solution: Here, a11 = (u2)2,a22 = (u1)2,a12 = a21 = 0,a = (u1)2(u2)2.
We know
1 1  ∂  1 ∂a22  ∂  1 ∂a11  
κ=−  1 1  +  
2 a  ∂u  a ∂u  ∂u 2  a ∂u 2  

1 1  ∂  1 ∂(u1 )2  ∂  1 ∂(u 2 )2  
=−    +  
2 u1u 2  ∂u1  u1u 2 ∂u1  ∂u 2  u1u 2 ∂u 2  
1 1  ∂  1 1 ∂  1 2 
=− 1 2  1  1 2 2u  + 2  1 2 2u  
2 u u  ∂u u u ∂u u u 
1 1  ∂  2 ∂  2 
=− 1 2  1 2 +
2 u u  ∂u u ∂u 2  u1  
1 1
=− [ 0 + 0 ] = 0.
2 u1u 2
Since the Gaussian Curvature 0, the surface is developable.

Example 9.9.6. Show that the surface is defined by
y1 = f1(u1)
y2 = f2(u1)
y3 = u2,
where f1 and f2 are differentiable functions that are developable.

Solution:
 ∂ f1  2  ∂ f 2  2  1 2
ds = (dy ) + (dy ) + (dy ) =  1  +  1   (du ) + (du 2 )2
2 1 2 2 2 3 2
 ∂u   ∂u  
2 2
 ∂f  ∂f 
a11 =  11  +  21  = ( f 2′)2 + ( f 2′)2 , a12 = a21 = 0, a22 = 1
 ∂u   ∂u 
∂ f1 ∂f ∂f
and a = ( f1′)2 + ( f 2′)2 , where = f1′, 11 = f1′, 21 = f 2′
∂u1 ∂u ∂u
1 1
a11 = 2
22
2 , a = =1
( f1′) + ( f 2′) a22

1 1 1 1
R1212 a11 1 2
u 22 u 21 11 22
1 1 1 2 1 2
12 21 21 22 22 21
2 2 1 1 1 1
f1’ f 2’ 1 2
u 22 u 21 11 22
1 1 1 2 1 2
12 21 21 22 22 21
Here,
1 1 1 a11 1 1
a11 11,1 2 f1 f1 2 f 2 f 2 ,
11 ( f1 ) 2
( f 2 ) 2 u1
2 2
2 ( f1 ) ( f 2’ )2
i
all 0
jk
,
R1212 = ( f1′) + ( f 2′) [ 0 − 0 + 0 − 0 + 0 − 0 ] = 0

2 2

R1212
κ= = 0.
a
Therefore, the surface is developable.
9.10 The Geodesic Curvature

We study the intrinsic geometry of surfaces with a derivation of a formula
describing the behavior of the tangent vector to a surface curve, which is
analogous to the Serret-Frenet formula.
Let C:uα = uα(s) (9.46)

be a curve on S whose metric is given by
duα du β
aαβ = 1. (9.47)
ds ds
Here, the tangent vector λα to C:
duα
λα = (9.48)
ds
is a unit vector. If we differentiate the quadratic relation aαβλαλβ = 1 intrinsi-

δλ β
cally with respect to s, we obtain aαβ λ α = 0, from which it follows that
δλ α δs
surface vector is orthogonal to λα.
δs
Now, we introduce a unit surface vector ηα normal to λα, so that
δλ α
= χ g ηα , (9.49)
δs
where χα is a suitable scalar.
We choose the sense of ηα in such a manner that (λα, ηα) is positive,
i.e.,
ϵαβλαηα = + 1
(the choice of orientation of λ and η uniquely determines the sign of χg and

the sine of the angle between λ & η is + 1).
The vector ηα is the unit surface vector orthogonal to curve C and the
scalar χg is called the Geodesic Curvature of C.
We know  αβ λα = η β and  αβ ηβ = λ α .
δη β δλ δ δ
Now, =  αβ α =  αβ ( aαβ λ β ) =  αβ aαβ λ β =  αβ aαβ χ g η β
δs δs δs δs
= χ g  αβ ηα = − χ g  βα ηα = − χ g λ β

δηα
= − χ g λα . (9.50)
δs
We may refer to (9.48) and (9.49) as Serret-Frenet formulas for curve C

lying on S.
δλ
The transversion of α = χ g ηα and by ηα ,
δs
δλ δλ δλ
we get χ g = α λ α = α ( αβ λβ ) =  αβ λβ α
δs δs δs
λ1 λ2
1 αβ δλα
= e λα = a δλ1 δλ2 .
a δs
δs δs

δλ α
Alternatively, using = χ g ηα ,
δs
β
λ 1 λ 2
δλ
χ g = αβ λ α = a δλ 1 δλ 2
δs
δs δs
du1 du 2

ds ds
= a .
d 2u1  1  du β duγ d 2u 2  2  du β duγ
+  2 +  
ds 2  βγ  ds ds ds  βγ  ds ds
These expressions are known as Beltrami’s Formula for Geodesic

Curvature.
Theorem 9.10.1. A necessary and sufficient condition that a curve on a

surface S is a geodesic is that its geodesic curvature is zero.
Proof: The differential equation of a geodesic is
d 2u du du
0,
ds 2 ds ds

d d .u du du
or ds ds . 0,
ds ds

du du du du
or 0,
u ds ds ds ds

du du du
or = 0,
u ds ds ds
du du
or 0
ds ds
δ  duα 
∴   = 0 ,
δ s  ds 
or 0
s
∴ χ g ηα = 0.
Since ηα is a unit vector, we must have χ g = 0

and conversely, let χ g = 0
δ  duα  ∂ α α
  = λ = χ g η = 0.
δ s ds ∂s
duα
= constant , c
ds
or uα = cs + d.
This shows that geodesic of a surface is therefore analogous to a straight

line in a plane. Therefore, a curve C on a surface S is geodesic if its geodesic
curvature χg vanishes identically. A straight line on any surface is geodesic.
Example 9.10.1. Find the geodesic curvature of a small circle:
C : u1 = constant = u01 ≠ 0, u 2 = u 2

on the surface of the sphere
S: y1 = acosu1cosu2
y2 = acosu1sinu2
y3 = asinu1
tanu01
(or show that the geodesic curvature χ g = ).
a
Solution: If the arc-length s of C is measured from the plane u12 = 0, we
have
Y3
λ
C P
u01
Y2
u2 a s
Y1
Figure 9.5
 r = cosu1 , s = rθ = a cosu1u 2 ; ∴u 2 = s  .

 a 0 0
acosu01 

s
u2 =
acosu01
and the equation of C can be written in the form
2 s (i)
u1 u10 and u
acosu10
du
from i, we find the components of the unit tangent vector along
C: ds
1 du 2
2 1
λ = 0, λ = = , (ii)
ds acosu01
so that
1
d 1 1 du 1 2 du2
0
s ds ds 22 ds
2
1 11 1
cossu sinu tan u10
acosu10 a2
2
d 2 2 du 2 2 du2
0 0
s ds ds 22 ds

 dλ 2 d  1  
 since =  1  = 0 .
 ds ds  acosu0  

δλ α 1
We know = χ g ηα g
δs s
1
∴ tan u01 = χ g η1 (iii)
a2
δλ 2
= χ gη 2 = 0 (iv)
δs
Since C is not a geodesic, χg ≠ 0, from (iv) η2 = 0, and
1
η iη j =a12a(η)
ηi is a unit vector, we have aij ηiηj =1 ) =∴
2 2 2
(η=1, 1, η1 = .
a
1 1 1
From (iii), we get ∴ 2 tan u01 = χ g η1 or χ g = 2 tan u01 .a = . tan u01 .
a a a
Example 9.10.2. Consider the surface of the right circular cone
S: : x1 = u1cosu2
x2 = u1sinu2
x3 = u1
s
the curve C : u1 = a, u 2 = .
a 2
Show that the geodesic curvature is .
2a
Solution: Here, ds2 = 2(du1)2 + (u1)2(du2)2

a11 = 2,a22 = (u1)2, a12 = a21 = 0,
duα 1 du1 du 2 1
λα = ,λ = = 0, λ 2 = =
ds ds ds a
1 1 11 1 1 g 22 11 1 u1
g 22, g 22,1 2u
22 g 11 2 u1 22 2

2 1 g 22
g 2 22, g 22 22, 2 g 22 0,
22 2 u2

2
1
d 1
1 du 1 du 2 u1 1 1
so that 0 2
s ds ds 22 ds 2 a 2a
2
d 2 2 du 2 2 du2
0 0.
s ds ds 22 ds

Again, by the geodesic curvilinear formula,
δλ α α δλ
1
1
= χ g η , = χ g η1 or , χ g η1 = −
δs δs 2a (i)
and χ g η 2 = 0 since Cisnot a geodesic , χ g ≠ 0, η 2 = 0. (ii)
Since C is not a geodesic and

ηi is a unit vector, we have aijηiηη =1
j i j
)2 =∴1,η1 = − 12 .
η = 12.(η(η1)12=1,
1 1  1
From (ii), χ g η1 = − , or χ g  −  = −
2a  2 2a
2
∴ χg = Ans.
2a
[ds2 = a2(-cosu1 sinu2du2-cosu2sinu1du1)2 + a2(cosu1 cosu2du2-sinu2sinu1du1)2
+ a2cos2 u1(du1)2
= a2 [cos2 u1 (du2)2 + sin2 u1 (du1)2 + cos2 u1(du1)2] = a2(du1)2 + a2 cos2 u1
(du2)2. Here, a11 =a2,a12 = 0, a22 = a2cos2u1
 1  1α 1 1  ∂ g 22  1 1  ∂
  =g [22,α]=g11[22,1]= .  − 1  = 2  − 1 (a2 cos2 u1)
 2 2  g 11 2 ∂u a 2 ∂u
a2
( a2cos 2u1 ) = − 2cosu1 ( − sinu1 ) =cosu1sinu1
2a 2
 2  1 ∂ g 22
 =g [22,α]=g [22,2]= g 22 =0. ]
2α 22

 2 2  2 ∂u 2
Example 9.10.3. Show the geodesic curvature of the curve u = c on a sur-

face with metric
1 ∂µ .
φ 2 (du)2 + µ 2 (dv )2 =
φµ ∂u
Solution: Here, ds2 = ϕ2(du)2 + μ2(dv)2
a11 = ϕ2,a22 = μ2 and a12 = a21 = 0
a11 0
a= = a11a22 = φ 2 µ 2
0 a22

1 1 1 1
a11 = = 2 , a 22 = = 2 , a12 = a 21 = 0 .
a11 φ a22 µ
Applying the Beltrami Formula to find geodesic curvature χ g ,
du1 du 2

ds ds
χg = a
d 2u1  1  du β duγ d 2u 2  2  du β duγ
+  2 +  
ds 2  βγ  ds ds ds  βγ  ds ds

du1 d 2u2 2 du du du2 d 2u1 1 du du

ds ds 2 ds ds ds ds 2 ds ds
du
1 2
du
2
2 1
du du
1
2 2
du du
2
2 1
du du
2
2
2
ds ds 11 ds ds 22 ds ds 12 ds ds
du
2 2
du
1
1 1
du du
1
1 2
du du
2
1 1
du du
2
2
2
ds ds 11 ds ds 22 ds ds 12 ds ds

 2  22 1  1 ∂a11  1 ∂φ φ ∂φ
 11  = a [11,2 ] = 2  −  = − 2 2φ = − 2
  µ 2 ∂v 2µ ∂v µ ∂v
 2  22 1  1 ∂a22  1 1 ∂µ 1 ∂ µ
  = a [ 22,2 ] = 2   = 2 2
µ =
∂v µ ∂v
 2 2  µ 2 ∂v 2µ
 2  22 1  1 ∂a22  1 ∂µ 1 µ
  = a [12,2 ] = 2   = 2 2µ =
 1 2  µ 2 ∂u 2 µ ∂u µ ∂u

1 1 1 a11 1 1
a11 11,1 2
2
11 2 u 2 2 u u

 1  11 1  1 ∂a22  1 ∂µ µ ∂µ
  = a [ 22,1] = 2  −  = − 2 2µ =− 2
 2 2  φ 2 ∂u 2φ ∂u φ ∂u

1 1 1 a11 1
a11 12,1 2
2
12 2 v 2 2 v

Since for the parametric curve u = c is constant,
du 2  1  du 2 du 2  1   dv  3
χg = − a   =− a 
ds  2 2  ds ds  2 2   ds 
3
µ ∂ µ  dv  µ 1 ∂µ 1 ∂µ
= −φµ = φµ 2 3 =
2 
φ ∂u ds  φ µ ∂u φµ ∂u
2 2 2
 2  du  2  dv   dv  1 
1 = φ + µ , = 2 , since u = c 
  ds   ds   ds  µ 

Example 9.10.4. Show the condition that the u1−curve and u2−curve geo-
2 1
desics are 0 and 0, respectively.
11 22
Solution: We have
δλ β
χ g = αβ λ α .
δs
Now, for the u1−curve, we have
1
λ(α1) =  
,0
 a11 
δλ 2
λ g(1) = 12 λ 1 and others are zero
δs
1 d 2 2 1 du1
12 and all other Christoffel symbols are zero,
ds 11 ds
1 d 2 2 1 1 1 2 1 1
ae12 a .1 0
ds 1 1 a11 11 a11 a11

1 1 2
g a 3
.
11
(a11 ) 2

Similarly,
δλ 1 α  1 
λ g( 2) = 21λ 2 , where λ( 2) =  0,
δs  a22 
2 d 1 1 2 2
ae21 and all other Christoffel symbols are zero
ds 2 2
d 1 1
a 1 ( 2 )3
ds 22

d 1 1
a ( 2 )3
ds 22

1 1 1 1
a 3
0 a 3
.
22 22
(a22 ) 2 (a22 ) 2

A curve will be geodesic if the geodesic curvature is zero.

( ) ( )
Therefore, λ g1 = 0 and λ g2 = 0.
2 1
This implies that 0 and 0, respectively.
11 22
9.11 Exercises
1. Find the element of area of the surface of radius r if the equations of the
surface are given in the form x1 = a sin u cosv, x2 = a sin u sinv, and x3 =
acos u, where xi are the orthogonal Cartesian coordinates.
2. Find the differential equations for the geodesic in cylindrical coordinates:
x1 = u1 cosu2
x2 = u1 sinu2
x3 = u3.
3. Show that the formula for the Gaussian Curvature K can be written in
the form
1   ∂  a12 ∂a11 1 ∂a22  

K=   1  2 −
2 a   ∂u  a11 a ∂u a ∂u1  
 ∂  2 ∂a12 a ∂a11 1 ∂a11  

+ 2 − 12 1
− 
 ∂u  a ∂u
1
a11 a ∂u a ∂u 2  

4. Prove that the catenoid
ds2 = a2 cosh2 x1 (dx1)2 + a2 cosh2 x2 (dx2)2

and the right helicoid
ds2 = (dy1)2 + [(y1)2 + a2](dy2)2

are locally isometric.
5. Determine whether the surface of a helicoid given by
x1 = u1 cosu2
x2 = u1 sinu2
x3 = cu1
is developable, where is a constant and u1 and u2 are curvilinear coordi-
nates of the surface.
6. Show that the surface with metric
(i) ds2 = (du1)2 + [(u1)2 + a2](du2)2 is not developable.
(ii) ds2 = (u2)2(du1)2 + [(u1)2(du2)2 is developable.

7. Show that for the sphere of radius c, with the equation of the form
x1 = c cosu1 cosu2
x2 = c sinu1 sinu2
x3 = c cosu1, where c is a constant,
1
the total curvature is κ = .
c2
10
Surfaces in Space
10.1 Introduction
Our study of the geometry of surfaces was carried out from the point of
view of a two dimensional being whose universe is determined by the sur-
face parameters u1 and u2. The study of surfaces was based entirely on the
first quadratic differential form. In the discussion of isometric surfaces in
the previous chapter, we remarked that a pair of isometric surfaces, for
example, a cone and a cylinder which are indistinguishable in intrinsic
geometry, appear to be quite distinct to an observer examining them from
a reference frame located in the space in which the surfaces are imbedded.
An entity that provides a characterization of the shape of the surface as it
appears from enveloping space is the normal line to the surface.
In this chapter, the geometric shape and properties of a surface with the
aid of the quadratic form that depends fundamentally on the behavior of
the normal line is discussed.
10.2 The Tangent Vector

A surface S imbedded in E3 was defined by three parametric equations
yi=yi (u1,u2) , (i=1,2,3), (10.1)
∂ yi α
i
so that dy = α du ,
∂u
where the yi are orthogonal Cartesian coordinates of the reference frame

located in the space surrounding S. An element of arc ds of a curve on S is
determined by the formula
321
ds2=aαβ duα duβ, (10.2)
∂ yi ∂ yi
where aαβ = .
∂uα ∂u β
The choice of Cartesian variables yi in the space enveloping the surface is
clearly not essential and we could have equally referred the points of E3 to a
curvilinear coordinate system X related to Y by the transformation
xi=xi (y1,y2,y3).
Now, relative to the frame X, the line element in E3 is given by
ds2=gij dxi dxj, (10.3)
∂ yk ∂ yk
g = .
where ∂x i ∂x j
ij
Then, Equation (10.1) for the surface S can be written as
S:xi=xi (u1,u2), (10.4)
∂x i α
so that dx i = du . (10.5)
∂uα
Hence, Equation (10.3) can be formed as
ds2=gij dxi dxj
∂x i ∂x j α β .
= g ij du du .
∂uα ∂u β
Comparing this to Equation (10.2), we get

Surfaces in Space 323
∂x i ∂x j
aαβ = g ij , (i , j = 1,2,3) , (α , β = 1,2 ) , (10.6)
∂uα ∂u β
where dxi and gij’s are tensors with respect to the transformations induced
on the space variablesxi, whereas duα and aαβ are tensors with respect to the
transformation of Gaussian surface coordinates uα.
Since both aαβ and gij in Equation (10.6) are tensors, this formula sug-
∂x i
gests that α can be regarded either as a contravariant space vector or as
∂u
a covariant surface vector.
∂x i
From Equation (10.5), dx i = α duα .
∂u
The left member dxi is invariant relative to a change of the surface coor-
dinate uα. Since duα is an arbitrary surface vector, we conclude that
∂x i (10.7)
∂uα
is a covariant surface vector. On the other hand, if we change the space

coordinates, duα, being a surface vector, is invariant relative to this change,
so that Equation (10.7) must be a contravariant space vector. Hence, we
can write Equation (10.7) as
∂x i
x αi = (10.8)
∂uα
i ∂x i α
Again, dx = du .
∂uα
dx i ∂ x i duα
We have =
ds ∂uα ds
or λ i = xαi λ α (10.9)

This equation tells us that any surface vector Aα (that is, a vector lying
in a tangent plane to S) can be viewed as a space vector with components
Ai determined by
Ai = xαi Aα . (10.10)
We shall refer to a vector Ai determined by this formula as a tangent

vector to surface S.
10.3 The Normal Line to the Surface

Let A and B be a pair of surface vectors drawn at point P of S, as shown in
Figure (10.1). According to Formula (10.10),
Ai = xαi Aα (Aα is lying in the tangent plane to S)
 i  ∂ x 1 ∂ x 2 , ∂ x 3  
 xα =  α α 
 ∂u , ∂u ∂uα  
.
They can be represented in the form
Ai = xαi Aα Bi = xαi Bα (10.11)
 ∂x i i i ∂u β i 
 ∂u β = x β , xα = ∂uα x β 

The vector product:

B u2
θ
P A
u1
S
r
Figure 10.1
A×B is the vector normal to the tangent plane determined by the vectors
A and B and the unit vector n perpendicular to the tangent plane, so ori-
ented that A,B, and n from a right handed system, are
A×B A × B , (10.12)
n= =
A × B AB sinθ
where θ is the angle between A and B.

n is called the unit normal vector to the surface S at P. Clearly, n is a
function of coordinates (u1,u2) and, as point P (u1,u2) is displaced to a new
position P(u1+du1,u2+du2), the vector n undergoes a change:
∂n α
dn = du , (10.13)
∂uα
∂r
where the position vector r is changed by the amount dr = β du β .
∂n ∂r ∂u
Let us use the scalar product dr .dn = α β duα du β . (10.14)
If we define ∂u ∂u
1 ∂n ∂r ∂n ∂r
bαβ = −  α β + β α  ,

2 ∂u ∂u ∂u ∂u 
(10.14) becomes
dr.dn=−bαβ duα duβ. (10.15)
The left hand of (10.15), being the scalar product of two vectors, is obvi-
ously an invariant. Moreover, from symmetry with regards to α and β, it
is clear that the coefficient of duα duβ in the r.h.s. of (10.15) is defined as a
covariant tensor of rank two.
The quadratic form B ≡ bαβ duα duβ (10.16)
is just like the first fundamental quadratic form A ≡ dr.dr.
A ≡ aαβ duα duβ,
which is in the study of intrinsic properties of surface.
Differential form (10.16) was introduced by Gauss and is called the sec-
ond fundamental quadratic form of the surface.
This differential equation plays an important role in the study of the
geometry of a surface when it is viewed from the surrounding space.
We write (10.12) in terms of the components of xαi of the base vectors aα.
ijk A j B k
ni (10.17)
ABsin
Using the formula ϵαβ λα μβ=sinθ,
ABsinθ=ϵαβ Aα Bβ. (10.18)
Substitute this value and from (10.11) in (10.17), we get
niαβ A α B β = ijk xαi x βi Aα Aβ

or (niαβ − ijk xαi x βi )Aα Aβ = 0.
Since this relation is valid for all surface vectors, we can write
niαβ − ijk xαi x βi = 0

(10.19)
Multiplying (10.19) by ϵαβ and putting the value of ϵαβ ϵαβ=2,

1
ni =  αβ ijk xαi x βi (10.20)
2
It is clear from the formula that ni is a space vector which does not
depend on the choice of the surface coordinates.
Example 10.3.1. Find the Second Fundamental quadratic form of right

Helicoid
r =(ucosv,usinv,cv).
Solution:
∂r
ru = = ( cosv , sinv ,0 )
∂u
∂r
rv = = ( −usinv , ucosv , c )
∂v
i j k
ru × rv = cosu sinv 0 = (csinv , −c cosv , u)
−usinv ucosv c

1
∴ ru × rv = ( u + c ) 2
2 2
( csinv, −ccosv, u )
η= 1
( u 2 + c 2 )2
∂η  1 u
ηuu = = (00,0,1
, 0,1 ) − 33
csinv , −ccosv
((csinv, u)  .
ccosv,, � u)
∂u u2 + c2
2 2
 u c (u 2 )22
u22 + cc2 

∂η 1
ηv = = ( c.cosv,c.sinv, 0)
c.cosv , c.sinv , � 0
∂v u 2 + c2
u2 c2

1
bαβ = −
2
(ηα rβ + ηβ rα )

1
b11 = − (η1r1 + η1r1 )
2
1 u
r
u u 0, 0,1 3
csinv , ccosv , u . .
u 2
c 2
u2 c 2 2
cosv , sinv , 0 0, 0, 0

1 c
b12 = − (η1r2 + η2r1 ) = 2 2
2 u +c

1
b22 = − (η2r2 + η2r2 ) = −η2r2 = −ηv rv = 0
2
−2c
 ≡ bαβ duα du β = −2b12dudv = dudv.
u2 + c 2
Example 10.3.2. Find the second fundamental form for the paraboloid
given by
r=(u,v,u2−v2).
Solution: Here, r=(u,v,u2−v2)

∂r ∂r
So A = 1 = (1,0,2u ) , B = 2 = ( 0,1, −2v )
∂u ∂u
i j k
A×B= 1 0 2u = −2ui + 2vj + k
0 1 − 2v

A × B −2ui + 2vj + k
n= =
A×B 4u 2 + 4v 2 + 1

( 2 {4v 2 + 1} , −8uv ,4u ) .(1, o,2u ) = 2

b11 = −nu .ru = 3
(4u 2 + 4v 2 + 1) 2 4u + 4v 2 + 1
2
b22 = −nv .rv =

( −8uv , −2{4u2 + 1} ,4v ).(0,1, −2v ) = − 2
3
(4u 2 + 4v 2 + 1) 2 4u + 4v 2 + 1
2
b12=b21=0
 ≡ bαβ duα du β = b11 (du1 )2 + b22 (du 2 )2 + b12du1du 2

2 (du1 )2 − ( du 2 )2  .
= 2 2
4u + 4v + 1
10.4 Tensor Derivatives

We are introducing the tensor differentiation of tensor fields which are
functions of both surface and space coordinates.
Let us consider a curve C lying on a given surface S and vector Ai defined
along C, whose parameter is t:
δ Ai dAi  i  dx k
j
= +  A , (10.21)
δt dt  j k  dt
g  

i
where refers to space coordinates xi and are formed from metric
j k
g
coefficient gij.
Again, if we consider a surface vector Aα defined along the same curve
C, we can form the intrinsic derivative with respect to the surface variables
δ Aα dAα  α  dx γ
β
= +  A . (10.22)
δt dt  β γ dt
a  
 α 
 β γ  refer to Gaussian surface coordinates u and are formed from
α
a 
metric coefficient aαβ.
A geometric interpretation of these formulas is at hand when the fields
δ Ai δ Aα
Ai and Aαare such that = 0 and = 0. In the first equation, the vec-
δt δt
tors Ai form a parallel field with respect to C, considered a space curve,
δ Aα
whereas = 0 defines a parallel field with respect to C as a surface curve.
δt
Intrinsic derivatives of the covariant vectors Ai and Aα are
δ Ai dAi  k  dx j
= −   Ak
δt dt  i j  dt (10.23)
g 

δ Aα dAα  γ  dx β
and = −   Aγ (10.24)
δt dt  α β  dt
a
Consider a tensor field Tαi , which is a contravariant vector with respect

to a transformation of space coordinates xi and a covariant vector with
respect to a transformation of surface vector uα.
If Tαi , is defined over a surface curve C and the parameter along C is
t, Tαi , is a function of t. We introduce a parallel vector field Ai along C,
regarded as a space curve, and a parallel vector field Bα along C as a surface
curve and form an invariant
φ ( t ) = Tαi Ai Bα .
ddφ (t ) dTαi dA
dA dB dBα
Then, = Ai � BBα +TTαii ii� BBα +TTiαA
i
Ai �i , , (10.25)
dt dt dt
dt dtdt
δ Ai δ Bα
but since Ai (t) and Bα (t) are parallel, then = 0 and = 0.
From (10.23) and (10.22), we have δt δt
 k  dBα  α  duγ
dAi dx j β
=   Ak and = −  B .
dt
g 
 i j  dt dt
a
 β γ 
dt
(10.25) becomes
dφ ( t ) dTαi  k  dx j α  α  duγ
= Ai Bα + Tαi  A
 k B − Tαi Ai  B
β
dt dt
g 
 i j 
dt
a
 β γ 
dt
 dT i  i  dx k  δ  duγ 
=  α + Tαj   − Tδi    Ai Bα .
 dt
 g 
 j k  dt a
 α γ  dt 

(10.26)

By the quotient law we conclude that the expression in the bracket is a

tensor of the same character as Tαi . We call this tensor an intrinsic deriva-
tive of Tαi , with respect to t, and write
δ Tαi dTαi  i  dx k  δ  duγ

i 
= + Tαj   − Tδ   (10.27)
δt dx  j k  dt  α γ dt
g  a 
 ∂T i  i  k  δ   duγ
i 
=  αγ + Tαj   x γ − Tδ  
 ∂u
 g 
 j k   α γ
a
  dt

 k ∂ x k ∂ x k duγ dx k 
 since xγ = ∂uγ ; ∂u γ dt = dt 
 
duγ
Since is an arbitrary surface vector, we conclude that
dt
∂Tαi  i  j  δ  i (10.28)
Tαi ,γ = γ +  Tα xγk −  Tδ
∂u g 
j k  a  α γ 

i
and call Tα,γ a tensor derivative of Tαi with respect to uγ.
i
The tensor derivative of Tα,β with respect to uγ is given by
Ti i
Ti T j xk Ti Ti
u jk (10.29)
g a a

If the surface coordinates at any n-point P of S are geodesic and the

space coordinates are orthogonal Cartesian, we see that at that point the
tensor derivatives reduce to the ordinary derivatives. It is concluded that
the operations of tensor differentiatio of products and sums follow the
usual rules and the tensor derivatives of gij, a αβ, ϵ ijk, ϵαβ and their associated
tensors vanish.
10.5 Second Fundamental Form of a Surface

We begin by calculating the tensor derivative of the tensor xαi , ,represent-
ing the components of the surface base vectors aα. We have
∂2 x i  i   δ 
i j k i
xα,β = +  x
 α β x −   xδ (10.30)
∂uα ∂u β  jk  αβ 
g   a

 i ∂x i 
 since, x α = 
∂uα
from which we get x i , xi , (10.31)
Since the tensor derivative of aαβ vanishes, aαβ = g ij xαi x βj .

i
We obtain g ij xα,γ x βj + g ij xαi x βj ,γ = 0,(10.32)
interchanging α, β, and γ cyclically,
i
g ij x β,α , xγj + g ij x βi xγj ,α = 0 (10.33)

i
g ij xγ,β , xαj + g ij xαi xαj ,β = 0.
(10.34)
From (10.32), (10.33), and (10.34) and using (10.31), we get

i
g ij xα,β xγj = 0
(10.35)
i
This is the orthogonality relation which states that xα,β is a space vector
normal to the surface and it is directed along the unit normal ni.
Hence, there exists a set of functions bαβ such that

i
xα,β = bαβ ni . (10.36)
The quantities bαβ are the components of a symmetric surface tensor and
the differential quadratic form
 ≡ bαβ duα du β
(10.37)
is the desired second fundamental form.
10.5.1 Equivalence of Definition of Tensor bαβ

We define
1  ∂n ∂r ∂n ∂r 
bαβ = −  α β + β α  .

2 ∂u ∂u ∂u ∂u 

∂r
We note that n and aα = are orthogonal, hence
∂uα
∂r ∂r
n. α = 0 and n. = 0.
∂u ∂u β
Differentiating these two scalar products with respect to uβ and uα,

respectively, and adding, we get
 ∂n ∂r ∂n ∂r   ∂2 r 
 α β + β α  + 2n  β α  = 0
∂u ∂u ∂u ∂u ∂u ∂u
1  ∂n ∂r ∂n ∂r   ∂2 r 
or  +  = − n  β α  .
2  ∂uα ∂u β ∂u β ∂uα  ∂u ∂u
 ∂2 r 
Hence, bαβ = n  β α  , (10.38)
 ∂u ∂u 
∂r
but, = aα = bi xαi .
∂uα
Therefore,
∂2 r ∂ xαi ∂bi i
β α = bi β + xα
∂u ∂u ∂u ∂u β
∂ xαi ∂b
= bi β + βi xαi x βj
∂u ∂x
 i  i  
∂x
= bi  αβ +  x j k
x
 α β .
 ∂u
 g 
 j k 



Using (10.30) in the above equations, we have
  δ  
∂2 r   xδ 
i i
= bi x α , β +  (10.39)
∂u β ∂uα   α β  
a
Multiplying (10.39) scalarly by n and considering n is orthogonal to

aδ = bi xδi so that n aδ = 0,
∂2 r
therefore n. i
= n.bi xα,β
∂u β ∂uα
i
= xα,β ni

= bαβ [by (10.36)].
This establishes the equivalence of the two definitions of the second fun-
damental form.
Equation (10.36) is known as the formulas of Gauss.
10.6 The Integrability Condition

In order to get insight into the significance of the tensor bαβ, let us examine
more closely the Gauss formulas
i
xα,β = bαβ ni , (10.40)

∂2 y  i   δ 
i j k i
where x α,β = α δ +   xα xδ −   xδ
∂u ∂u
g 
 jk  a
 αβ 
1
and ni =  αβ ijk xαj x βk
2
∂x j
with x αj = .
∂u α
If we insert these expressions in Equation (10.40), we obtain a set of sec-
ond order partial differential equations in which variables xi are the func-
tions of surface coordinates uα.
The surface S is defined as xi = xi (u1,u2) (i=1,2,3),(10.41)
are immersed; they are also functions of
∂x i ∂x j i j
aαβ = g ij α β = g ij x α x β and bαβ
∂u ∂u
In order for tensors aαβ and bαβ to be related to some surface, it is neces-
sary that xi satisfies the integrability conditions
2 i 2 i
∂γ xα β = ∂β xα γ , (10.42)
∂u ∂u ∂u ∂u
whenever the functions xαj are of class C2, which is equivalent to
i i σ
xα,βγ − xα,γβ = Rαβγ xδ,i (10.43)

σ
where Rαβγ is the Riemann tensor of the second kind formed with the coef-
ficients aαβ of the first fundamental quadratic form. We shall see that inte-
grability conditions (10.42) impose certain restrictions on possible choices
of functions aαβ and bαβ. These restricted conditions are known as the equa-
tions of Gauss and Codazzi.
1
i
Example 10.6.1. Show that bαβ = g ij xα,β n j =  γδ ijk xα,β
i
xγj xδk .
2
Solution: Here, from Equation (10.10)
Ai = xαi Aα , B j = x βj B β .

If θ is the angle between Ai and Bj, we can write
ABsinθ = αβ Aα Bβ.

If n is the unit surface normal, then
A B ijk x j A x k B
ni
AB sin  A B |

or n ijk x j x k A B 0.
Since Aα and Bβ are arbitrary, then
ni ijk x j x k 0,

or ni ijk x j x k
or ni αβ αβ =  αβ ijk xαj x βk ,
ni .2 =  γδ ijk xγj xδk

or
1
∴ ni =  γδ ijk xγj xδk .
2
i
Using xα,β = bαβ ni, we have
i
bαβ = xα,β i
ni = xα,β g ijn j = g ij xα,β
i
nj

 1 γδ 1
i
∴ bαβ = xα,β i
ni = xα,β  ijk xγj xδk  =  γδijk xγj xδk xα,β
i
.
2  2
Combining the above two equations, we get
1
i
bαβ = g ij xα,β n j =  γδ ijk xγj xδk xα,β
i
.
2
10.7 Formulas of Weingarten

Let ni be the unit normal vector to surface S. We begin with the relation
gijni ni =1
and when we form its tensor derivative, we have gijni,α n j + gijni n j,α = 0
or g ij n.in.,αj = 0. (10.44)
This shows that n.,αj ,considered as a space vector, is orthogonal to the

unit normal ni and hence it lies in the tangent plane to the surface.
i
Accordingly, it can be represented as a linear form in the base vectors x ,α .
i
x ,α is also called a surface base vector.
i
n,α = cαβ x β.i (10.45)

Since ni is normal to the surface, we have the orthgonality relation

g ij xαi η .i = 0,
whose tensor derivative is
i
g ij xα,β ni + g ij xαi n,iβ = 0, (10.46)
i
but from Gauss’s Formula: xα,β = bαβ ni ,(10.47)
so that the substitution from (10.46) and (10.45) in (10.47) yields
g ijbαβ nin. j + g ij xαi cβγ xγi . = 0,

or 1.bαβ + g ij xαi xγ.j cβγ = 0,  g ijnin. j 1
or bαβ + g ij xαi xγ.j cβγ = 0
Since, aαβ = g ij x αi x βj
bαβ + aαγ cβγ = 0.

We have
∴ bαβ = −aαγ cβγ .
Solving this equation for cβγ , we get
cβγ = −aαγ bαβ ., (since , cδβ = −aαβ bαδ )
So that Equation (10.45) becomes
n,δi = cδβ x βi .

= −aαβ bαδ x βi .

n,iα = −a βγ bβα xγi . (10.48)
These are the Weingarten formulas.

We can use these equations in deriving the equations of Gauss and
Codazzi.
10.7.1 Third Fundamental Form

If we write
cαβ = g ij n,iα n,jβ ,

we see that cαβ is a symmetric covariant surface tensor of type (0,2) and we
call the quadratic form C ≡ cαβ duα du β the third fundamental form of the
surface.
Using Weingarten formula (10.48), we get
Cαβ = g ij n,iα n,jβ = g ij ( −aδγ bδα xγi . )( −a µϑ bµβ xϑ.j ) = aγϑ aδγ bδα a µϑ bµβ

= a µδ bδα bµβ .

This is the relation between three fundamental forms on a surface,

but here the third fundamental form is not an actual fundamental form
because this can be obtained from first and second fundamental forms.
10.8 Equations of Gauss and Codazzi

Now we will establish Codazzi’s Equation using the Weingarten Equation.
We follow the integrability conditions
i i σ
xα,βγ − xα,γβ = Rαβγ xδi . (10.49)

Now, from tensor derivative Equation (10.47), we use (10.48) to obtain:

differentiating (10.47) and substituting (10.48)
i
xα,βγ = bαβ,γ ni + bαβ n,γi

= bαβ,γ . ni − bαβ aδλ bδγ x λi .
Similarly, i
xα,γβ = bαγ,β . ni − bαγ aδλ bδβ x λi . (10.50)
Substituting it in (10.49), we get

( xα,βγ − xα,γβ ) = (bαβ,γ − bαγ,β ). n − aδλ (bαβ bδγ − bαγ bδβ ) x λ
i i i i
.
Hence,
(bαβ,γ − bαγ,β ). ni − aδλ (bαβ bδγ − bαγ bδβ ) x λ.i = Rαβγ
σ
xδi . (10.51)

To obtain the equation of Codazzi, we multiply (10.51) by ni and

since xαi ni =0 (by using g ij xαi η .i = 0)
(bαβ,γ − bαγ,β ). ni ni − aδλ (bαβ bδγ − bαγ bδβ ) x λ.i ni = Rαβγ
σ
xδi ni ,

we get desire result: (bαβ,γ − bαγ,β ) = 0 .(10.52)

This is known as the Codazzi Equation,
which will constitute all integrability conditions of the formula of
Weingarten.
To obtain the equations of Gauss,
we multiply (10.51) by g ij x ρj and get
g ij x ρj (bαβ,γ − bαγ,β ) . ni − aδλ (bαβ bδγ − bαγ bδβ ) x λ.i g ij x ρj = Rαβγ
σ
xδi g ij x ρj .

Since g ij x ρj xσi = aσ p and 1st term be zero, we have

σ
aσ p Rαβγ = −aδλ (bαβ bδγ − bαγ bδβ ) aλρ .

Since aδλ aλρ = δ ρδ
Rραβγ = bαγ bρβ − bαβ bργ (10.53)

which is known as the Gauss Equation.
Since α,β assumes values 1,2 and bαβ=bβα, we see that there are two
independent equations of Codazzi and only one independent equation of
Gauss.
The independent equations of Codazzi are
bαα,β−bαβ,α=0 (α≠β) (no sum on α).
We know
b
b b b
u
a
a
[ ααβγ αβγγ
∴ R = R = 0 ; R1212 = R2121 = − R2112 = − R1221 ] .
We have
b
b b b
u a a
b
b b . b
u
a a

b b
Therefore, b b 0 (10.54)
u u a a
α≠β no sum on α
and the equation of Gauss,
Rραβγ=bαγ bρβ−bαβ bργ.

Here, ρ=1, α=2, β=1, γ=2,
therefore, R1212=b22 b11−b21 b12=b22 b11−b212 (10.55)

[since other Rραβγ=0].
This equation relates the coefficients bαβ and aαβ in the two fundamental
quadratic forms.
10.9 Mean and Total Curvatures of a Surface

We know the total curvature (from 9.39)
R1212
, where a = aαβ .
a
We can write Equation (10.55) in the form
R1212 b22b11 b212 b

. (10.56)
a a a
The Gaussian Curvature is equal to the quotient of the determinants of

the second and first fundamental quadratic forms.
We can define another important invariant H, given by the formula
1
H ≡ aαβ bαβ, (10.57)
2
which is called the mean curvature of the surface.
Example 10.9.1. Prove that C − 2 HB + κA = 0, where the notations have

their usual meaning.
Solution: Here, the third fundamental form is (C) = cαβ duα duβ,
the second fundamental form is (B) = bαβ duα duβ,
the first fundamental form is (A) = aαβ duα duβ
R1212 1 αβ
, and the mean curvature is H = a bαβ..
a 2
The Gauss formula is Rραβγ=bαγ bρβ−bαβ bργ and Rραβγ = κραβγ
∴ κραβγ = bαγ bρβ − bαβ bργ.

Multiplying both sides by aργ, we get
κραβγ a ργ = a ργ bαγ bρβ − a ργ bαβ bργ ,

ργ
or a a b b a b b as −a αβ = ραβγ a ,
or κaαβ = 2 Hbαγ − cαβ since cαβ = a ργ bαβ bργ ,
or cαβ − 2Hbαγ + κa aβ = 0,
or cαβ duα du β − 2 Hbαγ duα du β + κa αβ duα du β = 0
∴ C − 2 HB + κA = 0.
Example 10.9.2. Show that cαβ aαβ = 4 H 2 − 2κ .
Solution: Using the relations from above,
κaαβ = 2 Hbαγ − cαβ ,
or κaαβ λ α λ β = 2 Hbαγ λα λ β − cαβ λα λ β ,
or κ = 2 Hbαβ λ α λ β − cαβ λ α λ β .
Using this relation κaαβ = 2 Hbαβ − cαβ

and multiplying aαβ, we get
κaαβaαβ = 2 Hbαβ aαβ − cαβ aαβ

2κ = 2 H .2 H − cαβ aαβ as aαβ aαβ = 2.
Therefore, cαβ aαβ = 4 H 2 − 2κ .
Example 10.9.3. Find the Gaussian and mean curvature of the surface
x1 x2=x3.
Solution: Let the parametric representation of the surface be given by
x1=u1, x2=u2 and x3=u1 u2

∴ r = ( x 1 , x 2 , x 3 ) = (u1 , u 2 , u1u 2 )
∂ yi ∂ yi
aαβ = α β
∂u ∂u
2 2 2
∂ y i ∂ y i  ∂ y1   ∂ y 2   ∂ y 3 
∴ a11 = 1 1 =  1  +  1  +  1  = 1 + (u 2 )2
∂u ∂u  ∂u   ∂u   ∂u 
2 2 2
∂ y i ∂ y i  ∂ y1   ∂ y 2   ∂ y 3 
a22 = 2 2 =  2  +  2  +  2  = 1 + (u1 )2
∂u ∂u  ∂u   ∂u   ∂u 
∂ y i ∂ y i ∂ y1 ∂ y1 ∂ y 2 ∂ y 2 ∂ y 3 ∂ y 3
a12 = = + + = u1u 2.
∂u1 ∂u 2 ∂u1 ∂u 2 ∂u1 ∂u 2 ∂u1 ∂u 2
1 + (u 2 )2 u1u 2
Therefore, a = 1 2 1 2
= 1 + (u1 )2 + (u 2 )2
u u 1 + (u )
1 + (u1 )2 1 + (u 2 )2 −u1u 2
a11 = , a 22
= , a12
= .
1 + (u1 )2 + (u 2 )2 1 + (u1 )2 + (u 2 )2 1 + (u1 )2 + (u 2 )2
Since r = ( x 1 , x 2 , x 3 ) = (u1 , u 2 , u1u 2 ) ,

here we have A = (1,0,u2) and B = (0,1,u1)
A× B 1
The normal vector is n = = ( −u2 , −u1 ,1) .
A× B 1 2 2 2
1 + (u ) + (u )
1  ∂n ∂r ∂n ∂r 
Now, bαβ = −  α . β + β . α 
2  ∂u ∂u ∂u ∂u 
1  ∂n ∂r ∂n ∂r  ∂n ∂r 1
b11 = −  1 . 1 + 1 . 1  = − 1 . 1 = −

2 ∂u ∂u ∂u ∂u  ∂u ∂u 1 + (u )2 + (u 2 )2
1
( 0, −1,0 ).(1,0, u ) = 0
2
1  ∂n ∂r ∂n ∂r  ∂n ∂r 1
b22 = −  2 . 2 + 2 . 2  = − 2 . 2 = −

2 ∂u ∂u ∂u ∂u  ∂u ∂u 1 + (u )2 + (u 2 )2
1
( −1,0,0 ).( 0,1, u ) = 0

1
1  ∂n ∂r ∂n ∂r  1  1 + (u1 )2 + (u 2 )2 1 + (u1 )2 + (u 2 )2 
b12 = −  1 . 2 + 2 . 1  = −  − − 3 
2  ∂u ∂u ∂u ∂u  2 3
 {1 + (u1 )2 + (u 2 )2 } 2 {1 + (u1 )2 + (u2 )2 } 2 

1
= = b21.
1 + (u ) + (u 2 )2
1 2
1
0
1 + (u ) + (u 2 )2
1 2
Therefore, b =
1
0
1 + (u ) + (u 2 )2
1 2
1
1 + (u ) + (u 2 )2
1 2
1
κ= 2 2 = .
1 + (u ) + (u ) [1 + (u1 )2 + (u 2 )2 ]2
1 2

The mean curvature H is given by
1 1
H = aαβ bαβ = ( a11 b11 + a 22 b22 + 2a12 b12 ) = a12 b12
2 2
1 2
−u u 1 −u1u 2
× = 3 .
= 1 + (u1 )2 + (u 2 )2 1 + (u1 )2 + (u 2 )2 {1 + (u1 )2 + (u 2 )2 } 2
Example 10.9.4. Show that the given surface
x 1 = u1 cos u 2, x 2 = u1sinu 2 and x 3 = cu 2
is a minimal surface.
Solution: The first fundamental form of the surface is
∂ yi ∂ yi
aαβ =
∂uα ∂u β
2 2 2
∂ y i ∂ y i  ∂ y1   ∂ y 2   ∂ y 3 
∴ a11 = 1 1 =  1  +  1  +  1 
∂u ∂u ∂u ∂u ∂u
2 2 2 2
= (cos u ) + (sin u ) = 1
2 2 2
∂ y i ∂ y i  ∂ y1   ∂ y 2   ∂ y 3 
a22 = 2 2 =  2  +  2  +  2 
∂u ∂u  ∂u   ∂u   ∂u 
= (−u1sinu 2 )2 + (u1cosu 2 )2 + c 2 = (u1 )2 + c 2
∂ y i ∂ y i ∂ y1 ∂ y1 ∂ y 2 ∂ y 2 ∂ y 3 ∂ y 3
a12 = = + +
∂u1 ∂u 2 ∂u1 ∂u 2 ∂u1 ∂u 2 ∂u1 ∂u 2
= cos u 2 ( −u1 sin u 2 ) + ( sinu 2 )(u1 cos u 2 ) = 0
1 0
∴ a= 1 2 2
= (u1 )2 + c 2
0 (u ) + c
(u1 )2 + c 2 1
a11 = 1 2
22
2 = 1, a = , a12 = a 21 = 0.
(u ) + c (u ) + c 2
1 2
Here, r = ( x 1 , x 2 , x 3 ) = (u1 cos u 2 , u1sinu 2 , cu 2 ) .
 ∂x1 ∂x 2 ∂x 3   ∂x1 ∂x 2 ∂x 3 
We have A =  1 , 1 , 1  = ( cos u 2 , sinu 2 ,0 ) ; B =  2 , 2 , 2 
 ∂u ∂u ∂u   ∂u ∂u ∂u 
= ( −u1 sin u 2 , u1cosu 2 , c )
A × B = ( csinu 2 , −c cos u 2 , u1 )
1
n= 1 2 2
(csinu2 , −c cos u2 , u1 ).
(u ) + c
1  ∂n ∂r ∂n ∂r 
Now, bαβ = −  α . β + β . α 

2 ∂u ∂u ∂u ∂u 
1  ∂n ∂r ∂n ∂r  ∂n ∂r 1
b11 = −  1 . 1 + 1 . 1  = − 1 . 1 = − ( 0,0,1).
2  ∂u ∂u ∂u ∂u  ∂u ∂u (u )2 + c 2
1
(cos u2 , sinu2 ,0) = 0
∂n ∂r 1
b22 = − 2. 2 =−
(ccosu2 , c sin u2 ,0).
∂u ∂u 1 2
(u ) + c 2
( −u1 sin u2 , u1cosu2 , c ) = 0
1  ∂n ∂r ∂n ∂r  1 1
b12 = −  1 . 2 + 2 . 1  = −

2 ∂u ∂u ∂u ∂u  2 (u )2 + c 2
1
( 0,0,1).( −u1 sin u 2 , u1cosu 2 , c ) + ( ccosu2 , c sin u 2 ,0 ) .( cos u 2 , sinu 2 ,0 ) 

 
−c
= 1 2 2
= b21
(u ) + c
1 1
H = aαβ bαβ = ( a11 b11 + a 22 b22 + 2a12 b12 ) =
2 2
1 1 −c 
= 1.0 + 1 2 2 .0 + 2.0. = 0.
2  (u ) + c 1 2 2 
(u ) + c 

Since the mean curvature H=0 at every point of the given surface, it is a
minimal surface.
10.10 Exercises
1. Show that in the usual notations
1  ∂n ∂r ∂n ∂r 
bαβ = −  α . β + β . α  ,
2  ∂u ∂u ∂u ∂u 
where n is the unit normal and r is the position vector of the point on the
surface.
2. Find the mean curvature of the surface r=(u,v,u2−v2).
3. Calculate the Gaussian and mean curvature of the given surface
y1=u1 cosu2
y2=u1 sinu2
y3= f (u2 ).
4. When the equation of surface S, referred to a set of orthogonal Cartesian

axes, is taken in the form y 3 = f ( y 1 , y 2 ) ,
in parametric form y 1 = u1 , y 2 = u 2 , y 3 = f (u1 , u 2 ) ,
and partial derivatives of f ( y 1 , y 2 ) are denoted by
f y1 ≡ p , f y 2 ≡ q , f y1 y1 ≡ r , f y1 y 2 ≡ s , f y 2 y 2 ≡ t ,
then the coefficients of ααβ in ds 2 = aαβ duα du β
a11 = 1 + p 2 , a12 = pq, a22 = 1 + q 2

and the coefficients of bαβ of the second fundamental form are
r s t
b11 = 2 2
, b12 = 2 2
, b22 = .
1+ p + q 1+ p + q 1 + p2 + q2

(i) Show these.
(ii) Compute ααβ and bαβ for the surface of the sphere: y 3 = a 2 − ( y 1 )2 − ( y 2 )2 .
11
Curves on a Surface
11.1 Introduction
We assumed in last chapter that if the surface coordinates are uα and space
coordinates xi, then the equations of a surface are given by
S: xi = xi (u1, u2). (11.1)
If the coordinates uα are given as functions of a parameter, then the

point represented by these coordinates describes a curve on the surface
as the parameter varies. If we shall take the arc length s of the curve as the
arc parameter, then the equations of a smooth curve C lying on surface S is
C: uα = uα (s). (11.2)
If the values of uα (s) are inserted in (11.1), we obtain the space coordi-
nates xi of C in the form
S: xi = xi (u1, u2),
S: xi = xi (u1, u2 = xi (u1(s), u2(s)) = xi(s), (11.3)
which is the equation of C regarded as a space curve.

The properties of C can then be studied with the aid of the Serret-Frenet
formulas by analyzing the rates of the unit tangent vector λ, the unit princi-
pal normal μ, and the unit binormal v. Then, its curvature χ and its torsion
τ are connected with these vectors by Serret-Frenet formulas as
349
δλ i
= χµ i χ > 0
δs
δµ i
= τ v i − χλ i
δs
δ vi
= −τµ i .
δs
11.2 Curve on a Surface: Theorem of Meusnier

If we regard C as a surface curve, defined by equation (11.2), the compo-
nents λα of unit tangent vector λ are related to the space components λi of
the same vector by the formulas
dxi ∂ x i d uα
i
λ = = α ≡ x ia λ α , (11.4)
ds ∂ u ds

d uα
where λ α = .
ds
The unit surface vector ηα normal to λα and χg is geodesic curvature of C,

as shown in Figure (11.1), so that
λ
θ C
P
v
η µ
Figure 11.1
Curves on a Surface 351
∂ λα
= χ g ηα , (11.5)
∂s
where ηα is the unit normal to C in the tangent plane to the surface and χg
is the geodesic curvature of C.
If we differentiate (11.4) intrinsically with respect to s, we obtain
δλ i du β δλ α .
= xαi ,β λ α + xαi
δs ds δs
or x µ = xα ,β λ λ + xα χ g η (using the Frenet Formula and (11.5))
i i α β i α
The space components ηi of η are η i = xαi ηα and the Gauss formula is

x = bαβ ni .
i
α ,β
The equation becomes
χμi = bαβ ni λαλβ + χg ηi, (11.6)
where ni is the unit normal to the surface S.

Equation (11.6) states that the principal normal μ to C lies in the plane
of the vectors n and η. Since n, η and λ are orthogonal, n × η = λ.
We get єijk nj ηk = λi. (11.7)
and since λ is orthogonal to the plane of n and μ,
∴ μ × n = − sinθ. λ. (11.8)
or μ × n = єijk μj nk = − sinθ. λ
(θ is the angle between n and μ and μ × n is a vector along λ)

On multiplying (11.8) by χ, we get
єijk μj nk χ = − χsinθ λi.
Substituting the value of χμi from (11.6) in above equation,
єijk nk (bαβ nj λα λβ + χg ηj) = − χsinθ λi,

but єijk nj nk = 0 and єijk ηj nk = − λi
χg (− λi) = − χsinθ λi.
We have
χg = χsinθ. (11.9)
On the other hand, if we form the scalar product of both members of

Equation (11.6) with ni and note that
ni. μi = cosθ
(since, ni ni = 1, ni. ηi = 0),
ni. χμi = ni. bαβ ni λα λβ + ni. χ g ηi
χcosθ = bαβ λα λβ. (11.10)
The invariant bαβ λα λβ in (11.10) has the same value for all curves on S
with the same tangent vector λ at P. In particular, it has this value for the
curve formed by the intersection of the normal plane containing n and λ,
but every normal plane section angle θ is either 0 or π radians, so the nor-
mal plane section is
χcosθ = χ or – χ.
Since the rhs of (11.10) is an invariant, the value of χcosθ for every curve
C, tangent to λ is equal to the curvature χ(n) of the normal section in the
direction λ.
The curvature χ(n) is called the normal curvature of the surface S in the
direction λ.
We can thus write (11.10) as
χ(n) = bαβ λα λβ, (11.11)
where χ(n) = χcosθ.

According to (11.6), we can write
χμi = χ(n) ni. + χ g ηi.

This equation states that χ(n) and χg are the components of the curvature
vector χμi in the directions of the vectors n i and ηi.
11.2.1 Theorem of Meusnier

1
The radius of curvature R = of any curve at a given point on the surface
χ 1
is equal to the product of the radius of curvature R(n ) = of the corre-
χ (n )
sponding normal section at that point by the cosine of the angle between
the normal to the surface and the principal normal to the curve.
In symbols, we have R = ± R(n) cosθ.
If S is a sphere, every normal section is a great circle of the sphere and
if C is any circle drawn on the sphere, then the preceding result becomes
obvious from the elementary geometry consideration, as shown in Figure
(11.2).
This is known as Meusnier’s Theorem.
duα
We know ds 2 = aαβ duα du β and = λα .
ds
We see that the formula
bαβ duα du β 
χ (n ) = = . (11.12)
aαβ duα du β 

Result 11.2.1. If the surface is a plane, the normal curvature χ(n) = 0 at all points
1
of the plane, and if it is a sphere, χ (n ) = , where R is the radius of the sphere.
R
n
R = Rn cos θ
θ
Rn
Figure 11.2
Accordingly, we conclude from (11.12) that for the plane bαβ = 0 and for the
1
sphere bαβ duα du β = aαβ duα du β that aαβ = Rbαβ at all points of the sphere.
R
1
Therefore, for the sphere bαβ = aαβ ,
R
It is implied that bαβ and aαβ are proportional.
Result 11.2.2. Since ηi is perpendicular to λi, it lies in the plane con-

taining ni and μi. It is tangent to the surface and the angle between μi
π
and ηi is − θ .
2
Multiplying (scalar) (11.6) by ηi, we get
χµ i = bαβ ni λ α λ β + χ g η i
π 
χ cos −θ = χg (11.13)
2 
∴ χ g = χ sin θ.

Result 11.2.3. From (11.12), we have
bαβ duα du β 
χ (n ) = = .
aαβ duα du β 

Squaring and adding (11.11) and (11.13), we get
(χ g)2 + (χ(n))2 = (χ)2.
Also, we have χμi = χ(n) ni. + χ g ηi.

Multiplying it by gijμj,
g ij µ jχµ i = g ij µ jbαβ λ α λ β ni . + χ g η i g ij µ j
π 
χ = bαβ λ α λ β cosθ. + χ g cos −θ
2 
= χ (n )cosθ + χ g sin θ.

Example 11.2.1. If θ is the angle between the principal normal and the
surface normal, show that
μi = cosθni + sinθηi.
Solution: From (11.6) we have
χμi = bαβ ni λα λβ + χ g ηi,
using (11.13) and (11.10), i.e., χcosθ = bαβ λα λβ and χ g = χ sin θ
χμi = χcosθ ni + χ sin θ ηi
or μi = cosθ ni + sin θ ηi.
Example 11.2.2. Prove that the normal curvatures in the direction of the
b b
coordinate curves are 11 and 22 , respectively.
a11 a22
α β
Solution: We know the unit vector λ(1) along u1- curve and λ(2) along
u2- curve are
1 α 1
α
λ(1) = β
δ (1) and λ(2) = β
δ (2) , respectively.
a11 a22
Therefore, (11.11) gives
α β
χ (n ) = bαβ λ(1) λ(1)
1 α 1 β b11
χ (1n ) = bαβ δ (1) δ (1) =
a11 a11 a11
1 α 1 b
χ (2n ) = bαβ δ (2) β
δ (2) = 22 .
a22 a22 a22
Example 11.2.3. If the surface is plane, show that bαβ = 0.

Solution: If the surface is plane, then any normal section at any point of the
plane is a straight line and therefore its curvature is zero, i.e., χ(n) = 0
∴ χ(n) = bαβ λα λβ = 0
or bαβ duα duβ = 0 for all directions of duα (here duα is arbitrary).
Thus, for a plane, bαβ = 0.
Example 11.2.4. If a curve is a geodesic on the surface, prove that it is

either a straight line or its principal normal is orthogonal to the surface at
every point and conversely.
Solution: From (11.9), χ g = χ sinθ.
Since the curve is geodesic, χ g = 0
∴χ sinθ = 0.
Thus, either χ = 0 or sinθ = 0.

If χ = 0, the curve is a straight line and if sinθ = 0, then θ = 0 or π.
Hence, the principal normal and the surface normal are colinear and the
principal normal is orthogonal to the surface at every point.
Conversely it follows immediately.
Example 11.2.5. Find the normal curvature of the right helicoid.

Solution: The parametric representation of right helicoids is given by
x1 = u1 cos u2, x2 = u1 sinu2 and x3 = cu2
or r = (x1, x2, x3) = (u1 cos u2, u1 sinu2, cu2).
2 2 2
yi yi y1 y2 y3
a11 (cos u2 )2
u1 u1 u1 u1 u1
(sin u2 )2 1
2 2 2
yi yi y1 y2 y3
a22 ( u1sinu2 )2
u2 u2 u2 u2 u2
(u1cosu2 )2 c 2 ( u1 )2 c 2
yi yi y1 y1 y2 y2 y3 y3
a12 os u2 ( u1 sin u2 )2
co
u1 u2 u1 u2 u1 u2 u1 u2
(sinu2 ) (u1 cos u2 ) 0
1 0
a 1 2 2
(u1 )2 c 2
0 (u ) c

1  ∂n ∂r ∂n ∂r 
Now, bαβ = −  α . β + β . α 
2 ∂u ∂u ∂u ∂u
1  ∂n ∂r ∂n ∂r  ∂n ∂r
b11 = −  1 . 1 + 1 . 1  = − 1 . 1
2  ∂u ∂u ∂u ∂u  ∂u ∂u
1
=− 1 2 2
(0,0,1).(cos u 2 , sinu 2 , 0) = 0
(u ) + c
∂n ∂r 1
b22 = − 2. 2 =− (ccosu2 , c sin u 2 ,0).
∂u ∂u 1 2
(u ) + c 2
(−u1 sin u 2 , u1cosu 2 , c ) = 0

1  ∂n ∂r ∂n ∂r 
b12 = −  1 . 2 + 2 . 1 
2  ∂u ∂u ∂u ∂u 
1 1
=− [(0,0,1).(−u1 sin u 2 , u1cosu 2 , c )
2 (u1 )2 + c 2
−c
+ (ccosu2 , c sin u 2 ,0).(cos u 2 , sinu 2 , 0) = = b21
(u1 )2 + c 2
b11 b12 −c 2
b = bαβ = = −b12b12 = −
b21 b22 (u1 )2 + c 2

−c 2
| b | (u1 )2 + c 2 −c 2
Gaussian Curvature κ = αβ = 1 2 2 = 1 2 2 2 < 0.
| aαβ | (u ) + c [(u ) + c ]

It is a negative curvature and has minimal surface since H = 0.

−c 2
Thus, χ(1) + χ(1) = 2H = 0 and χ (1) . χ (2) = κ = 1 2 2 2 .
[(u ) + c ]
The normal curvature χ(n) of the given right helicoids is
b du du 2b12du1du2
(n )
a du du a11(du1 )2 a22 (du2 )2
2c 2
du1du2
1 2 2
(u ) c
1.(du ) [(u1 )2 c 2 ](du2 )2
1 2
1 2cdu1du2
1 2 1 2 2 2 2
(u1 )2 c 2 (du ) [(u ) c ](du )
11.3 The Principal Curvatures of a Surface

duα
We know tangent vector λ α = on the surface such that the normal
ds
curvature χ(n) is given by formula
χ(n) = bαβ duα duβ, (11.14)
assuming an extreme value.

Since vector λα is unit a vector, χ(n) in (11.14) has to be maximized sub-
ject to the considering relation
aαβ λα λβ = 1. (11.15)
Following the usual procedure of determining constrained maxima and

minima, we deduce that a necessary condition for an extremum is
bαβ λβ + Λaαβ λβ = 0, (11.16)
where Λ is the lagrange multiplier. If Equation (11.16) is multiplied by λα

and account is taken of relations (11.14) and (11.15), it follows at once that
Λ = − χ(n). Thus, Equation (11.26) for the determination of directions yield-
ing extreme values of χ(n) can be written as
(bαβ − χ(n) aαβ) λβ = 0 (11.17) (α = 1, 2, 3)

The set of homogenous equation (11.17) will possess nontrivial solu-

tions for λβ if the values of χ(n) are the roots of the determinal equation
|bαβ − ϑaαβ =| 0. (11.18)
The quadratic equation (11.18), when written out in expanded form, is
2 b
(a b ) 0, (11.19)
a
where b = | bαβ| and a = | aαβ|.

b
Since the Gaussian curvature κ is given by κ =
1 a
and the mean curvature H ≡ aαβ bαβ , ,
2
we see that Equation (11.19) assumes the form
ϑ2 − 2Hϑ + κ = 0. (11.20)
The roots ϑ = χ(1) and ϑ = χ(2) of (11.20) are called the principal curva-
α α
tures of the surface and the directions λ(1) and χ (2) , corresponding to these
extreme values of χ(n), are the principal directions of the surfaces.
From (11.20) it is clear that the principal curvature χ(1) and χ(2) are related
to mean and Gaussian curvatures by the formulas:
χ(1) + χ(2) = 2H
χ(1).χ(2) = κ (11.21)
From (11.16), it follows that the principal directions are determined by
β
(bαβ − χ (1)aαβ ).λ(1) =0
β
(bαβ − χ (2)aαβ ).λ(2) =0

α α
If the first of these equations is multiplied by λ(2) and the second by λ(1)
and the results subtracted, we obtain
α β
( χ (1) − χ (2) )aαβ λ(1) λ(1) = 0. (11.22)
If χ(1) ≠ χ(2), Equation (11.22) gives us
β β
aαβ λ(1) λ(2) = 0. (11.23)

That is, the principal directions are orthogonal. If the extreme values of
χ(n), are equal at a given point, then every direction is a principal direction,
and the point is called an umbillic point.
Theorem 11.1. At each point of a surface, there exist two mutually orthog-
onal directions for which the normal curvature attains its extreme values.
11.3.1 Umbillic Point

A point at which principal curvatures of the surface are equal, i.e., χ(1) = χ(2),
is called an umbillic or naval point. From Equations (11.20) and (11.21),
χ(1) + χ(2) = 2H
χ(1).χ(2) = κ.
Therefore, 2 χ(1) = 2H and (χ(2))2 = κ
H 2 = κ.
2
1 
From (10.56) and (10.57), aαβ bαβ = κ,
2 
b
or (aαβ bαβ )2 = 4 ,
a
b
or (a11b11 + a 22b22 + a12b12 + a 21b21 )2 = 4 [ a11 = aa 22 ,
a
a12 = −aa 21 , a22 = aa11 ],
or 4a(a11b12 − a12b11 )2 + [a11 (a11b22 − a22b11 ) − 2a12 (a11b12 − a12b11 )]2 = 0.
Since aij dxi dxj is positive definite and a is positive,
a11 b12 − a12 b11 = a11 b22 − a22 b11 = 0,
b11 b12 b22 .

implying = = (11.24)
a11 a12 a22
Thus, at the umbillic point, we have the above condition (11.24). We know
bαβ duα du β .
χ (n ) =
aαβ duα du β

α
Therefore, χ(n) is independent of the direction of du .
ds
At the umbillic point, the normal curvature is the same in every direc-
tion. At all other points where χ(1) ≠ χ(2), we have
α β
aαβ λ(1) λ(2) = 0. (11.25)

At each point of a surface that is non-umbillic, there exist two mutually

orthogonal directions for which the normal curvature attains its extreme
values.
11.3.2 Lines of Curvature

A curve on a surface such that the tangent line to it at every point is directed
along a principal direction is called a line of curvature.
From (11.17), the equations for determination of directions yielding
extreme values of χ(n) are
(bαβ− χ(n) aαβ)λβ = 0
or bαβ λβ = χ(n) aαβλβ.

du β
For eliminating χ(n) from these equations and setting λβ = :
ds
α = 1 gives b1β λ β = χ (n )a1β λ β

α = 2 gives b2 β λ β = χ (n )a2 β λ β .

b1β λ β a1β λ β
We get =
b2 β λ β a2 β λ β
b1β du β a1β du β
= ,
b2 β du β a2 β du β
b11du1 + b12du 2 a11du1 + a12du 2
or = ,
b21du1 + b22du 2 a21du1 + a22du 2
or (b11a12 − b12a11 )(du1 )2 + (b12a22 − b22a12 )(du 2 )2 + (b11a22
− b22a11 )du1du 2 = 0 (11.26)
Equation (11.26) represents the lines of curvature of the surface.
11.3.3 Asymptotic Lines

The directions on the surface given by
bαβ λα λβ = 0 (11.27)
are called asymptotic directions and the curves whose tangents are asymp-
totic directions are called the asymptotic lines of the surface.
Example 11.3.1. Show that a straight line on a surface is an asymptotic
line.
Solution: As the curve is a straight line, its curvature is κ = 0.
From χcosθ = bαβ λα λβ,
implying that bαβ λα λβ = 0.
Thus, the straight line is an asymptotic line.
Example 11.3.2. Show that the parametric curves are asymptotic lines if
b11 = b22 = 0.
Solution: Let the curves be asymptotic.
We have,
bαβ λα λβ = 0. (i)
α β
For u1 − curve, bαβ λ(1) λ(1) = 0,
α 1 α β 1 β
using λ(1) = δ (1), λ(1) = δ (1),
a11 a11
1 α 1 β
∴ bαβ δ (1) δ (1) = 0,
a11 a11
1 1 1 1
or b11 δ (1) δ (1) = 0,
a11 a11
b
or 11 = 0 ⇒ b11 = 0.
a11
α β
Similarly, for u2 −curve, bαβ λ(2) λ(2) = 0.
b22
From this equation, we get = 0 ⇒ b22 0.
a22
Conversely, suppose b11 = b22 = 0.
Using (i), it gives
b11 λ1 λ1 + b22 λ2 λ2 + b21 λ2 λ1 + b12 λ1 λ2 = 0
or b12 λ1 λ2 = 0, or b12 du1 du2 = 0.

Here, du1 du2 = 0 as b12 ≠ 0.
The curves are parametric.
Example 11.3.3. Show that the coordinate lines are asymptotic lines if
a12 = b12 = 0.
Solution: At each point of S where either bαβ duα duβ ≠ 0 or is not propor-
tional to aαβ duα duβ, then (11.26) specifies two orthogonal directions,
du 2
= φα (u1 , u 2 ) , (11.28)
du1
which coincide with the direction of the principal curvatures. Each equa-
tion of (11.27) determines a family of curves on S covering the surface
without gap. These two families of curves are orthogonal and, if they are
taken as a parametric net on S, the first fundamental form has the form
ds 2 = a11 (du 1 )2 + a22 (du 2 )2 .

According to (11.26), in the coordinate system u α takes the form
−b12a11 (du 1 )2 + b12a22 (du 2 )2 + (b11a22 − b22a11 )du 1du 2 = 0
and its solutions are u 1 = constant , and u 2 = constant

If we take du 1 ≠ 0 and du 2 0 , we see that b12 = 0, since a11 = 0.
Thus, a necessary condition for the net of lines of curvature to be orthog-
onal is that a12 = b12 = 0.
Consequently, we may always choose coordinates u1, u2 on S so that the
lines of curvature are the coordinate curves of this system which is allow-
able at any point of S which is not umbilic.
Conversely, if a12 = b12 = 0, then equation (11.26) has the solutions
u1 = constant, u2 = constant,
so that the coordinate lines are the lines of curvature.
Example 11.3.4. Show that the parametric curves are the lines of curvature
if a12 = b12 = 0.
Solution: Let us assume that parametric curves are the lines of curvature.
Then, they form an orthogonal net and will satisfy (11.26),
(b11 a12 − b12 a11) (du1)2 + (b12 a22 − b22 a12) (du2)2 + (b11 a22 − b22 a11)
du1 du2 = 0.
For u1 curve du2 = 0.
∴b11 a12 − b12 a11 = 0 (since du1 ≠ 0) (i)
and for u2 curve du1 = 0,
b12 a22 − b22 a12 = 0, as du2 ≠ 0. (ii)
Multiplying (i) by a22, and (ii) by a11, and adding these two equations,
we get
a22 (b11 a12 − b12 a11) + a11 (b12 a22 − b22 a12) = 0,
or a22 b11 a12 − a11 b22 a12 = 0,
or a12 (a22 b11 – a11 b22) = 0,
Implying that
either a12 = 0 or a22 b11 – a11 b22 = 0.
For the parametric curve, a22 b11 ≠ a11 b22 and so a12 = 0.
The condition a12 = 0 is that of orthogonality satisfied by all lines of
curvature.
Similarly, multiplying (i) by a22 and (ii) by a11, and adding these two
equations, we get
b12 (a22 b11 – a11 b22) = 0
⇒ b12 = 0.
Conversely, let a12 = b12 = 0.

(b11 a22 – b22 a11) du1 du2 = 0.
This is true if the lines of curvature b11 a22 ≠ b22 a11
du1 du2 = 0.
Hence, when du1 ≠ 0 and du2 = 0 ⇒ u2 = constant.

When du2 ≠ 0 and du1 = 0 ⇒ u1 = constant, the parametric curves are
the lines of curvature.
Example 11.3.5. Show that the lines of curvature on a minimal surface

form an isometric system.
Solution: If the parametric curves are the lines of curvature, then
a12 = b12 = 0.
1
For a minimal surface, H = aαβ bαβ = 0.
2
⇒ aαβ bαβ = 0
a a a
or (a11 b11 + a12 b12 + a 22 b22 + a 21 b21 = 0. a11 = 22 , a 22 = 11 , a12 = 21 
 a a a
Here, a12 = b12 = 0
a22 a
∴ b11 + a12 b12 + 11 b22 + a 21 b21 = 0
a a
1
or (a22b11 + a11b22 ) = 0
a
⇒ b11 = b22 = 0
∴ bαβ = 0.
Therefore, the surface is a plane, i.e., the surface is isometric with the
Euclidean plane.
Example 11.3.6. Find the principal curvature of the surface defined by
x1 = u1 cos u2 ,x2 = u1 sin u2 ,x3 = f(u1).
Find the condition that it is a minimal surface.

Solution: The parametric representation of the surface is
x 1 = u1 cos u 2 , x 2 = u1 sin u 2 , x 3 = f (u1 )

or r = (u1 cos u 2 , u1 sin u 2 , f (u1 ))
2
 ∂f 
ds = (dy ) + (dy ) + (dy ) = (du ) + (u ) (du ) +  1  (du1 )2
2 1 2 2 2 3 2 1 2 1 2 2 2
 ∂u 
  ∂ f 2
=  1 +  1   (du1 )2 + (u1 )2 (du 2 )2 .
  ∂u  
The first fundamental magnitudes aαβ are

2
f
a11 1 1 ( f )2 , a22 (u1 )2 , a12 a21 0
u1
a11 0
a a11a22 (u1 )2[1 ( f )2 ],
0 a22
2
f f
where f1 and f11
u1 u u2
1
a22 1 1 1 1
a11 2
, a22 1 2
, a12 a21 0.
a11a22 a11 1 (f ) a22 (u )
 ∂ x1 ∂ x 2 ∂ x 3 
Here, we have A =  1 , 1 , 1  = (cos u 2 , sin u 2 , f1 ) and
 ∂u ∂u ∂u 
 ∂ x1 ∂ x 2 ∂ x 3 
B =  2 , 2 , 2  = (−u1 sin u 2 , u1 cos u 2 ,0),
 ∂u ∂u ∂u 
A × B = (− f1u1 cos u 2 , − f1u1 sin u 2 , u1 ),
A× B 1
and n = = 1 (− f1u1 cos u 2 , − f1u1 sin u 2 , u1 )
| A × B | u 1 + ( f1 )2
1
= 2
(− f1 cos u 2 , − f1 sin u 2 ,1).
1 + ( f1 )

The symmetric covariant tensors bαβ are
1  ∂n ∂r ∂n ∂r 
bαβ = −  α . β + β . α 
2  ∂u ∂u ∂u ∂u 
1  ∂n ∂r ∂n ∂r  ∂n ∂r f11
b11 = −  1 . 1 + 1 . 1  = − 1 . 1 =
2  ∂u ∂u ∂u ∂u  ∂u ∂u 1 + ( f1 )2
1  ∂n ∂r ∂n ∂r  ∂n ∂r u1 f1
b22 = −  2 . 2 + 2 . 2  = − 2 . 2 =
2  ∂u ∂u ∂u ∂u  ∂u ∂u 1 + ( f1 )2
1  ∂n ∂r ∂n ∂r 
b12 = −  1 . 2 + 2 . 1  = 0 = b21.
2  ∂u ∂u ∂u ∂u 
f11
0
1 + ( f1 )2 u1 f1 f11
Therefore, b = = .
u1 f1 1 + ( f1 )2
0
1 + ( f1 )2
The equation of the principal curvature is
b
χ p2 − (aαβ bαβ )χ p + = 0
a
 f11 f1  u1 f1 f11
or χ p2 −  3 + 2 
χ p + = 0.
2 2 u 1
1 + ( f1)
u1[1 + ( f1 )2 ]2
 (1 + ( f1 ) ) 

f11 f1
Here, χ (1) = 3
and χ (2) = .
2 2 u 1 + ( f1 )2
1
(1 + ( f1 ) )
If ρ1 and ρ2 are the corresponding radii of curvatures, then
3
1 (1+( f1 )2 ) 2 u1 1 + ( f1 )2
ρ1 = = and ρ2 = .
χ (1) f11 f1

The condition for the surface is minimal if H = 0,
or 2H = χ(1) + χ(2) = 0,
f11 f1
+ =0
or 3
2 2 u 1+ ( f1 )2
1
(1+ ( f1 ) )
f1 (1 + ( f1)2) + u1 f11 = 0.
Example 11.3.7. Find the principal curvature of the surface defined by
x1 = u1, x2 = u2, x3 = f(u1, u2).
Solution: The parametric representation of the surface is
x1 = u1, x2 = u2, x3 = f(u1, u2),
or r = (u1, u2, f(u1, u2)).
2 2
 ∂ xi  ∂f 
a11 =  1  = 1 +  1  = 1 + f12
 ∂u   ∂u 
2 2
 ∂ xi  ∂f 
a22 =  2  = 1 +  2  = 1 + f 22
 ∂u   ∂u 
∂ xi ∂ xi ∂f ∂f
a12 = 1 2 = 1.0 + 0.1 + = f1 f 2 = a21
∂u ∂u ∂ u1 ∂ u 2
a11 a12 1 + f12 f1 f 2
∴a = = = 1 + f12 + f 22
a21 a22 f1 f 2 1+ f 2
2
1 + f 22 1 + f12 f1 f 2
a11 = 2 2 , a 22
= 2 2
, a12 =
1 + f1 + f 2 1 + f1 + f 2 1 + f12 + f 22
Now, for calculating the second fundamental bαβ

 ∂ x1 ∂ x 2 ∂ x 3   ∂ x1 ∂ x 2 ∂ x 3 
let A =  1 , 1 , 1  = (1,0, f1 ) and B =  2 , 2 , 2  = (1,0, f 2 )
 ∂u ∂u ∂u   ∂u ∂u ∂u 
∴ A × B = (− f1 , − f 2 ,1)
A× B 1
n= = (− f1 , − f 2 ,1)
| A × B| 1 + f12 + f 22
1 n r n r
b . .
2 u u u u
1 n r n r n r
b11 1
. 1 1
. 1 1
. 1
2 u u u u u u
1
1, 0, f1 [((1 f12 f 22 ) 2 ( f11 , f12 , 0)
3
1
( f1 , f 2 ,1) (1 f12 2
2
f ) 2( f1 f11
2 f2 f12 ) ]
2
f11
1
2 2 2
(1 f 1f ) 2
1 n r n r n r f 22
b22 2
. 2 2
. 2 2
. 2 1
2 u u u u u u
(1 f12 f 22 ) 2
1 n r n r f12
b12 1
. 2 2
. 1 1
b21
2 u u u u 2 2 2
(1 f 1 f )
2
1 f11 f12 1
f11 f 22 f122 2 2 2
b where k (1 f 1 f )
2
k 2 f12 f 22 1 f12 f 22

The equation of principal curvature is
b
χ p2 − (aαβ bαβ )χ p + = 0
a
 f 2 f + f 22 f 22 + 2 f1 f 2 f12 
or χ p2 − ( f11 + f 22 ) − 1 11 2 2
2
 χ p + f11 f 22− f 12 = 0.
 1 + f1 + f 2 

There is a quadratic equation in χp which gives the values of χ(1) and χ(2).
Example 11.3.8. Show that any curvature on a sphere is a line of curvature.

Solution: The parametric representation of a sphere of radius a is given
by
x1 = asinu1 cos u2, x2 = asinu1 sinu2 and x3 = acosu1.
Therefore, r = (asinu1 cos u2, asinu1 sinu2, acosu1).
The second order tensors aαβ for the first fundamental form are
ds2 = (dx1)2 + (dx2)2 + (dx3)2 = (acosu1 cos u2 du1 − asinu1 sinu2 du2)2
+ (acosu1 sin u2 du1 + asinu1 cosu2 du2)2 + (− asinu1 du1)2
ds2 = a2 (du1)2 + a2 sin2 u1 (du2 )2
∴ a11 = a2, a22 = a2 sin2 u1.
a12 = a21 = 0
The second fundamental form are given
b11 = −a,b22 = −aSin2u1 b12 = b21 = 0
Since a12 = a21 = 0, then the given curve has the line of curvature.
b b12 b22 1
Also, 11
a11 a12 a22 a
That is, bαβ are proportional to ααβ
Now, (b11 a12 − b12 a11)(du1)2 + (b11 a22 − b22 a11)du1 du2 + (b12 a22 − b22 a12)
(du2)2
= (−a Sin2u1 + a Sin2u1) = 0

Implies that any curve on a sphere is a line of curvature
Example 11.3.9. Find the equation of principal curvatures of the given
surface (right helicoid)

and show that it is a minimal surface.
Solution: The parametric surface is

∴ r = (u1 cos u2, u1 sinu2, cu2)
The first fundamental form of the surface
yi yi
a
u u
2 2 2
yi yi y1 y2 y3
a11
u1 u1 u1 u1 u1
(cos u2 )2 (sin u2 )2 1
2 2 2
yi yi y1 y2 y3
a22
u2 u2 u2 u2 u2
( u1sinu2 )2 (u1cosu2 )2 c 2 (u1 )2 c 2
yi yi y1 y1 y2 y2 y3 y3
a12
u1 u2 u1 u2 u1 u2 u1 u2
cos u2 ( u1 sin u2 ) (sin u2 ) (u1 cos u2 ) 0
1 0
a 1 2 2
(u1 )2 c 2
u)
0 (u c
(u1 )2 c 2 1
a11 1, a22 , a12 a21 0.
(u1 )2 c 2 (u )1 2
c2
Here, r = (x1, x2, x3) = (u1 cos u2, u1 sinu2, cu2)
 ∂x1 ∂x 2 ∂x 3 
We have A =  1 , 1 , 1  = (cos u 2 , sinu 2 ,0);
 ∂u ∂u ∂u 
 ∂x1 ∂x 2 ∂x 3 
B =  2 , 2 , 2  = (−u1 sin u 2 , u1cosu 2 , c )
 ∂u ∂u ∂u 
A × B = (csinu 2 , −c cos u 2 , u1 )
1
n= 1 2 2
(csinu 2 , −c cos u 2 , u1 ).
(u ) + c
1  ∂n ∂r ∂n ∂r 
Now, bαβ = −  α . β + β . α 
2  ∂u ∂u ∂u ∂u 
1 n r n r n r
b11 1
. 1 1
. 1 1
. 1
2 u u u u u u
1 1n r n r2 n r
b11 1
. (01 , 0,1).(cos 1
. u1 , sinu2 , 01) . 0 1
2(u1 )2u c 2 u u u u u
1
n r(0, 0,1).(cos1u2 , sinu2 , 0) 2 0
b22 (u1 )2 2 . c 2 2 (ccosu , c sin u2 , 0).( u1 sin u2 ,
u u 1
(u ) c 2 2
n r 1
b22 u1cosu22., c ) 2 0 (ccosu2 , c sin u2 , 0).( u1 sin u2 ,
u u 1
(u ) c 2 2
1 2 n r n r
b12 u cosu , c )1 . 0 2
1
2
. 1
2 u u u u
1 n 1r 1 .n r
b12 . [(0, 0,1).( u1 sin u2 , u1cosu2 , c)
2 u1 2u2 (u1 )2u2 c 2 u1
1 1
2 , u cosu , c )c
1 2 1 2
2 2[(0, 0,1).( u 2 sin u
(ccosu
2 (u ) c 1, c2sin u
2 , 0 ).(cos u , sinu , 0) b21
1 2 2
(u ) c
c 2
(ccosu2 , c sin u2 , 0).(cos c u2 , sinuc 2 , 0) c b21
b b11b22 b12b21 0.0 . (u ) 1 2 c 2 2
1 2
1 2 2 1 2 2
(u ) c (u ) c (u )2 c
c c c
ba b(11ub2)2 bc12b21 0.0
1 2 2
.
(u1 )2 c 2 (u1 )2 c 2 (u1 )2 c 2
1 1 11 22 12
a (u1 )2 c 2 H a b (a b11 a b22 2a b12 )
2 2
1 1 H 1a b 1c 11
(a b11 a022. b22 2a12b12 )
1.0 .0 2.00.
2 2
2 1 2
(u ) c 2
(u ) c 2
1 2
1 1 c
1.0 1 2 2
.0 2.00. 0.
2 (u ) c (u ) c 2
1 2
Since the mean curvature is H=0 at every point of the given surface, it
is a minimal surface.
The equation of principal surface is
b
χ p2 − (aαβ bαβ )χ p + = 0.
a
Here, aαβ bαβ = a11b11 + a 22b22 + a12b12 + a 21b21

1
= 1.0 + .0 + 0 + 0 = 0
(u ) + c 2
1 2
c2
b (u1 )2 + c 2 c2
and = − 1 2 = − .
a (u ) + c 2 [(u1 )2 + c 2 ]2
c2
The equation is χ p2 − =0
[(u1 )2 + c 2 ]2
c
χ (1)'s = ± .
(u1 )2 + c 2
Hence, 2H = χ(1) + χ(2) = 0.

Therefore, the surface is a minimal surface.
Example 11.3.10. Give a surface of revolution S,
y1 = rcos ϕ. y2 = rsin ϕ, y3 = f(r)
with f(r) of class C2. Prove that the lines of curvature on S are the meridi-
ans, ϕ = constant, and the parallels are r = constant.
Solution: Here, u = (rcos ϕ., rsin ϕ, f(r))
ds 2 = (1 + f12 )(dr )2 + r 2 (dφ )2

a11 = 1 + f12 , a22 = r 2 , a12 = a21 = 0
a = r 2 (1 + f12 ).
 ∂ y1 ∂ y 2 ∂ y 3 
We have A =  , , = (cosφ , sinφ , f1 );
 ∂r ∂r ∂r 
 ∂ y1 ∂ y 2 ∂ y 3 
B= , , = (−r sinφ , rcosφ ,0)
 ∂φ ∂φ ∂φ 
A × B = (−rf1 cosφ , −rf1 sinφ , r )
A× B 1
n= = (− f1cosφ , − f1sinφ ,1)
|A × B| 1 + f12
∂n ∂u
b11 = ⋅
∂r ∂r
1 1
= −(cos φ ., sin φ , f1 )  ( − f11cosφ , − f11sinφ ,0) − 2 f1 f11
2 2
 1 + f1
−3
2
(1 + f ) (− f1 cosφ − f1 sinφ ,1)
1
2
f11
=
1 + f12
∂n ∂u rf1
b22 = − ⋅ = , b12 = b21 = 0
∂φ ∂φ 1 + f12
b11 b12 rf1 f11

b= = b11 , b22 = .
b21 b21 1 + f12
The equation of the line of curvature is
(b11a12 − b12a11 )(dr )2 + (b12a22 − b22a12 )(dφ )2 + b11a22 − b22a11 )drdφ = 0

or (b11a22 − b22a11 )du1du 2 = 0
 f11 2 rf1 2 
 1 + f 2 r − 1 + f 2 (1 + f1 ) drdφ = 0
 1 1 
∴drdφ = 0 or r = constant or φ = constant .
11.4 Rodrigue’s Formula

A line of curvature is characterized by
∂ ni dx i
+ χ (n ) 0,
∂s ds
where χ(n) is the principal curvature of the surface.

Proof: From the Weingarten formula, we have
n,iα = −a βγ bβα xγi .,

duα
i βγ i du
α
 i duα ∂ ni duα ∂ ni 
or n ,α= −a bβα xγ . n,α ds = ∂ uα ds = ∂ s  ,
ds ds
i
∂n
or = −a βγ bβα xγi .λ α . (11.29)
∂s

We know from the line of curvature
bαβ
χ (n ) =
aαβ
or bαβ λ β = aαβ χ (n )λ β and bαβ λ α = aαβ χ (n )λ α .
∂ ni
Now, = −a βγ (bβα λ α )xγi . = −a βγ (aαβ χ (n )λ α )xγi .
∂s
= −δ αγ χ (n )λ α xγi . = − χ (n )λ α xαi .
duα ∂ x i dx i
= − χ (n ) = − χ ( n )
ds ∂uα ds
i i
∂n dx
or = − χ (n )
∂s ds
i i
∂n dx
∴ + χ (n ) = 0.
∂s ds
Conversely, let
∂ ni dx i
+ χ (n ) = 0.
∂s ds
Using the Weingarten formula, we have
n,iα = −a βγ bβα xγi .

∂ ni
= −a βγ bβα xγi .λ α
∂s
dx i
χ (n ) − a βγ bβα xγi .λ α = 0 (substituting the above result).
ds
Taking the inner product with g ik x kp , we get
dx i
χ (n ) g ik x kp − a βγ bβα g ik x kp xγi .λ α = 0
ds
and
dx k
χ (n ) g ki x kp − a βγ bβα aγ p λ α = 0.
ds
Interchanging dummy indices i and k in the 1st term, we get
∂ x k duα
χ (n ) g ik x kp − bpα λ α = 0,
∂ uα ds
or − bpα λ α + χ (n ) g ik x kp xαk λ α = 0,
or − bpα λ α + χ (n )a pα λ α = 0,
or ( χ (n )a pα − bpα )λ α = 0,

which is the equation of the line of curvature.

The Rodrigues formula is characteristic for lines of curvature.
Theorem 11.4.1. (Euler’s Theorem on Normal Curvature)

If the lines of curvature are not indeterminant at a given point P on the sur-
face and if θ is the angle between a given direction and a principal direction
at P, then the normal curvature is given by the formula
χ(n) = χ(1) cos2 θ + χ(2) sin2 θ.
Proof: We assume that P is not an umbilic point. If the parametric curves

are taken as the lines of curvature, then the principal curvature is given by
b11 b
χ (1) = and χ (2) = 22 .
a11 a22
Let θ be the angle between a given direction (δu1, δu2) and the principal
direction at a given point P. Since the coordinate curves are orthogonal, we
have
du1 du 2
cosθ = a11 and sinθ = a22 .
ds ds
Also, let χ(1), χ(2) be the principal curvatures at P. If the parametric curves
are taken as lines of curvature, then the normal curvature at P is given by
2 2
b11 (du1 )2 + b22 (du 2 )2  du1   du 2 
χ (n ) = = b11  + b22 
ds 2  ds   ds 
2 2
 cosθ  b  sinθ 
= b11  + 22 
 a11   a22 
b11 2 b
= cos θ + 22 sin 2θ
a11 a22
= χ (1)cos 2θ + χ (2)sin 2θ .

This theorem states that the normal curvature corresponding to any

direction can be simply represented in terms of the principal curvatures.
The total curvature in this case is given by
b11 b22 1 b b 
κ= and H =  11 + 22  .
a11 a22 2  a11 a22 
We conclude that the lines of curvature on a minimal surface form an

isometric system.
11.5 Exercises
1. Find the principal curvature of the surface defined by
x1 = u1 cos u2, x2 = u1 sin u2, x3 = f(u2).

Hence, find the Gaussian and the mean curvature.
2. Find the principal curvature of the surface defined by
(
x 1 = u1 cos u 2 , x 2 = u1sinu 2 and x 3 = a log u1 + (u1 )2 − a 2 )
3. ind the conditions for the meridians to be lines of curvature of
F
the helicoid
x1 = u1 cos u2, x2 = u1 sinu2 and x3 = f(u1) + cu2.

12
Curvature of Surface
12.1 Introduction
We discuss the determination of the shape of the surface S in a neighbor-
hood of any of its points considering arbitrary curves on S. We enable
restriction of S and the planes which are orthogonal to the tangent plane
to S at a point P. We introduce the normal Curvature χ (n) and consider the
behavior of χ (n) as a function of direction du2: du1 of the tangent plane to S.
We also study the Gauss-Bonnet theorem with geometric interpretation of
a formula that suggests an alternative definition of Gaussian curvature and
introduces a few concepts from geometry of n-dimensional metric mani-
folds, which are of interest in applications to dynamics and relativity.
12.2 Surface of Positive and Negative Curvature

The first fundamental form is positive definite (since ds2 = aαβduαduβ),
hence the sign of χ (n) depends on the second fundamental form only.
A surface at all points of which the Gaussian curvature κ (> 0) is posi-
tive, is called a surface of positive curvature,
b b22b11 − b 212
i.e., κ = = ⇒ b22b11 − b 212 > 0 as a > 0 on the surface
a a
1
and since χ (n) = bαβ λα λβ, we see that the principal radii R(n ) = to all
χ (n )
normal sections of the surface have positive curvature and do not differ in
sign. It will differ in sign if κ < 0.
Then, the equation
bαβ λα λβ = 0 (12.1)
defines two directions for which the radii of curvatures are infinite.
381
A surface at all points of which the Gaussian curvature κ (< 0) is neg-

ative is called a surface of negative curvature. If κ = 0 at a given point, the
directions given by (12.1) coincide and for this direction R is infinite.
From geometrical considerations, it is clear that ellipsoid, biparted
hyperboloid, and elliptic paraboloids are surfaces of positive curvature and
hyperboloids of one sheet and hyperbolic paraboloids are surfaces of neg-
ative curvature.
A Point on S is said to be elliptic if the signs of the principal curvatures
χ(1), χ(2) are the same. It follows that χ(n) at an elliptic point does not change
for any direction of the normal section. A point is hyperbolic if χ(1), χ(2) have
opposite signs. At a hyperbolic point there are two directions for which
χ(n) = 0. A point is parabolic if one of the values of χ(1) or χ(2) is zero. In spe-
cial cases, if χ (1) = χ(2) and all values of χ(n) are equal, such points are called
spherical or umbilical.
In the neighborhood of a spherical point, the surface looks like a sphere
and we can prove that if all points of S are spherical, then the surface is a
sphere.
Example 12.2.1. Show that the helicoids y1 = u1cosu2
y2 = u1sinu2
y3 = au2 are a surface of negative curvature.
Solution: Here, y1 = u1cosu2, y2 = u1sinu2 and y3 = au2
or r = (u1cosu2, u1sinu2, au2)

2 2
 ∂ yi   ∂ yi 
a11 =  1  = 1, a22 =  2  = a 2 + (u1 )2 ,
 ∂u   ∂u 
∂ y1 ∂ y1 ∂ y 2 ∂ y 2 ∂ y 3 ∂ y 3
a12 = 1 2 + 1 2 + 1 2 = 0 = a21 ,
∂u ∂u ∂u ∂u ∂u ∂u
2 2 1 2
a = a11a22 − a12 = a + (u ) > 0
(cosu2 , sinu 2 ,0) × (−u1sinu2 , u1cosu2 , a)
n=
(u1 )2 + a 2
1
= 1 2 2
(asinu 2 , −a cos u2 , u1 )
(u ) + a
Curvature of Surface 383
∂n ∂r  1 1 2u1 2 1 
b11 = − 1
. 1 = −  (0,0,1) − 2 
3 (asinu , − a cos u , u ) 
∂u ∂u 1 2
(u ) + a 2 2 1 2 2 2
 [(u ) + a ] 
((cosu2 , sinu 2 ,0) = 0
∂n ∂r −a
b22 = − . = 0 and b12 = − 1 2 2
∂u 2 ∂u 2 (u ) + a
a2
b = b11b22 − b12 2 = − ,
(u1 )2 + a 2
| bαβ | a2
κ= = − 1 2 2 2 < 0 and the given surface is a surface of neg-
| aαβ | [(u ) + a ]
ative curvature.
12.3 Parallel Surfaces

Definition 12.3.1. Let S be a smooth surface defined by equations
yi = yi(u1, u2), (i = 1, 2) (12.2)
where the coordinates yi are orthogonal Cartesian. A surface S is deter-

mined by the equations
y i (u1 , u 2 ) = y i (u1 , u 2 ) + hni (u1 , u 2 ), (12.3)
where ni is a unit normal to S and h is the distance measured along the

normal n, called a parallel surface to S.
Parallel surfaces figure prominently in the theory of elastic plates and
shells, where relations connecting the Gaussian curvatures κ and the mean
curvature H of S with the corresponding invariants for the surface S are
important.
12.3.1 Computation of aαβ and bαβ

We have
∂ yi
aα = bi , (12.4)
∂uα
where the base vector aα along the curves uα = constant and base vector bi
along the yi − axes.
∂ yi ∂yi
We introduce the notations α = yαi and α = yαi .
∂u ∂u
Differentiating (12.3), we get
yαi = yαi + hn,iα (12.5)
Multiplying ni on both sides,
yαi ni = yαi ni + hn,iα ni ,

but yαi ni = 0 for aα is orthogonal to n and n,iα ni = 0 since nini = 1

[aα lies on curves uα = constant]
∴ yαi ni = 0. (12.6)

Also, the unit vector ni to S is orthogonal to the base vector yαi on S ,

so that
yαi ni = 0. (12.7)
From (12.6) and (12.7), it is implied that ni and ni are collinear and
since they are unit vector
ni = ni ,
the metric coefficients aαβ of S are given by
aαβ = yαi y βi .
Using (12.5) in the above equation,
a ( yi hn,i )( y i hn,i )
( yi yi hn,i y i hn,i y i h 2n,i n,i ).

Applying the Weingarten formula in this expression,
n,iα = −aδγ bδα yγi

aαβ = ( yαi y βi + h(−aδγ bγα xγi ) y βi + h(−aδγ bγβ xγi ) yαi + h 2n,iα n,iβ )
= aαβ − 2hbαβ + h 2n,iα n,iβ [since yαi y βi = aαβ ]
aαβ = aαβ − 2hbαβ + h 2n,iα n,iβ (12.8)

The last term of the right hand side of (12.8) can be expressed in
terms of Gaussian curvature κ and the mean curvature H. Also using the
Weingarten formula, we get
n,iα n,iβ = aδγ bαδ yγi a λµbβλ y µi (Weingarten formula: n,iα = −aδγ bγα xγi )
= aδγ bαδ a λµbβλ aγµ , since yγi y µi = aγµ
n,iα n,iβ = a µλ bαµbβλ . (12.9)

The Gauss Equation is
Rαβγδ = bβδbαγ − bβγbαδ,
where Rαβγδ = κєαβ єγδ.

κєαβ єγδ = bβδ bαγ − bβγ bαδ.
Multiplying both sides by aαδ, summing on δ
κєαβ єγδ aαδ = bβδ bαγ aαδ − bβγ bαδ aαδ,
Using aαδ єαβ єγδ = −aβγ
−κaβγ = aαδ bβδ bαγ − 2Hbβγ, (12.10)

1 αδ
since H = bαδ a
2
Substituting the right-hand side of the 1st term of (12.10) from (12.9),
we get
− κaαβ = n,iα n,iβ − 2 Hbαβ [changing suffices α , β in the place of β and γ ]

or n,iα n,iβ = −κaαβ + 2 Hbαβ . (12.11)

Now we can write (12.8) by substituting the values of (12.11):
a a 2hb h2 ( a 2Hb )
or a a (1 h 2 ) 2hb (1 hH ). (12.12)

The above formula (12.12) enables us to compute aαβ at a given point
P (u1 , u 2 ) on S with the values of aαβ, bαβ, κ, H at the P (u1, u2) on S.
To compute the coefficients of bαβ on S and using Equation (10.36),
yαi ,β = bαβ ni or bαβ = yαi ,β ni

and we get from (12.5)
yαi = yαi + hn,iα .

Differentiating y i , yi , hn,i ,
so that b y i , ni
( yi , hn,i ) ni
y i , ni hn,i ni
b b hn,ia ni . (12.13)

Since the coordinates yi are rectangular Cartesian, ni = ni, nini = 1 we
have n,iα ni = 0.
On differentiating this orthogonality relation, we find
n,iαβ ni = −n,iα n,iβ

(12.13) becomes
bαβ = bαβ + hn,iαβ ni

= bαβ + h(−n,iα n,iβ ) nini
= bαβ − hn,iα n,iβ .
Using (12.11) in the above equation, i.e., using n,i n,i a 2Hb ,
bαβ = bαβ − hn,iα n,iβ

= bαβ − h(−κaαβ + 2 Hbαβ )
= bαβ (1 − 2Hh) − κaαβ h
∴bαβ = bαβ (1 − 2 Hh) − κaαβ h (12.14)

The result of the mentioned monograph by T. Y. Thomas is
κ
κ= 2
1 + h κ − 2hH
(12.15)
H − hκ
and H= .
1 + h 2κ − 2hH
It follows that when S is a developable surface, then the parallel surfaces

S are also developable.
12.4 The Gauss-Bonnet Theorem

The classical result, however, relates the integral of the Gaussian curvature
evaluated over the area of an arbitrary smooth surface to the line integral
of the geodesic curvature computed over the curve that bounds the area.
Gauss viewed this result as the most important theorem of the geometry
of surfaces at large.
Let D be the region bound by a closed piecewise smooth curve C over
surface S, as shown in Figure (12.1). D is homeomorphic to a circular disc.
We know that a unit tangent vector λα to a surface curve C is related to the
unit vector ηα to λα by
αi
C
Figure 12.1
δλ α
= χ g ηα . (12.16)
δs
where χg is geodesic curvature of C. Moreover, if C is a geodesic, then

χg = 0 at all points of C.
Conversely, since ηα is orthogonal to λα, it follows
αβ ηα λ β = 1
or ηα = αβ λ β
δλ α δλ α
∴χg = ηα = αβ λ β .
δs δs (12.17)
Now, integration over C is expressed as
δλ α
∫ C χ g ds = ∫ C αβ λ β ds (12.18)
δs
and when the line integral in the right-hand member of (12.18) is trans-
formed into a surface integral by Green’s formula, we get
∫ C χ g ds = − ∫ ∫ C k dσ + 2π − ∑(π − α i ), (12.19)

where the αi are interior angles of the contour C and dσ = adu1du 2 is the
element of the surface area of D. If C is smooth, the sum is Σ (π – αi) = 0.
Formula (12.19) gives us the statement of the Gauss-Bonnet theorem.

With the aid of Green’s theorem, we give a geometrical interpretation of
(12.19) as an alternative definition of Gaussian curvature.
Consider a sphere S of radius R and a spherical triangle P1 P2 P3 on S, as
shown in Figure (12.2), formed by the arcs P1 P2, P2P3, P1 P3 of three great
circles. We denote that the interior angles made by them at P1, P2, and P3
are α1, α2, and α3, respectively and cover the sphere by some coordinate net
(u1, u2).
If the base vector along the u1 −coordinate line at P is a1 and A (P1) is an
arbitrary surface vector at P1, we denote the angle between a1 and A (P1) as
φ. Let θ be the angle between A (P1) and geodesic arc P1 P2. When A (P1)
is propogated in a parallel manner along the geodesic triangle P1 P2 P3, it
assumes the position Aʹ (P1). Its object is to determine the angle φʹ between
a1 and Aʹ (P1).
During the parallel propagation of A (P1) along P1 P2, if the angle θ is
unchanged and vector A assumes the position A (P2) with geodesic arc P2
P3, then
A(P3)
β
γ
P3
α3
A’(P1)
α1
’
P1 A(P1)
A(P2)
θ
β
α2
a1 θ
P2
Figure 12.2
β = π – (α2 + θ).
In the course of parallel propagation of A (P2) along P2 P3, vector A con-

tinues making angle β with arc P2 P3 and assumes position (P3). Let γ be the
angle between A (P3) and geodesic arc P1 P3.
∴ γ = α3 – β = α3 − [π − (α2 + θ)] = α3 + α2 + θ − π
On continuing propagations of A along P1 P3, vector A maintains angle

γ with P1 P3 until it reaches point P1 when it assumes position Aʹ (P1). Now
φʹ is made by a1 and Aʹ (P1) is
φʹ = γ + α1 + φ − θ
= α3 + α2 + θ – π + α1 + φ − θ
= α1 + α2 + α3 + φ − π.
Now, the angle between A (P1) and Aʹ (P1) is φʹ − φ
φʹ − φ = α1 + α2 + α3 −π. (12.20)
This difference of angles is called the spherical excess of spherical trian-

gle P1 P2 P3.
If instead of interior angles αi we introduce the exterior angles θi = π − αi,
(12.20) becomes
ϕ ′ − ϕ = [3π − (θ1 + θ 2 + θ 3 )] − π = 2π − ∑ni =1 θ i .
When vector A is propagated along a geodesic polygon of n sides,

entirely similar computations yield for spherical excess of a polygon.
ϕ ′ − ϕ = ∑ni =1 α i − (n − 2)π (sum of interior angles of polygon of n sides
= (n − 2)π ) or ϕ ′ − ϕ = 2π − ∑ni =1 θi , if using exterior angles θi = π – αi, but
it is known for spherical trigonometry that spherical excess of a geodesic
polygon is equal to σ2 , where σ is an area of a polygon and R is the radius
of a sphere. R
σ
∴ ∴ ϕ ′ − ϕ = 2π − ∑ni =1 θ i =
R2
1
Since Gaussian curvature κ = ,
R2
2π − ∑ni =1 θ i (12.21)
κ= .
σ
This formula can be generalized to obtain the Gauss-Bonnet formula

(12.19) for the case where C is a geodesic polygon.
Thus, if the region D (Figure 12.12) is subdivided by a geodesic polygon
into regions of area dσ, then
∫ ∫ D kd σ = 2π − ∑in=1 θ i , (12.22)
which coincides with (12.19) since χg = 0 when C is a geodesic polygon.

Formula (12.22) was first obtained by Gauss, which is generalized
by Bonnet to yield (12.19), which is directly derived from (12.18) using
Green’s formula.
The left-hand side of (12.22) ∫ ∫ D κd σ is called the integral curvature of
D. It turns out that integral curvature is a topological invariant. Two sur-
faces are said to be topologically equivalent if they can be mapped into one
another by a continuous one to one transformation.
It can be shown by using (12.19) that
∫ ∫ D κd σ = 4π for all regular surfaces topologically equivalent to a
sphere
and ∫ ∫ D κd σ = 0 for all regular surfaces topologically equivalent to a
trous.
12.5 The n-Dimensional Manifolds

In this section, we introduce a few concepts from geometry of n-
dimensional metric manifolds which are of interest to apply in dynamics
and relativity. Here, many of these concepts are direct generalizations of
ideas introduced in this chapter related to the study of surfaces imbedded
in three-dimensional Euclidean manifolds.
The elements of distance between two neighboring points in n-
dimensional manifolds is
ds2 = gij dxi dxj (i, j = 1, 2,…n) with gij ≠ 0. (12.23)

The space is Euclidean if there exists a transformation of coordinates xi

such that the transformation of ds2 is a quadratic form with constant coef-
ficients since every real quadratic form with constant coefficients can be
reduced by a real linear transformation to the form
ds2 = λi (dxi)2, where λi = ± 1. (12.24)
This form of (12.24) can be used to define a Euclidean n-dimensional

manifold.
If, in particular, the form (12.24) is definite, we shall say that the man-
ifold is purely Euclidean, but if it is indefinite, the manifold will be called
pseudo-Euclidean.
A linear manifold determined by a set of n equations
C: xi = xi (t), t1 ≤ t ≤ t2
with suitable differentiability properties will be said to define a curve C in
an n-dimensional manifold.
If (12.23) is positive definite, we shall say that the positive number
dx i dx j
s = ∫ tt12 g ij dt is the length of the curve C.
dt dt
dx i
If vector λ i = defines the direction of the curve, then
ds
gij λi λj = 1 (from 12.23),
so that λi is a unit vector. The length of any vector Ai is given by
A = g ij Ai A j .

If λ and μ are two unit vectors, we define the cosine of the angle between
i i
them as
cosθ = gij λi μj. (12.25)
Example 12.5.1. The angle between unit tangent vector λi and normal vec-
tor μj is real if the form gij dxi dxj is positive definite.
Proof: By Cauchy-Schwarz inequality, we can write
(gij xi yj)2 ≤ (gij xi xj) (gij yi yj), (12.26)
where form gij dxi dxj ≥ 0.

Let Q (x) = gij xi xj be positive definite.
If we replace xi with xi + λyi in it where λ is an arbitrary scalar, we obtain
Q (x + λy) = gij (xi + λyi) (xj + λyj)
= gij (xi + λyi) (xj + λyj)
= gij xi xj + 2λxi yj + gij yi yjλ2
= gij xi xj + 2λ gij xi yj + gij yi yjλ2
= Q (x) + 2Q (x, y) λ + Q (y) λ2
Q (x + λy) is a quadratic equation of λ.

It is clear that Q (x + λy) ≥ 0.
The sign of equality holds if xi + λyi = 0.
Hence,
f (λ) Q (x) + 2Q (x, y) λ + Q (y) λ2 = 0
possessing no distinct real roots if
4 [Q (x, y)]2 – 4Q (x) Q (y) ≤ 0
or [Q (x, y)]2 – Q (x) Q (y) ≤ 0.
That is, (gij xi yj)2 ≤ (gij xi xj) (gij yi yj).

This equation establishes (12.26).
If we now set xi = λi and yi = μi, we get
( g ij λ i µ i )2
≤ 1.
( g ij λ i λ j )( g ij µ i µ j )

Since λi and μi are unit vectors, it is concluded that (gij λi μi)2 ≤ 1.

That implies that cos2θ ≤ 1
or − 1 ≤ cosθ ≤ 1.
That is, the angle between λi and μi are real.
12.6 Hypersurfaces
A set of n − equations
xi = xi (u1, u2,….un), (i = 1, 2, ….n), m ≤ n, (12.27)
is said to define a hypersurface over a neighborhood of variables uα if (a) the

i
xi in (12.27) are of class C2 and (b) the Jacobian matrix ∂ xα (α = 1,2,m)
∂u
is of rank m at each point of the neighborhood.
We also know that there are n – m linearly independent normals to Vm. If
we take n = m + 1, then Vm. is said to be the hypersurface of the enveloping
space Vm + 1.
Now, under these circumstances an m-dimensional hyper surface with
a Riemannian metric
ds2 = aαβ duα duβ, (α, β = 1, 2,….m) (12.28)
can be imbedded in an n-dimensional Euclidean manifold with
ds2 = dxi dxi. (12.29)
Now, it requires that
∂ xi ∂ xi
aαβ = and (i = 1,2n). (12.30)
∂ uα ∂ u β
1
The set of m(m + 1) partial differential equations of (12.30) in n-
2 1
variables x will not be expected to possess a solution unless n ≥ m(m + 1).
i
Thus, if m = 2, n ≥ 3, if m = 3, n ≥ 6, and so on. 2

Thus, it is possible to prove that a neighborhood of Rm can be imbedded
1
in En if n ≥ m(m + 1).
2
12.7 Exercises
1. Given a surface of revolution S
y1 = rcos ϕ. y2 = rsin ϕ, y3 = f(r)
ith f(r) of class, show that the points on a surface of revolution S for
w
which f1 f11 > 0 are elliptic, those for which f1 f11 < 0 are hyperbolic, and
if f11 = 0, then S is a cone.
2. Find the nature of a point on a unit sphere
(x1)2 + (x2)2 + (x3)2 = 1.

Part III
ANALYTICAL MECHANICS
13
Classical Mechanics
13.1 Introduction
Classical mechanics originated with the work of Galileo and were devel-
oped extensively by Isaac Newton. It deals with the motion of particles
in a fixed frame of reference. In order to develop the science of mechan-
ics of a universe consisting of more than two particles, it is necessary to
adjoin Newtonian laws with the principle of superposition of effects. The
basic premise of Newtonian mechanics is the concept of absolute time
measurement between two reference frames at a constant velocity. In this
system, other coordinate systems may be used so that the metric remains
Euclidean.
13.2 Newtonian Laws of Motion

Newton’s Laws of Motion are stated in the following form:
1. Everybody continues in its state of rest or of uniform motion

in a straight line until it is compelled by impressed forces to
change that state.
2. The change of motion is proportional to impressed motive
force and takes place in the direction of the straight line in
which that force is impressed.
3. To every action there is always an equal and opposite
reaction.
The first law depends on the dynamical concept of force and the kine-
matical idea of uniform rectilinear motion.
The second law of motion introduces the kinematical concept of motion
and the dynamic idea of force. To understand its meaning, it should be
noted that Newton uses the term motion in the sense of momentum as the
399
product of mass and velocity. Thus, the change of motion means “the rate
of change of momentum with respect to time.”
d(mv )
∴F = (13.1)
dt
If we postulate the invariance of mass then, (13.1) can be written as
dv
F = ma , where a = . (13.2)
dt
d(mv )
If F = 0, then = 0 implies that mv = constant.
dt
Hence, v is a constant vector.
Therefore, the first law is a consequence of the second law.
The third law of motion states that accelerations always occur in pairs.
In terms of force, we may say that if a force acts on a given body, the body
itself exerts an equal and opposite directed force on some other body.
Newton called the two aspects of the force action and reaction.
Newtonian Law is sometimes called inertia to distinguish it from the
Newtonian Law of Gravitation.
This law states that the force of attraction between a pair of particles is
proportional to the product of their masses, is inversely proportional to the
square of the distance between them, and directed along the line joining
the particles,
M1 M 2
F =k r, (13.3)
r3
where k is a universal constant and r is a vector directed from mass M1 to

mass M2.
In order to develop the science of mechanics of a universe consist-
ing of more than two bodies, it is necessary to include the principles of
superposition of effects and make assumptions regarding the nature of
constraints.
Classical Mechanics 401
13.3 Equations of Motion of Particles

Theorem 13.3.1. The work done in displacing a particle along its trajectory
is equal to the change in the kinetic energy of particle.
Proof: Let the position of moving particle P be determined by a vector r.
The equation of the path C of the moving particle can be written as
C: xi = xi (t) (13.4)
and we call the curve C the trajectory of the moving particle.

dr
The velocity of P is a vector v = , whose components are
dt
i dx i
v = (13.5)
dt
d 2r
and acceleration a = and its components
dt 2
vi dv i i dx k
ai vj
t dt j k dt
d2xi i dx j dx k (13.6)
dt 2 j k dt dt

δ vi i
Where is the intrinsic derivative and the second kind of
δt j k
Christoffel symbol calculated from the metric tensor gij.
If the2mass of a particle is m, then by Newton’s Second Law of Motion
d r
F = m 2 , where its components express
dt
δ vi
Fi = m = mai. (13.7)
δt
The first clear introduction of the idea of the energy of mechanics as

a quantity equal to the product of mass and the square of velocity of the
particle was made by Huygens in the 17th century. The full use of this idea
in the concept of work did not come until the 19th century.
We define that the element of work done by the force F is producing a

displacement dr by the invariant dW = F. dr and its components of F and
dr are Fi and dxi respectively.
Therefore, dW = gij Fi. dxj = Fjdxj, (13.8)
where Fj = gij F i.
The work done in displacing a particle along the trajectory C, joining a
pair of points P1 and P2 is line integral:
W = ∫ dW = ∫ PP12 Fi dx i. (13.9)

Now, by (13.7) we get
W = ∫ PP12 g ij F i dx i
δ vi i
= ∫ PP12 g ijm dx
δt (13.10)
i j i
δ v dx δv j
W = ∫ PP12 mg ij dt = ∫ PP12 mg ij v dt,
δ t dt δt

δ ( g ij v i v j ) δ vi j
but = 2 g ij v since gij is an invariant.
δt δt
δ ( g ij v i v j ) d( g ij v i v j )
=
δt dt
and
d( g ij v i v j ) δ vi j
= 2 g ij v.
dt δt
Using this result in (13.10), we get
j
1 d( g ij v i v ) m
W = ∫ PP12 m dt = g ij v i v j |PP12 = T2 − T1,
2 dt 2
m 1
where T = g ij v i v j = mv 2 = Kinetic Energy of a particle.
2 2
Thus, we conclude that the work done in displacing a particle along the
trajectory is equal to the change in the kinetic energy of the particle.
13.4 Conservative Force Field

The force field Fi is such that the integral (13.9) is independent of the path.
Therefore, the integrated Fidxi is an exact differential, i.e.,
dW = Fidxi (13.11)
of the work function W. The negative of the work function W is called the
force potential or potential energy. We get from (13.11),
∂V
Fi = − , (13.12)
∂ xi
where the potential energy V is the function of coordinates xi.

∂V
Hence, fields of force are called conservative if Fi = − i .
∂x
Theorem 13.4.1. A necessary and sufficient condition that a force field Fi,
defined in a simple connected region, is conservative if Fi, j = Fj, i.
∂V
Proof: Suppose Fi is conservative. Then, Fi = − i .
∂x
∂ Fi  k 
Now, Fi , j = −  Fk
∂x j  i j 

 ∂V 
∂ − i   k 
 ∂x  
= −  Fk
∂x j i j
 
∂2 V  k 
=− −   Fk
∂x j ∂x i  i j 

∂ Fj  k 
and Fj ,i = −  Fk
∂x i  j i 

∂2 V  k 
=− −   Fk .
∂x i ∂x j  j i
 

From these two relations, it is implied that
Fi, j = Fj, i.
Conversely, let Fi, j = Fj, i.

Fi k Fj k
Then, Fk Fk
xj i j xi j i
∂ Fi ∂ Fj  k 
or j = i , as   is symmetric due to i , j.
∂x ∂x
 i j 
∂V
Take Fi = −
∂ xi
∂ Fi ∂ 2V ∂ 2V
∴ j =− j i =− i j
∂x ∂x ∂x ∂x ∂x
∂  ∂V 
= i − j 
∂x  ∂x 
∂ Fj
=
∂ xi
∂ Fi ∂ Fj
=
∂ x j ∂ xi
∂V
It implies that we can take Fi = − .
∂ xi
Therefore, Fi is conservative.
13.5 Lagrangean Equations of Motion

Consider a particle moving on the curve C: xi = xi(t) and curve C is the
trajectory of the particle.
At time t, the particle is at point P (xi).
1 2
The kinetic energy T mv
2
(13.13)
m
g ij x i x j ,
2
Since x i = v i .
Differentiating (13.13),
∂T
now = mg ij x j
∂ x i
∂ T m ∂ g ij i j
and = x x
∂ xi 2 ∂ xi
d  ∂T  d j
or  i  = (mg ij x )
dt ∂ x dt
 ∂ g ij 
= m  g ij x j + k x k x j  .
 ∂ x 
Subtracting the above two equations, we get
d  ∂T  ∂T  j ∂ g ij k j  m ∂ g ij i j
 i  − i = m  g ij x + k x x  − x x
dt ∂ x ∂x  ∂x  2 ∂ xi
  1 ∂ g ij 1 ∂ g ik 1 ∂ g jk  j k 
= m  g ij x j +  + − x x 
  2 ∂ x k 2 ∂ x j 2 ∂ x i  

= m( g ij x j + [ jk , i]x j x k )

= mg il ( xl + g il [ jk , i]x j x k )

l
mg il xl x j x k ,
j k

d x 
2 i i  dx j dx k  l 
i l j k
Since a = +   = 
x +   x x (from (13.6)).
dt 2  j k  dt dt  j k 
 
Therefore, in parentheses on the right side of the above equation is
acceleration and since mgilal = mai = Fi, it is a component of the force field.
Therefore,
d  ∂T  ∂T
 − = Fi (13.14)
dt  ∂ x i  ∂ x i
Equation (13.14) is the Lagrangean Equation of Motion.

This equation gives the statement of Newton’s 2nd Law of Motion in the
form of Lagrange.
For a conservative system of forces,
∂V
Fi = − .
∂ xi
d  ∂T  ∂T ∂V
 i  − i = − i
dt ∂ x ∂x ∂x
or
d  ∂ T  ∂ (T − V )
 − = −0 .
dt  ∂ x i  ∂ xi
Since the potential V is a function of the coordinate xi alone, let L = T − V,
and the Lagrangean function
d  ∂ (T − V )  ∂ (T − V )
 − =0
dt  ∂ x i  ∂ xi
d  ∂ L)  ∂ L (13.15)
or  i  − i = 0
dt  ∂ x  ∂ x
Equation (13.15) is known as Lagrange’s Equation of Motion for conser-
vative, holonomic systems.
Example 13.5.1. Show the covariant components of the acceleration vector
in a spherical coordinate system with
ds2 = (dx1)2 + (x1dx2)2 + (x1)2sin2x2 (dx3)2 are
a1 = x 1 − x 1 ( x 2 )2 − x 1 ( x 3 sin 2 x 2 )2
d
a2 = [( x 1 )2 x 2 ] − ( x 1 )2 sin x 2 cos x 2 ( x 3 )2
dt
d
a3 = [( x 1 sin x 2 )2 x 3 ]
dt
Solution: In a spherical coordinate system,
ds2 = (dx1)2 + (x1dx2)2 + (x1)2sin2x2 (dx3)2.
If v is the velocity of the particle,
2 2 2 2
 ds   dx 
1 2 3
2 1 2  dx  1 2 2 2  dx 
v = = + (x )  + ( x ) sin x 
 dt   dt   dt   dt 
= ( x 1 )2 + ( x 1 )2 ( x 2 )2 + ( x 1 )2 sin 2 x 2 ( x 3 )2 .
If T is kinetic energy,
1 1
T = mv 2 = m( x 1 )2 + ( x 1 )2 ( x 2 )2 + ( x 1 )2 sin 2 x 2 ( x 3 )2 ] (i)
2 2
Using the Lagrangean Equation of Motion,
d ∂T ∂T
 − = Fi , where F1 = ma1 .
dt ∂ x 1 ∂ x 1
Take i = 1,
d  ∂T  ∂T
ma1 =  − .
dt  ∂ x 1  ∂ x 1
From i, we have
ma1 =
d ∂ 1
 1{
dt  ∂ x 2 } 
m[( x 1 )2 + ( x 1 )2 ( x 2 )2 + ( x 1 )2 sin 2 x 2 ( x 3 )2 

∂ 1
{ }
− 1 m[( x 1 )2 + ( x 1 )2 ( x 2 )2 + ( x 1 )2 sin 2 x 2 ( x 3 )2 
∂x 2


d 1 1 1 1 2 2
( )
2
a1 =  2 x  − 2 x [( x ) + sin 2 x 2 ( x 3 )2 ] = x1 − x 1 ( x 2 )2 − x 1 sin x 2 x 3 .
dt  2  2
Take i = 2,
d ∂T ∂T
ma2 =  −
dt ∂ x 2 ∂ x 2

From (i), we have
ma2 =
d ∂ 1
 2
dt  ∂ x 2 { } 
m[( x 1 )2 + ( x 1 )2 ( x 2 )2 + ( x 1 )2 sin 2 x 2 ( x 3 )2 

∂ 1
{
− 2 m[( x 1 )2 + ( x 1 )2 ( x 2 )2 + ( x 1 )2 sin 2 x 2 ( x 3 )2 ]
∂x 2
d 1
a2 =  2( x 1 )2 x 2  − ( x 1 )2 sin x 2 cos x 2 ( x 3 )2 ]
dt  2 
d
= [( x 1 )2 x 2 ] − ( x 1 )2 sin x 2 cos x 2 ( x 3 )2
dt
Take i = 3,
d ∂ 1
ma3 = [ 3 { m[( x 1 )2 + ( x 1 )2 ( x 2 )2 + ( x 1 )2 sin 2 x 2 ( x 3 )2 ]
dt ∂ x 2
∂ 1
− 3 { m[( x 1 )2 + ( x 1 )2 ( x 2 )2 + ( x 1 )2 sin 2 x 2 ( x 3 )2 }]
∂x 2
d 3 1 2 2 2
a3 = [x ( x ) sin x ].
dt
Example 13.5.2. Use the Lagrangean equations to show that if a particle is
not subjected to the action of forces, then its trajectory is given by
yi = ait + bi,
where the ai and bi are constants and the yi are orthogonal Cartesian
coordinates.
Solution: If v is the velocity of a particle, then we know
v 2 = g ij y i y j ,

where yi are the orthogonal Cartesian coordinates.
Since gij = 0, if i ≠ j
= 1, if i = j,
so v 2 = ( y i )2
1 1
We know T = mv 2 = m( y i )2 .
2 2
The Lagrangean Equation of Motion is
d ∂T ∂T
− = Fi
dt ∂ y i ∂ y i

Since the particle is not subjected to the action of forces, i.e., Fi = 0,

d ∂T ∂T
therefore, − =0
dt ∂ y i ∂ y i
d
or (my i ) − 0 = 0
dt
d 1
or ( y ) = 0
dt
i i
or y = constant = a
∴ yi = ait + bi.
Example 13.5.3. Prove that if a particle moves so that its velocity is con-
stant in magnitude, then its acceleration vector is either orthogonal to the
velocity vector or it is zero.
Solution: We know
gijvivj = v2 = constant.
Taking the intrinsic derivative,
δ
( g ij v i v j ) = 0,
δt
δ i j δ j i
or g ij  (v )v + (v )v  = 0,
δt δt 
δ i j δ
or g ij (v )v + g ij (v j )v i = 0,
δt δt
δ i j
or 2 g ij (v )v = 0
δt
(interchanging i and j in 2nd term and gij = gji and s is symmetric metric)
δ i j
g ij (v )v = 0
δt
δ i
This shows that the acceleration vector (v ) is either orthogonal to vj
δ i δt
or zero, i.e., (v ) = 0.
δt
13.6 Applications of Lagrangean Equations

We consider several examples of the application of Lagrangean Equations
for determining the trajectories which include the cases of particles mov-
ing on smooth curves and surfaces.
(a) Free-moving particle: If a particle is not subjected to the action of

forces, then we have
d  ∂T  ∂T
− =0
dt  ∂ y i  ∂ y i (13.16)

1 i i
and T = my y , implying that yi = 0.
2
Integrating this, we get that yi = ait + bi represents a straight line.
(b) Simple Pendulum:
Let a pendulum bob of mass m be suspended by a string (Figure 13.1). In
spherical coordinates,
O Y2
ф
R
mg cos ф
Y1 θ P
mg sin ф
Y3 mg
Figure 13.1
ds2 = dr2 + r2dϕ2 + r2sin2ϕdθ2
1 1
and T = mv 2 = m(r 2 + r 2φ 2 + r 2 sin 2φθ 2 ) (i)
2 2
Lagrangean Equation of Motion:
d  ∂T  ∂T
 − = Fi
dt  ∂ x i  ∂ x i
Here, x1 = r, x2 = ϕ, x3 = θ.
1 d  ∂T  ∂T
For x = r ,  − = mgcosφ − R.
dt  ∂ r  ∂ r
R
From i, r + r (φ 2 + sin 2φθ 2 ) = gcosφ − . (ii)
m
2
For x = φ , rφ + 2rφ − rθ sinφcosφ = − gsinφ
2
(iii)

d 2 2 
and x 3 = θ (r sin φθ ) = 0. (iv)
dt
If the motion in one plane is from (ii), (iii), and (iv), by taking θ = 0
. R
r − r φ 2 = gcosφ −
m
rφ + 2rφ = − gsinφ .

 g
If r = 0, we get φ = −   sinφ which is the equation of a simple
r
pendulum.
For small angles of oscillation, the vibration is a simple harmonic. For
large vibrations, the solution is given in terms of elliptic functions.
(c) Constant Gravitational Field
We take a Cartesian reference frame and Y 3-axis to the normal to the plane
of the earth, so the potential V of the constant Gravitational field is = mgy3
if the positive Y 3-axis is directed upward. Therefore,
y1 = 0, y 2 = 0, y3 = − g .

The equation of the path of trajectory is
ya = aat + ba (a = 1, 2)
1
y 3 = − gt 2 + at + b.
2
This trajectory is a parabola whose axis is parallel to the Y 3-axis.

(d) Motion of a Particle on a Curve
Let a particle move on a curve
xi = xi(s) (i = 1, 2, 3), (13.17)
with s being the arc parameter.

We suppose that C has a continuously turning tangent, so the curve is
of C2.
The components vi of velocity vector v of the particle are
dx i dx i ds
vi = = = vλ i , (13.18)
dt ds dt
dx i ds
Where λ i = is the unit tangent vector to C and v = is the mag-
nitude of v. dt dt
Now acceleration vector a, by intrinsically differentiating (13.18), is
δ vi δ v i δλ i dv i δλ i
ai = = λ +v = λ +v , (13.19)
δt δt δ t dt δt
δ v dv
and since v is a scalar, = .
Again, δ t dt
δλ i δλ i ds δλ i
= =v = vκµ i , (13.20)
δ t δ s dt δs
δλ i
using the Serret-Frenet formula = κµ i, κ > 0, where κ is curvature and
μ is principal normal unit. δ s
δλ i
Now, substituting of (13.20) in (13.19), we get
δs
dv i 2 i
ai = λ + v κµ , (13.21)
dt
which states that an acceleration vector a lies in the osculating plane (λi, μi).
2
Here, the component in the direction of the principal normal is, v where
1 R
is R = the radius curvature of C.
κ
The force
dv i
F i = mai = m λ + mv 2κµ i . (13.22)
dt
Since F lies in the osculating plane of the curve, the component of all
external forces normal to this plane is zero. This condition enables us to
compute that reaction R is normal to C, i.e.,
Riλi = 0.
If R = 0, the curve C is called the natural trajectory of the particle.
Example 13.6.1. Establish the energy equation T + V = constant.

Solution: Let a bead of mass m slide under gravity along a smooth curve C
lying in the vertical Y1Y 2 -plane, as shown in Figure (13.2).
Y2
C
R
α λ
µ
F
mg
Y1
Figure 13.2
The force F acting on m is
F = mg + R,
where R is the pressure exerted by the curve on the particle and mg is
the gravitational force. Since the curve is smooth, R is normal to C. If α
is the angle between the direction of R and the Y 2-axis, the components
of F in the directions of the tangent λ and the principal normal μ are
F(λ) = −mgsignα and F(μ) = −mgcosα + R
Using (13.22), we get
dv
m mgsign and mv 2 mgcos R. (13.23)
dt
dy 1 dy 2 dv dv ds
Here, cosa = , sina = and =
ds ds dt ds dt
dv dv dy 2
∴m = mv = −mgsignα = −mg
dt ds ds
dv dy 2
or mv = −mg . Integrating this, we get
ds ds
1 2
mv = −mgy 2 + constant (13.24)
2
Since R is zero in the direction of the path, we can write directly from
(13.24).
T + V = Constan, where V is potential energy.
Example 13.6.2. Let a particle of mass m move under gravity along a

smooth cycloid (Figure 13.3)
y1 = a(θ − sinθ)
y2 = a(1 + cosθ), 0 ≤ θ ≤ 2π (i)
dv dy 2
Solution: The equation m = −mg (from 13.23)
dt ds
d 2s dy 2
or m 2 = −mg . (ii)
dt ds
Here,
s = ∫ θ0 (dy 1 )2 + (dy 2 )2 = ∫ θ0 (dy 1 )2 + (dy 2 )2 dθ = a ∫ θ0 2(1 − cosθ )dθ

θ  θ
= 2a ∫ θ0 sin dθ = 4a 1 − cos
2  2
(dy ) + (dy ) = [adθ − cosθ dθ ]2 + [−asinθ dθ ]2
1 2 2 2
= a 2 (2 − 2cosθ )dθ 2 ),
2 2
 2 θ  θ
or (s − 4a) = −4acos = 16a 2 cos ,
 2   2
Y2
F µ
O λ
2πα Y1
mg
Figure 13.3
2 2
or (s − 4a) = 2a  cos θ  = a(1 + cosθ ) = y 2 .
8a  2

From (ii), we get
dy 2 d  (s − 4a)2  − g −g
s = − g = −g   = 2(s − 4a) = ( s − 4a )
ds ds  8a 8a 4a

g
or s + s=g
4a
The general solution is
 g 
S = C1 cos  t + C 2  + 4a . (iii)
 4a 

The integration constants c1 and c2 are determined by initial conditions

of motion on a cycloid.
 g 
The period of motion  2π /
  is independent of amplitude c1.
4a 
This fact first was discovered by Christian Huygens about 300 years ago.
Huygens proposed the use of a cycloidal pendulum in the construction of
Isochronous clocks.
Using the second equation of (13.23), it is shown that R = 2mgcosα.
(d) Motion of a Particle on a Surface
Let the equations of regular surface S be given in parametric form
S: xi = xi (u1, u2) (i = 1, 2, 3). (13.25)
The force F is the resultant of all external forces acting on the particle,
thus including the reaction R of the surface on the particle. When surface
is smooth, R is normal to S and represents pressure that constrains the
particle to remain on S.
The space component vi of the velocity v of the particle is related to the
surface components vα by the formula
i dx i ∂ x i dua
v = = a = x aiu a ,
dt ∂u dt (13.26)
v = x aiv a
i

where v a = u a
δ vi
The acceleration ai = ,
δt
δ vi δ va δ xi
∴ ai = = x ai + va a (13.27)
δt δt δt
= x aa + v xα ,β v β ,
i a a i
δ va
where aa =
δ t.
We take the Gauss formula
xαi ,β = bαβ ni , substuting in (13.27), we get
ai = xαi aα + v α v β bαβ ni . (13.28)

 α duα ∂ uα dx 
Thus, ai = xαi aα + bαβ ni v 2 λ α λ β  v = = = v λ α  since
the normal curvature κ (n ) = bαβ λ α λ β  dt ∂ x dt 
ai = xαi aα + κ (n )ni v 2 .
Since Fi = mai, we have
F i = mxαi aα + mκ (n )v 2ni . (13.29)
The first term of the right hand side of (13.29) is the component of F in
the tangent plane to S and second term is the component of F along the
normal n.
The component of F in the direction of the normal n is
F ini = mxαi aα ni + mκ (n )v 2nini

(from 13.29)
= 0 + mκ (n )v 2
Fini = mκ(n)v2, (13.30)

i
since surface vector xα is orthogonal to normal ni and nini = 1.
The components of F in the plane tangent to S, are given by
g ij x j F i mg ij x j x i a m g x j v 2ni
(n) ij
ma a 0,

j i j i j
Since g ij xγ xα = aγα and g ij xγ n = 0 because the surface vector xγ are
orthogonal to nj.
Thus, xγj Fj = maγ
j
and let Fγ ≡ xγ Fj . From this we obtain a pair of Newtonian equations
Fγ = maγ (13.31)
relating the surface force vector Fγ to the surface acceleration vector aγ.
It can be written in Lagrangean form,
m 2
Kinetic Energy T = v is
2
m m
T = aαβ v α v β = aαβ u α u β .
2 2
The Lagrangean Equation of motion,
d  ∂T  ∂T
 − = Fα , (13.32)
dt  ∂ u α  ∂ uα
Where Fα is defined in (13.31). When the force is conservative in

∂V
Fα = − α , V is potential energy.
∂u
The velocity vα of the particle along the trajectory is vα = vλα, hence
δ vα dv δλ α
aα = = λα +v
δt dt δt
dv δλ α
= λα + v2 .
dt δs
δλ α
We know = κ g ηα
δs
Where ηα is the unit normal to the trajectory in the tangent plane and κg
is the geodesic curvature, we can write
v dv 2
a v g
t dt
dv 2
v v g
ds
1 dv 2 α
Therefore, aα = λ + v 2κ g ηα .
2 ds
It follows from the result of (13.22),
dT α
Fα = λ + 2Tκ g ηα (taking unit mass)
ds
dT
If the vector Fα vanishes identically, then = 0 and κg = 0 along the
trajectory. ds
The first of these equations states that v = constant and if v ≠ 0, then
the trajectory is a geodesic [curve on a surface S if its geodesic curvature
is zero].
Example 13.6.3. Let a particle of mass m be constrained to move under

gravity on a smooth paraboloid of revolution.
Solution: The equation of a paraboloid of revolution, as illustrated in Figure
(13.4), is
1
y3 [( y1 )3 ( y 2 )2 ], a constant (13.33)
4a
Y3
R
C
F
mg
Y2
θ
Y1
Figure 13.4
Introducing cylindrical coordinates (r, θ, z), the coordinates are
y1 = rcosθ, y2 = rsinθ, and y3 = z.
Substitute in (13.33) and we get
r2
z= (13.34)
4a
1
and kinetic Energy T = my i y i becomes
2
1  r2  
T = m  1 + 2  r 2 + r 2θ 2 
2   4a  

The potential energy is
r2
V = mgy 3 = mg
4a
Since the surface is smooth and the reaction R is normal to S’,

we use the Lagrangean Equation of Motion
d  ∂T  ∂T ∂V
 − = Fα , with Fα = − α (since components of R in
dt  ∂ u α  ∂ uα ∂u
tangent plane to S are zero) and parametrize the surface by setting u1 = r,
u2 = θ.
For α = 1
d  ∂T  ∂T ∂V
 1  − 1 = − 1 ,
dt ∂ u ∂u ∂u
d  ∂T  ∂T ∂V
or 
  − =−
dt ∂ r ∂r ∂r
d  r2   rr 2 r
mr  1 + 2   − m 2 = −mg
dt  4a  4a 2a
 r 2  rr 2 r
r 1 + 2  + 2 − rθ 2 = − g
 4a  4a 2a
 r2  rr 2 gr
 1 + 2  r + 2 − rθ = −
2
4a 4a 2a

For α = 2
d 2
(r θ ) = 0 (13.35)
dt
The 2nd equation of (13.35) gives
r 2θ = h, constant (13.36)

The 1st equation becomes
 r2  rr 2 h 2 gr
 1 + 2

r + − 3 =− , (13.37)
4a  4a r2
2a
which has a unique solution with initial values.

If the particle is constrained to move on a horizontal circle r = constant,

then (13.37) becomes
h2 gr gr 4
h2
r3 2a 2a
and
gr 4
h h2 g
θ = 2 ⇒ θ 2 = 4 = 2a4 = ,
r r r 2a
so that the angular velocity θ is independent of the radius of the circle.

If we replace the surface of the paraboloid in this example by the surface
of the sphere, we have the problem of a spherical pendulum.
13.7 Himilton’s Principle

Consider a particle of mass m in a three-dimensional Euclidean manifold,
referring to a curvilinear system of coordinates X. The particle is in motion
under the effect of force F and our problem is to determine the trajectory.
C: xi = xi (t), (i = 1, 2, 3) t1 ≤ t ≤ t2,
where t denotes the time.
Theorem 13.7.1. If a particle is at point P1 at time t1 and is at P2 at a time t2,

then the motion of a particle takes place in such a way that
∫ tt12 (δ T + Fiδ x i )dt = 0, (13.38)
where xi = xi (t) are the coordinates of the particle along a trajectory and
xi + δxi are the coordinates along a varied path beginning at P1 at time t1
and ending at P2 at a time t2.
Proof: Consider a particle moving on the curve
C: xi = xi (t), (i = 1, 2, 3) t1 ≤ t ≤ t2
1
The kinetic energy of the particle, T = mg ij x i x j .
2
i i
Here, T is a function of ( x , x ). Let C′ be another curve,
joining at t1 and t 2 to show that if C is
C ′ : x i (, t ) = x i (t ) + δ x i (t ),
with δ x i (t ) = ξ i (t ) and ξ i (t1 ) = ξ i (t 2 ) = 0,

then
∂T i ∂T i
δ (T ) = δ x + i δ x . (13.39)
∂ x i ∂x
 ∂T ∂T 
Now, ∫ tt12 (δ T + Fiδ x i )dt = ∫ tt12  i δ x i + i δ x i + Fiδ x i  dt
 ∂ x ∂x 
∂T i t2 ∂ T
= ∫ tt12 δ
i x dt + ∫ t1
δ x i dt + ∫ tt12 Fiδ x i dt
∂ x ∂ xi
t
∂T i
t2  ∂T i
2
d  ∂T 
=∫ δ x dt +  δ x  − ∫ tt12  i  δ x i dt
dt  ∂ x 
t1
i i
∂x  ∂ x t1
+ ∫ tt12 Fiδ x i dt .
t2
 ∂T 
Since δ x i (t1 ) = 0,δ x i (t 2 ) = 0, then  i δ x i  = 0
 ∂ x t1
∂T i t2 d  ∂ T 
∴∫ tt12 (δ T + Fiδ x i )dt = ∫ tt12 i δ x dt − ∫ t1
i t i
 i  δ x dt + ∫ t12 Fiδ x dt .
∂x dt ∂ x
 ∂T d  ∂T  
⇒ ∫ tt12 (δ T + Fiδ x i )dt = ∫ tt12  i −  i  + Fi δ x i dt
 ∂ x dt  ∂ x  
We know the Lagrangean Equation of Motion
d  ∂T  ∂T
 − =F
dt  ∂ x i  ∂ x i
∂T d  ∂T 
or −   = − Fi , (13.40)
∂ x i dt  ∂ x i 
so ∫ tt12 (δ T + Fiδ x i )dt = 0 is proven.

In particular, if this field is conservative, then there exists a potential

function V (x1, x2, x3), such that
∂V
= − Fi .
∂ xi
From (13.38) we get ∫ tt12 (δ T + Fiδ x i )dt = 0,
t ∂V i
or ∫ t12 (δ T − δ x )dt = 0,
∂ xi
or
∫ tt12 (δ T − δ V )dt = 0 or ∫ tt12 δ (T − V )dt = 0. (13.41)
We defined the Lagrangean Function L = T −V, so that we can write
∫ tt12 δ L dt = 0.

Since the limits of integration are fixed, we have a concise formulation

of Himilton’s principle for a conservative field in the form
δ ∫ tt12 L dt = 0. (13.42)
We can state from this result:

In a conservative field of force, a particle moves so that the integral
∫ t1 L dt , evaluated along the trajectory xi = xi(t), t1 ≤ t ≤ t2 has a stationary
t2
value in comparison with its values for all neighboring paths beginning at
point P1 at t = t1 and ending at point P2 at t = t2.
Equation of Motion (13.15)
d  ∂L  ∂L
 − = 0 follows at once from Formulation (13.42).
dt  ∂ x i  ∂ x i
Theorem 13.7.1. Integral of Energy: The motion of a particle in a conser-

vative field of force is such that the sum of its kinetic and potential energies
is a constant.
Consider a particle moving on the curve
C: xi = xi(t), t1 ≤ t ≤ t2,
where t denotes the time. The kinetic energy T of the particle is given by
1
T = mg ij x i x j ,
2
since Kinetic energy is invariant,
dT δ T δ  1 
= = mg ij x i x j
dt δ t δ t 2  
1 δ
= mg ij ( x i x j )
2 δt
1  δ x i j δ x j i 
= mg ij  x + x  (13.43)
2  δt δt 
δ x i j
= mg ij x = mg ij ai v j
δt
dT
= mai v i ,
dt
where ai is the acceleration and vj is the velocity of the particle.

∂V
For a conservative field of force, Fi = − i = mai , substituting in (13.43),
we get ∂x
dT ∂V ∂ V dx i
= mai v i = −v i i = − i .
dt ∂x ∂ x dt (13.44)
∴ dT = − dV
dt dt
Integrating this, we get T + V = h, where h is the constant of integration.
13.8 Principle of Least Action

In 1744, Euler showed that the integral ∫ mv ds has a stationary value
along the trajectory of a particle moving in a central field of force. In
1760, Lagrange extended Euler’s result by demonstrating that the inte-
P
gral A = ∫ P12 mv.ds has a stationary value along the trajectories of particles
moving in a conservative force field, provided that the constraints are not
functions of time. This lead him to formulate the principle of least action.
Himilton, attempted to understand Lagrange’s formulation of the principle.
Let us consider the integral
A = ∫ PP21 mv.ds (13.45)

evaluated over path
C: xi = xi(t), t1 ≤ t ≤ t2
Where C is the trajectory of the particle of mass m moving in a conser-

vative field of force. Here, we suppose that neither kinetic energy T nor
potential energy V is a function of time t. In a three-dimensional space
with curvilinear coordinates, the integral (13.45) can be written as
A = ∫ PP21 mv.ds
P2 dx i j
= ∫ mg ij
P1 dx
dt
dx i dx j
= ∫ tt (( PP21)) mg ij dt.
dt dt
1 dx i dx j
Here, Kinetic Energy T = mg ij
2 dt dt
∴ A = ∫ tt (( PP12 )) 2Tdt (13.46)
The integral has a physical meaning only when evaluated over the tra-
jectory C, but its value can be computed along any varied path joining the
points P1 and P2. Let us consider a particular set of admissible paths Cʹ
along which the function T + V, for each value of parameter t, has the same
constant value h. The integral A is called the action integral.
The principle of least action states that “of all curves of Cʹ passing through
P1 and P2 in the neighbourhood of the trajectory C, which are traversed at a
rate such that, for each C ʹ, for every value of t, T + V = h, that one for which
the action integral A is stationary is the trajectory of the particle.”
When stated in the form of the variational equation, the principle reads
∫ tt (( PP12 )) 2Tdt = 0, (13.47)
with the auxiliary condition
T + V – h = 0 on Cʹ. (13.48)
We construct a function = 2T + λϕ, where ϕ = T +V − h and determines

the solution of the system of four equations
 ∂F d  ∂F  ∂F
 −   = 0, (i = 1,2,3)
 ∂ x i dt  ∂ x i  ∂ x i
 T +V − h = 0

An investigation of this system shows that λ(t) = −1 and it follows from

this fact that the trajectory C is determined by the solution of the system
d  ∂T  ∂T ∂V
 i  − i = − i (i = 1,2,3) . (13.49)
dt ∂ x
 ∂x ∂x
These are the Lagrangean equations of motion.

We consider this variational problem with fixed end points by change of

variable. Since the kinetic energy
2
1 1  ds 
T = mg ij x i x j = m
2 2  dt 
2
 ds  2T ds 2T
or = ⇒ = (13.50)
 dt  m dt m
m m
∴ dt = ds = ds ,
2T 2(h − V )
(13.46) can be written as
m
A = ∫ tt (( PP12 )) 2Tdt = ∫ tt (( PP12 )) 2(h − V ) ds = ∫ tt (( PP12 )) 2m(h − V )ds
2(h − V )
A = ∫ tt (( PP12 )) 2m(h − V )ds (13.51)
Since the integrand in the preceding integral is clearly independent of t,

we now parametrize our varied path C ʹ, so that
C: xi = xi(u) and u1 ≤ u ≤ u2, where P1: xi(u1) and P2: xi(u2)

dx i
and we write ds = g ij x'ix' j du,
du,where x' i = .
du
We can write the action integral (13.51) in the form
A = ∫ uu12 2m(h − V ) g ij x′ i x′ j du. (13.52)
Since the limits of integration in (13.52) are fixed, we see that the deter-
mination of the trajectory is equivalent to finding the geodesics in a three-
dimensional manifold with arc element
dS2 = 2m(h – V) gijdxidxj. (13.53)

13.9 Generalized Coordinates

In the solutions of most of the mechanical problems it is more convenient
to use some other set of coordinates instead of Cartesian coordinates. For
example, in the case of a particle moving on the surface of a sphere, the
correct coordinates are spherical coordinates r, θ, ϕ and are only two vari-
able quantities.
Let there be a particle or system of n particles moving under possible
constraints. For example, a point mass of the simple pendulum or a rigid
body moving along an inclined plane. Then, there will be a minimum
number of independent coordinates required to specify the motion of the
particle or system of particles. The set of independent coordinates suffi-
cient in number to specify unambiguously the system configuration are
called generalized coordinates and are denoted by q1, q2,……qn where n is
the total number of generalized coordinates or degree of freedom.
In a system where the system is defined by n-generalized coordinates q1,
q ,……qn and a number of the degree of freedom that is n, that such system
2
is called an un-connected holonomic system.

Let there be N particles composing a system and let
i
x(α ) (i = 1,2,3),(α = 1,2, N ) be the positional coordinates of these par-
ticles referred to some convenient reference frame in E3. The system of N
free particles is described by 3N parameters. If the particles are constrained
in some way, there will be certain relations among the coordinates x(iα )
and suppose that there are r such independent relations.
f i ( x(1)
1 2
, x(1) 3
, x(1) 1
; x(2) 2
, x(2) 3
, x(2) ;x(1N ) , x(2N ) , x(3N ) ) = 0,(i = 1,2,.r )
(13.54)
If these r equations of constraints in (13.54) can be solved for some r

coordinates in terms of the remaining 3N − r coordinates, the latter can
be viewed as the independent generalized coordinates qi. It is more con-
venient however to assume that each of the 3N coordinates is expressed in
terms of 3N – r = n independent variables qi and written in 3N equations
x(iα ) = x(iα ) (q1 , q 2 ,qn , t ), (13.55)
where we introduce the time parameter t which may enter in the problem
explicitly if one deals with moving constraints. If t does not enter explicitly
in Equation (13.55), the dynamic system is called a natural system.
The velocity of the particles is given by the differential Equation (13.55)

with regards to time. Now,
∂ x(iα ) j ∂ x(iα )
x (iα ) = q + . (13.56)

∂qj ∂t
The time derivative q j of generalized coordinates qj are the generalized

velocities.
For symmetry reasons, it is desirable to introduce a number of superflu-
ous coordinates qi, describing the system with the aid of k > n coordinates
q1, q2,……qk. In this event, there will exist certain relations of the form
f j (q1, q2,……qk, t) = 0, (13.57)

i
so that the quantities qi and q are no longer independent.
Differentiating it, we get
∂fj i ∂fj
q + = 0. (13.58)

∂ qi ∂t
It is clear that they are integrable, so that one can deduce Equation
(13.57) and use it to eliminate the superfluous coordinates.
In some problems, the functional relations of the type
F j (q1 , q 2 ,q k , q 1 , q 2 ,q k , t ) = 0 ( j = 1,2.m) (13.59)
arise, which are non-integrable. If the non-integrable relations in (13.59)

occur in the problems, we shall say that the given system has k − m degrees
of freedom, where m is the number of independent non-integrable relations
in (13.59) and k is the number of independent coordinates. The dynamic
systems involving non-integrable relations in (13.59) are called non-
holonomic to distinguish them from holonomic systems in which the
number of degrees of freedom is equal to the number of independent gen-
eralized coordinates. In other words, a holonomic system is one in which
there are no non-integrable relations involving the generalized velocities.
13.10 Lagrangean Equations in Generalized

Coordinates
Let there be a system of particles which requires n independent gen-
eralized coordinates or a degree of freedom to specify the states of its
particles.
The position vectors xr are expressed as a function of generalized coor-
dinates qi, (i = 1, 2,…n) and the time t, i.e.,
xr = xr(q1, q2,..qn, t)
and assumes that the functions xr(q, t) are of class C2.

The velocity x r of any point of the body is given by
∂ xr ∂q j ∂ xr
x r = +
∂q j ∂t ∂t
∂ xr j ∂ xr
= q + ( j = 1,2,n),

∂qj ∂t
where q j are generalized velocities.

Consider the relation, with n degree of freedom
xr = xr (q1, q2, ..qn) (13.60)
involving n independent parameters qi.

The velocities x r are given by
∂ xr j
x r = q , (r = 1, 2, 3; j = 1, 2,n), (13.61)

∂qj
where q j transform under any admissible transformation.
q k = q k (q1 , q 2 ,..qn ); (k = 1,2,n) (13.62)
in accordance with contravariant law.

The kinetic energy of the system is expressed by
1
T = ∑ mg rs x r x s , (r , s = 1,2,3), (13.63)
2
where m is the mass of the particle located at the point xr and grs are the
components of the metric tensor.
Substituting the values of x r from (13.61) in (13.63), we get
1 ∂ xr ∂ xs i j
T = ∑ mg rs i q q
2 ∂q ∂qj
1
= aij q i q j ,
2
∂ xr ∂ xs
where aij ≡ ∑ mg rs i (r , s = 1.2.3), (i , j = 1,2,n)
∂q ∂qj
1
since T = aij q i q j (13.64)
2
is invariant and the quantities aij are symmetric, we conclude that aij are
components of a covariant tensor of rank two with respect to the transfor-
mations in (13.62) of generalized coordinates.
Since the kinetic energy T is a positive form in the velocities q i ,| aij | > 0 ,
we construct the reciprocal tensor aij.
Using the Lagrangean Equation of Motion articulated in (13.5), we get
d T T l
ail ql q j q k . (13.65)
dt q i qi j k

l
Put ql q j q k Q l in (13.65) and it becomes
j k
d  ∂T  ∂T l (13.66)
 i − i = ail Q = Qi (i = 1,2,n ).
dt  ∂ q  ∂ q

Now, from the relations,

∂ x r ∂ x r ∂ x r ∂ 2 xr j ∂ x r d  ∂ x r 
= , = 
q and = and using equa-
∂ q i ∂ qi ∂ qi ∂ x i ∂ q j ∂ qi dt  ∂ qi 
tions (13.61) and (13.63), now the left-hand side of (13.65) becomes
d  ∂T  ∂T ∂ xr
− = ∑ ma ,
dt  ∂ q i  ∂ qi (13.67)
r
∂ qi

in which aj = gijaj is the acceleration of point P

Newton’s 2nd law gives us
mar = Fr (13.68)
where Fr are the components of F acting on the particle located at point P.

From Equation (13.68), we have
∂ xr ∂ xr
∑ mar = ∑ Fr

∂ qi ∂ qi
and Equation (13.67) can be written as
d  ∂T  ∂T ∂ xr
− = ∑ Fr i .
dt  ∂ q i  ∂ qi ∂q (13.69)

Comparing (13.66) with (13.67), we get
∂ xr ∂ xr
Qi = ∑ mar = ∑ Fr ,

∂ qi ∂ qi
where vector Qi is called a generalized force.

The equations
d  ∂T  ∂T
− = Qi
dt  ∂ q i  ∂ qi (13.70)

are known as Lagrangean equations in generalized coordinates.

The solution of these equations is in the form
C: qi = qi (t).
If there exists a function V(q1, q2,..qn) such that the system is said to be
conservative, for such systems Equation (13.70) assumes the form
d  ∂L  ∂L
− = 0,
dt  ∂ q i  ∂ qi (13.71)

where L = T − V is the kinetic potential.

Since L(q, q ) is a function of both the generalized coordinates and
velocities,
dL  ∂ L  i ∂ L i
= q + i q .
dt  ∂ q i  ∂q (13.72)

∂L d  ∂L 
From (13.71), we get =
∂ qi dt  ∂ q i 
Then, Equation (13.72) becomes
dL  ∂ L  i d  ∂ L  i d  ∂ L i 
= q +  i  q =  i q  . (13.73)
dt  ∂ q i  dt  ∂ q  dt  ∂ q 

i
Since L = T − V but potential energy is not a function of q ,
∂ L i ∂T i
q = i q = 2T ,

∂ q i ∂ q
1
since T = aij q i q J .
2
Thus, Equation (13.73) can be written as
d(L 2T ) d(T V )
0,
dt dt
which implies that T + V = h(constant).
Thus, along the dynamical trajectory, the sum of the kinetic and poten-
tial energies is a constant.
Example 13.10.1. A simple pendulum consists of a bob of mass m sup-
ported by a light inextensible cord of length l. Suppose that the pendulum
is set in vibration in some plane which we take as a Y1 Y 2-plane, as shown
in Figure (13.5).
Solution: The Lagrangean Equations are expressed as
d  ∂T  ∂T
− = Qi .
dt  ∂ q i  ∂ qi (i)

The Kinetic energy is expressed as
1
T = my i y i . (ii)
2
The equations of the pendulum are
 q
 y 1 = lsinθ = lsin
 l

 y 2 = l(1 − cosθ ) = l  1 − cos q  , (iii)
  l

O
Y1
l1
q1 = θ
m1 (y11, y12)
l2
m2
(y21, y22)
Y2
q2 = φ
Figure 13.5
where the arc length q = lθ is a generalized coordinate.

Since
q
y 1 = qlcos

l
q
y 2 = qlcos
 , equation (ii) becomes
l
1
T = m(q )2,
2
The work is expressed as Wδ = −mgsinθδq
q
= −mgsin δ q
l
q
and the generalized force is expressed as Q = −mgsin . Thus, Equation (i)
l
yields
q
q + gsin = 0. (iv)
l
Since, for a small displacement, sinθ = θ,
q
q + k 2q = 0  since sinθ ⇒ θ =  ,
 l
g
where k 2 = .
l
The solution of this equation is
q = acos (kt + α).
13.11 Divergence Theorem, Green’s Theorem,

Laplacian Operator, and Stoke’s Theorem
in Tensor Notation
(i) Divergence Theorem
Let F be a vector point function in a closed region V, bound by the regular
surface S. Then,
∫ V divF = ∫ S F.n ds, (13.74)
where n is unit normal to S.

In orthogonal Cartesian coordinates, the divergence of F is given by
∂ Fi (13.75)
divF = .
∂ xi
If the components of F relative to an arbitrary curvilinear coordinate

system X are denoted by Fi, then the covariant derivative of Fi is
Fi i
F,ij Fk.
xj k j

The invariant F,ij in Cartesian coordinates represents the divergence of

the vector field F.
Also,
F. n = gij Fi nj = Fi ni = gij Fi nj = Fi ni.
Then, (13.74) can be written as
∫ V divF = ∫ S F.n ds = ∫ S F ini ds. (13.76)

(ii) Green’s Theorem

Let u (x1, x2, x3) and v (x1, x2, x3) be two scalar functions of class C2 in V and
of class C1 in the closed region. We denote the gradients of u and v by ui
and vi, respectively.
∂u ∂v
So, ∇u = ui = i and ∇v = v i = i .
∂x ∂x
If we take Fi = uvi,
from the divergence of Fi, we get
F,ii = g ij Fi , j = g ij (uvi , j + u j vi ).

We put this in Equation (13.75) and obtain the desired formula
∫ V g ij (uvi , j + u j vi )dV = ∫ S Fi ni ds = ∫ S uvi ni dS. (13.77)
The invariant gijvi, j appearing on the left-hand side of Equation (13.76),

∂ 2v
when expressed in Cartesian coordinates, is the Laplacian of v , i i and if
∂y ∂y
we denote the Laplacian operator by the symbol ∇2, we can write gij vi, j = ∇2v
and the inner product of gij vid uj can be written as
gij uj vi = ∇u. ∇v,
where we denote ∇ as the operator of the gradient.

From (13.76) we can write
∫ V (ug ij vi , j + g iju j vi )dv = ∫ S uni vi ds ,

or ∫ V (u∇ 2 v + ∇u. ∇v )dv = ∫ S un.∇v ds , (13.78)
2
or ∫ V (u∇ vdv = ∫ S un.∇v ds − ∫ ∇u.∇v dv ,
∂v
where n. ∇v = ni vi = .
∂n
Interchanging u and v in (13.77) and subtracting the resulting formula
from (13.77) yields the symmetrical form of Green’s Theorem
 ∂v ∂u 
∫ V (u∇ 2 v − v∇ 2 u)dv = ∫ S  u − v  ds. (13.79)
 ∂n ∂n 

(iii) Expansion Form of the Laplacian Operator

The Laplacian of v is
gij vi, j = ∇2v. (13.80)
When it is written out in Christoffel symbols associated with curvilinear

coordinates xi, it is
2 ij v 2 k v
v g
x xj
i
i j xk
(13.81)
and as a divergence of vector Fi,
Fi i
F,ii F j. (13.82)
xi j i

i
We know log g , hence divergence F,ii can be written as
j i xj
∂ Fi  ∂ 
F,ii = i + log g  F j
∂x  ∂x j 
=
1 ∂ gF
i
.
( (13.83)
g ∂ xi

∂v
If we put, F i = g ij , we get
∂xj
 ∂v 
∂  g g ij j 
1  ∂x  . (13.84)
∇ 2 v = g ij v j ,i =
g ∂x

It is the expansion form of the Laplacian operator.
(iv) Stoke’s Theorem

Let a portion of regular surface S be bound by a closed regular curve C
and F be any vector point function defined on S and on C. The theorem of
Stoke’s states that
∫ S n. curlF ds = ∫ C F. λ ds, (13.85)
where λ is the unit tangent vector to C and curlF is the vector whose com-
ponents in orthogonal Cartesian coordinates are determined from
e1 e2 e3
∂ ∂ ∂
curlF = = ∇ × F, (13.86)
∂ y1 ∂ y2 ∂ y3
F1 F2 F3

where ei is the unit base vector in a Cartesian frame.
We consider the covariant derivative Fi, j of the vector Fi and from a
contravariant vector
Gi = − є ijk Fj, k, (13.87)
we define the vector G to be the curl of F.

Since n. curlF = niGi = − є ijk Fj, k ni
dx i
and the components of unit tangent vector λ and
Equation (13.84) may be written as ds
dx i
− ∫ S ijk Fj,k ni ds = ∫ C Fi ds. (13.88)
ds
The integral ∫ C Fi dx i is called the circulation of F along contour C.

13.12 Hamilton’s Canonical Equations

Consider
J = ∫ tt12 L(q, q )dt , (13.89)

Where L = T − V is the kinetic potential.
d  ∂L  ∂L
We know − = 0, which is in the form
dt  ∂ q i  ∂ qi
d
L i − Lqi = 0 (i = 1,2, n) (13.90)
dt q
by using the subscript notation for partial derivatives of L(q, q ).
It is convenient to rewrite the system of n Lagrangean Equation (13.90)
in the form of an equivalent set of 2n first order equations, known as
Hamilton’s equations.
The function L(q, q ) = T (q, q ) − V (q) depends on n generalized coordi-
nates qi and n generalized velocities q i . Instead of q i , we can introduce a set
of n new variables pi defined by
pi = Lqi (q, q ), (i = 1,2,n), (13.91)

where (13.91) is solvable for q i in terms of pi and qi and the condition is
∂ Lqi
that ≠ 0.
∂ q i
We construct
H ( p, q) = q i pi − L(q, q ) . (13.92)

Differentiating (13.92) with respect to qj, we get
∂ q i ∂ q i
H q j ( p, q) = pi − Lqj
− Lq i

∂qj ∂qj
and since pi = Lqi ,
H q j ( p, q) = − Lq j . (13.93)
Similarly, we have
∂ q i ∂ q i
H p j ( p, q) = q j + pi − Lqi ,

∂ pj ∂ pj
which on using (13.91) reduces to
H p j ( p, q) = q j , (13.94)
but from the Lagrangean Equation in (13.90), we get
d
L i = Lqi .
dt q
From (13.91) and (13.93), we obtain
dpi dLqi
= = Lqi = − H qi (i = 1,2,n)
dt dt
(13.95)
dpi
or = − H qi ,
dt
which, together with n equation (13.94), yields
dqi
= H pi . (13.96)
dt
Equations (13.95) and (13.96) constitute 2n first order Hamilton’s
Canonical Equations.
13.12.1 Generalized Momenta

A function H (p, q), is known as Hamilton’s function.
∂L
Here, H = q i pi − L(q, q ) = q i Lqi − (T − V ) = q i i − T + V , where
∂ q
1 ∂T ∂L ∂ (T − V ) i ∂ T
T = aij q i q j , i = aij q i , so that q i i = q i i = q i  j
i = q aij q = 2T
2 ∂q ∂q  ∂q  
∂q
∴ H = 2T − T + V = T + V .
Thus, H is the total energy of the system.
∂L
The variables pi = Lqi = = aij q i are called the generalized momenta
∂ q i
and the square of the magnitude of pi is
p 2 = aij pi pi = aij aik q k a jl q l = akl q k q l = 2T .
13.13 Exercises
1. Let a particle of mass m be constrained to move on the surface of
a sphere of radius a. Relate the orthogonal Cartesian coordinates
yi to the surface coordinates uα by the formulas
 y 1 = asinu1 cos u 2

2 1 2
 y = asinu asinu

 y 3 = acosu1 .

d T T
Show that equations F becomes
dt u u
F1
u1 − (u 2 )2 sinu1 cos u1 =
ma 2
F2
u2 sin 2u1 + 2u 1u 2 sinu1 cos u1 = .
ma 2
Solve these equations for the case when Fα = 0, and show that the trajec-
tory is an arc of a great circle and the speed v = constant. [Hints: The first
integral of the second equation is u 2 sin 2u1 = constant . Use this result
in the first equation and observe that v 2 = a 2[(u 1 )2 + (u 2 )2 sinu1 .)
2. Find the dynamic equation in spherical coordinates with
ds 2 = (dr )2 + r 2 (dθ )2 + r 2 sin 2θ (dφ )2 .

 g
(If r = 0 , we get θ = −   sinφ .)
r
3. Find the dynamic equation in cylindrical coordinates with
ds 2 = (dr )2 + r 2 (dθ )2 + (dz )2.

4. Show that:
(a) In plane polar coordinates with ds2 = (dr)2 + r2 (dθ)2
1  ∂ (rFr ) ∂ Fθ 
div F =  +
r  ∂r ∂θ 
1  ∂  ∂v  ∂  1 ∂v  
∇2 v =   r  +   ,
r  ∂ r  ∂ r  ∂θ  r ∂θ  
where Fr and Fθ are the physical components of the vector F, that is,
F = Frr1 + Fθθ1,
where r1 and θ1 are unit vectors.

(b) In cylindrical coordinates with
ds 2 = (dr )2 + r 2 (dθ )2 + (dz )2

 1 ∂ (rFr ) 1 ∂ Fθ ∂ Fz 
div F =  + +
r ∂r r ∂θ ∂ z 
 1 ∂  ∂ v  1 ∂ 2 v ∂ 2V 
∇2 v =   r  + 2 2 + 2 ,
 r ∂ r  ∂ r  r ∂θ ∂z 

where F = Fr r1 + Fθ θ1 + Fz z1, and r1, θ1, z1 are unit vectors, so that Fr, Fθ,
Fz are the physical components of vector F.
14
Newtonian Law of Gravitations
14.1 Introduction
Sir Isaac Newton was perhaps the first to bring powerful dynamic ideas
and necessary mathematics to bear on motion in the solar system.
Newton built on the scientific contributions of his predecessors, par-
ticularly Kepler and Galileo. The validity of Kepler Laws depends upon
the fact that the masses of the planets are very small compared with the
mass of the sun. The empirical laws enunciated by Kepler laid Newton,
among others, to the conclusion that the force which keeps a planet in
its orbit around the sun varies inversely as the square of the distance
from the sun to planet. As soon as Newton proved that the gravitational
attraction between two homogeneous spheres could be calculated as if
the masses of the spheres were concentrated at their centers, the prog-
ress of dynamic astronomy was clear and rapid. In this chapter we also
study the Gauss theorem and problems of two bodies and restricted
three bodies.
14.2 Newtonian Laws of Gravitation

The inverse square law of attraction had its origin in Newton’s studies of
motion of planetary bodies in what he termed the eccentric conic sections.
We state this law as follows:
Two material particles attract each other with a force which is directly
proportional to the product of their masses and inversely proportional to the
square of the distance between them. The line of action of the force is along
the line joining the particles.
447
Thus, the law, in the form of a vector equation is
m1m2
F =γ r12,
r123
where m1 and m2 are the masses of the particles and r12 is the vector from
P1 and P2.
The constant γ depends on the choice of units; in a cgs system, its value
6.664 × 10−8 and its physical dimensions are M−1L3T−2. We usually take it
as γ = 1, so we write
m1m2 .
F= r12 (14.1)
r123
We deal with the particles with continuous distributions of matter and

one can subdivide the bodies into infinity. This procedure is for two bodies
τ1 and τ2 leads to the formula
ρ1 ρ2
F = ∫ τ1 ∫ τ 2 r12 dτ 1τ 2 , (14.2)
r123
where dτ1 and dτ2 are the volume elements of bodies τ1 and τ2, ρ1 and ρ2,
and their density functions and r12 is the position vector of dτ2 relative to
dτ1. We consider that ρ1 and ρ2 are piecewise, continuous.
It is necessary to verify that the generalized law of gravitation in (14.2)
reduces to (14.1) and yields no vanishing couples when the bodies τ1 and
τ2 are allowed to shrink to a point.
We introduce an orthogonal Cartesian reference frame Y and denote the
coordinates of points of the bodies τ1 and τ2 by ( y1i ) and ( y 2i ), respectively,
as shown in Figure (14.1).
We replace the distributed mass ρ1 ∆τ1 by the concentrated mass m1 at
P1 ( y11 , y12 , y13 ) and the mass ρ2 ∆τ2 by m2 at P1 ( y 12 , y 22 , y 23 ).
For components of force,
y 2i − y1i
∆F i = ρ1 ρ2∆τ 1∆τ 2
r3
and for the components of moments, ∆Li is relative to the origin O,

Newtonian Law of Gravitations 449
∆Li = eijk y1j ∆F k

y 2k − y1k
= eijk y1j ρ1 ρ2∆τ 1∆τ 2
r3
Adding these vectorially gives the resultant force
y 2k − y1k
F i = ∫ τ1 ∫ τ 2 ρ1 ρ2 dτ 1dτ 2 (14.3)
r3
and the resultant moment
y 2k − y1k
Li = ∫ τ1 ∫ τ 2 eijk y1j ρ1 ρ2 dτ 1dτ 2. (14.4)
r3
We choose the origin O of the coordinate system at P1, and let τ1 shrink
towards O and τ2 shrink towards P2. Since ρ1 and ρ2 in Equations (14.3) and
(14.4) are nonnegative functions, the first mean value theorem for integrals
is applicable and we obtain Figure (14.1).
 yk − yk 
F i =  2 3 1  ∫ τ1 ∫ τ 2 ρ1 ρ2dτ 1dτ 2
 r 
Y3
∆τ2
r12
∆τ1 P2 τ2
τ1
P1
O Y2
Y1
Figure 14.1
and
 yk − yk 
Li = eijk y1j 2 3 1  ∫ τ1 ∫ τ 2 ρ1 ρ2dτ 1dτ 2
 r 
As dimensions of τ1 are allowed to approach zero, y1i → 0 and Li → 0,

whereas the first of the above integrals reduces to
y2i
Fi m1m2 .
r3
This is the Law of Gravitation for two particles at (0, 0, 0) and ( y 12 , y 22 , y 23 ).

We observed that a material body integrating with a point mass pro-
duces no resultant moment L. Moreover, direct calculations show that this
is also true when the point mass is replaced by a sphere τ whose density ρ
is a continuous function of the radius alone. The resultant force F, exterted
by the body acting on a point mass m = ∫ τ ρdτ , is located at the center of
the sphere.
Consider a body τ with piecewise continuous density ρ and let P (y1, y2,
y ) be fixed either within or outside τ. The gravitational potential V (P) at
3
the point P due to τ is defined by the integral
ρ(ξ 1 , ξ 2 , ξ 3 )
V (P ) = ∫τ dτ (ξ ), (14.5)
r
1 1 2 2 2 2 3 3 2
where r = ( y − ξ ) + ( y − ξ ) + ( y − ξ ) is the distance between P
(y1, y2, y3) and the variable point (ξ1, ξ2, ξ3) is associated with the volume
element dτ (ξ) of τ.
If P is outside the body, the integral (14.5) is proper.
In particular,
V
Fi , (14.6)
yi
where the Fi are components of gravitational force

ρ(ξ )
Fi ( P ) = ∫ τ dτ (14.7)
r
exerted by the body τ on the particle of unit mass at P (y).

If P (y) is within τ, the integral (14.5) is improper since r = 0. Although,
V (P) is of class C∞ whenever P is exterior to τ.
14.3 Theorem of Gauss

The integral of the normal component of the gravitational flux computed
over a regular surface S containing gravitating masses within it is equal to
4πm, where m is the total mass enclosed by S.
Proof: According to Newton’s Law of Gravitation, a particle P and Let
θ are the angle between the unit outward normal to n to S and the axis of
a cone with its vertex at P. This cone subtends an element of surface dσ
[Figure (14.2)].
The flux of the gravitational field produced by m is
mcosθ r 2dω ,
∫ S F.ndσ = ∫ S
r 2 cosθ
r 2dω
where dσ = and dω are the solid angle subtended by dσ.
cosθ
Thus, we have
n
θ
F
dσ
r
dω
P
m
Figure 14.2
∫ S F.ndσ = ∫ S m. dω = 4π m. (14.8)
If there are n discrete particles of masses mi located within S, then
mi cosθ
F.n = ∑ni =1
r2
and the total flux is
∫ S F.ndσ = 4π ∑in=1 mi . (14.9)

The result of (14.9) can be easily generalized to continuous distributions
of matter whenever such distribution is nowhere near the surface S.
The contribution to the flux integration from the mass element ρdτ,
contained within τ, is
ρcosθ dτ
∫ S F.ndσ = ∫ S dσ
r2
and the contribution from all masses contained easily within S is
 ρcosθ dτ 
∫ S F.ndσ = ∫ S  ∫ τ  dσ , (14.10)
 r2 
where ∫ τ denotes the volume integral over all the bodies interior to S.
Since all masses are assumed to be interior to S, r never finishes, so that the
integrand in Equation (14.10) is continuous and one can interchange the
order of integration to obtain
 ρcosθ dσ 
∫ S F.ndσ = ∫ τ ρ ∫ S dτ ,
 r2 
cosθ dσ
but ∫ S = 4π . Since it represents the flux due to a unit mass con-
r2
tained within S,
∫ S F.ndσ = 4π ∫ τ ρdτ = 4π m,
where m denotes the total mass contained within S.

14.4 Poisson’s Equation

By the divergence theorem, we have
∫ S F.ndσ = ∫ τ divF dτ

and by Gauss’s theorem
∫ S F.ndσ = 4π ∫ τ ρdτ . (14.11)
Therefore,
∫ τ (divF − 4πρ )dτ = 0. (14.12)

This relation is true for an arbitrary τ and since the integrand in (14.11)
is continuous, we conclude that
divF − 4πρ = 0 throughout τ. (14.13)
In the definition of potential function V, we have
F = − ∇V and from (14.13),

divF = ∇F = − ∇2 V = 4πρ.
∴ ∇2 V = − 4πρ (14.14)
is the equation of Poisson.

If P is not occupied by the mass, then ρ = 0 and the potential energy
satisfies Laplace’s Equation.
∇ 2V = 0
We note in conclusion that the formula
ρdτ
V (P ) = ∫τ
r
gives a solution of Equation (14.14) at all points in τ.

14.5 Solution of Poisson’s Equation

We know that Green’s symmetrical formula is
 ∂v ∂u 
∫ τ (u∇ 2v − v∇ 2 u)dτ = ∫ S  u − v  dσ (14.15)
 ∂n ∂n 

where V is the volume enclosed by S and u and v are scalar point functions.
1
Put u = where r is the distance between the points P(x1, x2, x3) and
r
(ξ1, ξ2, ξ3) and V is gravitational potential.
Since 1 has a discontinuity at (xi) = (ξi), we delete P(x) from τ by enclos-
r
ing it by a sphere σ of radius δ at P. Apply Green’s formula to region τ − ϵ,
within which 1 and τ possess the desired properties of continuity.
r 1
In region τ − ϵ, ∇ 2u = ∇ 2 = 0 and formula (14.15) yields
r
 1  1
1 2 1 ∂V ∂ 1 ∂ V ∂

∫ τ − ∇ Vdτ = ∫ S   
− V r  dσ + ∫ σ  − V r  dσ , (14.16)
r  r ∂n ∂n   r ∂n ∂n 
where n is the unit exterior normal to the surface S + σ, since on σ the

nomal n is directed towards P, and .
Now, n r
 1  1
1 ∂V ∂ 1 ∂V ∂
∫ σ  − V r  dσ = ∫ σ  − − V r  dσ
 r ∂n ∂n   r ∂r ∂r 
 1 ∂V V  2
= ∫σ  − − r dω
 r ∂r r 2 
 ∂V 
= − ∫σ  r +V  dω
 ∂r 
 ∂V 
= −δ ∫ δ  dω − 4πV, (14.17)
 ∂r  r =δ

where V is the mean value of V over the sphere σ and ω denotes the solid
angle.
Let δ → 0, the right-hand member of (14.17) yields − 4πV (P) and it
follows
 1
1 ∂ V ∂
∫ σ  − V r  dσ = −4π V ( P ).
 r ∂n ∂n 

Thus, (14.15) becomes
 1
1 2 1 ∂V ∂

∫ σ ∇ Vdτ = ∫ σ  − V r  dσ − 4π V ( P )
r  r ∂n ∂n 

1 2
Since  → 0, ∫  ∇ Vdτ = 0.
r
Therefore,
1
1 1 1 1 ∂V 1 ∂
V (P ) = − ∫ σ ∇ 2Vdτ+ ∫σ dσ − ∫ σ V r dσ . (14.18)
4π r 4π r ∂n 4π ∂n
This gives the solution of Poisson’s Equation at the origin.

If V is regular at infinity, i.e., for a sufficiently large value of r, V is such
that
m  ∂V  m
[V ] ≤ and  ≤ 2, (14.19)
r  ∂r  r
Where m is constant.
If the integration in Equation (14.18) is extended over all spaces so that
r → ∞, then using equation (14.18), (14.19) becomes
1
V (P ) = − ∫ ∞ ∇ 2dτ , (14.20)
4π
but V is a potential function satisfying Poisson’s equation i.e.,∇2V = − 4πρ.

Hence, from (14.20), we get
dV
V (P ) = ∫ ∞ ρ .
r
14.6 The Problem of Two Bodies

Here, we consider a much simpler, very well-known problem in physics: an
isolated system of two particles which interact through a central potential.
This method is often referred to simply as the two-body problem.
The problem of two bodies can be stated as follows: Given a system of
two particles interacting in accordance with the law of universal gravitation,
what is the trajectory of the system? This problem was solved by Newton in
the Principia, Book I. It lies at the basis of all considerations in astronomy.
We refer our system to a set of orthogonal Cartesian axes. The coordinates
of m1 and m2 (at a given instant of time) by ( x11 , x12 , x13 ) and ( x 12 , x 22 , x 23 ), as
We also introduce another Cartesian reference frame Y, moving with m1
in such a way that m1 is in origin 0 and axis Y i always remains parallel to
axes Xi. The coordinate of mass m2 relative to the Y − axes are denoted by
yi, = and we have
X3
Y3
m2
(u1, u2, u3)

O Y2
m1
Y1
X2
X1
Figure 14.3
y i = x 2i − x1i (i = 1,2,3). (14.21)

We choose the coordinates as generalized coordinates, taking them as

the center of masses. Thus,
m1 x1i + m2 x 2i
ui = (i = 1,2,3). (14.22)
m1 + m2
Our choice of generalized coordinates is as follows:
q1 = y1, q2 = y2, q3 = y3, q4 = u1, q5 = u2, q6 = u3.
Solving, we get
x1i x 2iyi [m2 x 2i (m1 m2 )ui m1x1i

m1 i m1 i
ui (u x1i ) y i x 2i ui (u x1i )]
m2 m2
m2
ui yi
m1 m2
m1
and x 2i ui yi . (14.23)
m1 m2

This will help us find the positional coordinates xi in terms of general-

ized coordinates qi. Since the magnitude of the force of attraction F is given
m1m2
by F = 2 , where r is the distance between the particles.
r
The potential energy V is
m1m2 m1m2
V= =
r ( x1 − x 2 ) + ( x12 − x 22 )2 + ( x13 − x 23 )2
1 1 2

V is the function of y i = x 2i = x1i .

We recall the Lagrangean equations
d T T V
(α)
dt q i qi qi
For computation
1 1
T = m1 x 1i x 1i + m2 x 2i x 2i
2 2
2 2
1  i m2 i 1  i m1 i
= m1 u −  
y + m2 u −  
y
2  m1 + m2  2  m1 + m2 
1 1 m1m2 i i
= (m1 + m2 )u iu i + y y ,
2 2 m1 + m2
V
since 0, for i 4, 5, 6, (α) can be reduced to
qi
m1m2 1 V
y
m1 m2 yi
(14.24)
ui 0 (i 1, 2, 3)
These differential equations are the motion of our system.
We first note that the motion of mass m2 relative to m1 is same as if mass
m1 were fixed and m2 attracted toward it with a force whose potential is
m1 + m2
V.
m1
(14.24) becomes
m1 m2 V . (14.25)
m2 y i
m1 yi
Thus, our problem is reduced to a study of motion under the action of

central forces.
The second set of equations of (14.24) states that the center of mass
moves in a straight line with constant velocity.
If we carry out the integration in (14.24) under the assumption that m1
(mass of sun) is much larger than m2 (mass of earth). If m1 >> m2, the cen-
ter of mass ui will nearly coincide with those of mass m1.
Thus, x1i = ui and from the second set of equations in (14.24), we con-
clude that the mass m1 moves through space with constant velocity.
We need to examine the motion of m2 relative to m1.
m1 + m2
If m1 >> m2, then =1
m1
and accordingly (14.25) becomes
V
m2 y i .
yi
Let mass m1 and m2 be in the plane of Y1 Y 2, since a force field is a central

force so that there is no component of force at right angles to the plane.
Let the polar coordinate of m2 be (r, θ),
i.e., y 1 = rcosθ
y 2 = rsinθ
1
Kinetic energy, T = m2[( y 1 )2 + ( y 2 )2 ]
2
1
= m2[r 2 + r 2θ 2 ],
2
m1m2
and V =−
r
In Lagrangean Equation (α), with q1 = r and q2 = θ, we get
m1m2
m2r − m2rθ 2 = −
r2
d 2
(r θ ) = 0 and
dt
m
or r − rθ 2 + 21 = 0
r
2
r θ = h, constant (14.26)
These ordinary differential equations determine the trajectory. The

second equation states that sectorial velocity is constant, which is one of
Keplar’s Laws. The relation r 2θ = h can determine the time required to
describe the orbit.
If h ≠ 0, the trajectory is not a straight line and we have
r2dθ = hdt
or
1
t = ∫ θ0 r 2dθ .
h
df df dθ d d dθ d  h 
We know = ∴ = =   and
dt dθ dt dt dθ dt dθ  r 2 
d  dr  d  dr dθ  h d  h dr 
r =   =  =   and using it in (14.26),
dt  dt  dt  dθ dt  r 2 dθ  r 2 dθ 
2
h d  h dr   h  m1
−r + 2 = 0,
r 2 dθ  r 2 dθ   r 2  r
Multiplying by r2,
2
d  h dr  3  h 
h −r + m1 = 0,
dθ  r 2 dθ   r2 
d  h dr  h 2
or h − + m1 = 0,
dθ  r 2 dθ  r
d  1 dr  1 m1
or − + = 0. (14.27)
dθ  r 2 dθ  r h 2
1
Change the variable u = ,
r
2
du du dr 1 dr d u d  du  d  1 dr  d  1 dr  .
= =− 2 , 2 =   =  − 2  = −  2 
dθ dr dθ r dθ dθ dθ dθ dθ r dθ dθ r dθ 
Now. In (14.27) becomes
d 2u m1
2 +u = 2 .
dθ h
The solution of the differential equation is
1
u = [1 − e cos (θ − )]
l
l
or r = , (14.28)
1 − e cos (θ − )
h2
where l ≡ and e, ϵ are integration constants.
m1
This is an example of a simple use of Hamilton’s Equation.
Example 14.6.1. Consider a particle of mass m moving under the attrac-

tion of a central force field with the potential V(r), r being the distance of
the particle from the center of attraction.
Solution: We choose the polar coordinates r = q1, θ = q2 as generalized
coordinates, then
1 1
T = m[r 2 + (rθ )2 ] = aij q i q j ,
2 2
m 0 
where (aij ) =  2 ,
 0 mr 
1
but H = T + V = aij q i q j + V .
2
By (13.95), we get
p12 p22
H= + + V (r ).
2m 2mr 2
H p22 H H p1 H p2
Therefore, V (r ), 0, , .
r mr 3 p1 m p2 mr 2
Hence, Hamilton’s Equation in this problem is
dr p1 dθ p dp p2 dp
= , = 22 , 1 = 2 3 − V ′(r ), 2 = 0.
dt m dt mr dt mr dt
From these, we get
d
(mr 2θ ) = 0,
dt
which is Kepler’s Law of Planetary Motion.

m
From these equations, it can be shown that if V = − , the orbit is a
conic section. r
14.7 The Problem of Three Bodies

In general, in the three bodies problem we study the equations of motion
of three finite bodies. The problem is restricted in the sense that one of the
three masses is taken to be so small that the gravitational effect on the other
two masses by the third mass is neglected. Then, the general three body
problem reduces to a problem known as a restricted three body problem.
The small body is known as an infinitesimal mass and the remaining
two massive bodies are known as finite masses. The Earth, Moon, and sat-
ellite system constitute a good example of a restricted three body problem.
Let us consider a coordinate system, OXYZ with an origin at O rotat-
ing relative to the inertial frame with angular velocity ω about the Z-axis.
Without any loss of generality, we can choose the coordinate system such
that the X-axis lies along the line joining the finite masses m1 and m2 with
O as a bary center. Let the distance between m1 and m2 be denoted by ρ,
the position of infinitesimal mass m be denoted by (x, y, z), and the radius
vector from m to m1 and m to m2 be ρ1 and ρ2, respectively.
The motions of m1 and m2 are known. We are only to calculate the
motion of m.
Now, the kinetic energy of m is in reference to rotating frame OXYZ.
1
Kinetic Energy, T m[(x y )2 ( y x )2 z 2 ]
2
1
m[x 2 y 2 z 2 2 (xy yx ) 2
(x 2 y 2 0] T2 T1 T0
2
T T2 T1 T0 , (14.29)

1
where T2 = m( x 2 + y 2 + z 2 )
2
1
T1 = m(2ω ( xy − yx ))
2
1
T0 = m(ω 2 ( x 2 + y 2 )) (14.30)
2
m m
Potential Energy V = −mγ  1 + 2  , (14.31)
 ρ1 ρ2 
1
where ρ1 = [( x + α )2 + y 2 + z 2 ]2
1
ρ2 = [( x − b)2 + y 2 + z 2 ]2 , (14.32)
where (−a, 0, 0) and (b, 0, 0) are coordinates of m1 and m2.

The generalized momentum corresponding to the generalized coordi-
nate system is
T
Px m(x y)
x
T
Py m( y x)
y
T
Pz mz . (14.33)
z
Now, T2 can be written in the generalized terms of momenta:
2 2 2
1 1 Px Py Pz
T2 m(x 2 y 2 2
z ) m y x
2 2 m m m
1 Px2 Py2 Pz2 2 2
m (Px y Py x ) (x 2 y2)
2 m2 m
Px2 Py2 Pz2 1 2
T2 (Px y Py x ) m (x 2 y 2 ). (14.34)
2m 2

We know the Hamiltonian H is given by
H T2 T0 V
Px2 Py2 Pz2 1 2
(Px y Py x ) m (x 2 y2 )
2m 2
1
m( 2 (x 2 y 2 )) V
2
Px2 Py2 Pz2
H (Px y Py x ) V . (14.35)
2m
The canonical equation of motion of the infinitesimal mass is given by
H Px
x y
Px m
H Py
y x
Py m
H Pz
z
Pz m
H V
Px Py
x x
H V
Py Px
y y
H V
Pz . (14.36)
z z
Equations (14.36) are six first order ordinary differential equations

which describe the motion of infinitesimal mass in general terms.
Here, ω will be the function of time. If the finite mass m1 and m2 move
along a circular orbit, then ω is the angular velocity of m1 and m2. Then, H
is independent of time. Therefore, there exists a Jacobi integral.
Equations (14.36) are six differential equations of first order. They can be
put in the form of three equations of the second order. We have from (14.33)
Px = m( x − ω y )
∴ Px = m( x − ω y )
Similarly, Py = m( y − ω x )
and Pz = mz.

Substituting these values in the last three equations of (14.36), we get
V
m(x y ) m ( y x) ,
x
1 V
or (x y ) ( y x) ,
m x
2 1 V
or x 2 y x .
m x
2 1 V
Similarly, y 2 x y
m y
1 V
and z . (14.37)
m y
Now, let us introduce the modified potential energy, U:
U = −V + T0
m m 1
= −mγ  1 + 2  + mω 2 ( x 2 + y 2 ) (14.38)
 ρ1 ρ2  2
U 1 U
m 2 x or 2 x .
x m x
Equation (14.37) can be written as
1 U
x 2 y
m x
1 U
y 2 x
m y
1 U
z . (14.39)
m y
The Jacobi integral can be obtained from Equation (14.39) and multiply-
ing the above equation, x , y , z , respectively, and adding
1 U dx U dy U dz 1 U,
 yy
xx  zz

m x dt y dt z dt m t

integrating
1 2 1
( x + y 2 + z 2 ) = U + h,
2 m
1 1 m m 1 
or ( x 2 + y 2 + z 2 ) −  −mγ  1 + 2  + mω 2 ( x 2 + y 2 ) = h,
2 m  ρ1 ρ2  2 
or
1 2
( x + y 2 + z 2 ) − γ  m1 + m2  − 1 ω 2 ( x 2 + y 2 ) = h.
 ρ1 ρ2  2 (14.40)
2

This is known as the Jacobi Integral.

14.8 Exercises
1. If a particle of mass m is constrained to move on a smooth surface, show
that the system of Hamilton’s equation is
du H dp H
, , ( 1, 2) with p ma u
dt p dt u
1
and H ma p p V.
2
dH
2. Show that along the dynamic trajectory = 0 that H = constant is an
dt
integral of Hamilton’s equations.
1 2 p2
3. If T = (q ) and V = kq2, k > 0, show that H = + mω 2q 2 /2, where
2 2m
k 2h
ω 2 = . Deduce that q = sin (ω t + α ).
m mω 2
Appendix A: Answers to
Even-Numbered Exercises
Exercise 1.9
1. (a) ai (b) ( x i )2 (c) n (d) n (f) δ ii = n
2. (b) ( a11 + a12 +…. + a1n )x 1 +….(a11 + an2 +…. + ann )x n
∂ ∂
(c)
∂x 1 ( )
gb1 +……. + 1 gbn
∂x
( )
5. (a) aik
(b) aijp x i x j + aipk x i x k + a pjk x k x j
7. (a) u1v1 +…+ un v n + u1v 2 +……+ u1v n +…+ un v1 +……+ un v n
(b) aijk x i x j x k
8. (a) Aklj (b) Bik (c) ai ai (d) δ ii
Exercise 3.9
 −3 
 2 3 
 2 
ij  −5 
1. 4, g =  3 5 
2
 
 −3 −5 3 
 2 2 2 
2. ds 2 = (dy 1 )2 + ( y 1 )2 (dy 2 )2 + (dy 3 )2

 1 0 0   1 0 0 
 0 −2 −4  1  0 3 4 
4.   , −22  
 0 −4 3   0 4 2 
469
470 Appendix
1  1 cosα 
8. sin 2α  cosα 1 

Exercise 4.6  2  1  3  1  3
5. (i) [12,2 ] = 2 x 1 , [ 22,1] = − x 1 ,   = 1 2 ,  = 1 ,
 1 2  ( x )  1 3  x  2 3
1  3  1  3  2
 1  1
 1  1 2
 2
1 2 ,  = 1 ,  = cotx ,   = −x ,   = − x sinx , 
x  1 3  x  2 3   2 2   3 3   3 3
 1  1 2
 2  2 2
x ,  = − x sinx ,   = − sinx cosx
 3 3   3 3 
1 ∂f 1 ∂f 1 ∂f  2  1 ∂
(ii) [12,2 ] = , , [ 22,1] = − 1
,{22,2] = , =
2 ∂x 1
2 ∂x 2 ∂ x 1  1 2  2 f ∂ x
∂f  2  1 ∂f  1  1 ∂f  2  1 ∂f
, = 1 , =− , = ,
∂ x  1 2  2 f ∂ x  2 2  2 ∂ x 1  2 2  2 f ∂ x 1
 i   i  1  2   2  1
4. (i)   = 0 (ii)  j k  = −x ,  =  = 1
j k   1 2   2 1  x
  
 1  1  2  1  3  1 2 2
(iii)   = −x ,   = 1 ,  = − x (sinx )
 2 2   1 2  x  3 3 
 3  1  2  2 2
  = 1 ,  = sinx cos x
 1 3  x  3 3 
 3   3  2
 =   = cot x
 2 3   3 2 
 2 4
11.  −φcotθ , − r φ , r 
Exercise 5.7
1
5. Rij ,k − Rik , j 7. a = −
2
Appendix 471
Exercise 8.8
1
5. χ = , µ i = ( −1,0,0, ) and τ = 0; νi = ( 0,0,1)
a
Exercise 9.11
2
1. d σ = a sinu dudv
2
d 2u1 1  du 2  d 2u 2 2 du1 du 2
2. ds 2 − u  ds  = 0, ds 2 + u1 ds ds = 0
Exercise 10.10
4 (v 2 − u2 )
2. 3
(1 + 4v 2 + 4u 2 ) 2
(u1 f 2 )2 − f1
3. κ = − ,H= 3
[(u1 )2 + f12 ]2
2[(u1 )2 + f12 ]2
Exercise 11.5
 f1  (u1 f 2 )2 (u1 f 2 )2
1. X 2p +  3 X −
 p {(u1 )2 + f12 }2 = 0 , κ = − ,
{(u1 )2 + f12 }2
 {(u1 )2 + f12 } 2 
1 − f1
H= 3
2
{(u1 )2 + f12 } 2
−a a
2. X (1) = 1 2 , X ( 2) = 1 2
(u ) (u )
3. [ u1 f1 f 2 + 1 + f12 = 0 ]
Exercise 12.5
2. κ > o, elliptic
References
Arthur E. Fischer, An Introduction to Conformal Ricci Flow, Classical and Quantum

Gravity, Institute of Physics Publishing (Online, 2004).
Bennett Chow, Peng Lu, Lie Ni, Hamilton’s Ricci Flow, Vol. I (Preliminary version,
2005).
Chaki, M.C., Tensor Calculus (Calcutta Publishers, India, 1987).
De, U.C., Differential Geometry of Curves and Surfaces in E3, (Anamaya Publishers,
New Delhi, 2007).
Eisenhart, L.P., An Introduction to Differential Geometry, (Princeton University
Press, 1940)
Eisenhart, L.P. Riemannian Spaces of Class Greater than Unity, Annals of Math.
(vol. 38).
Goreux S.J., Introduction to Tensor Calculus (New Central Book Agency, India,
1984).
Ishwar, B., Basics of Dynamical Systems, (DSTA=2018, ISM, India, 2018).
Levi-Civita, T., The Absolute Differential Calculus (Blackie & Sons, London, 1927).
Millman, R.S. and Parker, G.D., Elements of Differential Geometry, (Academic
Press, 1965).
Nayak, P., Mechanics: Newtonian, Classical, Relativistic, (Asian Books, New Delhi,
2008).
Resnik, R., Introduction of Special Relativity, (John Wiley, New York, 1968).
Sokolnikoff, I.S. Tensor Analysis theory and Applications to Geometry and Mechanics
of Continua (John Wiley & Sons, Inc, New York, 1951).
Sokolnikoff, I.S., Mathematical Theory of Elasticity (New York, 1946).
Weatherburn, C.E., Riemannian Geometry and the Tensor Calculus, (Cambridge
University Press, 1938).
Weatherburn, C.E., Differential Geometry of Three Dimensions, (Cambridge
University Press, London, 1966).
Willmore, T.J., An Introduction to Differential and Riemannian Geometry, (Oxford
University Press, UK, 1965).
473
Index
є-Systems, Codazzi equation, 335, 338, 339–341

derivative, 351–352, 385, 441 Congruence of curves, 89
Conjugate tensor, 59, 75–77
Absolute derivative, 222 Conservative force field, 403–405
Absolute tensor, 60, 61, 232, 234, 275 energy of, 403
Action integral, 428, 429 Contraction of tensor, 53–54
Action, principle of least, 427–429 Contravariant and covariant laws, 33,
Admissible transformations, 28, 30–31, 34, 35
188, 196, 200, 432 Contravariant tensor, 32–34, 40–41,
Affine transformation, 111, 112, 114 76, 115–116, 271
tensor, 63–64 Contravariant transformation, 37
Algebra of tensors, 43–45 Contravariant vector, 33, 35–40, 130,
Angle, between two vectors, 86–88, 91, 131
207–210 Coordinate curves, 202–205, 209, 211,
between two coordinate 266, 364, 378
hypersurfaces, 89–94 Coordinate surface, 201
Angle between two curves on surface, Coordinates, curvilinear, 195–219
272–277 curve, 202–205, 364
Arc, length, 195–200, 228, 272, 279, cylindrical, 65, 82, 134, 204–205,
281, 292, 312, 349, 437 238, 243, 421
curve of, 195, 208, 265 Gaussian, 266, 267, 268, 280, 323,
Associate tensors, 77–84, 146, 165, 332 330
Asymptotic lines, 362–375 generalized, 430–438, 442, 457, 461
geodesic, 277–278, 289–291
Beltrami’s formula, 310, 316 orthogonal, 74, 206, 211, 268, 274,
Bertrand curves, 235–252 321, 438, 441
Bianchi identity, 152, 160–165, 166, 169 spherical, 28–29, 35–36, 80–82, 92,
Binormal, 4, 221, 231, 233, 234, 237, 195, 282–284, 411–412, 430
261, 349 surface, 205, 265–267, 269, 323, 327,
329, 330, 332, 335, 349
Central force, 458, 459, 461 transformation of, 25–26, 27–31, 32,
Christoffel symbols, 97–110, 150, 172, 45, 174, 196, 269, 330, 392
235, 277–278, 289–290, 295, 401, Covariant tensor, 32–35, 41–42, 77,
440 114–115, 180, 326, 433
transformation of, 110–113, 290 differentiation, 113–129
475
476 Index
law, 33, 34, 48 n-dimensional manifold, 4, 26, 31,

relative tensor of, 123–129 391–394
vector, 33, 34, 35, 84, 85, 86–87, 131, space, 170–171
137–141, 189–192 Einstein tensor, 166–170
Cramer’s rule, 14, 17 Energy, 443
Curl, 129–141 conservation of, 403, 426
of covariant vector, 189–190 equation of, 414–415
Curvature, 308–319, 349–375, 381–394 kinetic, 401, 403, 405, 408, 419, 421,
constant, 1, 158, 297 424, 426, 427, 428, 429, 433, 436,
Einstein, 296–308 459, 462–463
Gaussian, 293, 294–308, 341, 343, potential, 403, 415, 419, 421, 427,
357, 359, 381, 382 435, 453, 457, 463, 465
line of, 361–362, 364–366, 371, Equality of two tensor, 45
374–379 Euclidean n-dimensional manifold,
mean, 341–346, 359 175, 196, 197, 200, 280, 391, 392,
scalar, 165, 167–170, 171 394, 423
surface of, 1, 297, 381–383 Euclidean space, 2, 4, 53, 63, 73,
surface of negative of, 1, 381–383 174–175
total, 295, 300, 341–346
Curve of congruence, 89 First fundamental tensor, 75, 81
coordinates of, 202–205, 364 Free index, 10, 11
net of, 266 Frenet formula, 233–252, 308, 310,
surface of, 308, 330, 350, 387 349, 414
Curvilinear coordinate in E3, 195, Fundamental form, third, 338
200–210 second, 326, 332–334
on surface, 265–267 tensor, 75, 76, 97, 110, 120–122
Fundamental quadratic form, first,
Determinants, 15–18, 185–192 267–272, 292, 294–295
differentiation of, 18–19
expansion of, 150, 185 Gauss-Bonnet theorem, 381,
multiplication of, 187 387–391
Developable, 293–294 Gauss-Codazzi equation, 335, 338,
Differentiation, covariant, 143, 146 339–341
determinant of, 18–19 Gauss equation of surface, 335, 338,
intrinsic, 221–226 339–341, 385
tensor, 113–129 Gauss formula, 334, 337, 341, 351,
Divergence, 129, 130–136 418
Divergence theorem, 438, 453 Gauss theorem, 447, 451–452, 453
Dummy index, 11 Gaussian curvature, 293, 294–308, 341,
343, 357, 359, 381, 382
e-System, 177–181 Generalized coordinate, 430–438, 442,
application of, to determinants, 457
185–192 momenta of, 443–444, 463
Einstein curvature, 296–308 velocity, 431, 432, 442
Index 477
Geodesic coordinates, 277–278, Jacobian determinants, 27, 30–31, 60,

289–291 188
Geodesic curvature, 308–319, 350, 351
Geodesics, 277–289 Kepler’s law, 447, 462
coordinate systems, 278, 289–291 Kinetic energy, 401, 403, 405, 408, 419,
equation, 278, 279, 282–284, 421, 424, 426, 427, 428, 429, 433,
287–289, 292 436, 459, 462–463
trajectories, 420 Kinetic potential, 435, 442
Geometry, 1, 4, 73, 75, 143–175, 196, Kronecker delta, 11, 15, 120–122
200, 265–319, 326, 381, 387, 391 generalized, 181–183, 185–192
of space curves, 221, 222, 228–233
Gradient, 34, 130, 132 Lagrangean equation of motion,
Gravitation, Einstein’s law of, 170 405–411, 412, 419, 421–422
Newton’s law of, 170, 400, 447–451 application of, 411–423
Gravitational potential, 450, 454 generalized coordinates in,
Green’s function, 388, 391, 454 432–438
Green’s theorem, 389, 438–441 multiplier, 358
Laplacian
Hamilton’s principle, 423–427 of invariant, 136–137
canonical equation of, 442–444, 461, operator, 439, 440
462, 464 Length,
Helix, 221, 254–262 arc of, 195–200, 228, 269, 272, 279,
circular, 258–262 281, 292, 312, 349, 437
cylindrical of, 256–258 of a vector, 84–86, 206, 208, 392
Hypersurface, 88–89, 394 Line element, 205
angle between hypersurface, 89–94 curvature of, 361–362, 364–366,
orthogonal, 89, 90–91 371, 374–375, 376–379
Linear equation, 14–15, 17, 236, 238
Index, dummy, 11, 122, 133, 134
free, 11 Manifold, n-dimensional, 4, 26, 31,
Inner product of tensors, 54–55, 77 391–394
Integrability conditions, 4, 173, 174, Euclidean of, 175, 196, 197, 200,
334–336, 339 280, 391, 392, 394, 423
Integral of energy, 426–427 Riemannian, 196, 278, 280, 291
curvature, 387, 391 Mean curvature, 341–346, 359
Intrinsic differentiation, 221–226 Metric space, 1, 74
Intrinsic geometry, 267–272, 321 derivatives of, 27
geometry surface of, 4, 265, 292, 308 Metric tensor, 74–75, 76–77, 78–79,
Intrinsic properties, 4, 265, 267, 292, 162–163, 224, 270–272
297, 326 Meusnier theorem, 350–358
Invariants, 25, 26, 31, 56, 136–137 Minkowski space, 85
derivatives of, 130 space time, 85, 161, 162
Isometric surfaces, 292–294 Motion of particle, 5, 13, 401–403, 413,
Isotropic tensor, 64 417
478 Index
n-dimensional manifold, 4, 26, 31, Restricted three body problem, 447,

391–394 462
n-ply orthogonal system, 89 Ricci tensor, 143, 159–160, 162–163,
Newtonian law of motion, 399–400, 170
401, 406, 434 Ricci’s theorem, 120–122, 172
law of gravitation, 170, 400, 447–451 Riemannian Christoffel tensor,
Normal, 4, 89–90, 185, 230, 231, 232 143–149, 161, 294–308
Normal curvature, 352, 378, 381 curvature tensor, 158
Normal line to the surface, 321, properties, 150–158
324–329 Riemannian geometry, 143–175, 196,
Normalized cofactor, 17 200
Riemannian space, 73, 74, 75, 143, 158,
Orthogonal ennuple, 89 161, 167, 170, 171–174, 277, 289
Orthogonality two vectors, 87–88 Rodrigues formula, 376–379
Osculating plane, 230, 231, 414
Outer multiplication, 51–55 Second fundamental form of a surface,
326, 327, 332–334
Parallel surface, 383–387 Second fundamental tensor, 76, 80,
Parallel vector fields, 226–228 82
Parallel vectors on a surface, Serret-Frenet formula, 233–252, 308,
291–292 310, 349, 414
Partial derivative, 97, 187–188, Skew-symmetric, 46–51, 137
216–219 Space, 195–219, 221–262, 321–346
Plane, 198, 200, 205, 231–233, 293 constant curvature, 158
normal, 232, 352 Einstein, 170–171
osculating, 230, 231, 414 Euclidean, 2, 4, 53, 63, 73, 75,
rectifying, 232, 237 174–175
Poisson equation, 170, 453 flat, 161
solution of, 454–456 Minkowski, 85, 161, 162
Principal curvature, 358–375 n-dimensional, 26–27, 74, 195
direction, 358, 359 Riemannian, 73, 74, 75, 76, 143, 158,
Principal normal vector, 230 161, 167, 170, 171–174, 277, 289
Principle of least action, 427–429 Riemannian n-dimensional, 171,
Problem of three bodies, 462–466 196
Problem of two bodies, 456–462 Stoke’s theorem, 441
Pseudo-Euclidean, 392 Straight line, equations, 252–254, 280,
312, 399
Quotient laws of tensors, 56–58 Substitution factor, 13
Summation convention, 10–11
Reciprocal base systems, 195, 210–216 Einstein of, 10
Reciprocal tensor, 58–60, 76, 270 Surface curve, 330
Rectifying plane, 232, 237 Surface of positive curvature, 381–383
Relative tensor, 60–63, 123–129 negative curvature, 357, 381–383
Index 479
Symmetric tensors, 45–46 Total curvature, 295, 300, 341–346

System of different orders, 9 Trajectory, of particle, 401, 402, 403,
405, 427, 428
Tangent vector, 229, 230, 253, 321–324 dynamic, 436
Tensor derivative, 329–332, 337, 339 natural of, 414
Tensorial differentiation, 113 Transformation, admissible, 28, 30–31,
Tensors, absolute, 60, 61, 232, 234, 275 188, 196, 200, 432
affine, 63–64 affine, 111, 112, 114
algebra, 43–45 by invariance, 31–32
associated, 77–84, 146, 332 by contravariant, 37
Cartesian, 4, 63, 64 by covariant, 37, 65
concept, 34–43 Christoffel symbols, 110–113,
conjugate, 59, 75–77 290
contraction of, 53–54 of coordinate, 25–26, 27–31, 32,
contravariant, 32–34, 40–41, 76, 45, 174, 196, 269, 330, 392
115–116, 271
covariant, 32–35, 41–42, 77, Umbillic point, 360–361
114–115, 180, 326, 433
covariant differentiation of, 113–129 Vectors, null, 64, 86–87
derivatives, 329–332, 337, 339 contravariant, 33, 35–40, 130,
Einstein, 166–170 131
fundamental, 75, 76, 77, 97, 110, covariant, 33, 34, 35, 84, 85,
120–122 86–87, 131, 137–141, 189–192
higher order, 40–43 curl of, 129, 137–141, 189–190
intrinsic differentiation of, 221–226 derivatives of, 114, 225–226,
isotropic, 64 330
metric, 74–75, 76–77, 78–79, divergence of, 131, 136, 438, 440
162–163, 224, 270–272 normal, 230, 325, 337
mixed, 42–43, 51–52, 120, 132–136 null, 64, 86–87
pseudo, 64–65 orthogonal, 87–88, 94, 216
quotient law of, 56–58 principal normal, 230
reciprocal, 58–60, 76, 270 product of two covariant vectors,
relative, 60–63, 123–129 190–192
Riemann-Christoffel, 143–158, 161, reciprocal base, 195, 210–216
294–308 surface, 271, 291, 309, 323–324
skew-symmetric, 46–51, 137 tangent, 221, 229, 321–324
symmetric, 45–46 unit, 85, 86, 206, 207, 229, 230,
types of, 43–45, 46–47, 52, 54, 231
56–57, 58, 116–120, 130, 132, 137
zero, 43, 174, 175 Weingarten formula, 337–338, 339,
Torsion, 231, 233 376, 377, 385
Also of Interest
Check out these other titles from Scrivener Publishing

SIMULATION AND ANALYSIS OF MATHEMATICAL METHODS IN
REAL TIME ENGINEERING APPLICATIONS, Edited by T. Ananth
Kumar, E. Golden Julie, Y. Harold Robinson, and S. M. Jaisakthi, ISBN:
9781119785378. Written and edited by a group of international experts in
the field, this exciting new volume covers the state of the art of real time
application of computer science using mathematics. NOW AVAILABLE!
MATHEMATICS IN COMPUTATIONAL SCIENCE AND ENGINEERING,

Edited by Ramakant Bhardwaj, Jyoti Mishra, and Satyendra Narayan,
ISBN: 9781119777151. This groundbreaking new volume, written by
industry experts, is a must-have for engineers, scientists, and students
across all engineering disciplines working in mathematics and computa-
tional science who want to stay abreast with the most current and provoc-
ative new trends in the industry. DUE OUT IN SPRING 2022!
i-Smooth Analysis: Theory and Applications, by A.V. Kim, ISBN

9781118998366. A totally new direction in mathematics, this revolution-
ary new study introduces a new class of invariant derivatives of functions
and establishes relations with other derivatives, such as the Sobolev gener-
alized derivative and the generalized derivative of the distribution theory.
NOW AVAILABLE!
Open Ended Problems: A Future Chemical Engineering Approach, by

J. Patrick Abulencia and Louis Theodore, ISBN 9781118946046. Although
the primary market is chemical engineers, the book covers all engineer-
ing areas so those from all disciplines will find this book useful. NOW
AVAILABLE!
SYSTEMS WITH DELAYS: Analysis, Control, and Computations, By A.V.
Kim and A. V. Ivanov, ISBN: 9781119117582. This mathematical treatise
presents a constructive theory of systems with delays, also known as delay
differential equations (DDE), which includes standard theory and new
results on numerical treatment of time-delay systems and LQR control
design theory. NOW AVAILABLE!
WILEY END USER LICENSE AGREEMENT
Go to www.wiley.com/go/eula to access Wiley’s ebook EULA.

Introduction To Differential Geometry With Tensor Applications

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Introduction To Differential Geometry With Tensor Applications

Uploaded by

Copyright:

Available Formats

Introduction to Differential

Geometry with Tensor

Wiley Global Headquarters

Limit of Liability/Disclaimer of Warranty

Library of Congress Cataloging-in-Publication Data

Cover image: Geometry images provided by De

Printed in the USA

Part I: Tensor Theory 7

2.6.2 Contravariant Vectors 35

5.6.2 Euclidean Spaces 174

Part II: Differential Geometry 193

9 Intrinsic Geometry of Surfaces 265

12 Curvature of Surface 381

Part III: Analytical Mechanics 397

Appendix A: Answers to Even-Numbered Exercises 469

Differential Geometry is the study of geometric properties of curves and

An Introduction to Differential Geometry with Tensor Applications

Euclidean Geometry was one of the most important branches of mathe-

Hence, Equation (0.4) can be written as

Hence, we conclude that a tensor of E2 may be regarded as generaliza-

coordinate system is called a Cartesian Tensor and a tensor from a general

motion of particles in a fixed frame of reference. Within those frames, other

1.2 Systems of Different Orders

1.3 Summation Convention Certain Index

aijxj, (i = 1,2,3 and j = 1,2,3,4) (1.1)

represents three linear forms:

a11x1 + a12x2 + a13x3 + a14x4

a21x1 + a22x2 + a23x3 + a24x4

a31x1 + a32x2 + a33x3 + a34x4 (1.2)

= (a11u1v1 + a21u2v1 + a31u3v1) + (a12u1v2 + a22u2v2 + a32u3v2)

1.3.1 Dummy Index

a1x1 + a2x2 + a3x3 + … + anxn

1.3.2 Free Index

1.4 Kronecker Symbols

Such a system is called a Kronecker symbol or Kronecker delta.

We shall now consider some properties of this system.

Property 1.4.1. If x1, x2, … xn are independent variables, then

Property 1.4.2. From the summation convention, we get

Similarly, δii = δii = n

Property 1.4.3. From the definition of δi j, taken as an element of unit

Similarly, δ ij a jk = aik (1.7)

δ ijδ kj = δ 1iδ k1 + δ 2i δ k2 + …+ δ niδ kn = 1.δ k1 + δ k2 +  + δ kn = δ ki

Remark 1.4.1. If we multiply xk by δ ki , we simply replace index k of xk with

Solution: (a) We have δ ijδ lj = δ 1iδ l1 + δ 2i δ l2 +  + δ niδ ln (1.8a)

∴δ ijδ lj = δ 1iδ l1 + δ 2i δ l2 +  + δ niδ ln = δ l1 + δ l2 + δ l3 + . + δ ln = δ li (1.8b)

(b) δ ijδ ljδ kl = δ liδ kl = δ ki by 1.8b

Example 1.4.2. If xi and yi are independent coordinates of a point, it is

Solution: The partial derivative of ϕ in two coordinate systems are different

Hence, the result follows from (1.9a) and (1.9b).

1.5 Linear Equations

aijxj = bi, (1.10a)

where x1, x2, …. xn are n unknown variables.

where a = |ai j| and the cofactor of ai j is Ai j.

by (1.10b), we get, axj = biAi j.

a11A11 + a12A12 + a13A13 = a

Similarly, we have a2jA1j = 0 a2jA2j = a and a2jA3j = 0

a3jA1j = 0 a3jA2j = 0 and a3jA3j = a.

Using Kronecker Delta Notation, these can be combined into a single

a1 j Akj = aδ 1k a2 j Akj = aδ 2k a3 j Akj = aδ 3k [ δ ij = 1, when i = j

1.6 Results on Matrices and Determinants of Systems

 a11 a12 … a1n   a11 a12 … a1n   a11 a12  . an1 

 a11 a11   b1 0   a12 a12   b12 0 

 a1 a 2  b1 b1